United States
•5#^™ Environmental Protection
*rn Agency PB2005-106460
Predicting Toxicity to
Amphipods From Sediment
Chemistry
-------
PB2005-106460
EPA/600/R-04/030
March 2005
Predicting Toxicity to Amphipods
From Sediment Chemistry
National Center for Environmental Assessment
Office of Research and Development
U.S. Environmental Protection Agency
Washington, DC 20460
-------
DISCLAIMER
This document has been reviewed in accordance with U.S. Environmental Protection
Agency policy and approved for publication. Mention of trade names or commercial products
does not constitute endorsement or recommendation for use.
ABSTRACT
The contribution of contaminated sediments to effects on sediment-dwelling organisms
(including plants and invertebrates), aquatic-dependent wildlife (amphibians, reptiles, fish, birds,
and mammals), and human health has become more apparent in recent years. Sediments can
serve both as reservoirs and as potential sources of contaminants to the water column and can
adversely affect sediment-dwelling organisms by causing direct toxicity or altering benthic
invertebrate community structure. Although the results of sediment toxicity tests and benthic
invertebrate community assessments can be used directly to evaluate or infer effects on resident
sediment-dwelling organisms, effective interpretation of sediment chemistry data requires tools
that link chemical concentrations to the potential for observing adverse biological effects.
This report describes the development of logistic regression models that quantify
relationships between the concentrations of contaminants in field-collected sediments and the
classification of samples as toxic on the basis of tests using two species of marine amphipods,
Rhepoxynius ahronius and Ampc/isca abdita. Individual chemical logistic regression models
were developed for 37 chemicals of potential concern in contaminated sediments to predict the
probability that a sample would be classified as toxic. These models were derived from a large
database of matching sediment chemistry and toxicity data that includes contaminant gradients
from a variety of habitats in coastal North America. Chemical concentrations corresponding to a
20, 50, and 80% probability of observing sediment toxicity (T20, T50, and T80 values) were
calculated to illustrate the potential for deriving application-specific sediment effect
concentrations and to provide probability ranges for evaluating the reliability of the models.
The individual chemical regression models were combined into a single model to
estimate the probability of toxicity on the basis of the mixture of chemicals present in a sample.
The average predicted probability of toxicity closely matched the observed proportion of toxic
samples within the same ranges, demonstrating the overall reliability of the PMax model for the
database that was used to derive the model. The magnitude of the toxic effect (decreased
survival) in the amphipod test increased as the predicted probability of toxicity increased.
The logistic models have a number of applications, including estimating the probability
of observing acute toxicity in estuarine and marine amphipods in 10-day toxicity tests based on
sediment chemistry. The models can also be used to estimate the chemical concentrations that
correspond to specific probabilities of observing sediment toxicity. Most importantly, the
models provide a framework for site-specific and regional assessments and for evaluating other
saltwater and freshwater endpoints.
Preferred citation:
U.S. EPA (Environmental Protection Agency). (2005) Predicting toxicity to amphipods from sediment chemistry¦.
National Center for Environmental Assessment. Washington. DC: EPA/600/R-04/030. Available from: National
Technical Information Sen ice. Springfield. VA. as PB2005-106460.
ii
-------
CONTENTS
LIST OF TABLES vi
LIST OF FIGURES ix
LIST OF ABBREVIATIONS AND ACRONYMS xiv
PREFACE xv
AUTHORS, CONTRIBUTORS, AND REVIEWERS xvi
1. EXECUTIVE SUMMARY 1
1.1. SUMMARY 1
1.1.1. Individual Chemical Models 2
1.1.2. Multiple-Chemical Models 3
1.2. RECOMMENDATIONS 5
1.3. APPLICATIONS 6
2. INTRODUCTION 8
3. DATABASE DEVELOPMENT 11
3.1. INTRODUCTION 11
3.2. COMPILATION OF MATCHING SEDIMENT CHEMISTRY AND TOXICITY
DATA 11
3.3. DATA AUDITING 13
3.4. DATA TREATMENT TO SUPPORT MODEL DEVELOPMENT 14
3.4.1. Calculation of Total PCBs 14
3.4.2. Classification of Toxic Samples 14
3.4.3. Data Screening for Model Development 15
3.5. DATABASE CONTENTS 16
4. LOGISTIC REGRESSION MODELS FOR INDIVIDUAL CHEMICALS 19
4.1. INTRODUCTION 19
4.2. METHODS 20
4.2.1. Logistic Regression Modeling 20
4.2.2. Concentration Interval Plots 22
4.3. INDIVIDUAL CHEMICAL LRM RESULTS 22
4.3.1. Model Results 22
4.3.1.1. Models Based on Sig Only Toxicity Classification 22
4.3.1.2. Models Based on MSD Toxicity Classification 24
4.3.1.3. Models Based on Organic Carbon-Normalized Chemical
Concentrations 25
4.3.2. Model Comparisons 26
4.3.2.1. Goodness-of-Fit Comparisons 26
4.3.2.2. Reliability Comparisons 27
4.3.3. Summary of Toxicity Classification Evaluations 27
4.4. EVALUATION OF ALTERNATIVE SCREENING APPROACHES 28
4.4.1. Unscreened Versus IX Screening 29
4.4.2. 2X Screening Versus IX Screening 30
4.4.3. Summary of Screening Approach Evaluations 30
iii
-------
CONTENTS (continued)
4.5. COMPARISON OF TOXICITY TEST ENDPOINTS 3 1
4.5.1. Statistical Comparisons of Species-Specific LRMs 31
4.5.2. Comparison of Tp values for Separate Amphipod Models 32
4.5.3. Reliability Comparisons 32
4.5.4. Summary of Species Comparisons 32
4.6. SENSITIVITY OF MODELS TO ERRORS IN UNDERLYING DATA 33
4.7 INDIVIDUAL CHEMICAL MODELS AND SPIKED-SEDIMENT BIOASSAY
MEDIAN LETHAL CONCENTRATION (LC50J VALUES 33
4.8. SUMMARY AND CONCLUSIONS 34
5. MULTIPLE-CHEMICAL MODELS 37
5.1. INTRODUCTION 37
5.2. MODEL DEVELOPMENT 37
5.3. MODEL RESULTS 40
5.4 USING THE MULTIPLE-CHEMICAL MODEL AS AN ANALYTICAL
FRAMEWORK 41
5.4.1. Relationship Between Probability of Observing a Toxic Effect and
Magnitude of Toxicity 41
5.4.2. Chemicals that Serve as the Most Effective Surrogates for Toxicity 42
5.4.3. Effect of High Levels of Several Chemicals 42
5.4.4. Sensitivity of P Max Model to Individual Chemical Models 43
5.4.4.1. Effect of PCB Model Correction 43
5.4.5. Predictions of P Max Model for Individual Species 43
5.4.6. Predictions of P Max Model for Individual Studies 44
5.5. APPLICATION OF THE MODELS TO AN INDEPENDENT DATA SET 45
5.6. SUMMARY AND CONCLUSIONS 46
6. APPLICATIONS OF MODELS 49
6.1. INTRODUCTION 49
6.2. APPLICATION OF THE P MAX MODEL TO DATA FOR OTHER
ENDPOINTS 49
6.3. USING LRMs TO EVALUATE EXISTING EMPIRICAL GUIDELINES 50
6.4. APPLICATION OF MODELS TO EVALUATIONS OF SITE-SPECIFIC OR
REGIONAL DATA 53
6.5. SUMMARY AND CONCLUSIONS 55
7. CONCLUSIONS AND RECOMMENDATIONS 56
7.1. CONCLUSIONS 56
7.2. RECOMMENDATIONS 59
7.3. APPLICATIONS 60
7.4. FUTURE DIRECTIONS 61
iv
-------
CONTENTS (continued)
REFERENCES 155
APPENDIX A1: REFERENCES FOR MARINE SEDTOX02 DATABASE 158
APPENDIX A2: REFERENCES FOR FRESHWATER SEDTOX02 DATABASE 162
APPENDIX B: SEDIMENT TOXICITY DATABASE (SEDTOX02) STRUCTURE 168
APPENDIX C: DATA ACQUISITION SCREENING METHODS 173
APPENDIX D: DATA EVALUATION METHODS 178
APPENDIX E: SENSITIVITY OF P MAX AND P AVG MODELS TO DIFFERENT
BINNING SCENARIOS 182
V
-------
LIST OF TABLES
Table 1. Number of samples and percent toxic samples summarized by marine amphipod
species and data source 63
Table 2. Distribution of chemical concentrations in sediment samples with matching
toxicity data for marine amphipods 64
Table 3. Number of samples and percent toxic samples summarized by sea urchin
species and test endpoint 65
Table 4. Distribution of chemical concentrations in sediment samples with matching
sea urchin fertilization toxicity data 66
Table 5. Distribution of chemical concentrations in sediment samples with matching
sea urchin developmental toxicity data 67
Table 6. Endpoint, number of samples, and percent toxic samples for Chironomus spp.
(C. Icnlans and C. riparius) and Hyalclla azlcca 68
Table 7. Distribution of chemical concentrations in sediment samples with matching
toxicity data for the H. azteca 10-14-day survival test 69
Table 8. Distribution of chemical concentrations in sediment samples with matching
toxicity data for the C. Icnlans or C. riparius 10-14-day survival test 70
Table 9. Distribution of chemical concentrations in sediment samples with matching
toxicity data for the H. aztcca 28-day growth and survival test 71
Table 10. Normalized chi-square values and number of samples for individual chemicals
based on Sig Only classification of toxic samples for the screened marine amphipod
database (logistic regression model parameters are shown for models having
normalized chi-square values greater than 0.15) 72
Table 11. Logistic model point estimates of T20, T50, and T80 values (95% confidence
interval) for individual chemicals based on Sig Only classification of toxic
samples for the screened marine amphipod database 73
Table 12. Percent of toxic samples within ranges defined by Sig Only logistic model
T20, T50, and T80 values and number of samples used to derive the logistic
model for each chemical for the marine amphipod database 74
Table 13. Normalized chi-square values and number of samples for individual chemicals
based on MSD classification of toxic samples for the screened marine amphipod
database (logistic regression model parameters are shown for models having
normalized ch-square values greater than 0.15) 75
VI
-------
LIST OF TABLES (continued)
Table 14. Logistic model point estimates of T20, T50, and T80 values (95% confidence
interval) for individual chemicals based on MSD classification of toxic
samples for the marine amphipod database 76
Table 15. Percent of toxic samples within ranges defined by logistic model T20,
T50, and T80 values and number of samples in the database used to derive the
logistic model for each chemical based on MSD classification of toxic samples
for the marine amphipod database 77
Table 16. Normalized chi-square values and number of samples for individual chemicals
for organic carbon-normalized concentrations for the screened marine amphipod
database (logistic regression model parameters are shown for models having
normalized chi-square values greater than 0.15) 78
Table 17. Logistic model point estimates ofT20, T50, and T80 organic carbon-normalized
concentrations (95% confidence interval) for individual chemicals based on Sig
Only classification of toxic samples in the screened marine amphipod database 79
Table 18. Percent of toxic samples within ranges defined by logistic model T20, T50,
and T80 values for organic carbon-normalized concentrations and number of
samples in the database used to derive the logistic model for each chemical based
on Sig Only classification of toxic samples for the marine amphipod database 80
Table 19. Differences between percent predicted toxic samples and percent observed toxic
samples (observed minus predicted) within ranges defined by logistic
model T20, T50, and T80 values based on Sig Only classification of toxic
samples for the marine amphipod database 81
Table 20. Differences between percent predicted toxic samples and percent observed toxic
samples (observed minus predicted) within ranges defined by logistic model T20,
T50, and T80 values based on MSD classification of toxic samples for the marine
amphipod database 82
Table 21. Differences between mean percent predicted toxic samples and percent observed
toxic samples (observed minus predicted) within ranges defined by logistic model
T20, T50, and T80 values for organic carbon-normalized concentrations
based on Sig Only classification of toxic samples for the marine amphipod
database 83
Table 22. Differences between mean percent predicted toxic samples and percent observed
toxic samples (observed minus predicted) within ranges defined by logistic
model T20, T50, and T80 values for each chemical using a screening factor of 2X
the mean of nontoxic samples and Sig Only classification of toxic samples for the
marine amphipod database 84
vn
-------
LIST OF TABLES (continued)
Table 23. Statistical comparisons of the logistic regression models for A ahdila and
R. ahronius using the chi-square statistic (-2 log likelihood) 85
Table 24. Ratio of Tp values from species-specific A. abdita models to corresponding
Tp values from the combined amphipod models (only models with normalized
chi-square >0.15 are included) 86
Table 25. Ratio of Tp values from the combined amphipod models to corresponding
Tp values from the species-specific R. ahronius models (only models with
normalized chi-square >0.15 are included) 87
Table 26. Differences between mean percent predicted toxic samples and percent observed
toxic samples (observed minus predicted) for A. abdita and R. ahronius within
ranges defined by T20, T50, and T80 values 88
Table 27. Changes in logistic model point estimates of T20, T50, and T80 concentrations for
PCBs based on Sig Only classification of toxic samples based on corrected PCB
model for the screened marine amphipod database 89
Table 28. Estimated probability of toxicity from marine amphipod chemical-specific
logistic regression models for LC50 values (dry wt.) reported from 10-day spiked
sediment amphipod toxicity tests 90
Table 29. Number and percent of samples by chemical class that represented the maximum
probability of toxicity used in the PMax model derived from the marine
amphipod database 91
Table 30. Differences between mean percent predicted toxic samples and percent observed
toxic samples (observed minus predicted) by probability quartile for individual
studies with at least 20 samples 92
Table 3 1. List of studies (primary data source) combined for analysis of broader geographic
areas 93
Table 32. Percent of samples predicted to be toxic to amphipods at the chemical
concentrations defined by sediment quality guidelines 94
Table 33. Percent of samples predicted to be toxic to amphipods at the chemical
concentrations defined by the Final Chronic Value for individual polycyclic
aromatic hydrocarbons 95
Table 34. Percent of samples predicted and observed to be toxic to amphipods within
ranges defined by toxic units of polycyclic aromatic hydrocarbons (PAHs) 96
Vlll
-------
LIST OF FIGURES
Figure 1. Box plots (plotted as per Tukey, 1977) summarizing the distribution of total
organic carbon (TOC) (log 10) values for each study with greater than
20 samples from the marine amphipod database 97
Figure 2. Logistic regression models and proportion of toxic samples in concentration
intervals in screened marine amphipod database for 37 chemicals based on
Sig Only classification of toxic samples 98
Figure 3. Logistic regression models and proportion of toxic samples in concentration
intervals in the screened marine amphipod database for 33 chemicals based on
MSD classification of toxic samples 105
Figure 4. Logistic regression models and proportion of toxic samples in organic carbon-
normalized concentration intervals in the screened marine amphipod database
for 25 chemicals based on Sig Only classification of toxic samples 111
Figure 5. Comparison of logistic model goodness of fit for the marine amphipod survival
endpoint with different toxicity classifications: Sig Only versus MSD 116
Figure 6. Comparison of logistic model goodness of fit for the marine amphipod endpoint
survival with different approaches to the expression of chemical concentrations:
dry weight (DW) versus organic carbon-normalized (OC) concentrations 117
Figure 7. Comparison of logistic regression model Tp values (T20, T50, T80) for Sig Only
and MSD toxicity classification approaches for the marine amphipod database 118
Figure 8. Comparison of IX mean screened (left) and unscreened (right) logistic regression
models and proportion of samples toxic in concentration intervals for lead and
phenanthrene for the marine amphipod database 119
Figure 9. Comparison of logistic model goodness of fit for the marine amphipod survival
endpoint using different screening methods (IX screening vs. no screening) and the
Sig Only classification of toxic samples 120
Figure 10. Logistic regression models and proportion of samples toxic in concentration
intervals for the standard screening approach (1X mean) (left) and the 2X mean
screening approach (right) for lead and fluoranthene using the marine amphipod
database 121
Figure 11. Comparison of logistic model goodness of fit for the marine amphipod survival
endpoint using different screening methods (IX mean vs. 2X mean) and Sig
Only classification of toxic samples 122
IX
-------
LIST OF FIGURES (continued)
Figure 12. Ratio of logistic regression model Tp values (T20, T50, T80) for the standard
screening approach (1X mean) to the 2X mean screening approach for the
marine amphipod database 123
Figure 13. Logistic regression models and concentration interval plots showing the effect
of correction in PCB units for 15 samples in the marine amphipod database 124
Figure 14. Mean percent control-adjusted marine amphipod survival for toxic samples
within intervals defined by the Tp values for all individual chemical models 125
Figure 15. Probability density functions for PAvg, PMax, and P Prod 126
Figure 16. Proportion of observed toxic samples versus the P Max and P Avg value
probability intervals based on Sig Only classification of toxic samples in the
marine amphipod database 127
Figure 17. Mean predicted and observed proportion of toxic samples within probability
quartiles for P Max and P Avg models derived from the marine
amphipod database 128
Figure 18. Observed proportion of toxic samples versus the probability of toxicity
predicted using the P Max and P Avg models derived from the marine
amphipod database 129
Figure 19. Mean percent control-adjusted survival for toxic samples within probability
quartile intervals for P Max model derived from the marine amphipod
database 130
Figure 20. Mean control-adjusted percent survival for samples within probability intervals
for P Max model (R2 = 0.94) derived from the marine amphipod database 131
Figure 21. Comparison of the original marine amphipod P Max model with the model
derived from data excluding polycyclic aromatic hydrocarbon chemistry
(R2 = 0.88) 132
Figure 22. Comparison of the original marine amphipod P Max model with model
derived from data excluding metals chemistry (R2 = 0.93) 133
Figure 23. Comparison of the original marine amphipod PMax model with model
derived from data excluding pesticides and PCB chemistry (R2 = 0.92) 134
x
-------
LIST OF FIGURES (continued)
Figure 24.
Mean predicted and observed proportion of toxic samples for samples with a
predicted probability of >0.75 from the PMax model in the marine amphipod
database 135
Figure 25. Observed proportion of toxic samples versus P Max values including only
those chemicals with normalized chi-square values exceeding 0.27 136
Figure 26.
Median predicted probability of toxicity within probability intervals for the
marine amphipod P Max model compared with data for A abdita: proportion
of observed toxicity based on Sig Only classification (left), control-adjusted
survival (center), and proportion of observed toxicity based on the MSD
classification (right) 137
Figure 27.
Median predicted probability of toxicity within probability intervals for the
marine amphipod P Max model compared with data for R. ahronius: proportion
of observed toxicity based on Sig Only classification (left), control-adjusted
survival (center), and proportion of observed toxicity based on the MSD
classification (right) 138
Figure 28.
Median predicted probability of toxicity within probability intervals for the
marine amphipod P Max model compared with data from Hudson-Raritan/Long
Island NSTP and Regional EMAP (A. abdita): proportion of observed toxicity
based on Sig Only classification (left), control-adjusted survival (center), and
proportion of observed toxicity based on the MSD classification (right) 139
Figure 29.
Median predicted probability of toxicity within probability intervals for the
marine amphipod P Max model compared with data from Virginian Province
EMAP (A. abdita): proportion of observed toxicity based on Sig Only
classification (left), control-adjusted survival (center), and proportion of
observed toxicity based on the MSD classification (right)
.140
Figure 30.
Median predicted probability of toxicity within probability intervals for the
marine amphipod P Max model compared with data from NSTP and Carolinian
EMAP from the southeastern U.S. (A. abdita): proportion of observed toxicity
based on Sig Only classification (left), control-adjusted survival (center), and
proportion of observed toxicity based on the MSD classification (right) 141
Figure 3 1.
Median predicted probability of toxicity within probability intervals for the
marine amphipod P Max model compared with data from Puget Sound, WA
(R. ahronius)-. proportion of observed toxicity based on Sig Only classification
(left), control-adjusted survival (center), and proportion of observed toxicity
based on the MSD classification (right) 142
XI
-------
LIST OF FIGURES (continued)
Figure 32. Median predicted probability of toxicity within probability intervals for the
marine amphipod PMax model compared with data from California (R.
abronins and A. abdita): proportion of observed toxicity based on Sig Only
classification (left), control-adjusted survival (center), and proportion of
observed toxicity based on the MSD classification (right) 143
Figure 33. Median predicted probability of toxicity within probability intervals for the
marine amphipod P Max model compared with proportion of observed toxicity
based on the Sig Only classification from individual studies: Elliot Bay, Puget
Sound, WA (R. abronins, n = 97) (left) and Biscayne Bay, FL (A. abdita,
n 120) (right) 144
Figure 34. Median predicted probability of toxicity within probability intervals for the
marine amphipod P Max model compared with proportion of observed toxicity
from an individual study from San Diego Bay, CA (R. abronins, n = 93);
toxicity based on Sig Only classification and MSD classification 145
Figure 35. Median predicted probability of toxicity within probability intervals for
the marine amphipod P Max model compared with data from the Calcasieu
Estuary (A. abdita)-. proportion of observed toxicity based on Sig Only
classification (left), control-adjusted survival (center), and proportion of
observed toxicity based on the MSD classification (right) 146
Figure 36. Average predicted probability and proportion toxic within probability intervals
for the marine amphipod P Max model applied to the sea urchin development
and fertilization endpoints 147
Figure 37. Median predicted probability and proportion toxic within probability intervals
for the marine amphipod P Max model applied to sea urchin (A. pnnctnlata)
development endpoint: proportion of observed toxicity based on Sig Only
classification (left), control-adjusted response (center), and proportion of
observed toxicity based on the MSD classification (right) 148
Figure 38. Median predicted probability and proportion toxic within probability intervals
for the marine amphipod P Max model applied to sea urchin (A. pnnctnlata)
fertilization endpoint: proportion of observed toxicity based on Sig Only
classification (left), control-adjusted response (center), and proportion of
observed toxicity based on the MSD classification (right) 149
xn
-------
LIST OF FIGURES (continued)
Figure 39. Median predicted probability and proportion toxic within probability intervals
for the marine amphipod PMax model applied to sea urchin (.V. piirpiiratns)
development endpoint: proportion of observed toxicity based on Sig Only
classification (left), control-adjusted response (center), and proportion of
observed toxicity based on the MSD classification (right) 150
Figure 40. Median predicted probability and proportion toxic within probability intervals
for the marine amphipod P Max model applied to sea urchin (.V. piirpiiraliis)
fertilization endpoint: proportion of observed toxicity based on Sig Only
classification (left), control-adjusted response (center), and proportion of
observed toxicity based on the MSD classification (right) 151
Figure 41. Proportion of samples that were toxic based on the H. azlcca 10-14-day
survival endpoint versus median probability of toxicity predicted using the
marine amphipod P_Max model 152
Figure 42. Proportion of samples that were toxic based on the C. Icnlans or C. riparius
10-14-day survival endpoint versus median probability of toxicity predicted
using the marine amphipod P Max model 153
Figure 43. Proportion of samples that were toxic based on the H. azlcca 28-day growth
and survival endpoint versus the median probability of toxicity predicted using
the marine amphipod P_Max model 154
Xlll
-------
LIST OF ABBREVIATIONS AND ACRONYMS
AET Apparent effect threshold
ASTM American Society for Testing Materials
BEDS Biological Effects Database for Sediments (MacDonald Environmental Sciences)
DDD 5-dichlorodiphenyldichloroethane
DDE Dichlorodiphenyldichloroethylene
DDT Di chl orodi pheny 1 tri chl oroethane
EMAP Environmental Monitoring and Assessment Program (U.S. EPA)
ERL Effect range low
ERM Effect range median
LC50 Median lethal concentration
LRM Logistic regression model
MLML Moss Landing Marine Laboratory (California)
MSD Minimum significant difference (classification of samples as toxic)
NSTP National Status and Trends Program (of the National Oceanic and Atmospheric
Administration)
PMax The maximum probability of observing toxicity, taken from the set of
probabilities calculated for each individual chemical in a sample
P Max model A regression model that predicts the probability that a sample will be toxic to
amphipods on the basis of the maximum probability of observing toxicity
calculated for all individual chemicals in a sample
PAvg The mean probability of observing toxicity based on the set of probabilities
calculated for each individual chemical in a sample
P Avg model A regression model that predicts the probability that a sample will be toxic to
amphipods based on the probability of observing toxicity averaged over all of
the individual chemicals in a sample
PProd The product of the probabilities of surviving exposure to all individual
chemicals in the sample
PAH(s) Polycyclic aromatic hydrocarbon(s)
PCB(s) Polychlorinated biphenyl(s)
PEL Probable effect level
SEDQUAL Sediment Quality Information System (State of Washington Department of
Ecology's Puget Sound database)
Sig Only Significance only (classification of samples as toxic)
SQGs Sediment quality guidelines
TEL Threshold effect level
TOC Total organic carbon
Tp The concentration that corresponds to a toxic response of "p" percent according
to the single chemical logistic models; for example, the T50 is the concentration
of a chemical that corresponds to the probability that 50% of the samples would
be toxic
xiv
-------
PREFACE
The U.S. EPA's National Center for Environmental Assessment developed this report
jointly with the National Oceanic and Atmospheric Administration, Coastal Protection and
Restoration Division, with substantial contributions from the U.S. Geological Survey. The report
is intended for risk assessors, field biologists, and research scientists interested in the
development and application of methods for evaluating the ecological risks associated with
chemicals in sediments.
Effective interpretation of sediment chemistry data requires tools that link chemical
concentrations to the potential for observing adverse biological effects. This report describes the
development of logistic regression models that quantify relationships between the concentrations
of sediment-associated contaminants and toxicity in two species of marine amphipods. The
models were developed using a large database of matching whole-sediment chemistry and
toxicity data that contains data published up until 2000. Amphipod toxicity tests were used as a
surrogate for valued ecological attributes that are more difficult to test and measure, including
structure and function of benthic communities, population viability of wildlife that depend on
benthos, and ecosystem processes such as organic matter decomposition and water filtration.
Because amphipod sediment toxicity tests are conducted using documented, standardized
methods, they are particularly amenable to analyses that combine results across studies, such as
those conducted in this project.
The logistic regression model (LRM) approach described in this report is similar to other
empirical approaches for deriving sediment quality guidelines in its reliance on matching field-
collected sediment chemistry and biological effects data. In contrast to other approaches to
developing sediment quality guidelines, however, the LRM approach does not identify threshold
values. Instead, it develops models that enable users to select the probability of observing
toxicity that corresponds to the users' specific objectives or to estimate the probability of
observing effects at a particular chemical concentration. The models provide a nationwide
framework that can be used to evaluate site-specific data, guide data collection efforts, and
compare ecological risks across sites and regions.
xv
-------
AUTHORS, CONTRIBUTORS, AND REVIEWERS
The National Center for Environmental Assessment (NCEA) of the U.S. Environmental
Protection Agency's (EPA's) Office of Research and Development was responsible for the
publication of this document. It was prepared by NCEA and the National Oceanic and
Atmospheric Administration, Coastal Protection and Restoration Division, under Interagency
Agreements DW13937819-01 -2 and DW1393980401 -0.
Authors
L. Jay Field
National Oceanic and Atmospheric Administration
Office of Response and Restoration
Coastal Protection and Restoration Division
7600 Sand Point Way, NE
Seattle, WA 98115
Susan B. Norton
U.S. Environmental Protection Agency
National Center for Environmental Assessment
1200 Pennsylvania Ave. NW (8623-N)
Washington, DC 20460
Donald D. MacDonald
MacDonald Environmental Sciences Ltd.
4800 Island Highway, N.
Nanaimo, British Columbia V9T 1W6
Corinne G. Severn
Premier Environmental Services
10999 Pumpkin Ridge Avenue
Las Vegas, Nevada 89135
Christopher G. Ingersoll
U.S. Geological Survey
Columbia Environmental Research Center
4200 New Haven Road
Columbia, MO 65201
xvi
-------
Contributors
Dawn Smorong, MacDonald Environmental Sciences Ltd., Nanaimo, British Columbia
Rebekka Lindskoog, MacDonald Environmental Sciences Ltd., Nanaimo, British Columbia
Lorraine Read, TerraStat Consulting Group, Seattle, WA
Carolyn Hong, EVS Consultants, Seattle, WA
Reviewers
David Mount
U.S. Environmental Protection Agency
National Health and Ecological Effects Research Laboratory
6201 Congdon Boulevard
Duluth, MN 55804
Keith Sappington
U.S. Environmental Protection Agency
National Center for Environmental Assessment (8623-N)
1200 Pennsylvania Ave. NW
Washington, DC 20460
Scott Ireland
U.S. Environmental Protection Agency
USEPA REGION 5
77 West Jackson Boulevard (G-17J)
Chicago, 1L 60604-3507
Donald M. MacDonald
National Oceanic and Atmospheric Administration
Office of Response and Restoration
Coastal Protection and Restoration Division
7600 Sand Point Way, NE
Seattle, WA 98115
John H. Gentile
Harwell, Gentile & Associates, LC
98 Moss Lane
Brewster, MA 0263 1
Susan Kane Driscoll
Menzie-Cura and Associates
Winchester, MA 01890
James Shine
Harvard School of Public Health
Department of Environmental Health
Boston, MA 02215
xvii
-------
Acknowledgments
The authors gratefully acknowledge the intellectual contributions and support from Ed Long,
Rusty Fairey, Dave Mount, Scott Ireland, Jim Keating, Jon Harcum, Rick Swartz, Scott Carr,
Pam Haverland, Judy Crane, Sherri Smith, Linda Porebski, Gail Sloane, Steve Bay, Tom Gries,
Ron Gouguet, Peter Landrum, Eric Smith, Paul Pinsky, and Pat Cirone.
This project would not be possible without contributions of data from many individuals and
programs. Data providers: NSTP from Ed Long, California database from Rusty Fairey, Tom
Gries (special thanks to Tom for providing the replicate data for the Puget Sound studies), and
Phil Crocker. EMAP data were provided by D. Keith and M. Hughes (Virginian Province), J.
Macauley and L. Harwell (Lousianian Province), J. Hyland and T. Snoots (Carolinian Province),
and D. Adams (NY/NJ REMAP).
Supplementary funding for data compilation was provided by EPA's Office of Water and the
National Research Council of Canada.
XVlll
-------
1. EXECUTIVE SUMMARY
1.1. SUMMARY
The contribution of contaminated sediments to effects on sediment-dwelling organisms
(including plants and invertebrates), aquatic-dependent wildlife (amphibians, reptiles, fish, birds,
and mammals), and human health has become more apparent in recent years. Sediments can
serve both as reservoirs and as potential sources of contaminants to the water column and can
adversely affect sediment-dwelling organisms by causing direct toxicity or altering benthic
invertebrate community structure. Although the results of sediment toxicity tests and benthic
invertebrate community assessments can be used directly to evaluate or infer effects on resident
sediment-dwelling organisms, effective interpretation of sediment chemistry data requires tools
that link chemical concentrations to the potential for observing adverse biological effects.
This report describes the development of logistic regression models that quantify
relationships between the concentrations of sediment-associated contaminants and toxicity to two
commonly tested species of marine amphipods, Rhcpoxynius ahronius and Ampc/isca abdita.
Amphipod toxicity tests are used as a surrogate for valued ecological attributes that are
more difficult to test and measure, including the structure and function of benthic communities,
population viability of wildlife that depends on benthos, and ecosystem functions such as organic
matter decomposition and water filtration. Because amphipod sediment toxicity tests are
conducted using documented, standardized methods (ASTM, 2002a, b), they are particularly
amenable to analyses that combine results across studies, such as those conducted in this project.
This report describes logistic regression models for 37 individual chemicals. The results
of these individual models are then combined into a single explanatory variable for estimating
the proportion of toxic samples expected in field-collected sediment samples. In addition, the
report illustrates the applications of the individual logistic models for evaluating sediment quality
guidelines and the use of the multiple-chemical models to predict toxicity for other locations and
endpoints. The report is intended for risk assessors, field biologists, and research scientists
interested in the development and application of methods for evaluating the ecological risks
associated with chemicals in sediments.
1
-------
1.1.1. Individual Chemical Models
Logistic regression models for 37 chemicals were developed using a large database of
matching whole-sediment chemistry and toxicity data that encompass many different
contaminant gradients from a wide variety of habitats in coastal North America. Logistic
regression uses a categorical (e.g., yes/no) variable as the dependent variable. Each sample was
designated as toxic or not toxic on the basis of a statistical comparison of the number of
amphipods that survived in the test sample relative to the negative control sample. The toxicity
classification was the dependent variable for the models, and the chemical concentration in the
field-collected sample was the explanatory variable. The models combined results from tests
that used either marine amphipod, R. abroiiins or A. abdita.
The chemical-specific models provide a basis for estimating the proportion of samples
expected to be toxic at different chemical concentrations. In contrast to other approaches to
evaluating the potential for observing toxicity on the basis of sediment chemistry (e.g., Long et
al., 1995; MacDonald et al., 1996, 2000), the logistic regression modeling approach does not rely
on specific effects thresholds. Instead, users can use the models to select sediment
concentrations (Tp values) that most directly meet the needs of their specific application. For
example, the models can be used to estimate concentrations for individual contaminants that are
likely to be associated with a relatively low incidence of sediment toxicity (e.g., 10, 15, or 20%).
Such point estimates of minimal-effect concentrations might be used in a screening assessment
to identify sediments that are relatively uncontaminated and have a low probability of sediment
toxicity. Similarly, contaminant concentrations for which there is a high probability of observing
adverse effects could be estimated. These higher point estimates could be used to identify
sediments that are highly likely to be toxic to amphipods and have a greater magnitude of effect
(i.e., higher percent mortality).
The Tp values can be used in much the same way as other sediment guidelines, except
that the Tp value provides a specific probability of observing toxicity and is associated with an
estimate of variance based on the fit of the model. The logistic regression models do not
represent dose-response relationships for individual chemicals; rather, they should be considered
to be indicators of toxicity based on field-collected sediment chemical mixtures.
The logistic regression approach was used to evaluate several issues that form the basis
for our recommendations for using the models. We used the single-chemical models to evaluate
two approaches for designating samples as toxic: (1) less than 90% survival that was
2
-------
significantly different from negative control samples (Sig Only), and (2) control-normalized
survival less than 80% that was significantly different from negative control samples (minimum
significant difference [MSD]) (based on analyses by Thursby et al., 1997). The Sig Only
approach had a greater tendency to underestimate the toxicity observed at low concentrations;
however, this discrepancy may be explained by the presence of other chemicals in the sample.
The MSD approach had a greater tendency to overestimate the toxicity observed at higher
concentrations. We selected the Sig Only approach for further exploration and development.
We also evaluated two approaches for normalizing sediment chemistry: dry weight and
organic carbon. We selected the dry weight normalization approach because the models had
higher goodness-of-fit statistics than the organic carbon-normalized sediment chemistry models,
and they had smaller differences between observed and predicted toxicity.
The presence of multiple contaminants, many of which may be present at very low
concentrations, complicates the evaluation of relationships between individual contaminants and
toxicity in field-collected samples. A data screening procedure was used to exclude samples for
which the selected chemical would not serve as a good indicator of observed toxicity. We used
the single-chemical models to evaluate three alternative screening criteria: (1) include all
samples in the model data set for an individual chemical (unscreened), (2) exclude toxic samples
that were less than or equal to the mean of nontoxic samples from the same study (1X screening),
and (3) exclude toxic samples that were less than or equal to two times the mean of nontoxic
samples from the same study (2X screening).
We selected the IX screening approach for further exploration and development. The
models from the unscreened alternative had much lower goodness-of-fit statistics and appeared
to show a weaker relationship between chemistry and toxicity than was observed with the other
screening alternatives. The 2X screening approach yielded models with slightly higher
goodness-of-fit statistics, but the IX screening approach performed slightly better at
concentrations above the T80 value. The 1X approach screened out fewer samples in the model
derivation, which may prove important in the future development of models for less frequently
measured chemicals.
1.1.2. Multiple-Chemical Models
Because the individual chemical models were derived from field-collected sediments that
included mixtures of contaminants, to some extent each individual chemical model represents the
-------
overall toxicity of the mixtures. However, an individual model would be expected to
underestimate the probability of observing toxicity in sediments contaminated with multiple
chemicals. The results of the individual models were combined to better estimate the probability
that a sediment sample would be toxic, based on the mixture of chemicals present in the sample.
Two approaches for combining the individual chemical model results into a single explanatory
variable representing the chemical mixture—the PMax model and the PAvg model—
accurately predicted the frequency of toxicity to amphipods observed in the database:
• P Max is the maximum probability of observing toxicity, taken from the set of
probabilities calculated for each individual chemical in the sample, and
• P Avg is the mean probability of observing toxicity, based on the set of probabilities
calculated for each individual chemical in the sample.
The multiple-chemical models were used to evaluate several additional issues, including
the relationship between the probability of observing a toxic effect and the magnitude of toxicity,
the identification of chemicals most influential in model performance, the performance of the
models in predicting toxicity of the two amphipod species, and the performance of the models in
predicting toxicity observed in regional data sets or in individual studies.
The magnitude of the effect (decreased survival) in the amphipod test increased as the
probability of toxicity increased, demonstrating that samples that are estimated to have the
highest probability of toxicity are also likely to be associated with high mortality.
For approximately 70% of the samples, individual chemical regression models for metals
produced the maximum probability used in the P Max model. This should not be construed to
imply that metals were causing toxicity in these samples, only that metals appear to be a good
indicator of toxicity in field-collected samples. Indeed, removing metals (or other entire
chemical classes) from the suite of individual chemical models used to generate the P Max
model resulted in only minor changes in the model and model fit.
Models were developed by combining data from tests that used R. abroiiins or A. abdita
in order to encompass more areas of the country and a broader range of sediment chemistry. The
P Max model was used to examine differences in model performance for the two species. The
observed toxicity was frequently less than predicted for A. abdita and greater than predicted for
R. ahroiiins. Nevertheless, the observed proportion of toxicity in the data for both species was
4
-------
strongly related to the nationwide model, affording confidence that the combined-species PMax
model provides a common framework that is applicable to both species.
We examined the performance of models in predicting observed toxicity for individual
studies within the database. On a study-by-study basis, there was mixed agreement between the
frequency of observed toxicity with that predicted by the P Max model. The mixed performance
suggests that the nationwide models should not be applied to individual studies without first
evaluating their performance with matching site-specific toxicity and chemistry data. However,
the nationwide P Max model provided a useful, common basis for evaluating toxicity test results
for individual sites included in the database. Application of the model to regional subsets of the
database used to derive the model demonstrated significant relationships between the P Max
model predictions and both observed proportion toxicity and percent control-adjusted survival.
There was also a strong relationship between predicted toxicity and observed toxicity in the
Calcasieu Estuary, an independent data set not included in the original derivation of the models.
1.2. RECOMMENDATIONS
As a starting point for most evaluations, we recommend using the P Max model, which
uses the highest predicted probability from any of the individual chemical models as the
explanatory variable. We recommend using the P Max model based on data from all studies
(i.e., the nationwide model), on both marine amphipod species, and on 37 chemical-specific
models. For the chemical-specific models, we recommend the models that classified samples as
toxic on the basis of less than 90% survival that was significantly different from negative control
samples (Sig Only), and that screened the data set by excluding toxic samples that were less than
or equal to the mean of nontoxic samples from the same study. The bases for these
recommendations are summarized briefly below.
We recommend using the P Max model, which is based on the highest predicted
probability from any of the individual chemical models, because it explained slightly more
variation in the data set than did the PAvg model. However, the two models provide slightly
different insights into sediment toxicity. P Avg may better reflect the overall degree of
contamination and is less susceptible to overestimating the probability of toxicity at sites with
high concentrations of one chemical. In some cases, it may be valuable to use both models to
take advantage of the different perspectives that they provide.
5
-------
We recommend using the nationwide model that combines data for both species of
marine amphipods. Combining data across studies and species represents the fullest range of
chemical concentrations and environmental conditions. In addition, the nationwide combined
model provides a common basis for comparing site- and species-specific results.
We recommend basing the models on the Sig Only classification so that more subtle
changes can be retained, particularly at lower concentrations. It may be valuable to compare the
Sig Only nationwide model with site-specific data classified using the MSD approach when test
variability obscures the relationship between chemistry and response at lower concentrations.
Finally, the chemical-specific models were greatly improved by using a screened data set
that excluded toxic samples that were less than or equal to the mean of nontoxic samples from
the same study. Using a more stringent criterion, such as excluding toxic samples that were less
than or equal to two times the mean of nontoxic samples from the same study, resulted in
improved goodness of fit in the models for most chemicals. However, the more stringent
criterion excluded an average of 70% of the toxic samples, which may limit future development
of models for other endpoints, regions, or chemicals that have fewer total samples. We
concluded that the small improvements in model fit did not outweigh the associated reduction in
sample size.
1.3. APPLICATIONS
The chemical-specific models provide a basis for estimating the probability that a sample
will be toxic for 37 individual contaminants over a wide range of contaminant concentrations. In
addition, they are useful for evaluating the degree of risk associated with commonly used
sediment quality guidelines (SQGs). The probabilities of toxicity associated with SQG threshold
values are generally consistent with their narrative intent. However, logistic regression models
have several advantages over current guideline approaches: (a) they present risk on a continuous
quantitative scale rather than by defining discrete categories based on threshold values, (b) the
continuous estimates of risk allow users to match the degree of risk with their objectives, and (c)
they express risk on a common scale of 0 to 1 across all chemicals. The individual chemicals
models would be expected to underestimate the probability of observing toxicity in samples that
are contaminated with many chemicals. For this purpose, we recommend using the multiple-
chemical models that combine the individual model results into a single explanatory value for
estimating the probability that a sample will be toxic.
6
-------
The multiple-chemical models provide a useful basis for conducting screening-level
assessments that require classifying or prioritizing samples on the basis of sediment chemistry.
Because the models do not consider potential differences in bioavailability or exposure, the
probability of toxicity may be over- or underestimated for some locations. Before applying the
models to a particular site, we recommend first evaluating how well the models fit the local
situation by collecting a test set of matching sediment chemistry and toxicity test data. The
logistic regression models can be used to design effective test sampling programs, and they can
also suggest issues that require further investigation (e.g., bioavailability). They can be very
useful for classifying samples into broad categories of concern on the basis of sediment
chemistry. The models should not be considered a complete substitute for direct effects
assessment (e.g., toxicity tests).
We evaluated the relationship between model predictions and the results of other toxicity
endpoints, including those commonly used in freshwater systems. The P Max model predictions
appear to be useful for predicting sea urchin response for Arhacia pimctulata, based on
development or fertilization tests, but not for Slroiigy/occiilroliispiirpiiralns. The models may
also be useful for predicting the response of freshwater amphipods, particularly the 28-day
Hyallda azteca growth and survival endpoint. There is the potential for developing endpoint-
specific models as more data are acquired.
7
-------
2. INTRODUCTION
The contribution of contaminated sediments to effects on sediment-dwelling organisms
(including plants and invertebrates), aquatic-dependent wildlife (amphibians, reptiles, fish, birds,
and mammals), and human health has become more apparent in recent years (Long and Morgan,
1991; U.S. EPA, 1997). Many toxic contaminants, such as metals, polycyclic aromatic
hydrocarbons (PAHs), polychlorinated biphenyls (PCBs), chlorophenols, and pesticides, are
found in only trace amounts in water, but they can accumulate to elevated levels in sediments
(Ingersoll et al., 1997). Therefore, sediments can serve both as reservoirs and as potential
sources of contaminants to the water column.
Contaminants associated with sediments can adversely affect resident sediment-dwelling
organisms by causing direct toxicity or by altering benthic invertebrate community structure
(Chapman, 1989). Furthermore, contaminated sediments can adversely affect fish and wildlife
species, either through direct exposure or through bioaccumulation in the food web.
A variety of approaches are used to evaluate the hazard posed by contaminated sediments
to ecological receptors (Ingersoll et al., 1997). These approaches include sediment chemistry
measurements, ex situ toxicity tests, benthic invertebrate community surveys, sediment toxicity
identification and evaluation procedures, and bioaccumulation assessments.
The results of sediment toxicity tests and benthic invertebrate community assessments
can be used directly to evaluate or infer effects on resident sediment-dwelling organisms.
However, effective interpretation of sediment chemistry data requires tools that link chemical
concentrations to the potential for observing adverse biological effects. Sediment chemistry
values that have been linked to biological effects can provide an efficient means for evaluating
the risks of sediment contamination when biological tests or surveys are unavailable.
The equilibrium partitioning approach links sediment chemistry values with biological
effects by combining the results of controlled laboratory tests using manipulated concentrations
of chemicals with theory on the factors controlling bioavailability (Di Toro et al., 1991, 2000).
Numerical sediment quality guidelines (SQGs) empirically link biological effects with sediment
chemistry by combining the results of toxicity tests using field-collected samples with the
concentrations of chemicals in those same samples (Long et al., 1995; MacDonald et al., 1996,
2000).
8
-------
The logistic regression model (LRM) approach described in this report is similar to other
empirical approaches for deriving SQGs because it relies on matching field-collected sediment
chemistry and biological effects data (e.g., sediment toxicity or benthic invertebrate community
structure effects). In contrast to other approaches to developing SQGs, however, the LRM
approach does not develop threshold values. Instead, it develops models that capture the
relationship between sediment chemistry and the probability of observing a toxic response. By
representing the relationship in continuous form, users can select the probability of observing
sediment toxicity that corresponds to their specific objectives. The relationships can also be used
to estimate the probability of observing effects, given the mixture of chemicals at a particular
location (Field et al., 1999, 2002).
The primary objectives of this report are to describe the development of individual
chemical LRMs, based on the standard marine and estuarine amphipod 10-day lethality toxicity
test endpoint (Chapter 4), and to combine these individual models into a single model for
predicting toxicity in field-collected sediment samples (Chapter 5). In addition, the report
illustrates the applications of the individual logistic models for evaluating SQGs and the use of
the combined models to predict toxicity for other sites and endpoints (Chapter 6).
The development of LRMs requires a large database of matching sediment chemistry and
toxicity data that includes a broad range of concentrations. Chapter 3 describes the development
of the SEDTOX02 database, which contains more than 3200 samples with matching sediment
chemistry and toxicity test results.
This report contains detailed descriptions of the database and model development. It is
intended for risk assessors, field biologists, and research scientists interested in a deeper
understanding of the different data analysis and treatment options considered during model
development and the strengths, limitations, and recommended application of the final models.
Our major findings have been published in two journal articles (Field et al., 1999, 2002). We
also refer readers to a companion effort (Smith et al., 2003) that fit multiple regression models
using many chemical concentrations measured in the sample as explanatory variables
simultaneously. These models require chemistry results for all of the chemicals used in each
model. The LRMs presented here were developed for application to a greater variety of
sediment chemistry results. This flexibility was used in the U.S. Environmental Protection
Agency's (EPA's) National Sediment Quality Survey (U.S. EPA, 2004), which used the
9
-------
approach described in this report to classify sediments into three tiers of probability of adverse
effects on the basis of a wide variety of sediment chemistry results.
10
-------
3. DATABASE DEVELOPMENT
3.1. INTRODUCTION
This investigation compiled synoptically collected sediment chemistry and sediment
toxicity data from throughout North America into the SEDTOX02 database (NOAA, 2004),
http://www.response.restoration.noaa.gov/cpr/sediment/sed_tox.html. The database is divided
into separate marine and freshwater databases of identical structure. The primary sources of the
estuarine and marine data included the National Oceanic and Atmospheric Administration's
National Status and Trends Program (NSTP), EPA's Environmental Monitoring and Assessment
Program (EMAP), Moss Landing Marine Laboratory (MLML) (which compiled data for the
state of California), the State of Washington Department of Ecology's Puget Sound Database
(SEDQUAL), and MacDonald Environmental Sciences' Biological Effects Database for
Sediments (BEDS). Appendices A1 and A2 contain the references for the marine and the
freshwater SEDTOX02 databases, respectively.
Many geographic areas along the Atlantic, Gulf, and Pacific coasts are represented in the
database, and it includes information on several marine and freshwater toxicity endpoints.
However, this report focuses on analyses using data from the EPA and American Society for
Testing Materials standard 10-day amphipod survival toxicity tests with Ampe/isca abdita and
Rhepoxynius abronius (U.S. EPA, 1994a, b; ASTM, 2002a). The database for this project was
developed in Microsoft FoxPro; the relational database structure is provided in Appendix B.
3.2. COMPILATION OF MATCHING SEDIMENT CHEMISTRY AND TOXICITY
DATA
All of the candidate data sets considered for inclusion in the database were critically
evaluated. Application of acceptance criteria (see appendices C and D) to individual studies
provided a basis for determining whether experimental designs and measurement endpoints,
sample collection and handling procedures, toxicity testing protocols and environmental
conditions, control responses, and analytical methods were consistent with established
procedures (Long et al., 1995; MacDonald et al., 1996; Field et al., 1999; ASTM, 2002a, b). In
the case of the data sets from NSTP, EMAP, SEDQUAL, and MLML sources, the standard
protocols established under each program were evaluated, and individual studies were generally
examined to identify possible deviations from these protocols. All of the data that met the
11
-------
acceptance criteria were incorporated into the project database. Samples were excluded from the
database if the survival in the associated negative control sample was less than 85% (expressed
as a mean of the negative control replicates).
Data sets that met the screening criteria were compiled in spreadsheets. To facilitate data
entry, a template was designed to standardize the format of the matching sediment chemistry and
toxicity data. Each data file included the following fields: location of the investigation (country,
area, and site); date of sediment collection; sampling and sample handling protocols or
procedures; species and life stage tested; test type (e.g., static porewater); type of test water (e.g.,
saltwater, freshwater); source of control and reference sediments; endpoint measured; method
used to determine whether the sample was toxic or nontoxic; study citation; and additional
explanatory comments.
The results of the toxicity tests conducted and the concentrations of all chemical analytes
measured were compiled in the spreadsheet that was created for each study. These latter data
were compiled on a sample-by-sample basis, including the control, reference, and test samples.
All chemical concentrations were entered as normalized to dry weight, with concentrations
below analytical detection limits reported at the detection limit value and a below-detection-limit
data qualifier added. Toxic or nontoxic descriptors were also assigned to each endpoint for each
sediment sample in the database as the first step in toxicity classification. The objective was to
standardize the toxicity classification within the database wherever possible. For the marine
amphipods, statistical significance derived from a comparison of test samples to the negative
control was preferred over other methods for determining significance (e.g., comparison to field
reference).
Data originating from NSTP, EMAP, and MLML were received with statistical
significance already determined. For Puget Sound data obtained from SEDQUAL, statistical
comparisons (one tailed t-test, a = 0.05) between the appropriate negative control and the test
samples were conducted on the replicate data. Other studies (received from BEDS) were
evaluated on an individual study basis. For these studies, a sediment sample was considered
toxic if the original investigator conducted suitable statistical analyses and reported that the
sample was significantly toxic when compared with the negative control or appropriate reference
site. If statistical significance was not determined by the investigator but sufficient information
on the replicate sample results or on the standard deviations of the results was provided, then a
12
-------
modified Student's T-test was conducted to determine the statistical significance of the results
for each sample.
If no statistical analysis could be performed, then a sediment sample was considered to be
toxic if the measured response was substantially different from the negative control or
appropriate reference response. A 20% or greater difference was considered to be substantial in
this context, generally reflecting the results of power analyses conducted on the results of
numerous toxicity tests (Thursby etal., 1997; Long et al., 1998; Carr and Biedenbach, 1999).
3.3. DATA AUDITING
To ensure the overall integrity of the database, a data verification and auditing plan was
developed and implemented. This plan consisted of three main elements.
First, the candidate data set was reviewed to identify potentially erroneous data.
Specifically, individual data sets were reviewed to identify improbable or impossible results
(e.g., extremely high or low chemical concentrations, dissolved oxygen concentrations exceeding
saturation levels, survival of >100%). If anomalous data were identified in this initial review (an
infrequent event), the principal investigator on the study was contacted to either verify the
reported results or provide the correct data. The primary data source was subsequently corrected
to reflect the input provided by the principal investigator.
The second phase of the data auditing process was data verification. For data that were
acquired electronically (the majority of the data), a minimum of 10% were compared with the
electronic source files. The few candidate data sets obtained in hard copy format (i.e., not in
electronic data files), had substantial potential for data transcription errors. For this reason, all of
the hand-entered data that were compiled in spreadsheets were fully verified against the original
data source prior to importing the data into the database. Two individuals working cooperatively
conducted data verification. Any errors or omissions identified were corrected, and the data
corrections were subsequently verified in a similar manner.
The third phase of the data auditing process was designed to determine whether any data
had been corrupted during the data translation process (i.e., transferring the data from the Excel
spreadsheets into the electronic database). To confirm that the data translation subroutines were
functioning appropriately, the data for several studies were exported into a spreadsheet format
that resembled the spreadsheets that had been constructed initially. The information contained in
these recompiled spreadsheets was then verified against the original data source. After the entire
13
-------
database had been compiled in the relational database format, data screening procedures (e.g.,
identification of orphan records, general confirmation of the relational database structure, and
extreme value checks) were applied to identify potential errors and further ensure the internal
consistency of the data in the database.
3.4. DATA TREATMENT TO SUPPORT MODEL DEVELOPMENT
3.4.1. Calculation of Total PCBs
The total concentration of PCBs was calculated for each sediment sample represented in
the database. The procedure used to calculate total PCBs depended on how the data were
reported in the original study. If only total PCBs were reported, these values were used directly.
If the concentrations of PCBs were reported as individual Aroclors (e.g., Aroclor 1242, Aroclor
1248), then the concentrations of the individual Aroclors were summed to determine the
concentration of total PCBs. When the concentrations of individual congeners were reported,
these values were summed to determine the total PCB concentration. If fewer than 20 congeners
were reported, the sum of the congeners was multiplied by 2, following the approach used by
NSTP (NO A A, 1989). If both Aroclors and congeners were measured, total PCBs were based on
the congener concentrations.
In calculating the total PCB concentration, below-detection-limit values were treated as
zero values. If all of the individual chemicals to be summed were below detection or if the
detection limit of any one nondetected chemical exceeded the sum of detected values, the highest
detection limit of the chemical constituents for the sample was used as the total value and
qualified as a below-detection-limit value.
3.4.2. Classification of Toxic Samples
Standardizing the classification of toxic samples in the database was an important step,
because the studies included in the database used several methods to designate individual
sediment samples as toxic or nontoxic. We used two approaches for identifying a consistent
response level across studies and applied them to all studies that used R. abroiiins and A. abdila.
The first approach, referred to as the "significance only" (Sig Only) approach, classified samples
as toxic if the sample was statistically different (/;<0.05) when compared with the negative
control and absolute survival was less than 90%. The criterion of 90% was used to preclude
14
-------
classifying samples as toxic because of low variability in the negative control and was based on
the minimum acceptable mean survival for negative control response in 10-day marine amphipod
toxicity tests (ASTM, 2002a). The application of this criterion changed the classification of 119
samples from toxic to nontoxic. The second approach, referred to as the minimum significant
difference (MSD) approach, classified samples as toxic if the sample was significantly different
(j)<0.05) when compared with the negative control and the difference in survival between the test
sample and control was at least 20%, that is, the test sample had a control-adjusted survival of
less than 80%. The difference of 20% corresponded to a power (1 - (3) of 0.9, based on analysis
of A. abdita data; in this study 90% of the tests could distinguish a difference of 20% at a
statistical significance level (a) of 0.05 (Thursby et al., 1997).
3.4.3. Data Screening for Model Development
The presence of multiple contaminants, many of which may be present at very low
concentrations, complicates the evaluation of relationships between individual contaminants and
toxicity in field-collected sediments. Consequently, the data for samples that were identified as
toxic in this investigation were further screened before they were used to develop the logistic
models for each individual contaminant (Field et al., 1999). The objective of the screening
process was to exclude toxic samples for which the chemical under consideration would not
serve as a good indicator of observed toxicity.
Although there are many possible approaches to screening, for simplicity, we evaluated
two approaches (1X and 2X) that followed the general screening approach used by Ingersoll et
al. (1996) and were similar to approaches used by others (Long and Morgan, 1991; Long and
MacDonald, 1992; MacDonald et al., 1996). In both approaches, the concentration of the
selected chemical in each toxic sample was compared with the mean of the concentration of that
substance in the nontoxic samples collected in the same study and geographic area.
For the IX screening approach, if the concentration of a chemical in an individual toxic
sample was less than or equal to the mean concentration of that chemical in the nontoxic samples
from that study area, it was considered unlikely that the observed toxicity could be attributed to
that chemical. Therefore, these toxic samples were not included in the "screened" data set used
for developing the logistic model for that chemical. For the 2X screening approach, the
"screened" data set used to develop the logistic model for a particular chemical excluded toxic
15
-------
samples less than or equal to two times the mean concentration of that chemical in the nontoxic
samples from that study area. For the development of the organic-carbon normalized models, we
applied the same procedures to the organic-carbon normalized chemical concentrations. Because
all of the screening approaches are based on the chemical concentrations in the nontoxic
samples, an important underlying assumption is that the factors influencing bioavailability are
similar for both nontoxic and toxic samples.
All nontoxic samples were included in the screened data sets produced using both
approaches. Samples from reference stations were treated the same as other samples and
included in the analysis. The data for chemical concentrations that were less than the reported
detection limit were not used to develop the logistic models.
3.5. DATABASE CONTENTS
The final SEDTOX02 database includes matching sediment chemistry and toxicity data
for both marine and freshwater systems. For convenience, in the data verification and analysis
steps, the database was separated into marine and freshwater databases with identical structures.
The marine database includes matching sediment chemistry and toxicity data from the
Atlantic, Gulf, and Pacific coasts of North America. Data from 10-day toxicity tests with two
species of amphipods (Rhcpoxynius ahronius and Ampc/isca abdita), for which survival is the
measured endpoint, represent the largest component of the database (Table 1). The use of
species differed by location; in general, R. ahronius dominated the West Coast studies, whereas
the studies conducted in the East used A. ahdita. Most of the data originated from large
programs (EMAP, NSTP, SEDQUAL, MLML), which used standardized methods of chemical
analyses and toxicity tests. Overall, 1257 (39%) of the 3223 sediment samples in the database
that had matching chemistry and toxicity were toxic to amphipods (i.e., survival was <90% and
significantly different from that of the negative control). For A abdita, 24% of the 2012 samples
were toxic in 10-day tests (Table 1). A higher proportion of the samples tested with R. ahronius
(i.e., 64% of 1211 samples) were identified as toxic (Table 1). Using the MSD approach to
classifying samples as toxic (i.e., control-adjusted survival was <80% and significantly different
from that of the negative control), 12.6% of the A. abdita samples and 40.8% of the R. ahronius
samples were classified as toxic.
The database includes information on the concentrations of more than 300 chemicals of
potential concern at contaminated sediment sites. More than 90% of the samples were analyzed
16
-------
for at least 10 of the 37 chemicals for which we developed models (Chapter 4). More than 70%
of the samples were analyzed for at least 20 of the modeled chemicals. For many of these
chemicals, the assembled data span a broad range of chemical concentrations. Table 2 presents
the distributions of the chemistry data (10th, 50th, and 90th percentiles) for samples with matching
amphipod toxicity data for metals, PAHs, PCBs, and several organochlorine pesticides. These
data show that the 10th to 90th percentile concentrations of the individual contaminants typically
span two to three orders of magnitude, with ranges often spanning four to six orders of
magnitude.
The covariation among chemical concentrations was found to be substantial in a
companion analysis (Smith et al., 2003). Principal component analysis was conducted on a
reduced data set (n = 2219) that contained samples with complete data on 22 metals and PAHs.
Two principal components explained 83% of the variation in the chemical data. When rotated
(with varimax rotation), the first factor explained 50% of the variation and was highly correlated
with the PAHs. The second factor explained 33% of the variation and was highly correlated with
the metals.
The percent total organic carbon (TOC) in test sediments averaged 1.92% (standard
deviation = 2.05, n = 3 117) and ranged from 0.01 to 29.4%. Based on visual inspection, there
was no clear pattern in TOC differences among the different studies (Figure 1). Because only
629 samples had results for acid volatile sulfides (AVS) and simultaneously extracted metals,
AVS normalization methods (Hansen et al., 1996) were not used.
The marine database also includes data on 100% porewater embryological development
and fertilization endpoints for two species of sea urchin (Arbacia punclulala and
Slrongy/occnlroluspiirpiiralus) (Table 3). The MSD values for S. purpuratus 96-hour
embryological development and 1-hour fertilization endpoints were 78% and 88%, respectively,
corresponding to a statistical significance (a) level of 0.05 and a power (1 - (3) of 0.9 (Phillips et
al., 2001). We used the MSD values derived by Carr and Biedenbach (1999) for A punclulala
48-hour embryological development and 1-hour fertilization endpoints of 83.6% and 84.5%,
respectively. These values corresponded to a statistical significance (a) level of 0.05 and a
slightly more stringent power (1 - (3) of 0.95. Using either the Sig Only or the MSD approach, a
high percentage (>60%) of the 782 samples were toxic for the embryological development
endpoint for both species combined, and there was little difference between species in percent of
17
-------
toxic samples. For the fertilization endpoint, about 40% of the 612 A. puncliilala samples and
60% of the 212 S. puvpuvaius samples were toxic. For both species and endpoints, the results
showed little difference between the Sig Only and MSD classifications in percent of toxic
samples.
The distribution of the bulk sediment chemistry data for the urchin tests (Tables 4 and 5)
show 10th to 90th percentile concentrations of the individual contaminants ranging from one to
three orders of magnitude for most chemicals. The ranges were typically larger for the organic
compounds than for metals. Percent TOC in test sediments averaged 2.1% (2.4% for A.
pimctulata and 1.8% for S. purpuraius) and ranged from 0.02 to 15.8% for the embryological
development test endpoint. Percent TOC values were similar for the fertilization endpoint.
The freshwater database included data from several frequently tested toxicity test
endpoints. Samples classified as toxic by the original investigator were taken at face value
because less information was available to independently evaluate statistical significance (e.g.,
replicate data were lacking) and no analyses had been conducted to identify MSD values for the
freshwater endpoints. The growth endpoints were treated as growth and survival, so if either
growth or survival had a toxic result, the sample was classified as toxic. Approximately 20% of
the 585 short-term survival test samples (10-14-day tests were grouped together for analysis)
were toxic for the freshwater midge species (Chironomus Icnlans and Chironomus riparius), and
almost 40% of the 10-day growth samples were toxic (Table 6). For the freshwater amphipod
Hyalclla azlcca, 24% of 567 samples were toxic in the 10-14-day survival test, and almost 40%
of the 125 samples were toxic in the 28-day growth and survival test.
The chemistry associated with the freshwater toxicity endpoints exhibited ranges of one
to three orders of magnitude (Tables 7, 8, and 9). In general, the chemical concentrations in
samples from the freshwater database were higher than those in the samples from the marine
database. The freshwater short-term toxicity tests also had higher TOC concentrations, with
average concentrations >4% and ranges from <0.1 to >50%. The H. azlcca 28-day growth and
survival test samples averaged 3% TOC (standard deviation = 2.1) and ranged from 0.1 to
11.6%.
18
-------
4. LOGISTIC REGRESSION MODELS FOR INDIVIDUAL CHEMICALS
4.1. INTRODUCTION
This chapter presents logistic regression models (LRMs) that predict the probability of
toxicity on the basis of individual chemical concentrations. The toxicity endpoint modeled was
the 10-day survival test conducted using two species of marine amphipods (A. abdita and R.
ahronius). Our objective for this analysis was to develop single-chemical models that could
serve as screening-level concentration-response relationships for individual chemicals. These
relationships could then be used to identify concentrations of individual chemicals that represent
different degrees of risk. The relationships could also be combined into multiple-chemical
models (Chapter 5).
The individual chemical models provide an opportunity to explore and compare different
methods of toxicity classification and chemistry normalization. Specifically, we evaluated two
approaches for designating samples as toxic: (1) less than 90% survival that was significantly
different from negative control samples (Sig Only), and (2) control-adjusted survival less than
80% that was significantly different from negative control samples (MSD) (based on the analysis
by Thursby et al., 1997). We evaluated two approaches for normalizing sediment chemistry:
dry-weight normalized and organic carbon normalized. Models were evaluated using several
approaches: goodness-of-fit statistics, visual examination of plots of the model and the
underlying data, and comparison of the predicted probabilities with the proportion of toxic
samples observed within ranges of predicted probability for the entire data set.
We also used the models to evaluate the implications of several data treatment options.
We evaluated three alternative screening criteria used to decide whether a given sample was
included in the model data set for an individual chemical: (1) include all samples, (2) exclude
toxic samples with concentrations that were less than or equal to the mean concentration of
nontoxic samples from the same study (IX screening), and (3) exclude toxic samples with
concentrations that were less than or equal to two times the mean concentration of nontoxic
samples from the same study (2X screening). Finally, we evaluated the implications of
developing models that combined data for both amphipod species.
19
-------
4.2. METHODS
4.2.1. Logistic Regression Modeling
Statistical models fit to the marine amphipod data in SEDTOX02 describe relationships
between the probability of a toxic outcome in the amphipod tests and concentrations of the
chemicals of interest. Exploratory plots of the data generated distributions that resembled typical
sigmoidal dose-response curves. The shape of these curves indicated that it might be appropriate
to model these relationships using an LRM. Logistic regression is typically applied to dose-
response data, such as that generated by spiked-sediment bioassays, or laboratory tests with a
binary outcome (Morgan, 1992).
The individual chemical LRMs were developed from the screened data set for each
chemical. The data screening procedures used in this study were intended to identify the
chemicals that serve as useful indicators of the toxic response observed in individual sediment
samples. The screening procedures also transformed the underlying data into a form that is more
consistent with the sigmoidal form of LRMs.
The individual chemical LRMs used the dichotomous toxicity test result (toxic or
nontoxic) as the dependent variable and the chemical concentration as the explanatory variable.
The model parameters (slope, intercept) define the shape of relationship between the chemical
concentration (Log 10) and the probability of a toxic result. In its simplest form, the logistic
model can be described using the following equation:
cxp/HO BI(x)l
^ / + cxp/BO + Bl(x)/
where:
p = probability of observing a toxic effect,
BO = intercept parameter,
B1 = slope parameter, and
x = chemical concentration or log chemical concentration.
This logistic model was applied to the complete screened data for a number of substances
to develop relationships between the sediment chemistry and the toxicity test results. For each
substance modeled, the intercept (BO), slope (Bl), and chi-square statistic (-2 log likelihood)
20
-------
were determined. The data for each chemical were modeled independently. Thus, there was
only a single concentration variable (x) in each individual chemical model. However, for each
model, it is possible to consider the addition of various covariates, such as test species or
endpoints. If such a covariate is considered, then a separate intercept term, a separate slope term,
or both a distinct intercept and a distinct slope term can be fit for each level of the covariate. All
of the logistic regression analyses were conducted using the SAS Institute's logistic procedure
(SAS Institute, 1990). The slope and intercept parameters for the model were estimated using
the maximum-likelihood approach.
The chi-square statistic provides useful information for interpreting the results of the
logistic modeling. Specifically, the chi-square statistic was used to determine whether the slope
parameter, Bl, was significantly different from zero. For all of the models generated, the
probability (j) value) associated with the slope parameter was less than 0.0001; therefore, the null
hypothesis (slope = 0) can be rejected. Additionally, the chi-square statistic can be used to assess
how well the model fits the data. For data sets with similar sample sizes, a larger chi-square
statistic indicates a better fit of the model to the data. Note, however, that for a similar fit, the
chi-square statistic increases with sample size and thus cannot be used to compare the fit of data
sets that are not roughly the same size. Normalizing the chi-square statistic to the sample size
(N) provides a goodness-of-fit measure that could be applied across all the data sets. There are
no established criteria for considering a normalized chi-square statistic to be good. For the
purposes of this report, models that had a normalized chi-square value of greater than 0.15 were
considered to be a good fit. (The use of a stronger criterion is explored in Chapter 6.)
After the parameters are estimated, the model can be inverted to estimate the
concentrations that yield a certain response probability. The notation Tp (e.g., T50) is used to
denote the concentration that would give a toxic response of "p" percent according to the model
(e.g., the probability that 50% of the samples would be toxic). Confidence intervals for these
effect concentrations that describe the uncertainty associated with fitting the model were derived
using the delta method. The delta method is based on a truncated Taylor series expansion that
uses the variance-covariance matrix derived from the maximum-likelihood fit and the derivative
of the function of interest (in this report, T20, T50, and T80 were used as examples) with respect
to each parameter (Morgan, 1992).
21
-------
4.2.2. Concentration Interval Plots
Concentration interval plots were used to visualize the relationship between the matching
sediment chemistry and toxicity data for individual contaminants. The plots were prepared by
calculating the proportion of toxic samples within discrete concentration intervals. The
individual points represent the median of the chemical concentrations in the samples within the
interval and the proportion of the samples classified as toxic within the interval. Each point on
the plots represents a minimum of 15 individual samples (a greater number of samples was
included in the interval if more than one sample had the same concentration). The range
represented by each concentration interval was determined from an ascending list of unique
sample concentrations for each contaminant, with the number of intervals determined by the total
number of unique sample concentrations for the selected contaminant. The purpose of the plots
was only to help visualize the general relationship between chemical concentrations and the
probability of observing toxicity. The goodness of fit of each model was evaluated using the
normalized chi-square statistics discussed above.
4.3. INDIVIDUAL CHEMICAL LRIM RESULTS
4.3.1. Model Results
This section presents LRM results for individual chemicals. We describe models that
used the two toxicity classification approaches and the two chemistry normalization methods.
All of these models combined data for the two amphipod species and used a screening criterion
that excluded toxic samples with concentrations less than or equal to the mean of nontoxic
samples from the same study (IX screening approach).
4.3.1.1. Models Based on Sig Only Toxicity Classification
Acceptable logistic models were generated for 37 substances for the Sig Only approach,
including 10 trace metals, 22 individual PAHs, total PCBs, and 4 organochlorine pesticides
(Table 10). All of these models used dry-weight chemical concentrations. The slopes for all
models were positive, indicating that increased chemical concentrations were associated with
increased probability of toxicity. Although the models for all 37 substances had normalized chi-
square statistics exceeding our criterion of 0.15, the models for arsenic, nickel, and p,p'-DDE
had normalized chi-square values of 0.17, 0.18, and 0.16, respectively, indicating relatively
poorer fits.
22
-------
Concentration interval plots provide additional information for evaluating the
relationships between chemical concentration and the probability of observing sediment toxicity
in the screened data set used to derive the model (Figure 2). For example, the plots for lead,
mercury, and zinc confirm that logistic models provide good fits of the underlying amphipod
toxicity data. Importantly, the range of concentrations represented in the database appears to
span the effects range, as demonstrated by the low proportion of toxic samples (0%) observed at
the lowest concentrations and the high proportion of toxic samples (90 to 100%) observed at the
highest chemical concentrations. Similar results were obtained for many of the organic
compounds (e.g., fluoranthene and phenanthrene); however, the observed proportion of toxic
samples tended to be somewhat lower (roughly 90%) at the highest concentrations of these
substances. The plot for p,p'-DDE shows both high variability and the presence of several
outliers, consistent with its relatively low normalized chi-square value.
Although the logistic models provide effective tools for estimating the probability of
observing sediment toxicity at various chemical concentrations, point estimates of sediment
effect concentrations are also useful for assessing sediment quality conditions. As an example,
the chemical concentrations that correspond to the 20, 50, and 80% proportion of toxic samples
for amphipod survival were determined and designated as Tp values: T20, T50, and T80,
respectively (Table 11).
The reliability of the chemical-specific logistic models was evaluated by comparing the
probability of toxicity predicted by the models to the proportion of samples actually observed to
be toxic. This comparison differs from the concentration interval plots in that the data screened
out of the logistic model development process were included in the reliability evaluation.
Predicted versus observed values were compared for four ranges of chemical concentrations
defined by the Tp values (i.e., T20-T50, >T50-T80, and >T80). The percent of samples
within each concentration range that were toxic was determined (Table 12). The logistic models
and associated point estimates were considered reliable if the observed proportion of toxic
samples was consistent with the predicted probability of toxicity.
The results of this evaluation indicate that the logistic models and associated point
estimates of sediment effect concentrations generally provide a reliable basis for estimating the
observed proportion of toxic samples in the project database. The models underestimated the
proportion of toxic samples at concentrations below the T20 value for all 37 chemicals, although
35 of the 37 chemicals were within 10% of the top of the range. The underestimation of toxicity
23
-------
below the T20 value may be a consequence of the screening procedure, and is discussed further
in Section 4.4.1. Between the T20 and T50 values, the proportion of toxic samples observed for
most of the chemicals (30 of 37) was within the predicted range of 20 to 50%. The proportion of
toxic samples observed between the T50 and T80 values was within the predicted range of 50 to
80% toxicity for all 37 chemicals. Above the T80 value, the proportion of toxicity was equal to
or exceeded 80% for 22 chemicals. Arsenic and p,p'-DDE had a substantially lower proportion
of toxic samples than predicted. The models for these chemicals would be expected to
overestimate toxicity for high concentrations.
Among the logistic models for the various classes of contaminants, those for PAHs were
the most reliable. For 16 of 22 PAHs, the actual proportion of toxic samples was correctly
predicted within three of the four concentration ranges defined by the Tp values; however, a
higher-than-predicted proportion of toxic samples was observed above the T20 values for all
PAHs (Table 12). Among the logistic models for the trace metals, those for chromium, copper,
lead, mercury, and zinc were the most reliable, as indicated by the level of agreement between
the predicted and observed proportion of toxic samples to amphipods. Likewise, the logistic
model for total PCBs provided an accurate basis for predicting toxicity to amphipods in the
database. A somewhat lower level of reliability was observed for the organochlorine pesticide
models.
4.3.1.2. Models Based on MSI) Toxicity Classification
Acceptable logistic models were generated for 33 substances for the MSD approach,
including 7 trace metals, 22 individual PAHs, total PCBs, and 3 organochlorine pesticides (Table
13). All of these models used dry-weight chemical concentrations. The slopes for all models
were positive, indicating that increased chemical concentrations were associated with increased
probability of toxicity. The models for eight chemicals—chromium, silver,
1-methylnaphthalene, 2-methylnaphthalene, 2-6 dimethylnapthalene, biphenyl, naphthalene, and
perylene—had among the lowest normalized chi-square values, indicating relatively poorer fits.
Models for antimony, arsenic, nickel, and p,p'-DDE had normalized chi-square values of less
than 0.15.
Concentration interval plots for the MSD toxicity classification approach are shown in
Figure 3. The proportion of toxic samples is often less than 80% at the highest concentrations in
24
-------
the database (e.g., zinc and 1-methylphenanthrene). This truncation at the higher concentrations
reflects the more stringent criteria for classifying samples as toxic.
The T20, T50, and T80 values are shown in Table 14. The percent of toxic samples
within the ranges defined by these Tp values is shown in Table 15. The agreement between the
models and the observed proportion of toxic samples was very good below the T80 value. At
concentrations below the T20 value, the observed proportion of toxic samples was within the
predicted range for all of the chemicals except p,p'-DDT. The observed proportion of toxic
samples was within the predicted ranges at concentrations between the T20 and T50 values for
all of the 33 models. The proportion of toxic samples observed between the T50 and T80 values
was within the predicted range of 50 to 80% toxicity for 3 1 of the 33 chemicals. However,
above the T80 value, the proportion of toxic samples exceeded 80% for only 10 chemicals.
Above the T80 value, the models overestimated the proportion of toxic samples observed for 16
chemicals, and insufficient data were available to evaluate seven models.
4.3.1.3. Models Based on Organic Carbon-Normalized Chemical Concentrations
Acceptable logistic models based on organic carbon-normalized chemical concentrations
were developed for 25 organic chemicals, including 21 individual PAHs, total PCBs, and 3
organochlorine pesticides (Table 16). The number of samples available for developing these
models was slightly reduced from the total because a small percent of the samples was not
analyzed for organic carbon. (Models based on acid volatile sulfides and simultaneously
extracted metals were not pursued because of the low number of samples available with these
measurements). All organic carbon-normalized models used the Sig Only classification of toxic
samples. The models for four chemicals—1-methylnaphthalene, 2-methylnaphthalene, biphenyl,
and naphthalene—had the lowest acceptable normalized chi-square values, indicating relatively
poorer fits.
Concentration interval plots for the organic carbon-normalization approach are shown in
Figure 4. Contrary to the expectation that normalizing sediment chemistry for nonpolar organic
chemicals would reduce the variability in the concentration response relationships, the
concentration interval plots do not show less variability than the Sig Only dry-weight
normalization plots shown in Figure 2.
The T20, T50, and T80 values based on the organic-carbon normalized models are shown
in Table 17. The percent of toxic samples within the ranges defined by these Tp values is shown
25
-------
in Table 18. The models underestimated the incidence of amphipod toxicity at concentrations
below the T20 value for all 25 chemicals, although 21 of the 25 were within 10% of the top of
the range. Between the T20 and T50 values, the proportion of toxic samples observed for 8 of
the chemicals was within the predicted range of 20 to 50%, and 18 were slightly above the
predicted range. The proportion of toxic samples observed between the T50 and T80 values was
within the predicted range of 50 to 80% toxicity for all chemicals except biphenyl. Above the
T80 value, the proportion of toxicity exceeded 80% for fluorene only, and insufficient data were
available to evaluate 7 chemicals. The models for the other 17 chemicals would be expected to
overestimate toxicity for high concentrations.
4.3.2. Model Comparisons
We compared the performance of the models to provide additional insights into the
strengths and limitations of the different options and to explore the implications of those
differences. The models were compared on the basis of goodness of fit and the degree of
agreement between the predicted proportion of toxic samples and the proportion observed in the
database. The implications of different modeling options were further evaluated by comparing
the resulting Tp values.
4.3.2.1. (ioodness-of-Fit Comparisons
We compared normalized chi-square statistics across all the different chemical models
for the different modeling alternatives. We also evaluated the number of chemicals whose
models exceeded the 0.15 normalized chi-square criterion.
The Sig Only approach generated acceptable models for 4 more chemicals than did the
MSD approach. Models for antimony, arsenic, nickel, and p,p'-DDE had normalized chi-square
values of less than 0.15 for the MSD toxicity classification. Regression models based on toxicity
classification using the Sig Only approach consistently had higher normalized chi-square values
than did those using the MSD approach, with the exception of benzo(b)fluoranthene (Figure 5).
The organic carbon-normalized approach generated acceptable models for 25 of the 27 organic
chemicals that had acceptable models using dry-weight normalization. Regression models based
on dry-weight-normalized concentrations had higher normalized chi-square values than did those
using organic carbon-normalized concentrations for 25 of the 27 chemicals (Figure 6).
26
-------
4.3.2.2. Reliability Comparisons
Tables 19, 20, and 21 show the difference between the observed proportion of toxic
samples and the average predicted proportion of toxic samples within the four ranges defined by
the Tp values for the Sig Only, MSD, and organic carbon-normalization approaches,
respectively. Using antimony as an example, of the 1041 samples a with predicted probability of
toxicity of
-------
The two approaches for classifying toxicity had different strengths and limitations. A
greater number of models had acceptable goodness-of-fit statistics with the Sig Only approach
than with the MSD approach. Models fit using the Sig Only classification approach had slightly
better goodness-of-fit statistics than did those fit using the MSD approach. The differences
between observed and predicted proportions of toxic samples were smaller for the MSD
approach at concentrations less than the T50 value. The Sig Only approach had a greater
tendency to underestimate observed toxicity at low concentrations. In contrast, at concentrations
greater than the T80 value, differences between observed and predicted proportions of toxic
samples were smaller for the Sig Only approach. The MSD approach had a greater tendency to
overestimate toxicity at these higher concentrations.
It is not unexpected that individual chemical models would predict a proportion of toxic
samples lower than that observed at low concentrations. Chemicals in the sample other than the
one being modeled may be responsible for the observed toxicity. Predictions at high
concentrations have greater significance for evaluating risk at contaminated sites. For these
reasons, we selected the Sig Only approach for further evaluation and development.
4.4. EVALUATION OF ALTERNATIVE SCREENING APPROACHES
As discussed in Section 3.4.3, the data for each contaminant were screened prior to
applying the logistic model or plotting the data. The data screening procedure was designed to
exclude samples where the chemical under consideration would not serve as a good indicator of
observed toxicity. The standard screening approach (1X screening) that we used as the basis for
comparisons was described in the methods section (Section 4.2). In this approach, the
concentration of the selected chemical in each toxic sample was compared with the mean of the
concentration of that substance in the nontoxic samples collected in the same study and
geographic area. If the concentration of a chemical in an individual toxic sample was less than
or equal to the mean concentration of that chemical in the nontoxic samples from that study area,
it was considered unlikely that the observed toxicity could be attributed to that chemical.
Therefore, these toxic samples were not included in the screened data set used for developing the
logistic model for that chemical.
Although there are many possible ways to identify samples for screening (e.g., various
statistical criteria), we confined our evaluation to simple approaches similar to those used
previously (Ingersoll et al., 1996; Long and Morgan, 1991; Long and MacDonald, 1992;
28
-------
MacDonald et al., 1996). This section compares the standard method with two alternatives: (1)
using all of the data in developing single-chemical models (unscreened), and (2) eliminating
toxic samples having concentrations less than or equal to twice the mean concentration of that
chemical in the nontoxic samples from that study area (2X screening).
The models developed using the two alternative screening approaches were first
compared with our standard approach on the basis of concentration interval plots and goodness
of fit. The 2X screening alternative was further evaluated by comparing the degree of agreement
between the predicted proportion of toxic samples with the proportion observed in the database
and the resulting Tp values.
4.4.1. Unscreened Versus IX Screening
The unscreened data set includes all of the data for a chemical, whereas the 1X screening
removed 41.3 to 59.5 % (average of 48.5%) of the toxic samples from the derivation of the
models. The effects of including all data for a chemical are illustrated in Figure 8 using lead and
phenanthrene as representative examples. Each plot shows the LRMs and the concentration
interval data for the respective screening approach.
Visual inspection of these plots revealed that inclusion of the data for those toxic samples
in which the chemical of concern is a poor indicator of the observed response tended to scatter
and skew the data distributions, particularly at lower concentrations. The plots of the unscreened
data for lead and phenanthrene show very few intervals with <20% effects. Importantly, the
unscreened data for phenanthrene showed only a weak relationship between chemistry and
toxicity for concentrations <1000 mg/kg dry weight, and the incidence of effects is
approximately 20 to 60% below that concentration. However, there were few differences
between the distributions of the screened and unscreened data at higher concentrations.
Models fit with the unscreened data had much lower normalized chi-square values
(Figure 9). The models for only four chemicals—copper, fluorene, dieldrin, and DDT—
exceeded the normalized chi-square criterion of 0.15 that we used to identify models with fits
sufficient to support the calculation of the Tp values. The 1X screening approach yielded a
greater number of acceptable models describing the relationship between contaminant
concentrations and biological effects.
29
-------
4.4.2. 2X Screening Versus IX Screening
The 2X screening approach removed 63.8 to 84.2% (average of 70.4%) of the toxic
samples from the derivation of the model. The effects of the 2X screening alternative are
illustrated in Figure 10 using lead and fluoranthrene as representative examples. Each plot
shows the LRM and the concentration interval data for the respective screening approach.
The concentration interval plots show that the 2X screening approach reduced variability
somewhat. The reduced variability was also reflected in slightly higher normalized chi-square
values for the models fit using the 2X approach for most chemicals (Figure 11). In addition, the
concentration interval plots indicate that the logistic regression curves for the 2X screening
approach were shifted to the right at lower concentrations.
Another way to examine the differences between the models developed from the two
screening approaches is to compare the Tp values. Figure 12 shows the relationship between the
Tp value concentrations calculated using the different screening approaches. The greatest
difference was seen in the T20 values, which is consistent with the patterns observed in
concentration interval plots. The difference in Tp values decreased with concentration; the T80
values were almost identical.
Comparing differences between the observed and predicted proportions of toxic samples
in the IX and the 2X screening approaches (Tables 19 and 22, respectively), the differences in
the 2X screening approach were smaller for predictions below the T50 values. Differences were
comparable for the T50-T80 value range. For concentrations above the T80 value, the
differences in the 1X screening approach were slightly smaller than those in the 2X approach
(absolute average 9.1% difference compared to 10.1%).
4.4.3. Summary of Screening Approach Evaluations
We evaluated alternative screening approaches used to exclude samples where the
chemical under consideration would not serve as a good indicator of observed toxicity. The
models from the unscreened alternative showed a weaker relationship between chemistry and
toxicity than was observed with the other screening alternatives. The models generated using the
2X screening approach had better goodness-of-fit statistics. The 2X approach had lower
differences between observed and predicted toxicity at concentrations below the T50 value;
however, it had a slightly greater tendency to overestimate toxicity at concentrations above the
30
-------
T80 value. In addition, the 2X approach excluded approximately 22% more toxic samples from
the model derivation.
The 2X screening approach has some advantages over the 1X screening approach,
particularly at concentrations less than the T20 value. However, the IX screening approach
performed slightly better at concentrations above the T80 value and screened out considerably
fewer toxic samples in the model derivation. We selected one modeling approach to manage the
many permutations associated with further model development and exploration. We retained the
IX model because of the better performance at high Tp values and retention of more samples.
4.5. COMPARISON OF TOXICITY TEST ENDPOINTS
We used the LRM approach to investigate the implications of using a model that
combines the response of the two amphipod species. We explored more complex models that
included different slopes or intercepts for the two species. We compared the Tp values that
would result from individual species models with those of the combined models. In addition, we
compared the toxicity observed for each species with predictions based on the combined model.
4.5.1. Statistical Comparisons of Species-Specific LRIMs
Using the LRM approach, it is possible to fit a separate slope, an intercept, or both for
each of the two amphipod species. If the two species responded with different sensitivities to the
chemicals, a separate slope or intercept would be statistically significant. We tested the
significance of separate slope, intercept terms using a sequential chi-square comparison (Neter et
al., 1996). The first test compared a common slope, common intercept model (Model A) with a
common slope, different intercept model (Model B). The results of this analysis indicated that
an additional intercept term was statistically significant (a = 0.05, 1 degree of freedom) for all of
the chemicals considered (Table 23). The second test compared Model B with a different slope,
different intercept model (Model C). The distinct slope term was not significant (a = 0.05, 1
degree of freedom) for 20 of the 37 chemicals. For these chemicals, parallel models with a
different intercept for each species were preferred.
31
-------
4.5.2. Comparison of Tp values for Separate Amphipod Models
We developed separate logistic models for each amphipod species and calculated Tp
values for comparison with the combined amphipod models (Tables 24 and 25). The results
showed that the T20, T50, and T80 values for A. abdita were higher than those of the combined
model by a factor of approximately 2, indicating that A. abdita is slightly less responsive than the
combined model would predict. In contrast, the T20, T50, and T80 values for R. ahronius were
lower than those of the combined model by factors of 3.9 for the T20 value and 3.1 for the T80
value, indicating that R. ahronius is more responsive than the combined models would predict.
4.5.3. Reliability Comparisons
We compared the observed proportion of toxic samples for the two species in the
unscreened data set with predicted values from the combined model within the four ranges
defined by the combined model Tp values (i.e., T80). On
average, the difference between observed and predicted proportion of toxic samples for A. abdita
was less than 10% for all concentration ranges (Table 26). The observed proportion of toxic
samples using R. ahronius was much greater than predicted below the T50 value. Above the T50
value, the difference between observed and predicted proportions of toxic samples using R.
ahronius averaged 11%.
4.5.4. Summary of Species Comparisons
The results of the species-specific analyses suggest that there are substantial differences
in the chemical-specific models for the two species. The model comparison indicated that
parallel models with different intercepts was the preferred model for more than half of the
chemicals. Comparing the Tp values derived from separate species models with those derived
from the combined model suggests that R. ahronius has a greater response than A. abdita at
similar concentrations. Still, on average, the T50 values differed by approximately a factor of 2
for A. abdita and a factor of 3 for R. ahronius. The combined model consistently and
substantially underpredicted the observed proportion of toxic samples for R. ahronius below the
T50 value. Above the T50 value, differences between the observed proportion of toxic samples
and that predicted using the combined individual chemicals models were minimal for most of the
A. abdita data and for R. ahronius data.
32
-------
The observed species differences have several possible explanations: unmeasured
chemicals or factors other than chemistry may have influenced R. abroniiis results at low
chemical concentrations, or there may be inherent differences in sensitivity between the two
species. However, the greatest difference between the species-specific observations and the
combined models are at lower concentrations for R. abroniiis. As discussed above, individual
chemical models may underestimate observed toxicity at low concentrations because chemicals
in the sample other than the one being modeled may be responsible for the observed toxicity.
Therefore, the species-specific differences could also be explained if the database for R. abroniiis
contains a disproportionate amount of data from areas with a high degree of contamination from
multiple chemicals. This possibility is difficult to investigate with single-chemical models, but
the issue is revisited in Chapter 5.
4.6. SENSITIVITY OF MODELS TO ERRORS IN UNDERLYING DATA
We did not conduct a formal analysis to evaluate the sensitivity of the models to potential
errors in the underlying database. However, after initial model development and evaluation, we
discovered an error in PCB concentration units for 15 samples. These samples had erroneously
high concentrations, and, in addition, a relatively small number of other samples in the database
had similarly high PCB concentrations. This situation provided an opportunity to evaluate the
degree of change in the models resulting from errors in a small number of highly influential
values.
The correction in concentrations for these 15 samples changed the LRM for PCBs,
particularly at high concentrations (Figure 13). It also improved the model fit: the normalized
chi-square value changed from 0.24 to 0.27. As expected, the Tp values changed most at high
concentrations; the T80 was reduced by 42% (Table 27).
4.7. INDIVIDUAL CHEMICAL MODELS AND SPIKED-SEDIMENT BIOASSAY
MEDIAN LETHAL CONCENTRATION (LC50) VALUES
Dose-response data from laboratory spiked-sediment bioassays provide additional
perspective on the concentrations of individual chemicals that can be considered to cause
toxicity. However, results from spiked-sediment bioassays are not immediately comparable with
predictions from the LRMs. Toxicity from spiked-sediment bioassays is more confidently
attributed to the chemical added to the sample. In contrast, the response to any individual
-------
chemical measured in field-collected sediments may be confounded by the presence of other
chemicals.
In addition, the magnitude of response reported in spiked-sediment bioassays is much
greater than the magnitude of response required to classify a sample as toxic in this study. Most
of the studies in the literature on spiked-sediment toxicity report LC50s. An LC50 value
represents the concentration corresponding to 50% survival of test organisms. In this study,
many samples with much higher test survival were classified as toxic. The control-normalized
survival averaged across all samples and chemicals decreases with increasing probability of
toxicity defined by the model Tp values (Figure 14). The relationship is such that an average
survival of 50% (LC50) corresponds to concentrations exceeding T50 values.
Reported LC50 values for 10-day, spiked-sediment toxicity tests conducted with marine
amphipods were compared with the probability of toxicity estimated from the individual
chemical models (Table 28). Using the logistic models, the probability of toxicity at the reported
LC50 values ranged from 0.54 for zinc to 0.97 for mercury, with most estimates falling within
the 0.8 to 0.9 range. This is consistent with the average percent survival observed at high
probability of toxicity, as shown in Figure 14.
4.8. SUMMARY AND CONCLUSIONS
This section presents LRM results for models that predict the probability of toxicity on
the basis of individual chemical concentrations. The toxicity endpoint modeled was the 10-day
survival test conducted using two species of marine amphipods (A. abdita and R. ahronius).
The LRM approach was used to explore and compare different methods of toxicity
classification and chemistry normalization. We evaluated two approaches for designating
samples as toxic: (1) less than 90% survival that was significantly different from negative
control samples (Sig Only), and (2) control-normalized survival less than 80% that was
significantly different from negative control samples (MSD) (based on the analysis by Thursby
et al., 1997). We evaluated two approaches for normalizing sediment chemistry: dry-weight
normalized and organic carbon normalized.
Models were evaluated using several approaches: goodness-of-fit statistics, visual
examination of plots of the model and the underlying data, and reliability (predicted vs. observed
proportion of toxic samples within ranges of probability) for the entire data set. The MSD
approach had a greater tendency to overestimate the proportion of toxic samples observed at
34
-------
higher concentrations. Although the Sig Only approach had a greater tendency to underpredict
observed toxicity at low concentrations, this discrepancy may be explained by the presence of
other chemicals in the sample. The models based on organic carbon-normalized sediment
chemistry had lower goodness-of-fit statistics than did the dry-weight-normalized models, and
they also had larger differences between observed and predicted toxicity. We selected the Sig
Only approach with dry-weight normalization for further evaluation and development.
We also used the LRM approach to evaluate the implications of several data treatment
options. We evaluated three screening criteria used to decide whether a given sample was
included in the model data set for an individual chemical: (1) include all samples (unscreened),
(2) exclude toxic samples with concentrations that were less than or equal to the mean
concentration of nontoxic samples from the same study (IX screening), and (3) exclude toxic
samples with concentrations that were less than or equal to two times the mean concentration of
nontoxic samples from the same study (2X screening). The models from the unscreened
alternative showed a weaker relationship between chemistry and toxicity than was observed with
the other screening alternatives. The 2X screening approach has some advantages over the 1X
screening approach, particularly at concentrations less than the T20 value. However, the IX
screening approach performed slightly better at concentrations above the T80 value and screened
out fewer samples in the model derivation, which may prove important in the future development
of models for less-frequently measured chemicals. We concluded that the small improvements
in model fit did not outweigh the associated reduction in sample size, and retained the IX
approach for further development and exploration.
Finally, we evaluated the implications of developing models that combined data for the
two amphipod species. Agreement between the observed proportion of toxic samples and that
predicted using the combined individual chemicals models was good for both species above the
T50 values. Below the T50 value, the combined models performed well for A abdita but
consistently and substantially underpredicted the observed proportion of toxic samples for R.
ahronius. The discrepancies may be explained by inherent differences in sensitivity between the
two species, by a greater responsiveness of R. ahronius to nonchemical factors, or by the
database for R. ahronius containing a disproportionate amount of data from areas with a high
degree of contamination from multiple chemicals. These issues will be investigated further using
multiple-chemical models.
35
-------
The single-chemical models presented in this chapter can be used to develop screening-
level concentration-response relationships for individual chemicals. These relationships could
then be used to identify concentrations of individual chemicals that correspond to different
degrees of risk, depending on the objectives of the user. This application is similar to the current
use of SQGs; the use of the single-chemical models in evaluating SQGs is discussed in Chapter
7. The risks of a toxic response posed by the mixture of chemicals present in a particular sample
are best evaluated using a multiple-chemical approach, which is discussed in the next chapter.
36
-------
5. MULTIPLE-CHEMICAL MODELS
5.1. INTRODUCTION
One of the major challenges in assessing the ecological risk associated with exposure to
contaminated sediments is the presence of chemical mixtures. Field-collected sediments, as a
rule, contain complex mixtures of chemicals and other factors that influence toxic response.
Because the individual models described in Chapter 4 were derived from field-collected
sediments rather than from laboratory dose-response studies, to some extent the individual
models incorporate the overall toxicity of the mixture of chemicals in the samples. We sought to
improve the estimates of response by combining the information contained in the individual
chemical models into a single estimate of toxic response. The multiple-chemical model could
then be applied to predict whether new samples with known sediment chemistry would be
expected to produce a toxic response in an amphipod test.
This chapter describes the approach used to develop and evaluate multiple-chemical
models. Sections 5.2 and 5.3 discuss model development and results, respectively. The models
were used to evaluate several issues (Section 5.4): the relationship between the probability of
observing a toxic effect and the magnitude of toxicity, the identification of chemicals that most
influence model performance, the effect of reducing the number of individual chemical models
used in developing the multiple-chemical models, the performance of the models in predicting
toxicity of the two amphipod species, and the performance of the models in predicting toxicity
observed in individual studies. Finally, the performance of the models was evaluated using an
independent data set.
5.2. MODEL DEVELOPMENT
Most evaluations of the effects of mixtures on aquatic toxicity endpoints such as survival
and growth have focused on two empirical models of noninteractive joint action: concentration
addition and response addition (Broderius, 1991). Concentration addition, which is also referred
to as "simple similar action," assumes that contaminants act independently but by a similar mode
of action. Toxic unit models, which are a specialized case of concentration addition, have been
applied to the assessment of the toxicity of PAH mixtures in sediment (Swartz et al., 1995; Di
Toro et al., 2000; Lee et al., 2001), but they are unlikely to be applicable to complex mixtures of
contaminants commonly found in the environment that have different modes of toxic action. In
37
-------
response addition, or independent action, which is expected to apply to cases where
contaminants have a different mode of action, toxicity would be predicted only when one or
more contaminants exceeds its toxicity threshold.
We explored three alternative approaches for combining the individual chemical models
into a single explanatory variable that could be used to estimate the probability of observing
toxicity in a given sample:
• the maximum probability of observing toxicity for a sample, taken from the set of
probabilities calculated for each individual chemical in the sample (PMax),
• the mean probability of observing toxicity for a sample, based on the set of
probabilities calculated for each individual chemical in the sample (P Avg), and
• the product of the probabilities of surviving exposure to all individual chemicals in
the sample (PProd), calculated by multiplying the values of one minus the
probability of observing toxicity for each chemical in the sample and then subtracting
the resulting product from one.
All three approaches can be considered similar to response-addition models. A model
based on PMax would predict toxicity on the basis of the individual chemical model with the
highest probability of toxicity. This approach would be expected to be most effective in
predicting toxicity when the degree of correlation between responses to individual chemicals is
high. The PProd and PAvg approaches include all chemicals in the estimate of toxicity.
Models using these explanatory variables would be expected to be most effective when the
degree of correlation between responses to individual chemicals is intermediate and low,
respectively (U.S. EPA, 2000).
A possible disadvantage of P Avg as an explanatory variable is that it gives more weight
to chemical classes that have many individual chemicals that tend to co-occur. For example,
P Avg incorporates the output from models for 22 individual PAHs. Because individual PAHs
are likely to co-occur in environmental samples, P Avg may be influenced more by the
concentrations of PAHs than by the concentrations of other chemicals. In addition, the P Avg
approach may appear to reduce the influence of chemicals associated with high probabilities of
toxicity by averaging them with chemicals having low probabilities of toxicity.
P Max, P Avg, and P Prod were all developed using the simple chemical parameter
estimates shown in Table 10, except for PCBs. As discussed in Section 4.5, we discovered an
38
-------
error in PCB units for 15 samples after the models had been fit and evaluated. This change
resulted in a minor change in the PCB model parameter estimates. However, because the effects
of the correction on the multiple-chemical models were extremely small (as discussed further in
Section 5.4.4), the multiple-chemical models were not changed.
Preliminary plots of P Avg and PMax versus the percent of samples that were toxic in
sequential intervals of increasing contamination appeared nearly linear, so a multiple linear
regression modeling approach was pursued. The dependent variable for the models was the
frequency of toxicity observed within the sequential intervals of increasing contamination (i.e.,
probability intervals). The probability intervals were defined by first sorting the samples by the
predicted probability (i.e., P Avg and P Max) and then combining samples into groups
containing 50 unique predicted probability values. The median predicted probability associated
with each interval was used as the explanatory variable. Because the dependent and explanatory
variables both depend on how samples are binned, we evaluated the sensitivity of the models to
different binning approaches (Appendix E). The results of this analysis indicated that, above a
minimum number of regression samples (i.e., the number of bins) and samples within the bins,
the regression model coefficients showed little change with variations in binning approach.
We initially considered a fourth alternative, the sum of the probabilities of observing
toxicity for each individual chemical in the sample (PSum model). This approach was
eliminated from further consideration because the resulting range of predicted probabilities had
no upper bound, making the results very sensitive to the number of chemicals measured in a
particular sample.
We developed the multiple-chemical models using all samples with matching chemistry
and toxicity (i.e., no additional data screening procedures were employed). To minimize the
potential impact of samples with partial chemistry, only samples with measured values for at
least 10 chemicals were included in the data set used to derive the multiple-chemical models.
We evaluated the alternative approaches by examining goodness of fit (as indicated by R-
squared values), concentration interval plots (described in Section 4.2.2), and the agreement
between the observed proportion of toxic samples and that predicted using the multiple-chemical
models (described in Section 4.3.2.2).
A multivariate approach to the problem uses all chemical data simultaneously to predict
the probability of a toxic test result was explored in a companion effort (Smith et al., 2003). The
multivariate approach requires complete data for all chemicals included in the model. The single
39
-------
explanatory variable approach developed in this chapter is potentially applicable to a wider array
of chemistry data.
5.3. MODEL RESULTS
Probability density functions of P Avg, PMax, and PProd values are shown in Figure
15. The probabilities generated using the P Avg and P Max approaches were fairly evenly
distributed across the range of probabilities. The distribution of P Prod values was so heavily
skewed toward higher values that P Prod would have limited ability to discriminate among
moderately to highly contaminated samples. We eliminated this model from further
consideration.
The regression models generated using P Avg and P Max are shown in Figure 16. A
quadratic term was significant for the P Max model, indicating that the relationship is
curvilinear. Both models explained a large amount of the variation in the observed frequency of
toxic samples in the probability intervals (R-squared values were 0.89 and 0.93 for P Avg and
P Max models, respectively). Regression diagnostic plots (e.g., Cooks distances, residual
distributions [not shown]) revealed no overly influential values or severely nonnormal residual
distributions.
We compared the differences between the frequency of observed toxicity with that
predicted by the P Avg and P Max models within quartiles of the predicted probability (Figure
17). All samples in the database were used in the comparison (including samples with fewer
than 10 chemicals measured). When applied to individual samples, the P Avg model could yield
predictions of probability of toxicity greater than 1, so the predicted probability was capped at 1.
The mean predicted probability of toxicity within probability quartiles closely matched the
observed proportion of toxic samples within the same probability quartiles, demonstrating the
overall reliability of both the P Max and the P Avg models within the database that was used to
derive the model.
The PMax and PAvg models explained 92% and 88%, respectively, of the variation in
the frequencies of toxicity observed in the probability intervals derived from the entire database
(Figure 18). The P Max model adjusts for the difference between the maximum probability
from the individual chemical models and the observed proportion of toxic samples within the
same probability interval. For example, for a maximum probability of 1 from the individual
chemical models (x-axis), the observed proportion of toxic samples (and the predicted
40
-------
probability from the PMax model) is 0.84 (Figure 16). The data used to derive the PAvg
model show the opposite situation, where the mean probability value from all of the individual
chemical models is somewhat lower than the corresponding observed proportion of toxic
samples. Thus, mean probabilities of 0.5 and 0.75 correspond to proportions of toxic samples of
0.7 and 0.9, respectively; the PAvg adjusts the final estimate of toxicity accordingly (Figure
16).
5.4. USING THE MULTIPLE-CHEMICAL MODEL AS AN ANALYTICAL
FRAMEWORK
The multiple-chemical models provide a consistent analytical framework that can be used
to evaluate several issues that are relevant to evaluating sediment toxicity. In the following
sections, we demonstrate how the models can be used to
• evaluate the relationship between the probability of observing a toxic effect and the
magnitude of toxicity,
• identify the chemicals that serve as the most (or least) effective surrogates for
toxicity,
• evaluate the effect of reducing the number of individual chemical models combined
into the multiple-chemical model,
• evaluate the performance of the models in predicting toxicity of the two amphipod
species, and
• evaluate the performance of the models in predicting toxicity observed in specific
studies.
We used the P Max model in the evaluations discussed in the remainder of this report
because it explained a slightly higher amount of variation, it is less influenced by the number of
chemicals analyzed in a sample, and we could reduce the number of analytical permutations.
5.4.1. Relationship Between Probability of Observing a Toxic Effect and Magnitude
of Toxicity
The magnitude of the effect (decreased survival) in the amphipod test increased as the
probability of toxicity increased (Figures 19 and 20). Toxic samples with a probability of
41
-------
toxicity less than or equal to 0.25 had an average control-adjusted survival of greater than 75%,
whereas samples with a probability of toxicity greater than 0.75 had an average control-adjusted
survival of less than 50%. Figure 20 shows the strong relationship between the probability of
toxicity predicted by the PMax model and control-adjusted survival. This demonstrates that
samples that are estimated to have the highest probability of toxicity are also likely to cause a
high magnitude of mortality.
5.4.2. Chemicals that Serve as the Most Effective Surrogates for Toxicity
The P Max model is based on the individual chemical that has the highest probability of
toxicity. For approximately 70% of the samples, individual chemical regression models for
metals produced the maximum probability used in the P Max model (Table 29). This should not
be construed to imply that metals were causing toxicity in these samples. It does indicate,
however, that metals appear to be a good predictor of toxicity in field-collected samples where
mixtures of contaminants are likely to be present.
The influence of different chemical classes on the model was explored by excluding the
individual chemical models for PAHs, metals, and pesticides and PCBs. The models produced
by excluding the different chemical classes are shown in Figures 21, 22, and 23, (PAHs, metals,
and pesticides and PCBs, respectively). The difference between the models was most evident
when metals were excluded, which is not surprising, as 70% of the samples had a metal
associated with the maximum probability for that sample. Still, there were only minor changes
in the models. This is consistent with the interpretation that the chemicals associated with the
P Max values are serving as surrogates for the overall degree of contamination present in a
sample.
5.4.3. Effect of High Levels of Several Chemicals
The number of chemicals in a sample that have a high probability of toxicity according to
the individual chemical models (e.g., /;>0.75) makes a difference in how well the model
predictions match the observed proportion of the samples that are toxic (Figure 24). As shown,
when only one chemical in a sample has a probability of toxicity greater than 0.75, the P Max
model tends to overestimate the incidence of observed toxicity. The P Max model slightly
underestimated the frequency of toxicity observed in samples that contained two or more
chemicals with a probability of toxicity greater than 0.75. The degree of underestimation,
42
-------
however, remained about the same even for samples containing many chemicals with
probabilities greater than 0.75.
5.4.4. Sensitivity of PMax Model to Individual Chemical Models
The PMax model described above was developed by including all 37 chemicals with
chemical-specific models with normalized chi-square values greater than 0.15. We investigated
model sensitivity to the inclusion criterion. In particular, we hypothesized that a more stringent
inclusion criterion (i.e., excluding chemicals with poorer fits) would improve the multiple-
chemical models. The P Max model fit using a normalized chi-square criterion of 0.27, which
represented the average normalized chi-square value for acceptable models, included 19
chemicals, contained only the linear term, and explained slightly more variation (R2 = 0.94,
Figure 25) than the original P Max models (Figure 16).
5.4.4.1. Effect of PCK Model Correction
We examined the influence of the updated PCB model on the PMax model. As
discussed in Section 4.6, the correction in PCB units for 15 samples resulted in a model
sufficiently different to cause us to update the PCB model parameters reported in Table 1.
However, the effect of updating the PCB model on the P Max model results was extremely
small. The model parameters changed slightly, to 0.11 (intercept), 0.34 (linear term), and 0.39
(quadratic term) from 0.11, 0.33, and 0.4, respectively. The maximum difference in predicted
probability of toxicity for an individual sample was 0.0025. Because these differences were so
small, we elected to continue using the original P Max model in the evaluations discussed in this
report.
5.4.5. Predictions of P Max Model for Individual Species
The P Max models also provide a framework for re-examining the species-specific
differences in model performance. Figures 26 and 27 plot the probability of toxicity predicted
from the nationwide P Max model against three species-specific variables: (1) the proportion of
observed toxicity based on the Sig Only classification, (2) control-adjusted survival, and (3) the
proportion of toxicity based on MSD classification. There was a strong relationship between the
predicted probability of toxicity and the three variables. The relationship between predicted and
observed proportion toxicity was slightly stronger using the MSD classification for both species.
43
-------
The relationship between predicted and observed toxicity was very different for the two
species. In general, R. abroiiins showed a higher response rate than did A. abdita at similar
predicted probabilities. This difference is consistent with that discussed for the individual
chemical models in Section 4.5 and may be explained by inherent differences in sensitivity
between the two species or a greater sensitivity of R. abroiiins to nonchemical factors, or the
database for R. abroiiins may contain a disproportionate amount of data from areas with a high
degree of contamination from multiple chemicals.
As shown in Figures 26 and 27, most of the A. abdita data fall between predicted
probabilities of 0 and 0.5, whereas the R. abroiiins data fall mostly between 0.25 and 0.75.
Although we cannot completely resolve the cause of differences in species responses, differences
in sediment chemistry cannot be discounted as a contributing factor. Although the variability in
the modeled response of R. abroiiins was higher than that in the A. abdita response, the model
still served as an effective indicator of the observed proportion of toxic samples. Combining
data from tests that used either species enables the development of models that encompass more
areas of the country and a broader range of sediment chemistry. The strong relationships of the
individual species' responses with the nationwide model affords confidence that the combined-
species model provides a common framework that is applicable to both species.
5.4.6. Predictions of P IMax Model for Individual Studies
The multiple-chemical models are a function of the covariation among chemicals on a
nationwide basis. Accordingly, the nationwide model may not accurately predict observed
toxicity at a particular site if the chemical mixture is very different from the average across the
nation or if site-specific factors greatly influence the degree of toxicity (e.g., by increasing or
decreasing chemical bioavailability).
To examine the application of the nationwide model to individual studies, we first
examined the performance of models in predicting observed toxicity for individual studies within
the database that are represented by more than 20 samples (Table 30). We compared the
differences between the frequency of observed toxicity with that predicted by the P Max models
within quartiles of the predicted probability. Of the 39 studies with sufficient data to evaluate,
14 had observed frequency of toxicity within 20% of that predicted in all quartiles having data.
Nine studies had observed frequency of toxicity that differed by more than 20% in all quartiles
having data. The remainder of the studies had mixed performances.
44
-------
We conducted a more in-depth evaluation of the performance of the models for
predicting toxicity in specific studies by comparing predicted and observed toxicity within
smaller intervals. To have sufficient data for this evaluation, we combined data from several
studies within a common geographic region (Table 3 1). Figures 28 through 32 plot the
probability of toxicity predicted from the nationwide PMax model against three site-specific
variables: (1) the proportion of observed toxicity based on the Sig Only classification, (2)
control-adjusted survival, and (3) the proportion of toxicity based on MSD classification.
Figures 28 through 32 show that the nationwide P Max model provides a useful basis for
evaluating toxicity test results for individual regions included in the database. There were
significant relationships between the P Max model predictions and observed proportion of
toxicity, but the relationships were not one-to-one, indicating that regions differed from the
nationwide model by intercept or slope or both (left-hand graphs in Figures 28 through 32). The
differences were not consistent across the regions. The probability of toxicity was strongly
related to percent control-adjusted survival for all regions except California (center graphs in
Figures 28 through 32). For the California data (right-hand graph in Figure 32), comparing the
P Max model (which was derived using the Sig Only toxicity classification) with observations
classified using the MSD approach greatly improved the strength of the relationship. For the
other regions, using the MSD approach only slightly clarified the relationship, mostly at low
probability of toxicity.
In theory, model predictions also could be compared with observations from studies with
fewer samples (e.g., from a smaller area). Comparisons are most useful if the observed data
represent a large range of predicted probabilities of toxicity (left-hand graph in Figure 33).
Studies with observations from a small range of predicted probabilities may show no apparent
relationship (right-hand graph in Figure 33). A consistently high proportion of toxic samples
(left-hand graph in Figure 34) may show a stronger relationship with the nationwide model by
classifying toxic samples using an MSD approach (right-hand graph in Figure 34).
5.5. APPLICATION OF THE MODELS TO AN INDEPENDENT DATA SET
Application of the models to independent data (data not used in model derivation) was an
important step in evaluating the models. The P Max model was applied to a small independent
data set consisting of three studies from the Calcasieu Estuary (Louisiana) that had matching
sediment chemistry and toxicity data for A abdita (Redmond et al., 1996; unpublished data set
45
-------
provided electronically by P. Crocker, U.S. EPA, Region 6, Dallas, TX; MacDonald et al.,
2001). The Calcasieu Estuary is a highly contaminated industrial waterbody that is included on
the National Priorities List of hazardous waste sites. The Calcasieu Estuary data were not
included in the database used to derive the models. The data set contains 170 matched chemistry
and toxicity test results using A. abdita and represents a wide range of contaminant
concentrations.
There was a strong linear relationship between the predicted probability of toxicity and
the proportion of toxicity observed, based on both the Sig Only and the MSD classifications
(left-hand and right-hand graphs in Figure 35). In general, the frequency of toxicity observed in
the Calcasieu Estuary samples was greater than that predicted by the PMax model. The
predicted probability of toxicity showed a strong relationship with control-adjusted survival
(center graph in Figure 35).
5.6. SUMMARY AND CONCLUSIONS
We developed multiple-chemical models to combine the individual chemical models into
a single prediction of the toxicity of sediment samples with known sediment chemistry. Two
approaches for combining the individual chemical models (the P Max and PAvg) produced
models that accurately predicted the frequency of toxicity to amphipods observed in the
database:
• P Max is the maximum probability of observing toxicity for a sample, taken from the
set of probabilities calculated for each individual chemical in the sample, and
• P Avg is the mean probability of observing toxicity for a sample, based on the set of
probabilities calculated for each individual chemical in the sample.
The P Max model explained 92% of the variation in the frequency of toxic samples
observed in probability intervals constructed using the entire data set. The R-squared value for
models using the P Avg value was slightly lower.
We used the P Max model to evaluate several issues: the relationship between the
probability of observing a toxic effect and the magnitude of toxicity, the identification of
chemicals most influential in model performance, the effect of reducing the number of individual
chemicals models used in developing the multiple-chemical models, the performance of the
46
-------
models in predicting toxicity of the two amphipod species, and the performance of the models in
predicting toxicity observed in individual studies and regional areas.
The magnitude of the effect (decreased survival) in the amphipod test increased as the
probability of toxicity increased, demonstrating that samples that are estimated to have the
highest probability of toxicity are also likely to be extremely toxic.
For approximately 70% of the samples, individual chemical regression models for metals
produced the maximum probability used in the PMax model. This should not be construed to
imply that metals were causing toxicity in these samples, only that metals appear to be a good
predictor of toxicity in field-collected samples. Indeed, removing metals (or other entire
chemical classes) from the suite of individual chemical models used to generate the P Max
model resulted in only minor changes in the model and model fit.
The P Max model also provides a useful basis for comparing the species-specific
differences in model performance. As observed with the individual chemical models, the
observed toxicity was almost always less than predicted for A. abdita and greater than predicted
for R. ahronius. However, the distribution of the data differs substantially for the two species
across the probability of toxicity predicted by the PMax model. Nevertheless, the observed
proportion of toxicity observed in both species was strongly related to the nationwide model,
affording confidence that the combined-species P Max model provides a common framework
applicable to both species.
Although the P Max model reliably predicted toxicity for the entire database used to
derive the models, it may not accurately predict observed toxicity at a particular site if the
chemical mixture is very different from the distribution of mixtures in the nationwide database or
if site-specific factors influence the degree of toxicity (e.g., bioavailability). To address this
issue, we examined the performance of models in predicting observed toxicity for individual
studies within the database. On a study-by-study basis, there was mixed agreement between the
frequency of observed toxicity and that predicted by the P Max model. The mixed performance
emphasizes the importance of calibrating the national models with site-specific data (discussed
further in Section 6.4). Still, the nationwide P Max model provided a useful, common basis for
evaluating toxicity test results for individual sites included in the database.
Application of the model to regional subsets of the database used to derive the model
demonstrated significant relationships between the P Max model predictions and both observed
proportion toxicity and percent control-adjusted survival. There was also a strong relationship
47
-------
between predicted toxicity and observed toxicity in the Calcasieu Estuary, a data set not included
in the original derivation of the models.
48
-------
6. APPLICATIONS OF MODELS
6.1. INTRODUCTION
This chapter discusses the application of the single-chemical and the multiple-chemical
LRMs to problems that are frequently encountered by risk assessors. Section 6.2 discusses the
relevance of the models to evaluating other toxicity endpoints and test systems, including those
commonly used in freshwater systems. Section 6.3 discusses the models in the context of other
empirical approaches for evaluating risks associated with individual chemicals found in
sediments. Finally, in section 6.4, we discuss how these models might best be used to evaluate
the risks of sediment contamination at specific sites or regions.
6.2. APPLICATION OF THE P MAX MODEL TO DATA FOR OTHER ENDPOINTS
The SEDTOX02 database contains matched sediment chemistry and toxicity test data for
endpoints other than the two marine amphipods. The data available for two sea urchins
(Slroiigy/occiilrolnspiirpiiraliis and Arbaciapimcliilala) are presented in Tables 3 through 5.
Data for the freshwater amphipod (Hyalclla azlcca) and midges (Chironomus teiitans and
Chironomus riparins) are shown in Tables 6 through 9. These data proved to be insufficient to
support models for many chemicals of concern for these test endpoints; instead, they were used
to evaluate the application of the marine amphipod models to these other endpoints.
Models for the marine amphipods were applied to the sea urchin embryological
development and fertilization data sets. The data for the two urchin species were combined for
each of the development (A. pimcliilala 48h and S. piirpiiraliis 96h) and fertilization (lh)
endpoints. The relationship between the average predicted probability of toxicity from the
PMax model and the proportion of toxicity observed within probability intervals is shown in
Figure 36. Note that there is a very weak relationship between the model and both urchin
response variables. However, if the data for the two urchin species are evaluated separately, A.
pimcliilala shows a strong relationship for the development endpoint (Figure 37) and a weaker
relationship for the fertilization endpoint (Figure 38). The results for S. piirpiiraliis demonstrated
no relationship between the predicted probability of toxicity and either the development or the
fertilization endpoint (Figures 39 and 40).
Models for the marine amphipods were applied to freshwater data sets for three
freshwater endpoints, H. azlcca 10-14-day survival, Chironomus spp. 10-14-day survival, and
49
-------
H. aztcca 28-day growth and survival. The proportion of toxicity observed using H. aztcca
(Figure 41) and Chironomus (Figure 42) 10-14-day survival tests was less than that predicted by
the PMax model. Below the mean predicted probability of 0.75, observed proportion of toxic
samples was consistently low (<25%). This relationship suggested a threshold, so we used a
spline model to describe the relationships (General Additive Model with 2 degrees of freedom
[Insightful Corp., 2001]). The spline model explained 51% and 64% of the variability seen in
the H. aztcca and Chironomus data, respectively. One possible explanation for this type of
threshold relationship is that the marine amphipod toxicity tests respond at lower concentrations
of chemicals in sediments than do the freshwater short-term toxicity tests.
The H. aztcca 28-day growth and survival endpoint showed a much stronger relationship
between the predicted probability based on the P Max model and the observed proportion toxic
in a relatively small data set (n = 126) (Figure 43). There is a strong relationship (R2 = 0.95)
between the probability of toxicity based on the P Max model versus the H. aztcca 28-day
growth and survival endpoint (Figure 43).
In summary, the results of comparing other endpoints with the P Max model predictions
suggest that the marine amphipod models are useful for predicting sea urchin response on the
basis of embryological development or fertilization tests for A. pimctulata but not for S.
puvpuvatus. The marine amphipod models also have utility for predicting the response of
freshwater amphipods, particularly the 28-day H. aztcca growth and survival endpoint. There is
the potential for developing endpoint-specific models as more data are acquired.
6.3. USING LRIMs TO EVALUATE EXISTING EMPIRICAL GUIDELINES
Hazardous waste site evaluations often involve the collection of substantial quantities of
sediment chemistry data, and these data are frequently used to support screening-level ecological
risk assessments. To evaluate such data, sediment assessors often use numerical sediment
quality guidelines (SQGs) such as threshold effect levels (TELs) and probable effect levels
(PELs), effect range low (ERL) and effect range median (ERM), and apparent effect thresholds
(AETs) (Gries and Waldow, 1996; Long and Morgan, 1991; Long and MacDonald, 1992; Smith
et al., 1996; Ingersoll et al., 1996, 2001, 2002; MacDonald et al., 1996, 2001).
Although derivation methods for the different SQGs are well described and are consistent
for all the chemicals within a given type of SQG, there is no straightforward method that enables
the user to either evaluate the degree to which individual SQGs meet their objectives or compare
50
-------
the risk levels associated with different SQGs. The logistic model approach provides a way to
put the individual SQG values into perspective by estimating the probability of observing
toxicity to amphipods at the chemical concentrations defined by the SQGs. Examples are shown
in Table 32 for three commonly used sets of SQGs that represent a range of threshold values:
TELs and PELs, ERLs and ERMs, and AETs. ERLs and TELs represent chemical
concentrations below which toxicity would be expected to occur infrequently (<25%) (Long and
Morgan, 1991; MacDonald et al., 1996), whereas effects are expected to be frequently observed
at concentrations exceeding PEL and ERM concentrations. In contrast, endpoint-specific AET
values represent concentrations above which toxicity is always expected for that endpoint.
The results are generally consistent with the narrative intent of the SQGs for most of the
chemicals for which SQGs had been derived. The highest probability of observing toxicity to
amphipods was noted for the amphipod AETs, with an estimated proportion of toxic samples
ranging from 45 to 99% and a median value of 90%. The predicted probability of observing
toxicity was lower for the PEL and ERM values, with median values of 55% and 72%,
respectively. At concentrations corresponding to the TELs, predicted probabilities of observing
sediment toxicity ranged from 10 to 41% (depending on the chemical under consideration), with
the probability of toxicity below 25% for 24 of the 27 chemicals considered (Table 32). The
probability of observing sediment toxicity was a little higher at the ERL concentrations (ranging
from 11 to 47%), with a median value of 33%. The probabilities may be higher than expected
for ERLs and TELs because they are calculated for individual chemicals. In practice, these
guidelines are most appropriately applied jointly; that is, a sediment sample has a low probability
of causing toxicity if all chemicals are below the ERL or TEL.
The LRMs help users select the sediment effect concentrations that most directly meet
the needs of their specific application. T10, T15, or T20 values could be calculated and used to
identify concentrations for individual contaminants that are likely to be associated with a
relatively low incidence of sediment toxicity (10, 15, or 20%, respectively). Such point
estimates of minimal-effect concentrations might be used in a screening assessment to identify
sediments that are relatively uncontaminated and have a low probability of sediment toxicity. As
discussed above, this evaluation would best be conducted by evaluating all chemicals
simultaneously. For example, of the samples having all chemicals at concentrations below their
respective T20 value, 19% were toxic, based on the Sig Only classification.
51
-------
Similarly, contaminant concentrations for which there is a high probability of observing
adverse effects could be estimated by calculating T70, T80, or T90 values. These higher point
estimates could be used to identify sediments that are highly likely to be toxic to amphipods and
to have a greater magnitude of effect (i.e., higher percent mortality). The Tp values can be used
in much the same way as other sediment guidelines, with the difference that the Tp value is
associated with a specific probability of observing toxicity and can include confidence bounds on
the sediment concentrations associated with a given Tp value.
Individual chemical SQGs are useful for identifying thresholds below which sediment
toxicity is unlikely to be observed and above which sediment toxicity is likely to occur.
However, it is difficult to determine the extent to which risk increases with the magnitude of
exceedance of an SQG. In addition, individual SQGs must be combined to address risks
associated with mixtures. To address both of these issues, practitioners have applied hazard
quotient approaches, which sum the ratios of the measured concentration of each contaminant to
its corresponding toxicity threshold. Several investigators have applied mean SQG quotients to
evaluate mixtures of contaminants in field-collected sediment samples (Long et al., 1998;
MacDonald et al., 2000; Fairey et al., 2001, Ingersoll et al., 2001). Such evaluations are based
on an assumption that concentration-response relationships for each chemical are similar. The
logistic regression modeling approach avoids this assumption by fitting concentration-response
relationships separately for each chemical and then standardizing the response variables to values
between 0 and 1.
Another approach to evaluating risks associated with sediment chemistry is the
equilibrium partitioning (EqP) approach. EqP links sediment chemistry values with biological
effects by combining the results of controlled laboratory tests using manipulated concentrations
of chemicals with theory on the factors controlling bioavailability (Di Toro et al., 1991, 2000).
EqP values represent concentrations of individual chemicals below which effects would not be
expected to occur. Individual EqP values cannot be directly compared with SQGs derived using
field-collected data (including Tp values) because the EqP values reflect the toxicity attributable
to individual chemicals, whereas the SQGs reflect the toxicity of the mixture. For example, the
Tp values associated with EqP values for individual PAHs are very high (Table 33), a result that
is consistent with a high degree of covariation among PAHs in sediments.
The EqP approach can be used to estimate the toxicity associated with a mixture of PAHs
using a toxic unit approach (Di Toro et al., 2000). Again, any comparison with empirical
52
-------
approaches is imperfect because the latter reflect the contribution of chemicals other than PAHs
to toxicity. The PAH toxic units were calculated on the basis of EPA's Final Chronic Value
following the methods described in EPA's National Sediment Quality Survey (U.S. EPA, 2004).
Most of the samples in the database had PAH toxic units less than 1. Both the P Max values and
the proportion of toxic samples in the database increased with increasing PAH toxic units (Table
34). Below a PAH toxic unit of 1, 37% of the samples were toxic. At PAH toxic units of 10 and
greater, 81% of the samples were toxic.
In summary, the LRMs provide a useful framework for evaluating the degree of risk
associated with commonly used SQGs. Logistic models have several advantages over current
SQGs approaches: (a) they present risk on a continuous quantitative scale rather than by
defining discrete categories based on threshold values, (b) the continuous estimates of risk allow
users to match the degree of risk with their objectives, (c) they express risk on a common scale
of 0 to 1 across all chemicals, and (d) they provide a more direct avenue for assessing risk of
multiple chemicals.
6.4. APPLICATION OF MODELS TO EVALUATIONS OF SITE-SPECIFIC OR
REGIONAL DATA
The models described in this report were derived from a large database of matching
sediment chemistry and toxicity that included data from many different coastal areas of North
America and many different chemical gradients. The models make predictions based on the
central tendency of the relationship between sediment chemistry and a toxic outcome across this
broad gradient; therefore, the models can be used to support screening-level assessments that
roughly rank or prioritize samples on the basis of sediment chemistry, particularly when
concentrations of chemicals vary over a wide range. For example, the National Sediment
Quality Survey (U.S. EPA, 2004) used these models to help classify locations into three tiers
reflecting the probability of adverse effects.
However, because the models do not consider potential differences in bioavailability, test
species, or site-specific mixtures of chemicals, the probability of toxicity may be over- or
underestimated for some locations. As discussed in Section 5.4.6, the differences between
quantitative model predictions and site- or region-specific observations can be substantial.
Although the derivation of site-specific models may be desirable, data from an individual site are
rarely sufficient to support model derivation. Rather than deriving site- or regional-specific
53
-------
models with a small data set, we recommend using site-specific data to determine how well the
nationwide models fit the local situation.
The evaluation of the independent data set from the Calcasieu Estuary provides an
example. By comparing the proportion of toxic samples or the mean control-adjusted survival
with the mean predicted probability of toxicity within discrete probability ranges (e.g.,
probability quartiles, as shown in Table 30, or probability intervals if sufficient data are available
[Figure 35]), the performance of the models with data from the site can be evaluated. In the best
case, the observed proportion of toxic samples will closely match that predicted by the model. In
this case, the models can be applied with confidence to samples having only sediment chemistry.
If the relationship between observed and predicted proportions of toxic samples is strong but
different, the relationship can still be used to predict site-specific toxicity. In addition, a strong
relationship between the percent survival and the predicted probability of toxicity provides
important information on the magnitude of the response. If the relationship between predicted
and observed proportions of toxic samples or percent survival is weak, then the models are likely
to be less useful for the site.
Although there are no universal criteria for determining whether a model fit is acceptable,
standard regression techniques and diagnostic plots can be used to identify whether specific
values are overly influencing the relationship or the assumptions of linear regression are not met
(Neter et al., 1996).
To best compare model predictions with site-specific observations, we recommend
collecting matching chemistry and toxicity testing results from samples that have a wide range of
chemical gradients and predicted probabilities of toxicity. It is particularly important to collect
sufficient data from areas of high and low probability of toxicity to better define the
relationships. Therefore, LRMs have great utility in helping to design sampling programs.
The models can also suggest issues that require further investigation. If the models
predict a higher proportion of toxic samples than the proportion observed (false positives), then
issues related to bioavailability may be investigated further. The individual chemical models
could be used to determine whether specific chemical models are associated with the high false
positive rate. If toxicity occurs at a much higher frequency than predicted (false negatives), then
it may be important to consider chemicals not accounted for (e.g., no models available) or issues
related to the sediment matrix (e.g., grain size effects).
54
-------
6.5. SUMMARY AND CONCLUSIONS
In this chapter we discussed the application of the single-chemical and multiple-chemical
LRMs to problems that are frequently encountered by risk assessors. We evaluated the
relationship between model predictions and the results of other toxicity endpoints, including
those commonly used in freshwater systems. The PMax model predictions appear to be useful
for predicting sea urchin response for A. pimctulata, based on development or fertilization tests,
but not S. pnrpiiraliis. The models are also useful for predicting the response of freshwater
amphipods, particularly the 28-day H. azteca growth and survival endpoint. There is the
potential for developing endpoint-specific models as more data are acquired.
The LRMs provide a useful framework for evaluating the degree of risk associated with
commonly used SQGs. The probabilities of toxicity associated with SQG threshold values are
generally consistent with their narrative intent. However, LRMs have several advantages over
current guideline approaches. They present risk on a continuous quantitative scale rather than by
defining discrete categories based on threshold values. The continuous estimates of risk allow
users to match the degree of risk with their objectives, and risk is expressed on a standardized
scale of 0 to 1 across all chemicals. The logistic models provide a direct yet flexible avenue for
assessing risk of multiple chemicals and provide the basis for more quantitatively evaluating the
reliability and fit of these models to observations.
Finally, we discussed how these models might best be used to evaluate the risks of
sediment contamination at specific sites or regions. The LRM approach can be used to conduct
screening-level assessments that roughly classify or prioritize samples on the basis of sediment
chemistry. Because the models do not consider potential differences in bioavailability or
exposure, the probability of toxicity may be over- or underestimated for some locations. For
applications that require a greater degree of accuracy (e.g., remediation decisions), we
recommend first evaluating how well the models fit the local situation by collecting a test set of
matching sediment chemistry and toxicity test data. The LRMs can be used to design effective
test sampling programs, and they can also suggest issues requiring further investigation (e.g.,
bioavailability). The LRMs may be most useful for classifying samples into broad categories of
concern on the basis of sediment chemistry.
55
-------
7. CONCLUSIONS AND RECOMMENDATIONS
7.1. CONCLUSIONS
A large database of matching sediment chemistry and toxicity data was carefully
evaluated and assembled. The database encompasses many different contaminant gradients from
a wide variety of habitats in coastal North America. Using this database, LRMs were developed
for 37 individual chemicals that describe relationships between the concentrations of sediment-
associated contaminants and acute toxicity to two commonly tested species of marine
amphipods: Rhepoxynius ahronius and Ampc/isca abdita.
The chemical-specific models that were derived in this investigation provide a basis for
estimating the proportion of samples expected to be toxic over a wide range of contaminant
concentrations for 37 individual contaminants. As such, these models help users select the
sediment effect concentrations that most directly meet the needs of their specific application.
For example, T10, T15, or T20 values could be calculated and used to identify concentrations for
individual contaminants that are likely to be associated with a relatively low incidence of
sediment toxicity (10, 15, or 20%, respectively). Such point estimates of minimal-effect
concentrations might be used in a screening assessment to identify sediments that are relatively
uncontaminated and have a low probability of sediment toxicity. Similarly, contaminant
concentrations for which there is a high probability of observing adverse effects could be
estimated by calculating T70, T80, or T90 values. These higher point estimates could be used to
identify sediments that are highly likely to be toxic to amphipods and have a greater magnitude
of effect (i.e., higher percent mortality).
The Tp values can be used in much the same way as other sediment guidelines, with the
difference that the Tp value is associated with a specific probability of observing toxicity and an
estimate of variance based on the fit of the model. Although the LRMs do not represent dose-
response relationships for individual chemicals, they can be considered indicators of toxicity
based on field-collected sediment chemical mixtures.
Because the individual models were derived from field-collected sediments that include
mixtures of contaminants rather than from individual dose-response relationships, to some extent
they incorporate the overall toxicity of the mixtures. Combining the individual model results
into a single explanatory variable provided a way to estimate the probability that a particular
56
-------
sample will be toxic. Two combined explanatory values (PMax and PAvg) accurately
predicted the frequency of toxicity to amphipods observed in the database:
• P Max is the maximum probability of observing toxicity, taken from the set of
probabilities calculated for each individual chemical in the sample, and
• P Avg is the mean probability of observing toxicity, based on the set of probabilities
calculated for each individual chemical in the sample.
The logistic regression approach served as a framework for evaluating several issues that
form the basis for our recommendations for using the models. We used the single-chemical
models to evaluate two approaches for designating samples as toxic: (1) less than 90% survival
that was significantly different from negative control samples (Sig Only), and (2) control-
normalized survival less than 80% that was significantly different from negative control samples
(MSD) (based on analyses by Thursby et al., 1997). Although the Sig Only approach had a
greater tendency to underestimate the toxicity observed at low concentrations, this discrepancy
may be explained by the presence of other chemicals in the sample. The MSD approach had a
greater tendency to overestimate the toxicity observed at higher concentrations.
We also evaluated two approaches for normalizing sediment chemistry: dry-weight
normalized and organic carbon normalized. The models based on organic carbon-normalized
sediment chemistry had lower goodness-of-fit statistics than the dry-weight-normalized models
and larger differences between observed and predicted proportions of toxic samples.
We used the single-chemical models to evaluate three screening criteria used to decide
whether a given sample was included in the model data set for an individual chemical: (1)
include all samples (unscreened), (2) exclude toxic samples that were less than or equal to the
mean of nontoxic samples from the same study (IX screening), and (3) exclude toxic samples
that were less than or equal to two times the mean of nontoxic samples from the same study (2X
screening). The models developed using IX screening showed much stronger relationships
between chemistry and toxicity than was observed when all samples were included. The IX
screening approach performed slightly better than the 2X screening approach at concentrations
above the T80 value. In addition, the 1X approach screened out considerably fewer toxic
samples in the model derivation than did the 2X approach, which may prove important in the
future development of models for less frequently measured chemicals. We selected the IX
57
-------
screening approach to limit the many permutations with further model development and
exploration.
The multiple-chemical models were used as a framework for evaluating several
additional issues, including the relationship between the probability of observing a toxic effect
and the magnitude of toxicity, the identification of chemicals most influential in model
performance, the performance of the models in predicting toxicity of the two amphipod species,
and the performance of the models in predicting toxicity observed in regional data sets or in
individual studies.
The magnitude of the effect (decreased survival) in the amphipod test increased as the
probability of toxicity increased, demonstrating that samples that are estimated to have the
highest probability of toxicity are also likely to be extremely toxic.
For approximately 70% of the samples, individual chemical regression models for metals
produced the maximum probability used in the PMax model. This should not be construed to
imply that metals were causing toxicity in these samples, only that metals appear to be a good
predictor of toxicity in field-collected samples. Indeed, removing metals (or other entire
chemical classes) from the suite of individual chemical models used to generate the P Max
model resulted in only minor changes in the model and model fit.
Models were developed by combining data from tests that used two species of marine
amphipod (R. abroiiins or A. abdiia) in order to encompass more areas of the country and a
broader range of sediment chemistry. The P Max model provided a useful basis for examining
differences in model performance for the two species. The observed toxicity was frequently less
than predicted for A. abdita and greater than predicted for R. ahroiiins. Nevertheless, the
observed proportion of toxicity in the data for both species was strongly related to the nationwide
model, affording confidence that the combined-species P Max model provides a common
framework applicable to both species.
We examined the performance of models in predicting observed toxicity for individual
studies within the database. On a study-by-study basis, there was mixed agreement between the
frequency of observed toxicity and that predicted by the P Max model. The mixed performance
suggests that the national models should not be applied to individual studies without first
evaluating their performance with matching site-specific toxicity and chemistry data. However,
the nationwide P Max model provided a useful common basis for evaluating toxicity test results
for individual sites included in the database. Application of the model to regional subsets of the
58
-------
database used to derive the model demonstrated significant relationships between the PMax
model predictions and both the observed proportion of toxic samples and percent control-
adjusted survival. There was also a strong relationship between the predicted and observed
proportion of toxic samples in the Calcasieu Estuary, a data set not included in the original
derivation of the models.
7.2. RECOMMENDATIONS
Our analyses and model comparisons resulted in several recommendations. As a starting
point for most evaluations, we recommend using the P Max model, which is derived from the
highest predicted probability from any of the individual chemical models. We recommend using
the model based on data from all studies (i.e., the nationwide model), on both marine amphipod
species, and on 37 chemical-specific models. For the chemical-specific models, we recommend
the models that classified samples as toxic on the basis of less than 90% survival that was
significantly different from negative control samples (Sig Only) and that screened the data set by
excluding toxic samples that were less than or equal to the mean of nontoxic samples from the
same study (1X screening). The bases for these recommendations are summarized briefly below.
The P Max model, which is based on the highest predicted probability from any of the
individual chemical models, explained slightly more variation in the data set than did the PAvg
model. The two models provide slightly different insights into sediment toxicity. P Avg may
better reflect the overall degree of contamination, and it is less susceptible to overestimating the
probability of toxicity at sites with high concentrations of one chemical. PMax more accurately
predicted toxicity at sites having high concentrations of more than one chemical.
We recommend using the nationwide model that combines data for both species of
marine amphipods. Combining data across studies and species represents the fullest range of
chemical concentrations and environmental conditions. In addition, the nationwide combined
model provides a common framework for comparing site- and species-specific results.
We recommend basing the models on the Sig Only classification so that more subtle
changes can be retained, particularly at lower concentrations. It may be valuable to compare the
Sig Only nationwide model to site-specific data classified using the MSD approach when test
variability obscures the relationship between chemistry and response at lower concentrations.
Finally, the chemical-specific models were greatly improved by using a screened data set
that excluded toxic samples that were less than or equal to the mean of nontoxic samples from
59
-------
the same study. Using a more stringent criterion, such as excluding toxic samples that were less
than or equal to two times the mean of nontoxic samples from the same study, resulted in
additional improvements for most chemicals. However, the more stringent criterion excluded an
average of 70% of the toxic samples, which may limit future development of models for other
endpoints, regions, or chemicals that have fewer total samples. We concluded that the small
improvements in model fit did not outweigh the associated reduction in sample size.
7.3. APPLICATIONS
The chemical-specific models provide a basis for estimating the probability that a sample
will be toxic for 37 individual contaminants over a wide range of contaminant concentrations. In
addition, they provide a useful framework for evaluating the degree of risk associated with
commonly used SQGs. The probabilities of toxicity associated with SQG threshold values are
generally consistent with their narrative intent. However, LRMs have several advantages over
current guideline approaches: they present risk on a continuous quantitative scale rather than by
defining discrete categories based on threshold values, the continuous estimates of risk allow
users to match the degree of risk with their objectives, and they express risk on a common scale
of 0 to 1 across all chemicals. The individual chemical models would be expected to
underestimate the probability of observing toxicity in samples that are contaminated with many
chemicals. For this purpose, we recommend using the multiple-chemical models that combine
the individual model results into a single explanatory value for estimating the probability that a
sample will be toxic.
The multiple-chemical models provide a useful framework for conducting screening-
level assessments that require classifying or prioritizing samples on the basis of sediment
chemistry. Because the models do not consider potential differences in bioavailability or
exposure, the probability of toxicity may be over- or underestimated for some locations. Before
applying the models to a particular site, we recommend first evaluating how well the models fit
the local situation by collecting a test set of matching sediment chemistry and toxicity test data.
The LRMs can be used to design effective test sampling programs, and they can also suggest
issues that require further investigation (e.g., bioavailability). The LRMs should not be
considered a complete substitute for direct-effects assessment (e.g., toxicity tests).
We evaluated the relationship between model predictions and the results of other toxicity
endpoints, including those commonly used in freshwater systems. The P Max model predictions
60
-------
appear to be useful for predicting sea urchin response for A. punctulata, based on development or
fertilization tests, but not for S. pnrpiiraliis. The models may also be useful for predicting the
response of freshwater amphipods, particularly the 28-day H. aztcca growth and survival
endpoint. There is the potential for developing endpoint-specific models as more data are
acquired.
7.4. FUTURE DIRECTIONS
The results of this study suggest many promising avenues for future work. First and
foremost, this research provides evidence of the value of combining and standardizing
information from many different studies. Future efforts directed at encouraging investigators to
add data to the SEDTOX02 database will enable the investigation of additional species, test
endpoints, and chemicals. Providing funds and mechanisms to share the database with the
scientific community will advance our knowledge of the effects of sediment chemicals on
aquatic organisms.
We hope that the release of the SEDTOX02 database will prompt continuing refinement
and exploration of modeling approaches linking toxicity test results with sediment chemistry.
Possibilities include optimizing the screening approach, evaluating the minimum number of
chemicals that need to be included for acceptable performance of the multiple-chemical models,
exploring alternative multivariate and logistic modeling approaches, and investigating the
reasons for variable model performance in individual studies.
Additional guidance on applying these models to site-specific assessments would
increase their use and foster consistent application. Guidance development would be aided by
testing the models with additional independent data sets and developing case studies that
illustrate their application.
We hope that this study will prompt the collection of more data sets that match toxicity
test results with sediment chemistry. Additional data from freshwater systems are especially
needed. The most useful data would be broad-scale surveys that include contaminated sites.
Sediments should be analyzed for the full range of chemical classes and tested using high-
quality, consistent toxicity test methods.
The results of this study suggest that additional work is needed to characterize the
bioavailability of chemicals to sediment-dwelling organisms. Contrary to expectation,
normalizing chemical concentrations to TOC did not improve model fits. As additional data are
61
-------
acquired, we may be able to evaluate the effectiveness of acid volatile sulfides as a normalizing
factor for metal concentrations.
Finally, the toxicity endpoint modeled in this project serves as a surrogate for valued
ecological attributes that are more difficult to test and measure. These include the structure and
function of benthic communities, population viability of wildlife that depends on the benthos,
and ecosystem functions such as organic matter decomposition and water filtration. Additional
research is needed to improve the linkages between these valued endpoints, toxicity test results,
and sediment chemistry.
62
-------
Table 1. Number of samples and percent toxic samples summarized by
marine amphipod species and data source"
Data
source
Ampelisca abdita
Rhepoxynius abronlus
Number of
samples
Percent toxic
Number of
samples
Percent toxic
Sig Only
MSD
Sig Only
MSD
EMAP
1203
22.2
9.5
NA
NA
NA
NSTP
649
23 J
15.6
NA
NA
NA
MLML
43
1 1.6
7
465
72.7
52.3
SEDQUAL
NA
NA
NA
594
63.5
34
BEDS
117
41
30.8
152
36.8
32.2
Total
2012
23.6
12.6
1211
63.7
40.8
'Samples were classified as toxic if significantly different from control and less than 90% survival (Sig
Only) and if significantly different from control and less than 80% control-adjusted survival (MSD).
BEDS = Biological Effects Database for Sediments. MacDonald Environmental Sciences
EMAP = U.S. Environmental Protection Agency Estuarinc Monitoring and Assessment Project
MLML = Moss Landing Marine Laboratory (California)
NSTP = National Oceanic and Atmospheric Administration Status and Trends Program
SEDQUAL = Sediment Quality Information System. Washington State Department of Ecology
N A = no data
63
-------
Table 2. Distribution of chemical concentrations in sediment samples with
matching toxicity data for marine amphipods
Chemical
Number of
samples
Percentile
10'"
5<),h
90th
Metals fmg kg dry wt.)
Antimony
2173
0.2
0.7
2.9
Arsenic
2844
2.2
7.4
19
Cadmium
2958
0.05
0.3
1.9
Chromium
2827
9.1
50
130
Copper
3091
2.6
26
160
Lead
3010
5.5
23
130
Mercury
2788
0.02
0.1
0.8
Nickel
2916
2.4
19
44
Silver
2552
0.03
0.2
1.9
Zinc
3013
16
89
300
Polycyclic aromatic hydrocarbons fug kg dry wt.)
1-Mcthylnaphthalcnc
1677
0.5
6.5
60
1 -Mcthy lphcnanthrcnc
1697
0.3
11
130
2.6-Dimcthylnaphthalcnc
1505
0.4
6.3
67
2-Mcthylnaphthalcnc
2077
0.8
12
130
Accnaphthcnc
1795
0.2
7
130
Accnaphthylcnc
1747
0.2
7
120
Anthracene
2268
0.5
20
410
Bcn/(a)anthraccne
2574
1.2
40
760
Bcn/o(a)pyrcnc
2526
1.2
53
910
Bcn/o(b)fluoranthcnc
1645
0.9
48
1000
Bcn/o(g.h.I)pcrylcnc
2210
1.1
46
550
Bcn/o(k)fluoranthcnc
1691
0.5
26
620
Bi phenyl
1507
0.5
6.8
54
Chryscnc
2650
1.7
53
1000
Dibcn/(a.h)anthraccnc
1886
0.2
14
170
Fluoranthcnc
2734
->
81
1400
Fluorcnc
2011
0.4
11
160
Indcno( 1.2.3-c.d)pyrcnc
2212
0.9
47
600
Naphthalene
2201
1.8
16
220
Pcrylcnc
2174
1.5
36
370
Phcnanthrcnc
2688
1.7
44
660
Pyrcnc
2768
3.2
87
1500
Polychlorinated hiphenyls fug kg dry wt.)
PCBs. total
1989
2
39
640
Organochlorine pesticides fug kg dry wt.)
Dicldrin
770
0.04
0.8
5.1
p.p'-DDD
1672
0.08
1.8
20
p.p'-DDE
1899
0.08
2.2
59
p.p'-DDT
1176
0.06
1.2
18
64
-------
Table 3. Number of samples and percent toxic samples summarized by sea
urchin species and test endpoint
Urchin species
E
•evelopment
Fertilization
Number of
samples
Percent toxic
Number of
samples
Percent toxic
Sig Only
IMSD
Sig Only
IMSD
A. punc In lata
472
65
64.6
612
40.5
37.1
S. purpuralus
310
61.6
58.1
212
62.3
61.8
Total
782
63.7
62
824
46.1
43.5
65
-------
Table 4. Distribution of chemical concentrations in sediment samples with
matching sea urchin fertilization toxicity data
Chemical
Number of
samples
Percentile
10,h
50th
90th
Metals (mg kg dry wt.)
Arsenic
786
2.1
7.6
18
Cadmium
724
0.03
0.2
1.5
Chromium, total
788
5.9
43
110
Copper
788
2
21
150
Lead
781
3.4
23
120
Mercury
709
0.02
0.1
0.6
Nickel
768
1.3
14
28
Silver
622
0.03
0.3
1.5
Zinc
789
9.4
82
290
Polycyclic aromatic hydrocarbons
(jug kg dry wt.)
2-Methylnaphthalene
572
0.9
9.3
64
Acenaphthene
492
0.3
5.7
110
Acenaphthylene
C "> ">
J
0.3
6.5
69
Anthracene
642
0.7
16
340
Benz(a)anthracene
718
1.5
840
Benzo(a)pyrene
723
1.5
51
1030
Chrysene
749
2
46
1160
Dibenz(a,h)anthracene
595
0.3
12
150
Fluoranthene
770
3.9
81
1660
Fluorene
574
0.5
5.1
110
Naphthalene
604
2
9.8
84
Phenanthrene
736
2
27
600
Pyrene
754
4.7
86
1660
Polychlorinated hiphenyls (jug kg dry wt.)
PCBs, total
622
2
36
290
Organochlorine pesticides (jug kg dry wt.)
DDT, total of six isomers
590
0.6
6.9
82
p,p'- DDE
566
0.2
3.2
40
66
-------
Table 5. Distribution of chemical concentrations in sediment samples with
matching sea urchin developmental toxicity data
Chemical
Number of
samples
Percentile
10,h
50th
90th
Metals (mg kg dry wt.)
Arsenic
680
2.5
8.4
18
Cadmium
635
0.04
0.2
1.3
Chromium, total
689
6.1
46
140
Copper
689
2.2
25
150
Lead
682
3.2
23
120
Mercury
703
0.02
0.2
0.7
Nickel
671
1.2
15
31
Silver
573
0.03
0.3
1.6
Zinc
694
9.6
94
280
Polycyclic aromatic hydrocarbons
(jug kg dry wt.)
2-Methylnaphthalene
C "> ">
0.9
9.5
79
Acenaphthene
478
0.3
5.3
91
Acenaphthylene
525
0.3
6.7
60
Anthracene
588
0.6
15
320
Benz(a)anthracene
653
1.3
36
840
Benzo(a)pyrene
655
1.2
54
980
Chrysene
668
1.7
50
1100
Dibenz(a,h)anthracene
578
0.3
13
180
Fluoranthene
683
3.1
81
1450
Fluorene
537
0.4
5.3
100
Naphthalene
576
1.9
9.5
84
Phenanthrene
664
1.6
30
530
Pyrene
662
->
86
1460
Polychlorinated hiphenyls (jug kg dry wt.)
PCBs, total
637
2.2
-> ->
300
Organochlorine pesticides (jug kg dry wt.)
DDTS, total of six isomers
528
0.8
7.8
79
p,p'-DDE
520
0.2
3.4
43
67
-------
Table 6. Endpoint, number of samples, and percent toxic samples for
Chironomus spp. (C. tentans and C. riparius) and Hyalella aztecaK
Test species
Endpoint
Number of
samples
Percent toxic
Chironomus spp.
10-14-day survival
585
19.8
10-day growth and survival
286
37.8
H. azlcca
10-14-day survival
567
24.2
28-day survival
125
19.2
28-day growth and survival
125
38.4
'Samples were classified as toxic by the original investigator.
68
-------
Table 7. Distribution of chemical concentrations in sediment samples with
matching toxicity data for the H. azteca 10-14-day survival test
Chemical
Number of
samples
Percentile
10,h
50th
90th
Metals (mg kg dry wt.)
Arsenic
352
1.7
5.3
27.3
Cadmium
279
0.1
0.8
8.2
Chromium, total
350
9.5
31
111
Copper
396
8
32.7
207
Lead
325
8.4
35.8
362
Mercury
390
0.04
0.28
74.5
Nickel
299
6.1
15
50
Silver
146
0.04
0.2
4.2
Zinc
361
29.5
140
661
Polycyclic aromatic hydrocarbons
(jug kg dry wt.)
2-Methylnaphthalene
125
2.7
66.2
21,015
Acenaphthene
248
2.7
80
18,000
Acenaphthylene
201
2
53
1,828
Anthracene
309
8
172
12,000
Benz(a)anthracene
357
16
440
10,000
Benzo(a)pyrene
359
19
467
7,200
Chrysene
376
24
565
11,000
Dibenz(a,h)anthracene
199
3.7
101
1,156
Fluoranthene
406
37.7
950
16,802
Fluorene
276
6.65
111
14,000
Naphthalene
292
9.8
167
12,000
Phenanthrene
378
18
520
15,000
Pyrene
403
40
840
19,691
Polychlorinated hiphenyls (jug kg dry wt.)
PCBs, total
171
8.5
109
1,490
Organochlorine pesticides (jug kg dry wt.)
DDT, total of six isomers
110
2.2
17
143
p,p'-DDE
117
2
10
56
69
-------
Table 8. Distribution of chemical concentrations in sediment samples with
matching toxicity data for the C. tentans or C. riparius 10-14-day survival
test
Number of
Percentile
Chemical
samples
10,h
50th
90th
Metals (mg kg dry wt.)
Arsenic
392
2.2
7.5
28
Cadmium
266
0.3
1.4
5.9
Chromium, total
399
9
26
140
Copper
413
7
28
130
Lead
406
12
32
110
Mercury
388
0.04
0.2
1
Nickel
380
7.1
16
61
Silver
44
0.05
0.4
1.3
Zinc
403
-> ->
93
540
Polycyclic aromatic hydrocarbons
(jug kg dry wt.)
2-Methylnaphthalene
50
9.6
140
16,500
Acenaphthene
162
->
41
1,100
Acenaphthylene
155
4
49
1,230
Anthracene
205
14
120
2,460
Benz(a)anthracene
231
23
350
4,940
Benzo(a)pyrene
251
36
360
4,950
Chrysene
266
35
440
4,100
Dibenz(a,h)anthracene
157
6.9
120
920
Fluoranthene
275
59
640
7,510
Fluorene
185
8.4
69
1,800
Naphthalene
184
10
130
10,000
Phenanthrene
251
35
390
4,600
Pyrene
275
48
620
5,800
Polychlorinated hiphenyls (jug kg dry wt.)
PCBs, total
108
42
390
4,000
Organochlorine pesticides (jug kg dry wt.)
DDT, total of six isomers
10
18
76
3,570
p,p'-DDE
21
0.9
25
84
70
-------
Table 9. Distribution of chemical concentrations in sediment samples with
matching toxicity data for the H. azteca 28-day growth and survival test
Chemical
Number of
samples
Percentile
10,h
50th
90th
Metals (mg kg dry wt.)
Arsenic
57
3.9
24
93
Cadmium
71
0.5
2.4
12
Chromium, total
72
27
61
150
Copper
72
22
86
410
Lead
72
19
91
200
Mercury
68
0.1
0.3
0.8
Nickel
56
8.3
18
50
Silver
39
0.1
1
1
Zinc
72
94
270
940
Polycyclic aromatic hydrocarbons
(jug kg dry wt.)
2-Methylnaphthalene
54
16
150
1280
Acenaphthene
26
29
56
250
Acenaphthylene
44
10
46
280
Anthracene
48
10
120
1640
Benz(a)anthracene
88
12
99
690
Benzo(a)pyrene
75
20
190
1230
Chrysene
102
20
120
690
Dibenz(a,h)anthracene
22
10
48
110
Fluoranthene
103
26
190
1450
Fluorene
41
38
77
1530
Naphthalene
69
15
44
1080
Phenanthrene
92
16
95
1400
Pyrene
103
21
170
1480
Polychlorinated hiphenyls (jug kg dry wt.)
PCBs, total
45
250
2740
7700
Organochlorine pesticides (jug kg dry wt.)
DDT, total of six isomers
41
0.3
1.9
88
p,p'-DDE
45
0.1
1
84
71
-------
Table 10. Normalized chi-square values and number of samples for
individual chemicals based on Sig Only classification of toxic samples for the
screened marine amphipod database (logistic regression model parameters
are shown for models having normalized chi-square values greater than 0.15)
Number of
Chi-square
Chemical
samples
Intercept (B„)
Slope (B,)
value/N
Metals fmg kg dry wt.)
Antimony
1718
-0.9
2.41
0.25
Arsenic
2336
-4.14
3.17
0.17
Cadmium
2413
-0.34
2.51
0.3 1
Chromium
2399
-6.44
->
0.2
Copper
2580
-5.79
2.93
0.38
Lead
2481
-5.45
2.77
0.27
Mercury
2296
0.8
2.55
0.32
Nickel
2450
-4.61
2.77
0.18
Selenium
1655
0.07
Silver
2103
-0.11
1.97
0.25
Zinc
2516
-7.98
3.34
0.28
Polycyclic aromatic hydrocarbons
fug kg dry wt.)
1-Mcthylnaphthalcnc
1368
-4.14
2.1
0.24
1-Mcthylphcnanthrcnc
1401
-3.59
1.75
0.28
2.6-Dimcthylnaphthalcnc
1249
-4.05
1.9
0.2
2-Mcthylnaphthalcnc
1704
-3.76
1.78
0.25
Accnaphthcnc
1424
-3.62
1.75
0.33
Accnaphthylcnc
1447
-2.96
1.38
0.23
Anthracene
1823
-3.66
1.49
0.29
Bcn/.(a)anthraccnc
2099
-4.2
1.58
0.3
Bcn/o(a)pyrcnc
2053
-4.3
1.58
0.3
Bcn/o(b)fluoranthcnc
1348
-4.54
1.49
0.27
Bcn/o(g.h.i)pcrylcnc
1818
-4.28
1.59
0.25
Bcn/o(k)fluoranthcnc
1376
-4.28
1.57
0.29
Bi phenyl
1226
-4.11
2.21
0.26
Chryscnc
2126
-4.32
1.54
0.29
Dibcn/(a.h)anthraccnc
1546
-3.63
1.77
0.33
Fluoranthcnc
2189
-4.46
1.48
0.26
Fluorcnc
1668
-3.71
1.81
0.32
Indcno( 1.2.3-c.d)py rene
1837
-4.37
1.62
0.27
Naphthalene
1816
-3.78
1.62
0.24
Pen lene
1823
-4.68
1.76
0.22
Phenanthrcne
2173
-4.46
1.68
0.3
Pyrcnc
2240
-4.71
1.59
0.29
Polychlorinated hiphenyls fug kg dry wt.)
PCBs. total
1617
-3.46
1.35
0.27
Organochlorine pesticides fug kg dry wt.)
Dicldrin
633
-1.17
2.56
0.35
Gamma-hcxachlorocvclohcxanc
534
0.098
p.p'-DDD
1360
-1.9
1.49
0.27
p.p'-DDE
1552
-1.84
0.91
0.16
p.p'-DDT
931
-1.77
1.68
0.34
72
-------
Table 11. Logistic model point estimates of T20, T50, and T80 values (95%
confidence interval) for individual chemicals based on Sig Only classification
of toxic samples for the screened marine amphipod database11
Chemical
T20
T50
T80
Metals fmg kg dry wt.)
Antimony
0.63 (0.55-0.72)
2.4 (2-2.8)
8.9 (6.6-12)
Arsenic
7.4 (6.8-8.1)
20 (18-23)
56 (45-69)
Cadmium
0.38 (0.34-0.43)
1.4 (1.2-1.5)
4.9 (4-6)
Chromium, total
49 (44-53)
140 (130-160)
410(330-510)
Copper
32 (29-35)
94(86-100)
280(240-330)
Lead
30 (27-33)
94(84-100)
300 (240-360)
Mercury
0.14 (0.12-0.15)
0.48 (0.43-0.54)
1.7(1.4-2.1)
Nickel
15 (13-16)
47 (42-52)
150(120-190)
Silver
0.23 (0.19-0.26)
1.1 (0.98-1.3)
5.8 (4.4-7.6)
Zinc
94 (87-100)
240 (220-270)
640(540-750)
Polycyclic aromatic hydrocarbons
fug kg dry wt.)
1-Mcthylnaphthalcnc
21 (17-25)
94(73-120)
430 (280-670)
1 -Mcthy lphcnanthrcnc
18(15-23)
110(88-140)
700(450-1070)
2.6-Dimcthylnaphthalcnc
25(20-31)
130(96-180)
710(410-1230)
2-Mcthylnaphthalcnc
21 (18-26)
130 (100-160)
770 (510-1140)
Accnaphthcnc
19(15-24)
120(90-150)
710 (470-1090)
Accnaphthylcnc
14 (11-18)
140(100-190)
1420 (800-2520)
Anthracene
34 (27-42)
290(230-370)
2490(1630-3790)
Bcn/.(a)anthraccnc
61 (50-75)
470 (380-570)
3530 (2490-5020)
Bcn/.o(a)pyrcnc
69(57-85)
520 (430-630)
3910 (2750-5550)
Bcn/o(b)fluoranthcnc
130(100-170)
1110(810-1510)
9410 (5530-16000)
Bcn/o(g.h.i)perylene
67 (54-82)
500 (390-630)
3710(2440-5630)
Bcn/o(k)fluoranthcne
70 (55-90)
540 (410-710)
4120(2540-6680)
Bi phenyl
17 (14-21)
73(57-93)
310 (210-470)
Chryscnc
82 (67-99)
650 (530-800)
5190(3600-7480)
Dibcn/(a.h)anthraccne
19 (15-23)
110 (92-140)
690 (480-990)
Fluoranthcnc
120(98-150)
1030(830-1280)
8950 (6070-13200)
Fluorcnc
19(16-24)
110 (92-140)
660(460-950)
Indcno( 1.2.3-c.d)pyrcnc
68(56-84)
490(390-610)
3480 (2350-5160)
Naphthalene
30 (25-37)
220(170-280)
1570(1020-2410)
Pcrylcnc
74 (62-89)
450 (360-570)
2770 (1820—1210)
Phenanthrcne
68 (57-81)
460 (380-550)
3060 (2190—1260)
Pyrcnc
120(100-150)
930(770-1130)
6980(4940-9860)
Polychlorinated hiphenyls
fug kg dry wt.)
PCBs. total
35 (27—14)
370 (280-480)
3930 (2410-6390)
()rganochlorine pesticides
fug kg dry wt.)
Dicldrin
0.83 (0.65-1)
2.9 (2.3-3.6)
10 (6.9-15)
p.p'-DDD
2.2 (1.7-2.8)
19(14-25)
160(95-270)
p.p'-DDE
3.1 (2.2-4.4)
100 (61-180)
3410 (1280-9120)
p.p'-DDT
1.7 (1.3-2.2)
11 (8.3-15)
76 (45-130)
'The notation Tp (e.g.. T50) is used to denote the concentration that would give a response of "p" percent
according to the model (e.g.. the probability that 50% of the samples would be toxic).
73
-------
Table 12. Percent of toxic samples within ranges defined by Sig Only logistic
model T20, T50, and T80 values and number of samples used to derive the
logistic model for each chemical for the marine amphipod database11
Chemical
T8<)
Number of
samples
Metals (mg kg diy wtj
Antimony
103
48.5
67.9
82.5
2173
Arsenic
30
4.18
56.3
69.7
2844
Cadmium
27.6
50.9
62.7
78.7
2958
Chromium
24.5
42.7
55.7
80
2827
Copper
22.1
50.6
64.9
85
3091
I ,ead
28.5
45
60.6
90
3010
Mereurv
25.5
49
66.1
79.4
2788
Niekel'
253
44.7
60.4
NA
2916
Si h er
25.6
52.5
60.7
7.17
2552
Zinc
2.16
47.3
67.8
71.3
3013
I'olycyclic aromatic hydrocarbons
(ug kg dry ii7.)
1-Methylnaphthalene
23.7
48.3
60.3
75
1677
1 -Methy lphenanthrene
24.7
45.8
65.5
80
1697
2.6-1 )imethylnaphthalene
2.15
42.2
57.4
NA
1505
2-Methylnaphthalene
25.4
47.5
61.1
88
2077
Aeenaphthene
25.2
50.3
67.7
91.4
1795
Aeenaphthylene
24
44.7
67.6
NA
1747
Anthracene
26.5
48.8
66.9
77.1
2268
1ienz(a)anthraeene
28.5
45.3
65
82.9
2574
Men/.o(a)pyrene
27.7
48.5
64.2
8.18
2526
Ben/o(b) 11 uoranthene
24
46.3
67.4
NA
1645
Ben/o(g.h.i)perylene
253
46.6
6.16
86.7
2210
1ienzo(k) 11 uoranthene
25.5
44.2
68.3
9.13
1691
Miphenyl
25.2
46.3
56.7
8.13
1507
Chrysene
28.7
47.8
64.8
86.
2650
I )iben/(a.h)anthraeene
2.19
49
65.3
85.7
1886
I'luoranthene
28.4
47.1
64.9
87.9
2734
I'luorene
22.1
48.4
68.2
87.2
201 1
Indeno( 1,2.3-e.d)pyrene
25.6
44.5
64.9
90.5
2212
Naphthalene
26.4
4.18
64.1
89.7
2201
Perylene
26.9
39.4
59.8
NA
2174
Phenanthrene
28.2
47.3
64.1
85.7
2688
IVrene
28.3
46.2
65
87.2
2768
I'olychlorinated bipltenyls
tug kg dry m7. )
PC I is. total
26.8
46.3
72.7
81.5
1989
Chganochhrine pesticides
tug kg dry m7. )
Dieldrin
20.2
5.18
66.7
78.8
770
p.p'-l )l )l)
25.9
49.4
64.7
80.5
1672
p.p'-l )!)!•:
22.5
5.14
54.9
57.6
1899
p.p'-DDT
25.6
56.5
66.7
76.7
1 176
'The notation Tp (e.g.. T50) is used to denote the concentration that would give a response of "p" percent
according to the model (e.g.. the probability that 50% of the samples would be toxic).
NA = fewer than 10 samples
74
-------
Table 13. Normalized chi-square values and number of samples for
individual chemicals based on IMSD classification of toxic samples for the
screened marine amphipod database (logistic regression model parameters
are shown for models having normalized chi-square values greater than 0.15)
Number of
Slope
Chi-square
Chemical
samples
Intercept (B„)
(BO
value/N
Metals fmg kg dry wt.)
Antimony
1905
0.14
Arsenic
2551
0.11
Cadmium
2641
-1.17
2.27
0.2
Chromium, total
2609
-7.47
3.21
0.16
Copper
2803
-6.98
3.06
0.29
Lead
2718
-6.29
2.79
0.21
Mercury
2472
-0.06
2.68
0.25
Nickel
2653
0.13
Selenium
1822
0.05
Silver
2294
-0.85
2.03
0.19
Zinc
2742
-9.26
3.54
0.22
Polycyclic aromatic hydrocarbons
fug kg dry wt.)
1-Mcthylnaphthalcnc
1509
—1.94
2.24
0.19
1-Mcthylphcnanthrcnc
1523
-4.89
2.02
0.23
2.6-Dimcthylnaphthalcnc
1366
-5
2.15
0.17
2-Mcthylnaphthalcnc
1857
—1.37
1.72
0.17
Accnaphthcnc
1596
-4.29
1.7
0.23
Accnaphthylcnc
1570
-4.4
1.74
0.2
Anthracene
2000
-4.8
1.61
0.22
Bcn/.(a)anthraccnc
2264
-5.96
1.88
0.23
Bcn/o(a)pyrcnc
2212
-6.15
1.94
0.24
Bcn/o(b)fluoranthcnc
1451
-7.5
2.3
0.28
Bcn/o(g.h.i)pcrylcnc
1939
-6.55
2.12
0.22
Bcn/o(k)fluoranthcnc
1490
-6.63
2.18
0.27
Bi phenyl
1353
-4.53
2.1
0.18
Chryscnc
2307
-5.97
1.78
0.22
Dibcn/(a.h)anthraccnc
1677
-5.23
2.19
0.28
Fluoranthcnc
2373
-6.65
1.91
0.23
Fluorcnc
1817
-4.76
1.89
0.23
Indcno( 1.2.3-c.d)py rene
1962
-6.59
2.15
0.23
Naphthalene
1971
-4.82
1.72
0.18
Pen lene
1955
-6.29
2.15
0.19
Phcnanthrcnc
2335
-6.13
1.92
0.22
Pyrcnc
2404
-7.01
2.03
0.24
Polychlorinated hiphenyls fug kg dry wt.)
PCBs. total
1766
-4.41
1.48
0.24
Organochlorine pesticides fug kg dry wt.)
Dicldrin
682
-1.83
2.59
0.28
Gamma-Hcxachlorocvclohcxanc
573
0.08
p.p'-DDD
1480
-2.59
1.6
0.24
p.p'-DDE
1682
0.13
p.p'-DDT
1009
-2.51
1.64
0.27
75
-------
Table 14. Logistic model point estimates of T20, T50, and T80 values (95%
confidence interval) for individual chemicals based on IMSD classification of
toxic samples for the screened marine amphipod database11
Chemical
T20
T50
T80
Metals fmg kg dry wt.)
Cadmium
0.8 (0.71-0.91)
3.3 (2.7-4)
13 (9.8-18)
Chromium, total
78(72-85)
210(180-240)
570(440-730)
Copper
67 (61-74)
190(170-210)
540 (440-660)
Lead
57 (52-63)
180(160-210)
560(440-720)
Mercury
0.32 (0.29-0.36)
1.1 (0.91-1.2)
3.5 (2.7-4.4)
Silver
0.55 (0.47-0.63)
2.6(2.1-3.2)
13 (8.9-18)
Zinc
170(150-180)
410 (370-460)
1.020 (840-1.230)
Polycyclic aromatic hydrocarbons
fug kg dry wt.)
1-Mcthylnaphthalcnc
39 (32-47)
160 (120-220)
670 (410-1.090)
1-Mcthylphcnanthrcnc
54 (44-66)
260 (200-350)
1.270 (790-2.040)
2.6-Dimcthylnaphthalcnc
48(39-61)
210 (150-310)
950 (530-1.700)
2-Mcthylnaphthalcnc
55 (45-67)
350 (250-490)
2.260 (1.300-3.930)
Accnaphthcnc
51 (40-64)
330(230-460)
2.150 (1.220-3.770)
Accnaphthylcnc
54 (43-68)
330 (230-480)
2.080 (1.160-3.740)
Anthracene
130(100-160)
940(690-1280)
6.770 (4.040-11.300)
Bcn/.(a)anthraccnc
270(230-320)
1.460 (1.150-1.860)
7.980 (5.350-11.900)
Bcn/.o(a)pyrcnc
290(240-340)
1.510 (1.200-1.890)
7.840 (5.350-11.500)
Bcn/o(b)fluoranthcne
460 (370-560)
1.820 (1.420-2.340)
7.300 (4.860-11.000)
Bcn/o(g.h.i)pcrylcnc
270 (230-330)
1.240 (960-1.590)
5.580 (3.670-8.480)
Bcn/o(k)fluoranthcnc
250 (200-310)
1.080 (830-1.410)
4.660 (3.060-7.120)
Biphcnyl
3 1 (25-38)
140(100-200)
650 (380-1.130)
Chryscnc
370 (310-440)
2.200 (1.690-2.860)
13.200 (8.530-20.300)
Dibcn/(a.h)anthraccnc
57 (48-68)
250 (200-310)
1.070 (730-1.550)
Fluoranthcnc
570 (480-680)
3.050 (2.390-3.900)
16.200 (10.900-24.300)
Fluorcnc
61 (50-74)
330 (250-430)
1.770 (1.110-2.810)
Indcno( 1.2.3-c.d)pyrcnc
260 (220-310)
1.170 (920-1.480)
5.150 (3.480-7.630)
Naphthalene
100(81-120)
640 (460-890)
4.130 (2.430-7.020)
Pcrylcnc
190(160-230)
840 (650-1.090)
3.710 (2.420-5.680)
Phcnanthrcnc
290(250-350)
1.540 (1.200-1.980)
8.100 (5.370-12.200)
Pyrcnc
590 (500-700)
2.850 (2.270-3.570)
13.800 (9.480-19.900)
Polychlorinated hiphenyls
fug kg dry wt.)
PCBs. total
110(88-140)
940 (690-1.290)
8.120 (4.770-13.800)
()rganochlorine pesticides
fug kg dry wt.)
Dicldrin
1.5 (1.2-1.9)
5.1 (3.9-6.6)
18 (11-27)
p.p'-DDD
5.6(4.5-7.1)
41 (29-58)
300 (170-530)
p.p'-DDT
4.9 (3.6-6.5)
34(23-52)
240(120-470)
'The notation Tp (e.g.. T50) is used to denote the concentration that would give a response of "p" percent
according to the model (e.g.. the probability that 50% of the samples would be toxic).
76
-------
Table 15. Percent of toxic samples within ranges defined by logistic model
T20, T50, and T80 values and number of samples in the database used to
derive the logistic model for each chemical based on IMSD classification of
toxic samples for the marine amphipod database"
Chemical
T8<)
Number of
samples
Metals fmg kg dry wtj
Cadmium
17.4
39.9
61.5
69.2
2958
Chromium, total
15.4
37.7
54.1
75.9
2827
Copper
14.2
40.1
66.5
76.6
3091
Lead
16.6
37.8
63.9
61.5
3010
Mercury
16.7
44.4
56.4
80.8
2788
Silver
16.8
43.5
46.5
NA
2552
Zinc
14.6
39.9
65.8
58.3
3013
Polycyclic aromatic hydrocarbons
fug kg dry wt.)
1 -Mcthy 1 naphthalene
14.8
39.6
50
NA
1677
1 -Mcthy lphcnanthrcnc
15.5
40.7
63.6
50
1697
2.6-Dimcthylnaphthalcnc
14.5
51.7
NA
1505
2-Mcthylnaphthalcnc
17.1
37
55.6
90
2077
Accnaphthcnc
15.9
40.1
56.3
75
1795
Accnaphthylcnc
15.2
37.7
63.8
NA
1747
Anthracene
16.9
39.5
62.2
78.6
2268
Bcn/(a)anthraccnc
17.5
39.4
62.6
73.3
2574
Bcn/o(a)pyrcnc
18
40.4
63
81.8
2526
Bcn/o(b)fluoranthcnc
15.9
41.7
61.9
86.7
1645
Bcn/o(g.h.i)pcrylcnc
17.4
37.5
64.9
NA
2210
Bcn/o(k)fluoranthcnc
15.8
42
62.1
84.6
1691
Bcn/ofluoranthcncs. total
23.3
37.9
63.6
NA
617
Biphcnyl
16.3
36.5
50
NA
1507
Chryscnc
18.3
40.3
58.2
72.7
2650
Dibcn/(a.h)anthraccnc
16.5
38.4
69.7
68.4
1886
Fluoranthcnc
18.1
41.8
62.9
64.7
2734
Fluorcnc
14.8
38.4
63.2
69.6
2011
Indcno(1.2.3-c.d)pyrcnc
16.5
36.9
65.7
83.3
2212
Naphthalene
16.5
37.5
52.5
92.9
2201
Pcrylcnc
16.5
35.8
48.6
NA
2174
Phcnanthrcnc
18.4
40
63.4
69.2
2688
Pyrcnc
18.1
42.6
57.5
72.7
2768
I'olychlorinated hiphenyls
fug kg dry wt.)
PCBs. total
18.6
44.2
56.2
69.2
1989
()rganochlorine pesticides
fug kg dry wt.)
Dicldrin
17
45.5
60.9
72.7
770
p.p'-DDD
16.9
44.5
58
74.3
1672
p.p'-DDT
20.2
48.5
50.9
83.9
1176
'The notation Tp (e.g.. T50) is used to denote the concentration that would give a response of "p" percent
according to the model (e.g.. the probability that 50% of the samples would be toxic).
NA = fewer than 10 samples
77
-------
Table 16. Normalized chi-square values and number of samples for
individual chemicals for organic carbon-normalized concentrations for the
screened marine amphipod database (logistic regression model parameters
are shown for models having normalized chi-square values greater than 0.15)
Chemical
Number of
Intercept
Slope
Chi-square
(Hg/kg organic carbon)
samples
(Bo)
(B.)
value/N
Polycyclic aromatic hydrocarbons
1 -Methyl naphthalene
1291
-7
1.79
0.17
1 -Methylphenanthrene
1308
-7.36
1.88
0.27
2,6-Dimethylnaphthalene
1185
0.15
2-Methylnaphthalene
1573
-6.82
1.67
0.19
Acenaphthene
1329
-6.3
1.58
0.25
Acenaphthylene
1361
-5.55
1.4
0.2
Anthracene
1676
-6.86
1.57
0.26
Benz(a)anthracene
1921
-8.25
1.81
0.29
Benzo(a)pyrene
1897
-8.16
1.79
0.29
Benzo(b)fluoranthene
1245
-8.66
1.81
0.27
Benzo(g,h,i)perylene
1664
-7.73
1.68
0.22
Benzo(k)fluoranthene
1274
-7.92
1.75
0.27
Biphenyl
1157
-6.54
1.71
0.17
Chrysene
1974
-8.2
1.76
0.28
Dibenz(a,h)anthracene
1411
-7.3
1.86
0.3
Fluoranthene
2021
-8.15
1.67
0.25
Fluorene
1519
-7.31
1.84
0.28
Indeno( 1,2,3-c,d)pyrene
1685
-8.07
1.77
0.24
Naphthalene
1684
-6.27
1.44
0.16
Perylene
1709
-8.9
1.99
0.21
Phenanthrene
2005
-8.3
1.82
0.27
Pyrene
2092
-8.48
1.76
0.27
Po/ych/orinalcd biphenyls
PCBs, total
1530
-6.03
1.38
0.22
Organochlorine pesticides
Dieldrin
569
-4.33
1.85
0.24
Gamma-Hexachlorocyclohexane
523
0.07
p,p'-DDD
1261
-5.34
1.72
0.28
p,p'-DDE
1425
0.15
p,p'-DDT
854
-4.51
1.52
0.28
78
-------
Table 17. Logistic model point estimates of T20, T50, and T80 organic
carbon-normalized concentrations (95% confidence interval) for individual
chemicals based on Sig Only classification of toxic samples in the screened
marine amphipod database'1
Chemical
(^g/kg organic carbon)
T20
T50
T80
I'olycyclic aromatic hydrocarbons
1-Mcthylnaphthalcnc 1
1-Mcthylphcnanthrcnc 1
2-Mcthylnaphthalcnc 1
Accnaphthcnc 1
Accnaphthylcnc
Anthracene 3
Bcn/(a)anthraccnc 6
Bcn/o(a)pyrcnc 6
Bcn/o(b)fluoranthcnc 10
Bcn/o(g.h.i)pcrylene 6
Bcn/o(k)fluoranthcnc 5
Biphenyl 1
Chrysene 7.
Dibcn/(a.h)anthraccnc 1
Fluoranthcnc 11
Fluorcnc 1
Indeno( 1.2.3-c.d)pyrene 6
Naphthalene 2
Pcrylcnc 6
Phcnanthrcnc 6
Pvrcnc 10
I'olychlorinalecl hiphenyls
PCBs. total
.390 (1.120-1.730) 8.310 (5.870-11.800) 49.600 (27500-89.600)
.510 (1.220-1.850) 8.230 (6.430-10.500) 45.000 (29.200-69.200)
.800 (1.480-2.200) 12.200 (9.110-16.400) 82.600 (49.500-138.000)
.260 (990-1.610) 9.470 (6.900-13.000) 71.100 (41.600-122.000)
920 (710-1.190) 8.960 (6.320-12.700) 87.100 (46.900-162.000)
.010 (2.430-3.740) 22.900 (17.600-29.700) 174.000 (111.000-273.000)
.060 (5.070-7.250) 35.200 (29.000-42.700) 204.000 (146.000-287.000)
.230 (5.190-7.480) 37.200 (30.700-45.200) 223.000 (159.000-3 13.000)
.200 (8.150-12.900) 59.500 (45.200-78.300) 346.000 (217.000-550.000)
.000 (4.920-7.320) 40.200 (31.100-5.1900) 269.000 (171.000—+24.000)
.490 (4.360-6.910) 34.100 (25.800-45.000) 212.000 (132.000-338.000)
.020 (810-1.290) 6.580 (4.570-9.470) 42.400 (22.500-79.800)
380 (6.160-8.830) 45.200 (37.100-55.000) 277.000 (196.000-390.000)
.490 (1.210-1.830) 8.250 (6.610-10.300) 45.700 (31.000-67.400)
.200 (9.370-13.500) 76.000 (61.100-94.500) 514.000 (350.000-753.000)
.640 (1.350-2.000) 9.300 (7.320-11.800) 52.700 (35.000-79.200)
.050 (5.000-7.330) 36.900 (29.300-46.500) 225.000 (150.000-338.000)
.470 (1.990-3.070) 22.700 (16.200-31.700) 208.000 (114.000-378.000)
.000 (5.080-7.070) 29.800 (23.800-37.400) 149.000 (99.400-222.000)
.220 (5.250-7.360) 35.800 (29.400-43.800) 207.000 (146.000-293.000)
.500 (8.870-12.500) 64.300 (53.000-77.900) 392.000 (280.000-551.000)
2.290 (1.790-2.930) 23.100 (17.500-30.600) 233.000 (138.000-394.000)
( )rganochlorine pesticides
Dicldrin
p.p'-DDD
p.p'-DDT
39 (29-53)
200 (160-240)
110(85-160)
220(150-310)
1240 (950-1.630)
940 (650-1.380)
1.230 (650-2.320)
7.910 (4.930-12.700)
7.740 (4.040-14.900)
'The notation Tp (e.g.. T50) is used to denote the concentration that would give a response of "p" percent
according to the model (e.g.. the probability that 50% of the samples would be toxic).
79
-------
Table 18. Percent of toxic samples within ranges defined by logistic model
T20, T50, and T80 values for organic carbon-normalized concentrations and
number of samples in the database used to derive the logistic model for each
chemical based on Sig Only classification of toxic samples for the marine
amphipod database"
Chemical
Number of
(jig/kg organic carbon)
T80
samples
Polycyclic aromatic hydrocarbon
V
1 -Methyl naphthalene
26.3
47
55.3
NA
1612
1 -Methylphenanthrene
27
48.7
65.5
60
1641
2-Methylnaphthalene
29.8
48.2
57.8
62.5
2008
Acenaphthene
27.2
54.7
68.2
76.2
1743
Acenaphthylene
24.7
51.1
58.3
NA
1684
Anthracene
29.8
52.6
64.2
72
2189
Benz(a)anthracene
31
51.8
61.4
75
2489
Benzo(a)pyrene
29.3
53.7
61.2
78.3
2443
Benzo(b)fluoranthene
24.9
52.7
65
NA
1571
Benzo(g,h,i)perylene
21A
52.1
61.4
NA
2135
Benzo(k)fluoranthene
26.4
50.9
67.1
78.6
1621
Biphenyl
28
47.1
43.6
NA
1453
Chrysene
30.2
53.4
61.2
50
2564
Dibenz(a,h)anthracene
24.8
54.7
68.3
77.4
1815
Fluoranthene
30.1
53.2
61.9
78.3
2644
Fluorene
26
52.2
66.9
82.1
1936
Indeno( 1,2,3-c,d)pyrene
27.8
49.7
62.2
78.6
2135
Naphthalene
29.5
48
56.9
NA
2123
Perylene
28.4
45
50
72.7
2100
Phenanthrene
30.5
51.3
63.6
NA
2606
Pyrene
29.5
51.9
61.5
74.4
2678
Po/ych/orinalcd biphenyls
PCBs, total
27.2
51.5
64.3
77.8
1940
Organochlorine pesticides
Dieldrin
21.8
58.3
74.7
46.2
729
p,p'-DDD
27.8
54.9
61.2
70.5
1623
p,p'-DDT
28.6
59.2
61.4
78.8
1131
'The notation Tp (e.g.. T50) is used to denote the concentration that would give a response of "p" percent
according to the model (e.g.. the probability that 50% of the samples would be toxic).
NA = fewer than 10 samples
80
-------
Table 19. Differences between percent predicted toxic samples and percent
observed toxic samples (observed minus predicted) within ranges defined by
logistic model T20, T50, and T80 values based on Sig Only classification of
toxic samples for the marine amphipod databasea'b
Chemical
T8<)
Metals fmg kg dry wt.)
Antimony
19.2
17.4
6.1
-9.9
Arsenic
19.6
12.5
-3.6
-19.7
Cadmium
19.2
17.3
0.9
-12
Chromium, total
15.7
11.8
-3.4
-7.7
Copper
15.3
17.7
2
-2.7
Lead
19.1
14.2
-> ->
3.9
Mercury
17.5
16.2
2.4
-7.2
Nickel
16.5
11.4
-1.7
U 2
Silver
15.7
19.4
-2.7
-11.6
Zinc
15.1
14.7
5.8
-16.2
Polycyclic aromatic hydrocarbons
fug kg dry wt.)
1-Mcthylnaphthalcnc
17.6
16.3
-2.7
-11.5
1-Mcthylphcnanthrcnc
17.5
12.8
4.3
-7.8
2.6-Dimcthylnaphthalcne
17.3
10.9
-2
2-Mcthylnaphthalcnc
17.4
15.3
-0.3
-0.5
Accnaphthcnc
19.6
16.8
5
2.1
Accnaphthylcnc
16.4
11.5
7.5
-12
Anthracene
19.1
15.7
5.3
-10.1
Bcn/.(a)anthraccnc
:i
12
2.7
-3.2
Bcn/.o(a)pyrcnc
:u 1
14.5
1.1
-1.5
Bcn/o(b)fluoranthcnc
17.6
13.4
4.2
11
Bcn/o(g.h.i)pcrylcne
17.7
13.4
1.6
0.8
Bcn/o(k)fluoranthcnc
18.9
11.7
5.8
5.2
Biphcnyl
19.2
14.3
-4.3
-7
Chrysene
2o
14.1
1.9
-0.2
Diben/(a.h)anthraccne
r 4
15.9
2.7
-0.9
Fluoranthcnc
2(1 5
14.3
3.1
0.9
Fluorene
15.7
14.5
5.7
-2.3
Indcno( 1.2.3-c.d)pyrene
18.2
11.2
2.8
4
Naphthalene
17.9
11.6
2.3
0.9
Pen lene
19.2
7.5
-0.3
42
Phenanthrcne
2d 5
13.5
2.3
-2.8
Pvrcnc
2u (>
13.2
2.7
-0.2
Polychlorinated hiphenyls fug kg dry wt.)
PCBs. total
17.1
14.1
11.9
-9.5
Organochlorine pesticides fug kg dry wt.)
Dicldrin
12.9
21 -
4.2
-7.8
p.p'-DDD
16.9
1 (. <
2.7
-11.2
p.p'-DDE
11.3
-5.1
2'> S
p.p'-DDT
17.3
25 1
2
-16.4
'The notation Tp (e.g.. T50) is used to denote the concentration that would give a response of "p" percent
according to the model (e.g.. the probability that 50% of the samples would be toxic).
''Differences greater than 20% arc shaded. Blank cell indicates fewer than six samples.
81
-------
Table 20. Differences between percent predicted toxic samples and percent
observed toxic samples (observed minus predicted) within ranges defined by
logistic model T20, T50, and T80 values based on IMSD classification of toxic
samples for the marine amphipod databasea'b
Chemical
T8()
Metals fmg kg dry wt.)
Cadmium
10.5
8.8
0.3
:i:
Chromium, total
7
7.9
-8.6
11:
Copper
8.8
7.5
3.4
11:
Lead
9.8
5.8
2.4
:s i
Mercury
10.8
13.2
-5.9
(»^
Silver
9.6
11.4
-12.2
Zinc
8
9.1
2.9
}(i
Polycyciic aromatic hydrocarbons
fug kg dry wt.)
1 -Mcthy 1 naphthalene
9.9
9.5
-11.2
1-Mcthylphcnanthrcnc
10.1
9.5
2.7
'X (.
2.6-Dimcthylnaphthalcnc
9.9
1.4
-7
2-Mcthylnaphthalcnc
10.5
6
-4.3
2.5
Accnaphthcnc
10.7
10.1
-3.7
-14.3
Accnaphthylcnc
10
6.3
3.4
:n 4
Anthracene
11.6
8.5
1.9
-9.5
Bcn/(a)anthraccnc
12.6
7.3
1.3
-14.1
Bcn/o(a)pyrcnc
12.8
8.4
0.8
-4.5
Bcn/o(b)fluoranthcnc
12.2
9
-2.6
0
Bcn/o(g.h.i)pcrylcnc
12.5
6.8
5.2
Bcn/o(k)fluoranthcnc
12
9.4
0.1
-5.7
Biphcnyl
10.6
5.5
-9.9
-13.1
Chryscnc
13.1
9.3
-1.4
-15.3
Dibcn/(a.h)anthraccnc
11.5
6.6
6.4
-19.8
Fluoranthcnc
13.1
10.5
2.1
24 4
Fluorcnc
9.4
7.8
2
:<> S
Indcno(1.2.3-c.d)pyrcnc
11.7
6
4.2
-4.8
Naphthalene
10.6
6
-8.5
3.2
Pcrylcnc
11.5
3.9
-8.5
Phcnanthrcnc
13.3
9.2
2.6
-19.7
Pyrcnc
13.3
10.4
-3.4
-15
Polychlorinatedhiphenyls fug kg dry wt.)
PCBs. total
10.1
13
-6.6
-19.4
Organochlorine pesticides fug kg dry wt.)
Dicldrin
10.3
12.5
-1.9
-13
p.p'-DDD
9.7
14.4
-16.6
p.p'-DDT
13.3
18
-9.2
-8.2
'The notation Tp (e.g.. T50) is used to denote the concentration that would give a response of "p" percent
according to the model (e.g.. the probability that 50% of the samples would be toxic).
''Differences greater than 20% arc shaded. Blank cell indicates fewer than six samples.
82
-------
Table 21. Differences between mean percent predicted toxic samples and
percent observed toxic samples (observed minus predicted) within ranges
defined by logistic model T20, T50, and T80 values for organic carbon-
normalized concentrations based on Sig Only classification of toxic samples
for the marine amphipod databasea'b
Chemical
(jig/kg organic carbon)
T8<)
Polycyclic aromatic hydrocarbons
1 -Mcthy 1 naphthalene
18.7
16.4
-2.8
4< 5
1-Mcthvlphcnanthrcnc
19.7
17
2.7
2" 1
2-Mcthvlnaphthalcnc
2o
1 2
-1.2
U 4
Anthracene
2.1
-14.9
Bcn/(a)anthraccnc
-1
-11.5
Bcn/o(a)pyrcnc
-1.6
-6.7
Bcn/o(b)fluoranthcnc
1.1
-3.7
Bcn/o(g.h.i)pcrylcnc
1
Bcn/o(k)fluoranthcnc
19.8
19.1
5.2
-7.3
Biphcnyl
19.6
15.7
-15.3
'5 4
Chryscnc
22
2d S
-0.9
-8.8
Dibcn/(a.h)anthraccnc
IS 2
22 S
5.8
-6.1
Fluoranthcnc
22 1
21 ^
0.4
-4.9
Fluorcnc
l<> 1
19.8
5.4
-11.6
Indcno( 1.2.3-c.d)pyrcnc
2d 5
18
1.4
2d 2
Naphthalene
l<> 4
16.1
-2.9
-13.3
Pcrylcnc
2d 5
12.5
-8.9
Phcnanthrcnc
19
1.6
-14.6
Pvrcnc
21 -
19.5
-0.9
-14
I'olychlorinated hiphenyls
PCBs. total
17.1
19.1
4.5
-10.9
( )rganoc:hlorine pesticides
Dicldrin
11.7
25 '
13.5
41 5
p.p'-DDD
19.7
22 (.
0.3
2d 4
p.p'-DDT
19.6
28 (.
-1.6
-12.1
'The notation Tp (e.g.. T50) is used to denote the concentration that would give a response of "p" percent
according to the model (e.g.. the probability that 50% of the samples would be toxic).
''Differences greater than 20% arc shaded. Blank cell indicates fewer than six samples.
-------
Table 22. Differences between mean percent predicted toxic samples and
percent observed toxic samples (observed minus predicted) within ranges
defined by logistic model T20, T50, and T80 values for each chemical using a
screening factor of 2X the mean of nontoxic samples and Sig Only
classification of toxic samples for the marine amphipod databasea'b
Chemical
T8<)
Metals (mg kg diy wtj
Antimony
15.7
14 6
6.1
-1 1.8
Arsenic
16.3
9.9
-5.1
-22 7
Cadmium
16.3
14.7
-2.6
-13.1
Chromium, total
12
9.5
-4
-7.7
Copper
12.1
14 6
0.8
-4.1
I .ead
16
1 1.4
-4.8
2.6
Mereurv
14.2
13.4
0.5
-7.2
Niekel'
12.7
9
-3.4
-34 2
Silver
12.2
17.4
-4.5
-1 1.6
Zinc
1 1.9
1 1.8
4.1
-16.2
I'olycyclic aromatic hydrocarbons
(ug kg chy wtj
1-Methylnaphthalene
14.4
14.1
-2.7
-19.8
1 -Methy lphenanthrene
14
10
3.7
-7.8
2.6-1 )imethylnaphthalene
14.4
9.8
-5
2-Methylnaphthalene
14.5
12.6
-3
-4.5
Aeenaphthene
16.5
13.5
3.2
2.1
Aeenaphthylene
13.3
8.3
6.2
-12
Anthracene
16.4
12.2
4.5
-10.1
Ben/(a)anthracene
18
9
1.6
-3.2
Men/.o(a)pyrene
17.1
1 1.7
-0.3
-1.5
Men/.o(b)fluoranthene
14.5
12.1
4.2
11
1ienzo(g.h. I )pery lene
14.5
10.9
0.8
0.8
Men/.o(k)fluoranthene
15.6
10.3
5.8
5.2
Biphenyl
16.1
12.3
-7.4
-7
Chrysene
18
10.9
1.3
-0.2
I )iben/(a.h)anthraeene
14.6
12.9
1.5
-0.9
I'luoranthene
17.4
1 1.5
1.9
0.9
I'luorene
12.9
10.5
4.8
-2.3
lndeno( 1.2.3-c.d)pyrene
15.1
8.6
2.3
4
Naphthalene
15.1
8.6
-0.7
-2.5
Perylene
16.4
5.4
-2.1
-42 9
I'henanthrene
17.6
10.6
0.5
-2.8
I'yrene
17.7
10.1
1.9
-0.2
I'olvchlorinated biphenvls (ug kg drv wtj
PCBs. total
14.2
1 1.9
10.6
-21 3
Organochlorine pesticides (ug kg diy wtj
Dieldrin
10.6
19.4
1.7
-10.8
p.p'-l )l )l)
14.2
14.3
2.7
-1 1.2
p.p'-I )!)!•:
8.1
20 5
-6.3
-29 8
p.p-I)I)T
13.2
22 7
1.1
-16.4
'The notation Tp (e.g.. T50) is used to denote the concentration that would give a response of "p" percent
according to the model (e.g.. the probability that 50% of the samples would be toxic).
''Differences greater than 20% arc shaded. Blank cell indicates fewer than six samples.
84
-------
Table 23. Statistical comparisons of the logistic regression models for
A. abdita and R. abronius using the chi-square statistic (-2 log likelihood)
Common slope,
Common slope,
Separate slope,
common
separate
separate
Preferred
Chemical
intercept
intercept
intercept
model
Metals fmg kg diy wtj
Antimony
548.39
506.3
430.25
a
Arsenic
582.09
579.8
404.54
b
Cadmium
899.71
872.55
754.76
a
Chromium, total
645.99
618.91
468.43
a
Copper
1062.54
1061.94
987.62
b
I ,ead
928.02
91 1.34
679.52
a
Mercury
890.64
890.64
734.98
b
Nickel'
614.01
586.86
441.04
a
Si h er
75.184
753.56
530.67
b
Zinc
91.132
894.37
702.21
a
I'olycyclic aromatic hydrocarbons
fug kg dry ii7.)
1-Methylnaphthalene
357.46
350.53
327.48
a
1 -Methy lphenanthrene
487.05
486.26
397.74
b
2.6-1 )imethylnaphthalene
291.27
290.06
251.51
b
2-Methylnaphthalene
559.79
559.72
426.77
b
Aeenaphthene
496.32
487.75
475.79
a
Aeenaphthylene
381.16
380.26
332.24
b
Anthracene
599.92
599.88
526.43
b
1ienz(a)anthraeene
800.12
799.74
624.63
b
Hen/.o(a)pyrene
779.46
779.43
614.03
b
Henzo(b) 11 uoranthene
415.05
410.87
358.83
a
Ben/o(g-h.i)pery lene
614.51
61 1.4
454.16
b
Henz.o(k) 11 uoranthene
438.18
428.92
393.22
a
Biphenyl
352.24
346.47
322.79
a
Chrysene
775.03
775
608.87
b
I )ibenz(a.h)anthraeene
550.44
550.43
503.33
b
I;1 uoranthene
760.93
760.37
576.24
b
I'luorene
588.33
586.25
538.94
b
Indeno( 1.2.3-c.d)pyrene
619.06
617.96
494.56
b
Naphthalene
553.13
547.8
426.97
a
Perylene
535.85
526.83
398.32
a
Phenanthrene
830.04
824.66
647.82
a
I\rene
842.54
841.84
642.07
b
I'olvchloriiiated bipltenyls
-------
Table 24. Ratio of Tp values from species-specific A. abdita models to
corresponding Tp values from the combined amphipod models (only models
with normalized chi-square >0.15 are included)11
Chemical
T20
T50
T80
Metals
Antimony
1.46
1.04
0.74
Cadmium
1.71
1.4
1.15
Chromium, total
1.48
1.19
0.96
Copper
1.42
1.49
1.56
Lead
1.82
1.49
1.21
Mercury
1.77
1.92
2.09
Silver
2.63
2.66
2.68
Zinc
1.59
1.37
1.18
Polycyclic aromatic hydrocarbons
1 -Mcthy 1 naphthalene
1.21
1.29
1.38
1-Mcthylphcnanthrcnc
1.96
2.09
2.23
2,6-D i met hy 1 naphtha lc nc
1.32
1.26
1.21
2-Mcthy 1 naphthalene
2.11
2.09
2.07
Accnaphthcnc
1.24
1.72
2.4
Accnaphthylcnc
1.74
2.25
2.9
Anthracene
2.06
2.6
3.28
Bcn/.(a)anthraccnc
2.82
2.8
2.78
Bcn/.o(a)pyrcnc
2.71
2.91
3.13
Bcn/o(b)fluoranthcnc
1.92
2.01
2.11
Bcn/o(g.h.i)pcrylcne
2.48
2.46
2.44
Bcn/o(k)fluoranthcne
1.58
1.42
1.27
Bi phenyl
1.21
1.44
1.72
Chrysene
2.85
3.25
3.71
Diben/(a.h)anthraccne
1.64
1.97
2.37
Fluoranthene
3.13
3.21
3.29
Fluorene
1.53
2.05
2.74
Indcno( 1.2.3-c.d)pyrene
2.13
2.16
2.19
Naphthalene
2.05
1.83
1.63
Pen lene
1.86
1.42
1.08
Phenanthrcne
2.75
2.44
2.17
Pyrene
3.03
3.04
3.05
I'olychlorinated hiphenyls
PCBs. total
1.9
2.6
3.56
()rganoc:hlorine pesticides
Dicldrin
1.12
1.12
1.12
p.p'-DDD
1.83
1.3
0.92
p.p'-DDE
1.46
0.33
0.08
p.p'-DDT
1.36
1.07
0.85
MEAN
1.91
1.91
1.98
The notation Tp (e.g.. T50) is used to denote the concentration that would give a response of "p" percent
according to the model (e.g.. the probability that 50% of the samples would be toxic).
86
-------
Table 25. Ratio of Tp values from the combined amphipod models to
corresponding Tp values from the species-specific R. abronius models (only
models with normalized chi-square >0.15 are included)11
Chemical
T20
T50
T80
Metals
Cadmium
3.42
1.82
0.97
Copper
1.73
1.5
1.3
Lead
3.2
2.34
1.71
Mercury
2.32
2.13
1.97
Silver
3.37
->
2.67
Zinc
2.67
1.85
1.29
Polycyclic aromatic hydrocarbons
1 -Mcthy 1 naphthalene
1.68
3.39
6.82
1-Mcthylphcnanthrcnc
6.14
4.12
2.77
2.6-Dimcthylnaphthalcne
3.5
5.88
9.89
2-Mcthy 1 naphthalene
3.97
4.3
4.66
Accnaphthcnc
1.05
1.69
2.73
Accnaphthylcnc
3.18
3.72
4.36
Anthracene
3.44
2.89
2.43
Bcn/.(a)anthraccnc
4.48
3.88
3.36
Bcn/.o(a)pyrcnc
4.24
3.77
3.35
Bi phenyl
1.58
2.63
4.36
Chrysene
4.54
3.79
3.17
Diben/(a.h)anthraccne
2.72
2.2
1.78
Fluoranthcnc
5.1
4.06
3.22
Fluorene
2.07
2.27
2.49
Indeno( 1.2.3-c.d)pyrene
5.83
4.19
3.01
Naphthalene
8.52
4.57
2.45
Perylene
8.93
4.9
2.69
Phenanthrcne
5.48
3.49
2.22
Pyrene
4.77
3.89
3.17
()rganochlorine pesticides
Dicldrin
2.24
1.68
1.26
MEAN
3.85
3.23
3.08
The notation Tp (e.g.. T50) is used to denote the concentration that would give a response of "p" percent
according to the model (e.g.. the probability that 50% of the samples would be toxic).
87
-------
Table 26. Differences between mean percent predicted toxic samples and
percent observed toxic samples (observed minus predicted) for A abdita and
R. abronius within ranges defined by T20, T50, and T80 valuesa'b
Chemical
1. abdita
R. abronius
T80
T80
tchds (nig kg diy wt.j
Antimony
9
-1
—3
-19
54
34
8
-6
Arsenic
12
-10
-30
41
37
17
-29
Cadmium
8
-9
-7
1
47
6
-14
Chromium, total
7
-5
-7
3
41
34
5
-17
Copper
10
-9
-11
-7
37
31
15
-2
1 ,ead
7
-6
-15
6
47
31
13
4
Mercury
9
—3
-17
-14
43
19
4
Nickel'
9
-5
-8
-28
46
28
4
-14
Silver
7
-9
-16
-10
40
40
15
-5
/. inc
8
-7
-7
-13
39
34
20
-6
Polycyclic aromatic hydrocarbons
f/jg kg(by u l.j
1 -Methy^naphthalene
12
10
-6
-1
56
41
1 -Methy'lphenanthrene
10
1
-1 1
-12
51
34
21
9
2.6-Dimethy^naphthalene
12
1
-1
57
44
2-Methylnaphthalene
10
—3
-13
—3
45
40
25
0
Acenaphthene
15
2
-7
-8
45
14
13
Acenaphthylene
11
3
2
-
42
32
20
Anthracene
11
-1
-4
-18
46
13
2
Ben/(a)anthracene
9
-9
-9
-6
47
34
15
6
Ben/o(a)pyrene
10
-8
-1 1
-1
47
39
13
1
Ben/o( b )tl uoranthene
9
-1
2
-1
38
8
I Jen/.o(g.h.i )pery lene
9
-6
—3
-7
52
40
9
Ben/o( k )tl uoranthene
10
3
0
10
34
12
3
Biphenyl
13
6
-17
-17
59
46
22
15
Chrysene
8
-7
-7
-9
49
37
1 1
6
1 )iben/( a.h )anthracene
1 1
-4
-2
-1
48
36
6
2
ITuoranthene
9
-6
-7
-17
48
13
12
ITuorene
1 1
0
-4
-13
44
34
15
4
lndeno( 1,2.3-c.d)p> rene
9
-8
-2
0
49
40
9
4
Naphthalene
8
0
—3
-8
50
1 1
6
Pen lene
8
0
-8
-15
32
1 1
Phenanthrene
9
-2
-15
-8
48
31
18
-1
IVrene
9
-7
-11
-7
47
14
6
I'olychlorinak'd biplwnyls
kg diy \rl.j
PCIJs. total
8
—3
1
-16
52
29
24
4
C))^anocliloriiu' pesticides
kg diy \rl.j
Dieldrin
9
14
4
-26
48
44
15
6
p.p'-DDI)
8
0
2
-1
34
8
-16
p.p'-DDI •:
6
8
34
S')
-2
p.p'-DDI'
13
5
-5
1
50
45
12
-21
"Hie notation Tp (e.g.. T50) is used to denote the concentration that would give a response of "p" percent according to
the model (e.g.. the probability that 50% of the samples w ould be toxic).
''Percent predicted toxicity and Tp values were calculated using the combined species logistic model using the Sig
Only classification of toxicity and the screened marine amphipod database. Differences greater than 20% are shaded.
Blank cell indicates fewer than six samples.
88
-------
Table 27. Changes in logistic model point estimates of T20, T50, and T80
concentrations for PCBs based on Sig Only classification of toxic samples
based on corrected PCB model for the screened marine amphipod database11
Estimated
T20
T50
T80
Original
32.5
468
6750
Corrected
34.5
368
3930
'The notation Tp (e.g.. T50) is used to denote the concentration that would give a response of "p" percent
according to the model (e.g.. the probability that 50% of the samples would be toxic).
89
-------
Table 28. Estimated probability of toxicity from marine amphipod chemical-
specific logistic regression models for LC50 values (dry wt.) reported from
10-day spiked sediment amphipod toxicity tests
Chemical
LC50
Probability of
toxicity
Source
Cadmium (mg/kg)
9.81
8.8-10
8.2-11.5
6.9
0.9
0.88-0.9
0.88-0.91
0.85
Mearns et al., 1986
Kemp et al., 1986
Robinson et al., 1988
Swartz et al., 1985
Mercury (mg/kg)
13.1
0.97
Swartz et al., 1988
Zinc (mg/kg)
276
0.54
Swartz et al., 1988
Fluoranthene (mg/kg)
4.2
3.3-10.5
0.71
0.68-0.82
Swartz et al., 1988
Swartz et al., 1987
Phenanthrene (mg/kg)
3.68
0.82
Swartz et al., 1989
Total PCBs (|_ig/kg)
8.8
0.87
Swartz et al., 1988
p,p'-DDT (jxg/g)
11.2-125
0.5-0.85
Word et al., 1987
PCBs = polychlorinatcd biphcnyls
90
-------
Table 29. Number and percent of samples by chemical class that represented
the maximum probability of toxicity used in the P IMax model derived from
the marine amphipod database
Chemical class
Number
Percent
Metals
2234
69.3
PAHs
596
18.5
Pesticides-PCBs
393
12.2
PAHs = polycyclic aromatic hydrocarbons
PCBs = polychlorinatcd biphcnyls
91
-------
Table 30. Differences between mean percent predicted toxic samples and
percent observed toxic samples (observed minus predicted) by probability
quartile for individual studies with at least 20 samples"
Study
Number of
Probability
quartile
samples
<25
25-50
50-75
>75
.1. abc/ilo
New Bedford Harbor Monitoring. 1993
77
-20
-18
19
NSTP Hudson-Raritan Phase I. 1991
34
-10
4
NSTP Long Island Sound. 1994
63
35
}()
NSTP Boston Harbor. 1994
30
14
25
REMAP-Hudson/Raritan Bay. 1993
41
U
2()
REMAP-Long Island Sound. 1993
43
-7
-13
REMAP-Hudson/Raritan Bay. 1994
42
-6
-15
-4
REMAP-Long Island Sound. 1994
42
-14
48
EMAP-Dclawarc Bay. 1990
42
18
-5
EMAP-Chcsapcakc Bay. 1990
61
->
-19
-6
EMAP-Chcsapcakc Bay. 1991
62
9
12
EMAP-Chcsapcakc Bay. 1992
59
r
l<>
5<>
EMAP-Chcsapcakc Bay. 1993
62
'1
NSTP Charleston Harbor. 1993
79
IS
^(i
55
NSTP Savannah River. 1994
60
<>
M)
-
NSTP Biscaync Bay. Phase I. 1995
105
s
1
18
NSTP Biscaync Bay. Phase II. 1996
120
15
EMAP Virginia and N. Carolina. 1994
50
1 1
'1
EMAP Virginia and N. Carolina. 1995
50
l<>
NSTP Choctawhatchcc. 1994
21
l<>
-
NSTP St. Andrews Bay. 1993
31
l<>
4()
NSTP Tampa Bay Phase II. 1992
45
-35
58
R. abronius
Port of Tacoma. Blair Waterway
21
14
> /
Commencement Bay Remedial Invest.
50
l<>
Elliott Bay sediment survey. 1985
97
15
5
-
Pugct Sound Eight-Bay survey. 1985
48
'1
Everett Harbor. 1985
29
lu
Port of Tacoma Remedial Investigation
79
53
28
:i
Pugct Sound Ambient Monitoring. 1989
50
-8
-6
Pugct Sound Ambient Monitoring. 1990
50
49
(>4
Pugct Sound Ambient Monitoring. 1991
47
-5
2')
BPTCP. 1992 Q3. LA
58
->
2
BPTCP. Screening. 1992 Q4. San Diego
23
(• 1
^(.
BPTCP. Screening. 1993 Q2-3. San Diego
78
15
S
BPTCP. Screening. 1994 Ql. LA
45
(>()
35
Bptcp. Screening. 1994 Ql. Santa Ana
24
4<
Bptcp. Screening. 1994 Ql. San Diego
93
(i~
48
2?
EMAP So.CA. 1994 Q3. San Diego
25
> >
41
Palos Vcrdcs shelf and Santa Monica Bay
31
V)
8
(Swart/ct al.. 1991)
'Differences greater than 20% arc shaded. Blank cell indicates fewer than six samples.
92
-------
Table 31. List of studies (primary data source) combined for analysis of
broader geographic areas
Geographic area
Number of samples
Hudson-Raritan/Long Island Sound (NSTP and Regional EMAP)
280
Virginian Province (EMAP)
489
Southeastern US (EMAP Carolinian and NSTP)
636
Puget Sound (SEDQUAL)
594
California (MLML)
508
93
-------
Table 32. Percent of samples predicted to be toxic to amphipods at the
chemical concentrations defined by sediment quality guidelines
Chemical
ERL
ERM
TEL
PEL
AET
Metals
Antimony
NA
NA
NA
NA
99
Arsenic
22
85
20
73
99
Cadmium
46
89
32
77
93
Chromium (total)
-> ->
78
22
54
94
Copper
21
79
11
54
97
Lead
30
73
20
55
96
Mercury
22
60
19
60
85
Nickel
28
53
22
48
92
Silver
47
73
41
59
81
Zinc
-> ->
68
27
54
98
Polycyclic aromatic hydrocarbons
2-Mcthylnaphthalcnc
39
78
19
59
89
Accnaphthcnc
18
75
10
45
90
Accnaphthylcnc
-> ->
71
13
49
79
Anthracene
31
70
24
47
92
Bcn/.(a)anthraccnc
40
70
22
57
84
Bcn/.o(a)pyrcnc
47
68
23
57
79
Ben/o(g.h.I)perylene
NA
NA
NA
NA
78
Chryscnc
41
73
23
54
91
Diben/(rt,/?)anthraccne
39
66
10
53
90
Fluoranthene
41
74
19
56
90
Fluorene
20
77
21
55
94
Indcno( 1.2.3-c.d)py rene
NA
NA
NA
NA
83
Naphthalene
45
83
22
60
84
Phenanthrcne
39
70
23
53
94
Pyrene
44
67
22
57
88
I'olychlorinated biphemis
PCBs. total
16
40
16
40
78
()rganochlorine pesticides
Dicldrin
NA
NA
18
61
55
p.p'-DDD
NA
NA
15
36
69
p.p'-DDE
18
37
17
62
45
p.p'-DDT
NA
NA
16
35
91
AET = Apparent effect threshold foramphipod survival (Grics and Waldo. 1996)
ERL = Effect-range low (Long and MacDonald. 1992)
ERM = Effect-range median (Long and MacDonald. 1992)
PEL = Probable effect level (MacDonald ct al.. 1996)
TEL = Threshold effect level (MacDonald ct al.. 1996)
NA = No guideline value available
94
-------
Table 33. Percent of samples predicted to be toxic to amphipods at the
chemical concentrations defined by the Final Chronic Value for individual
polycyclic aromatic hydrocarbons
Final chronic value
Chemical
ppb dry weight,
assuming 1% OC
Percent predicted toxic
Acenaphthene
4,910
95
Acenaphthylene
4,520
89
Anthracene
5,940
88
Benz(a)anthracene
8,410
88
Benzo(a)pyrene
9,650
88
Benzo(b)fluoranthene
9,790
80
Benzo(k)fluoranthene
9,810
88
Benzo(g,h,i)perylene
10,950
89
Chrysene
8,440
85
Dibenz(a,h)anthracene
11,230
97
Fluoranthene
7,070
77
Fluorene
5,380
95
Indeno( 1,2,3-c,d)pyrene
11,150
90
Naphthalene
3,850
88
Perylene
9,670
91
Phenanthrene
5,960
87
Pyrene
6,970
80
95
-------
Table 34. Percent of samples predicted and observed to be toxic to
amphipods within ranges defined by toxic units for polycyclic aromatic
hydrocarbons (PAHs)
PAH toxic unit
Percent toxic
Number of
samples
Range
Average
Mean predicted
Observed
<1
0.09
36
37.8
2823
1-2
1.41
57
54.4
182
2-3
2.49
62
58.1
86
3-5
3.86
63
61.7
60
5-10
6.61
68
66.7
42
10-100
20.41
77
81
21
96
-------
• •
• •
Study
Figure 1. Box plots (plotted as per Tukey, 1977) summarizing the
distribution of total organic carbon (TOC) (log 10) values for each study with
greater than 20 samples from the marine amphipod database. Studies are
ordered from left to right by mean TOC. The top and bottom of each rectangular
box correspond to the upper and lower quartiles of the data, respectively; the dot
within each box corresponds to the median of the data.
97
-------
Antimony
10 10 10 10 10
Concentration (mg/kg)
Chromium
~T~
• .
4
/• ^
I V
/•
V? •
, ••
uxV
10 10 10 10 10'
Concentration (mg/kg)
Arsenic
~T~
•
J
jt% •
\/s
%•
:
10 10 10
Concentration (mg/kg)
Copper
10 10 10 10'
Concentration (mg/kg)
Cadmium
10 10 10 10
Concentration (mg/kg)
; •
W
'J •
| •
ji
* •
:•
10 10 10
Concentration (mg/kg)
Figure 2. Logistic regression models and proportion of toxic samples in
concentration intervals in screened marine amphipod database for 37
chemicals based on Sig Only classification of toxic samples. The individual
points correspond to the median of the sample concentrations within the interval
and the proportion of the samples that are toxic within the interval.
98
-------
Mercury
• S«
*
±J*
r i»
9L*.
V
7
10 10 10 10 10
Concentration (mg/kg)
Zinc
10 10 10 1C
Concentration (mg/kg)
Nickel
10 10 10 10 10'
Concentration (mg/kg)
1 -M ethylnaphthalene
••
10 10 10 10 1C
Concentration (ng/g)
Silver
10 10 10 10 10 10
Concentration (mg/kg)
1-Methylphenanthrene
O
£
10 10 10 10 10 1C
Concentration (ng/g)
Figure 2. (continued)
99
-------
2-M ethylnaphthalene
Acenaphthylene
f •
•/
; •J
' •/ •
•' I
:V
10 10 10 10 10 10'
Concentration (ng/g)
o
Q.
I I
r : • ,
•
• /
/•
; • /
:
:• Wm '¦
•
T*r i i
10 10 10 10 10 10' 10
Concentration (ng/g)
2,6-Dim ethylnaphthalene
Anthracene
10 10 10 10 10 10'
Concentration (ng/g)
10 10 10 10 10 10' 10
Concentration (ng/g)
Acenaphthene
Benz(a)anthracene
10 10 10 10 10 10'
Concentration (ng/g)
o
Q.
! I
•• .
<
| ; i
7
\T
i %
•
•
•
Mi
10 10 10 10 10 10' 10
Concentration (ng/g)
Figure 2. (continued)
100
-------
Benzo(a)pyrene
Benzo(g,h,i)perylene
••
•
a 4'
.. .
: •/
r •
Jf#
•
rl
* •
10 10 10 10 10' 10
Concentration (ng/g)
10 10 10 10 10' 10
Concentration (ng/g)
Benzo(b)fluoranthene
Biphenyl
O 04
: • / '¦
* V* '¦
• 7
vfs
; •
YJ
R
10 10 10 10 10' 10 10
Concentration (ng/g)
10 10 10 10 10 10' 10
Concentration (ng/g)
Benzo(k)fluoranthene
Chrysene
10 10 10 10 10' 10 10
Concentration (ng/g)
o
Q.
•
•
4
•
•
> T
•J
• J*
/• •
•
•
V%T"
• t *
10 10 10 10 10' 10
Concentration (ng/g)
Figure 2. (continued)
I0l
-------
Dibenz(a,h)anthracene
10 10 10 10 10 10' 10
Concentration (ng/g)
lndeno(1,2,3-c,d)pyrene
O 04
• 7
.«
%
/•
• *1
•
,/J/v
• u
IV
10 10 10 10 10' 10
Concentration (ng/g)
Fluoranthene
•
•
•7 •
• i
f •
« •
•
¦
10 10 10 10' 10 10
Concentration (ng/g)
Naphthalene
10 10 10 10 10 10' 10
Concentration (ng/g)
Fluorene
T"
~T~
P •
• •;
•
•A
t
10 10 10 10 10 10' 10
Concentration (ng/g)
Perylene
10 10 10 10 10' 10
Concentration (ng/g)
Figure 2. (continued)
102
-------
Phenanthrene
p,p'-DDD
10 10 10 10 10 10' 10
Concentration (ng/g)
o 04
•
• /
v/.;.
%[ •
• f
•
M •
10 10 10 10 10 10 10' 10
Concentration (ng/g)
Pyrene
"I I V
p,p'-DDE
• ,
•w •
Ik •
- /•
T */ ¦
«
fw
%
10 10 10 10 10' 10 10
Concentration (ng/g)
10 10 10 10 10 10 10' 10 10
Concentration (ng/g)
Dieldrin
p,p'-DDT
10 10 10 10 10 10
Concentration (ng/g)
10 10 10 10 10 10 10'
Concentration (ng/g)
Figure 2. (continued)
103
-------
Total PCBs
1
••
0 8
H 0 6
0 A
0 2
— U I L L
10 10 10 10: 10
Concentration (ng/g)
o
10
10
Figure 2. (continued).
104
-------
Cadmium
Silver
10 10 10 10 10
Concentration (mg/kg)
10 10 10 10
Concentration (mg/kg)
Chromium
Copper
10 10 10 10'
Concentration (mg/kg)
10 10 10 10 10'
Concentration (mg/kg)
Mercury
Lead
10 10 10 10
Concentration (mg/kg)
10 10 10
Concentration (mg/kg)
Figure 3. Logistic regression models and proportion of toxic samples in
concentration intervals in the screened marine amphipod database for 33
chemicals based on IMSD classification of toxic samples. The individual points
correspond to the median of the sample concentrations within the interval and the
proportion of the samples that are toxic within the interval.
105
-------
Zinc
2-Methylnaphthalene
10 10 10 10
Concentration (mg/kg)
1 -Methylnaphthalene
10 10 10 10 10
Concentration (ng/g)
1 -Methylphenanthrene
10 10 10 10 10 10
Concentration (ng/g)
0 8-
o
X
o
c
o
r
o
Q.
O
0 6-
• /
•
•/ #
'*/
4*
?
•
•
*/4
10 10 10 10 10 10
Concentration (ng/g)
2,6-Dimethylnaphthalene
10 10 10 10 10
Concentration (ng/g)
Anthracene
o
X
o
c
o
r
o
Q.
O
0 4
0 2
i I
1
.m.
•
•
•
7 ^
jf4
y.. m...: —
, •
' i i
10
10 10 10 10
Concentration (ng/g)
1 o
Figure 3. (continued)
106
-------
Acenaphthene
Benzo(a)py rene
10 10 10 10 10 10' 10
Concentration (ng/g)
o 04
j
"•y
/ •
• /
¦*
• <
»
<
J?'
Kfr,
10 10 10 10 10' 10 1C
Concentration (ng/g)
Acenaphthylene
Benzo(b)fluoranthene
O 04
•
• 7
!
*
10 10 10 10' 10
Concentration (ng/g)
•
7%
7'
••¦J
•7
• /:
>*
r •
••
«
/ •
10 10 10'
Concentration (ng/g)
Benz(a)anthracene
Biphenyl
«
•/
T
•>
•
• /
/•
• ¥
•
•
10 10 10 10' 10
Concentration (ng/g)
10 10 10 1C
Concentration (ng/g)
Figure 3. (continued)
107
-------
Benzo(k)fluoranthene
9
•/
• /
•
*•/
/•
•
7<
'
" •J' #
10 10 10 10'
Concentration (ng/g)
Dibenz(a,h)anthracene
1 r
10 10 10 10 10' 10
Concentration (ng/g)
Benzo(g,h,i)perylene
Fluoranthene
i ./•
•• /•
* /
>
••r
•
•vi/'
10 10 10 10 10' 10
Concentration (ng/g)
10 10 10 10' 10 10
Concentration (ng/g)
Chrysene
Naphthalene
•
• /
•
•j
• 7
w •
•
•mj
• %
I
• •V#
10 10 10 10 10' 10
Concentration (ng/g)
• .y
• 7 •
A
%/
•
w/ »
10 10 10 10' 10
Concentration (ng/g)
Figure 3. (continued)
108
-------
Fluorene
10 10 10 10 10' 10
Concentration (ng/g)
Phenanthrene
"I i I r
•
•4
•«
<
j L
10 10 10 10 101 10 10
Concentration (ng/g)
lndeno(1,2,3-c,d)pyrene
Pyrene
% /
/•
/
' •%
» .
AV
/ V
• .
10 10 10' 10
Concentration (ng/g)
10 10 10 10' 10 10
Concentration (ng/g)
Perylene
p,p'-DDT
• J '¦
• L ^
• % i
«#•/•....:
10 10 10 10 10' 10
Concentration (ng/g)
O 04
•
• f
• /
»
.%
•
• i
•
/. •
10 10 10 10 10 10' 10
Concentration (ng/g)
Figure 3. (continued)
109
-------
Dieldrin Total PCBs
1
0 8
H 0 6
0 A
• •
0 2
0
10
Concentration (ng/g)
1
0 8
H 0 6
0 A
0 2
0
10 10 10 10' 10
Concentration (ng/g)
10
10
p,p'-DDD
1
0 8
0 A
0 2
0
10'
10
10
10 10
Concentration (ng/g)
Figure 3. (continued).
110
-------
1-Methylnaphthalene
2,6-Dimethylnaphthalene
10 10 10 10' 10 10
Concentration (ng/g OC)
1-Methylphenanthrene
0 4 - r
10 10 10 10' 10 10
Concentration (ng/g OC)
Acenaphthene
: •
•
•
•
&
•
« _/
a
10 10 10 10 10 10
Concentration (ng/g OC)
10 10 10 10' 10 10 10
Concentration (ng/g OC)
2-Methylnaphthalene
Acenaphthylene
10 10 10 10' 10 10 10
Concentration (ng/g OC)
o
q!
10 10 10 10' 10 10 10
Concentration (ng/g OC)
Figure 4. Logistic regression models and proportion of toxic samples in
organic carbon-normalized concentration intervals in the screened marine
amphipod database for 25 chemicals based on Sig Only classification of toxic
samples. The individual points correspond to the median of the sample
concentrations within the interval and the proportion of the samples toxic within
the interval.
111
-------
Anthracene
Benzo(b)fluoranthene
O 04
« /
«
• mj
/
•
•<
••
f
•
**<
MJ
r*
•
<
•
10 10 10 10' 10 10 10
Concentration (ng/g OC)
o
Q.
•
•
•/
• /
> •
•7
¦
•j?r
10 10 10' 10 10 10
Concentration (ng/g OC)
Benz(a)anthracene
Benzo(k)fluoranthene
10 10 10 10' 10 10 10
Concentration (ng/g OC)
10 10 10 10' 10 10 10
Concentration (ng/g OC)
Benzo(a)pyrene
Benzo(g,h,i)perylene
10 10 10' 10 10 10
Concentration (ng/g OC)
10 10 10' 10 10 10
Concentration (ng/g OC)
Figure 4. (continued)
112
-------
Biphenyl
Fluoranthene
10 10 10 10' 10 10 10
Concentration (ng/g OC)
Chrysene
"i I r
o 04
•••
V
'
>•7
XT
•
• m
Ami*
10 10 10 10' 10 10 10
Concentration (ng/g OC)
Dibenzo(a,h)anthracene
1 r
10 10 10 10' 10 10 10
Concentration (ng/g OC)
0 8
o
X
o
H 0 6
c
o
o 04
o
a.
0 2
j
% j
• V
•u
•
• T
••»/ #
• •
ynv
••
w jrv ••
mmm
Ln . 1
10 10 10' 10 10 10 10
Concentration (ng/g OC)
Fluorene
~T
O 04
~T~
~T~
• *
• •
•/ •
•
•#
•
•
«y-;
10 10 10 10' 10 10 10
Concentration (ng/g OC)
lndeno(1,2,3-c,d)pyrene
1 1 r
10 10 10' 10 10 10 10
Concentration (ng/g OC)
Figure 4. (continued)
113
-------
Naphthalene Pyrene
0 8
o
X
o
H 0 6
c
o
o
Q.
O
Q_
0 2
10'
10
10
10
10
10
10
0 8
H 0 6
0 2
10' 10
10
10
10
10
10
10
Concentration (ng/g OC) Concentration (ng/g OC)
Perylene
10 10 10' 10 10 10
Concentration (ng/g OC)
Phenanthrene
o
£
y •
•/
V I
•
v /•
• ••
10 10 10' 10 10 10
Concentration (ng/g OC)
Dieldrin
> «
10 10 10 10 10' 10
Concentration (ng/g OC)
p,p'-DDD
o
£
9
;
•
•
• mjk
10 10 10 10 10' 10 10
Concentration (ng/g OC)
Figure 4. (continued)
114
-------
p,p'-DDE
10 10 10 10 10' 10 10 10 10
Concentration (ng/g OC)
Total PCBs
10 10 10 10 10' 10 10 10 10
Concentration (ng/g OC)
p,p'-DDT
0 8
0 2
10
10
10
10
10
10
10
Concentration (ng/g OC)
Figure 4. (continued).
115
-------
Cadmium
Chromium, total
Copper
Lead
Mercury
Silver
Zinc
1-Methylnaphthalene
1 -Methylphenanthrene
2,6-Dimethylnaphthalene
2-Methylnaphthalene
Acenaphthene
Acenaphthylene
Anthracene
Benz(a)anthracene
Benzo(a)pyrene
Benzo(b)fluoranthene
Benzo(g,h,i)perylene
Benzo(k)fluoranthene
Biphenyl
Chrysene
Dibenz(a,h)anthracene
Fluoranthene
Fluorene
lndeno(1,2,3-c,d)pyrene
Naphthalene
Perylene
Phenanthrene
Pyrene
Dieldrin
PCBS, total
p,p'-DDD
p,p'-DDT
000 005 0 10 0 15 0 20 0 25 030 035 040 045
Normalized Chi-square
¦ SigOnly ~ MSD
Figure 5. Comparison of logistic model goodness of fit for the marine
amphipod survival endpoint with different toxicity classifications: Sig Only
versus MSD.
-------
1-Methyl naphthalene
1 -Methylphenanthrene
2,6-Dmethyl naphthalene
2-Methyl naphthalene
Axnaphthene
Asenaphthylene
/Anthracene
Benz(a)anthracene
Benzo(a)pyrene
Benzo(b)fl uorant hene
Benzo(g,h,i)perylene
Benzo(k)fl uorant hene
Bi phenyl
Chrysene
Dbenz(a,h)anthracene
Fluoranthene
Ruorene
lndeno(1,2,3-c,d)pyrene
Naphthalene
Perylene
Phenanthrene
Pyrene
Dieldrin
PCBS, total
p,p'-DDD
p,p'-DDE
p,p'-DDT
0 00 0 05 0 10 0 15 0 20 0 25 0 30
Normalized Chi-square
0 35
040
045
IDW
~ oc
Figure 6. Comparison of logistic model goodness of fit for the marine
amphipod endpoint survival with different approaches to the expression of
chemical concentrations: dry weight (DW) versus organic carbon-
normalized (OC) concentrations. Both approaches use the Sig Only
classification of toxic samples.
117
-------
o
CM
Q
V)
1000
100
10
0 1
/'T •
T
/ •
/ •
•
0 1
o
ID
Q
w
J L
_L
1 10 100 1000
SIG Only T20
y = 3.77x
R2= 0.89
1000
100 -
o
oo
Q
W
1 10 100 1000 10000
SIG Only T50
y = 2.57x
R2= 0.90
1000
100
1 10 100 1000 10000
SIG Only T80
y = 1.65x
R2= 0.79
00
Figure 7. Comparison of logistic regression model Tp values (T20, T50, T80) for Sig Only and IMSD toxicity
classification approaches for the marine amphipod database.
-------
Lead
Lead
! !
: • /
^—-—
•: f
%i.fi
% •
%7
/_•
• •
,. i
1 10 100 1000 10000
Concentration (mg/kg)
X
o
o
¦tr
10
0 8
0 6
o 04
0
01
0 2
0 0
•
• *
•
• ; *7
*
• ;
i i
10 100 1000 10000
Concentration (mg/kg)
Phenanthrene
n i | r
Phenanthrene
10° 101 102 103 104 105 106
Concentration (mg/kg)
X
o
c
o
tr
o
Q.
O
10° 101 102 103 104 105 106
Concentration (mg/kg)
Figure 8. Comparison of IX mean screened (left) and unscreened (right)
logistic regression models and proportion of samples toxic in concentration
intervals for lead and phenanthrene using the marine amphipod database.
I 19
-------
Antimony
Arsenic
Cadmium
Chromium, total
Copper
Lead
Mercury
Nickel
Silver
Zinc
1-Methy Inaphthalene
1-Methylphenanthrene
2,6-Dimethylnaphthalene
2-Methylnaphthalene
Acenaphthene
Acenaphthy lene
Anthracene
Benz(a)anthracene
Benzo(a)pyrene
Benzo(b)fluoranthene
Benzo(g,h,i)perylene
Benzo(k)fluoranthene
Biphenyl
Chrysene
Dibenz(a,h)anthracene
Fluoranthene
Fluorene
lndeno(1,2,3-c,d)pyrene
Naphthalene
Perylene
Phenanthrene
Pyrene
Dieldrin
PCBS, total
P.P'-DDD
p.p'-DDE
000 005 0 10 0 15 0 20 0 25 0 30 0 35 040 045
Normalized Chi-square
¦ Screened ~ Unscreened
Figure 9. Comparison of logistic model goodness of fit for the marine
amphipod survival endpoint using different screening methods (IX screening
vs. no screening) and the Sig Only classification of toxic samples.
120
-------
Lead
Lead
10
0 8
0 6
0 4
0 2
0 0
-
• /
¦
•
%
%)
, •
»
•w t
... .^-V
•
•
•
•
\
10
0 8
0 6
0 4
0 2
0 0
1 10 100 1000 10000
Concentration (mg/kg)
Fluoranthene
!
• \/^
^—
•
• "A
V #
t •
•
»
•
10° 101 102 103 104 105 106
Concentration (ng/g)
O
X
o
I—
c
o
tr
o
Q.
O
1 10 100 1000 10000
Concentration (mg/kg)
Fluoranthene
O
X
o
I—
c
o
tr
o
Q.
O
10° 101 102 103 104 105 106
Concentration (ng/g)
Figure 10. Logistic regression models and proportion of samples toxic in
concentration intervals for the standard screening approach (IX mean) (left)
and the 2X mean screening approach (right) for lead and fluoranthene using
the marine amphipod database.
121
-------
Antimony
Arsenic
Cadmium
Chromium, total
Copper
Lead
Mercury
Nickel
Silver
Zinc
1-Methylnaphthalene
1-Methylphenanthrene
2,6-Dimethylnaphthalene
2-Methylnaphthalene
Acenaphthene
Acenaphthylene
Anthracene
Benz(a)anthracene
Benzo(a)pyrene
Benzo(b)fluoranthene
Benzo(g,h,i)perylene
Benzo(k)fluoranthene
Biphenyl
Chrysene
Dibenz(a,h)anthracene
Fluoranthene
Fluorene
lndeno(1,2,3-c,d)pyrene
Naphthalene
Perylene
Phenanthrene
Pyrene
Dieldrin
p,p'-DDD
p,p'-DDE
p,p'-DDT
PCBS, total
0 00 0 05 0 10 0 15 0 20 0 25 0 30 0 35 0 40 0 45
Normalized Chi-square
¦ 1X Mean Q2XMean
Figure 11. Comparison of logistic model goodness of fit for the marine
amphipod survival endpoint using different screening methods (IX mean vs.
2X mean) and Sig Only classification of toxic samples.
122
-------
1000
100 -
o
m
1 10 100 1000
Screening 1X T20
10000
1000
X
fsi
CD
'c
a?
0)
o
(/)
100
o
CO
h-
X
CM
O)
c
'c
-------
Original Corrected
•#
•>/
••
0 8
0 8
0 6
0 6
0 4
0 4
Q.
0 2
0 2
10"' 10° 101 102 103
5
6
10"' 10° 101 102 103
5
6
104
104
10
10
10
10
Concentration (ng/g) Concentration (ng/g)
Figure 13. Logistic regression models and concentration interval plots
showing the effect of correction in PCB units for 15 samples in the marine
amphipod database.
124
-------
100
03
>
T80
Probability Range
Figure 14. Mean percent control-adjusted marine amphipod survival for
toxic samples within intervals defined by the Tp values for all individual
chemical models.
125
-------
6
P_Prod
P_Avg
P_Max
A
2
0
0
Probability
Figure 15. Probability density functions for P Avg, P IMax, and P Prod.
The probability that a variable will have a value with a small interval around x
can be approximated by multiplying the value of y at x by the width of the
interval.
126
-------
0
0 0.2 0 4 0 6 0 8 1
P_Max Value
y = 0.11 +0.33x+0.40x2
R2 = 0.93
O
X
o
0.6
c
o
"tr
o
Q_
O
a.
0.4
0.2
0
0.2 0.4 0.6 0.8
1
P_Avg Value
y = 0.11 +1.42x-0.48x2
R2 = 0.89
Figure 16. Proportion of observed toxic samples versus the P IMax and
PAvg value probability intervals based on Sig Only classification of toxic
samples in the marine amphipod database. The regression lines are the P Max
and P Avg models. Each point represents the median sample probability of a
minimum of 50 individual samples within the interval and the proportion of the
samples that were toxic within the interval. Only samples having measurements
for 10 more modeled chemicals were used (n = 2856).
127
-------
Predicted
Observed
0.75
.1 0.5
¦e
o
Q.
O
CL „
0,25
<0.25 0.25-0.5 0.5-0.75 >0.75
N=990 N=1323 N=713 N=197
Probability Range
Predicted
Observed
<0.25 0.25-0.5 0.5-0.75 >0.75
N=943 N=1448 N=598 N=234
Probability Range
P Max Model
P_Avg Model
Figure 17. Mean predicted and observed proportion of toxic samples within
probability quartiles for P Max and PAvg models derived from the marine
amphipod database.
128
-------
o
tr
o
Q.
O
CL
"O
-------
100
(0
>
<= 0 25 >0 25 - 0 50 >0 5 - 0 75 >0 75
Probability Range
Figure 19. Mean percent control-adjusted survival for toxic samples within
probability quartile intervals for P IMax model derived from the marine
amphipod database.
130
-------
>
£
=J
"c
-------
1.00
0.75
0.50
0.25
-~
0.00
0
0.25
0.5
0.75
1
Maximum Probability
Figure 21. Comparison of the original marine amphipod P IMax model with
the model derived from data excluding polycyclic aromatic hydrocarbon
chemistry (R2 = 0.88). Data points represent the proportion of toxic samples
within unique probability intervals (minimum of 50 samples per interval) and the
dotted line is the model derived from the interval plot. The solid line is the
original P Max model.
132
-------
1.00
£ 0.25
¦c
o
Q_
O
O
X
o
I 0 50
0 .75
~
0 00
0 0 25 0 5 0 75
P Max Value
Figure 22. Comparison of the original marine amphipod P IMax model with
model derived from data excluding metals chemistry (R2 = 0.93). Data points
represent the proportion of toxic samples within unique probability intervals
(minimum of 50 samples per interval) and the dotted line is model derived from
the interval plot. The solid line is the original P Max model.
-------
0 00 1 1 1
0 0 25 0 5 0 75 1
P Max Value
Figure 23. Comparison of the original marine amphipod P IMax model with
model derived from data excluding pesticides and PCB chemistry (R2 = 0.92).
Data points represent the proportion of toxic samples within unique probability
intervals (minimum of 50 samples per interval) and the dotted line is model
derived from the interval plot. The solid line is the original P Max model.
-------
1 2 3-4 5-9 >9
Number of chemicals per sample with p>0.75
~
Predicted
¦
Observed
Figure 24. Mean predicted and observed proportion of toxic samples for samples
with a predicted probability of >0.75 from the P IMax model in the marine
amphipod database.
135
-------
o
X
0 0.2 0.4 0.6 0.8 1
P_Max Value
y= 0.09 + 0.73x
R2 = 0.94
Figure 25. Observed proportion of toxic samples versus P IMax values
including only those chemicals with normalized chi-square values exceeding
0.27. Data points represent the proportion toxic within unique probability
intervals (minimum of 50 samples per interval) and the line is the P Max model
derived from the interval plot.
136
-------
X
o
c
o
'tr
o
Q.
O
0.75 -
0.5
0.25
0.25 0.5 0.75
Probability of Toxicity
y = 0.29-1.10x+2.18x2
R2 = 0.94
03
>
£
D
if)
"O
£
00
D
"O
03
c
o
o
"c
-------
o
X
o
I—
c
o
tr
o
Q.
O
0.75 -
0.5
0.25
0.25 0.5 0.75
Probability of Toxicity
03
>
£
D
if)
"O
£
00
D
"O
03
c
o
O
"c
d)
o
-------
0 0 25 0 5 0 75 1
Probability of Toxicity
0 25 0 5 0 75 1
Probability of Toxicity
0 0 25 0 5 0 75 1
Probability of Toxicity
Figure 28. Median predicted probability of toxicity within probability intervals for the marine amphipod
P IMax model compared with data from Hudson-Raritan/Long Island NSTP and Regional EIMAP (A. abdita):
proportion of observed toxicity based on Sig Only classification (left), control-adjusted survival (center), and
proportion of observed toxicity based on IMSD classification (right). Each point represents the median sample
probability of a minimum of 25 individual samples within the interval (n = 280).
-------
o
o
X
o
I—
c
o
tr
o
Q_
O
0.75 -
0.5 -
0.25 -
0 0.25 0.5 0.75 1
Probability of Toxicity
y = 0.35-1.17X+2.05X2
R2= 0.81
03
>
£
D
if)
"O
£
00
D
"O
03
c
o
o
"c
d)
o
a3
Q.
100
75
50
25
o
X
o
I—
c
o
tr
o
Q.
O
0 0.25 0.5 0.75 1
Probability of Toxicity
y = 88.34+56.00X-105.09X2
R2= 0.91
0 75
0 5
0 25 -
0 0 25 0 5 0 75
Probability of Toxicity
y = 0.12-0.62x+1.36x
R2= 0.80
Figure 29. Median predicted probability of toxicity within probability intervals for the marine amphipod
P IMax model compared with data from Virginian Province EIMAP (A. abdita): proportion of observed toxicity
based on Sig Only classification (left), control-adjusted survival (center), and proportion of observed toxicity
based on IMSD classification (right). Each point represents the median sample probability of a minimum of 25
individual samples within the interval (n = 489).
-------
o
X
o
I—
c
o
tr
o
Q.
O
0.75
0.5
0.25
V ~
~ \
~
V ~
~ y
AA A
~ /
0 0.25 0.5 0.75 1
Probability of Toxicity
y = 0.19-1.08X+1.93X2
R2= 0.66
03
>
£
D
if)
"O
£
00
D
"O
03
c
o
o
"c
d)
o
-------
o
X
o
I—
c
o
tr
o
Q.
O
0.75 -•
0.5 -
0.25 -•
~
V
/
»
_L
0.25 0.5 0.75
Probability of Toxicity
03
>
£
D
if)
"O
£
00
D
"O
03
c
o
O
"c
d)
o
-------
o
X
o
I—
c
o
tr
o
Q.
O
0.75
0.5
0.25
~
~
~
~
•— ~
~
~ .
03
>
£
D
if)
"O
£
00
D
"O
03
c
o
O
"c
d)
o
-------
~ <
y
' ~
0.25 0.5 0.75
Probability of Toxicity
o
x
c
o
'¦c
o
Q.
O
0.75
0.5
0.25
~ ~
~ ~
0.25 0.5 0.75
Probability of Toxicity
y = -0.42+1.61 x
R2= 0.92
y = 0.07-0.20X
R2 = 0.28
Figure 33. Median predicted probability of toxicity within probability
intervals for the marine amphipod P IMax model compared with proportion
of observed toxicity based on Sig Only classification from individual studies:
Elliott Bay, Puget Sound, WA (R. abronius, n = 97) (left) and Biscayne Bay,
FL (A. abdita, n = 120) (right). Each point represents the median sample
probability of a minimum of 12 individual samples within the interval.
144
-------
Sig Only
MSD
0 75
0 5
0 25
~ ~<
~
~
0 0 25 0 5 0 75
Probability of Toxicity
X
o
c
o
tr
o
Q_
O
0.75
0.5
0.25
y
<
~
0 0.25 0.5 0.75
Probability of Toxicity
y = 0.77+0.20x
R2= 0.17
y = 0.22+0.98x
R2 = 0.76
Figure 34. Median predicted probability of toxicity within probability
intervals for the marine amphipod P IMax model compared with proportion
of observed toxicity from an individual study from San Diego Bay, CA (R.
abronius, n = 93); toxicity based on Sig Only classification and MSD
classification. Each point represents the median sample probability of a
minimum of 12 individual samples within the interval.
145
-------
o
X
o
I—
c
o
"tr
o
Q.
O
0.75 -•
0.5 -
0.25 -•
_L
0.25 0.5 0.75
Probability of Toxicity
03
>
£
D
if)
"O
£
00
D
"O
03
c
o
O
"c
d)
o
a3
Q.
100
75 -•
50 —
25 -•
4
_L
0.25 0.5 0.75
Probability of Toxicity
o
X
o
I—
c
o
"tr
o
Q_
O
0.75 -
0.5 -
0.25 -
0.25 0.5 0.75
Probability of Toxicity
y = 0.45+0.63x
R2= 0.82
y = 63.78-3.63x-85.80x
R2 = 0.97
y = 0.43+0.65x
R2= 0.82
On
Figure 35. Median predicted probability of toxicity within probability intervals for the marine amphipod
P IMax model compared with data from the Calcasieu Estuary {A. abdita): proportion of observed toxicity based
on Sig Only classification (left), control-adjusted survival (center), and proportion of observed toxicity based on
IMSD classification (right). Each point represents the median sample probability of a minimum of 12 individual
samples within the interval (n = 170).
-------
Urchin Development
Urchin Fertilization
0 75
0 5
0 25
~
~
~
~
. V
i
~
0 25 0 5 0 75
Probability of Toxicity
X
o
c
o
tr
o
Q_
O
0.75 -
0.5 -
0.25 -
0 0.25 0.5 0.75
Probability of Toxicity
y - 0.55+0.23x
R2=0.18
y = 0.15+1.52x-1.52x
R2 = 0.39
Figure 36. Average predicted probability and proportion toxic within
probability intervals for the marine amphipod P IMax model applied to the
sea urchin development and fertilization endpoints. Each point represents the
median sample probability of a minimum of 25 individual samples within the
interval and the proportion of the toxic samples within the interval (n = 782 and
824 for development and fertilization, respectively).
147
-------
X
o
c
o
'tr
o
Q.
O
0.75
0.5
0.25
I
~ J
~ y
i
0 0.25 0.5 0.75
Probability of Toxicity
~
0 0.25 0.5 0.75
Probability of Toxicity
oo
y = 0.40 + 0.68x
R2 = 0.74
y = 65.07 -67.46x
R2 = 0.79
y = 0.39 + 0.70x
R2 = 0.75
Figure 37. Median predicted probability and proportion toxic within probability intervals for the marine
amphipod P Max model applied to sea urchin (A. punctulata) development endpoint: proportion of observed
toxicity based on Sig Only classification (left), control-adjusted response (center), and proportion of observed
toxicity based on IMSD classification (right). Each point represents the median sample probability of a minimum of
25 individual samples within the interval and the proportion of the toxic samples within the interval (n = 472).
-------
X
o
c
o
tr
o
Q_
O
0.75
0.5
0.25 -
0 0.25 0.5 0.75
Probability of Toxicity
-------
X
o
c
o
tr
o
Q_
O
0.75
0.5
0.25
0 0.25 0.5 0.75
Probability of Toxicity
-------
X
o
c
o
tr
o
Q_
O
0.75
0.5
0.25 -
0.25 0.5 0.75
Probability of Toxicity
c
o
Q_
CO
<1>
or
"O
0)
"oo
D
"O
03
C
o
O
100
75 -
50 -
25
0.25 0.5 0.75
Probability of Toxicity
X
o
tr
o
Q_
O
0.75
0.5
0.25
I I i
0.25 0.5 0.75
Probability of Toxicity
y = 0.59 + 0.09x
R2= 0.03
y = 55.95 - 12.72x
R2= 0.04
y = 0.61 + 0.01x
R2= 0.00
Figure 40. Median predicted probability and proportion toxic within probability intervals for the marine
amphipod P IMax model applied to sea urchin (.V. purpuratus) fertilization endpoint: proportion of observed
toxicity based on Sig Only classification (left), control-adjusted response (center), and proportion of observed
toxicity based on IMSD classification (right). Each point represents the median sample probability of a minimum of
25 individual samples within the interval and the proportion of the toxic samples within the interval (n = 212).
-------
1.00
a 0.75
~
~
o 0.50
¦e
o
Q_
I 0.25
0.00
~
0.00 0.25 0.50 0.75 1.00
Probability of Toxicity
Figure 41. Proportion of samples that were toxic based on the H. azteca 10-
14-day survival endpoint versus median probability of toxicity predicted
using the marine amphipod P IMax model. Each point represents the median
sample probability of a minimum of 25 individual samples within the interval and
the proportion of toxic samples within the same interval (n = 567).
152
-------
1.00
o
5?
O
I—
c
o
'¦c
o
CL
o
0.75
0.50
0.25
0.00
0.00 0.25 0.50 0.75 1.00
Probability of Toxicity
Figure 42. Proportion of samples that were toxic based on the C. tentans or
C. riparius 10-14-day survival endpoint versus median probability of toxicity
predicted using the marine amphipod P IMax model. Each point represents
the median sample probability of a minimum of 25 individual samples within the
interval and the proportion of toxic samples observed within the same interval (n
= 585).
-------
1
0.75
0.5
0.25
0
0
0.25
0.5
0.75
1
Probability of Toxicity
y = 0.18 - 0.08x - 1,12x2
R2= 0.95
Figure 43. Proportion of samples that were toxic based on the H. azteca 28-
day growth and survival endpoint versus the median probability of toxicity
predicted using the marine amphipod P IMax model. Each point represents
the median sample probability of a minimum of 15 individual samples within the
interval and the proportion of the toxic samples observed within the same interval
(n = 126).
154
-------
REFERENCES
ASTM (American Society for Testing and Materials). (2002a) Standard guide for conducting 10-day. static
sediment toxicity tests with marine and cstuarinc amphipods. In: Annual book of standards. Vol. 11.05. E 1367-99;
pp. 693-719.
ASTM. (2002b) Standard test method for measuring the toxicity of sediment-associated contaminants with
freshwater invertebrates. E1706-00B. In: Annual book of standards. Vol. 11.05. E 1367-99; pp. 1125-1241.
Brodcrius. SJ. (1991) Modeling the joint toxicity of xenobiotics to aquatic organisms: basic concepts and
approaches. In: Aquatic toxicology and risk assessment. Vol. 14. Mayes. MA; Barron. MG. cds. Philadelphia. PA:
American Society forTcsting and Materials: pp. 107-127.
Carr. RS: Bicdcnbach. JM. (1999) Use of power analysis to develop detectable significance criteria for sea urchin
toxicity tests. Aquat Ecosys Health Manage 2:413-418.
Chapman. PM. (1989) Current approaches to developing sediment quality criteria. Environ Toxicol Chcm 8:589-
599.
Di Toro. DM; Zarba. CS: Hansen. DJ; ct al. (1991) Technical basis for establishing sediment quality criteria for
nonionic organic chemicals using equilibrium partitioning. Environ Toxicol Chcm 10:1541-1583.
Di Toro. DM; McGrath. JA; Hansen. DJ. (2000) Technical basis for narcotic chemicals and polycyclic aromatic
hydrocarbon criteria: II. mixtures and sediments. Environ Toxicol Chcm 19:1971-1982.
Faircy. R: Long. ER; Roberts. CA; ct al. (2001) An evaluation of methods for calculating mean sediment quality
guideline quotients as indicators of contamination and acute toxicity to amphipods by chemical mixtures. Environ
Toxicol Chcm 20:2276-2286.
Field. LJ; MacDonald. DD; Norton. SB; ct al. (1999) Evaluating sediment chemistry and toxicity data using
logistic regression modeling. Environ Toxicol Chcm 18:1311-1322.
Field. LJ; MacDonald. DD; Norton. SB; ct al. (2002) Predicting amphipod toxicity from sediment chemistry using
logistic regression models. Environ Toxicol Chcm 21(9): 1993-2005.
Grics. TH; Waldow. K.H. (1996) Progress re-evaluating Pugct Sound apparent effects thresholds (AETs). Vol. I.
1994 amphipod and cchinodcrm larval AETs. Pugct Sound dredged disposal analysis (PSDDA). U.S. Army Corps
of Engineers. Seattle District. U.S. Environmental Protection Agency. Region 10. State of Washington Departments
of Ecology and Natural Resources. Seattle. WA.
Hansen. DF; Bern . WJ; Mahony. JD; ct al. (1996) Predicting the toxicity of metal-contaminated field sediments
using interstitial concentration of metals and acid-volatile sulfide normalizations. Environ Toxicol Chcm 15:2080-
2094.
Ingcrsoll. CG: Havcrland. PS: Brunson. EL; ct al. (1996) Calculation and evaluation of sediment effect
concentrations for the amphipod llyolella azteca and the midge Chironomus riparius. J Great Lakes Res 22:602-
623.
Ingcrsoll. CG: Dillon. T; Biddingcr. RG. cds. (1997) Ecological risk assessments of contaminated sediment.
Pcnsacola. FL: SETAC Press.
Ingcrsoll. CG: MacDonald. DD; Wang. N; ct al. (2001) Predictions of sediment toxicity using consensus-based
freshwater sediment quality guidelines. Arch Environ Contain Toxicol 41:8-21.
155
-------
Ingcrsoll. CG: MacDonald. DD; Bnunbaugh. WG: ct al. (2002) Toxicity assessment of sediments from the Grand
Calumet River and Indiana Harbor Canal in northwestern Indiana. Arch Environ Contain Toxicol 43:156-167.
Insightful Corp. (2001) S-Plus 6 for windows guide to statistics. Seattle. W A.
Kemp. PF; Swart/. RC: Lambcrson. JO. (1986) Response of the photoccphalid amphipod. Rhcpoxynius ahronius.
to a small oil spill in Yaquina Bay. Oregon. Estuaries 9:340-347.
Lee. JH; Landrum. PF; Field. LJ; ct al. (2001) Application of a ZPAH model and a logistic regression model to
sediment toxicity data based on a species-specific water-only LC50 toxic unit for llyalclla azteco. Environ Toxicol
Chcm 20:2102-2113.
Long. ER: MacDonald. DD. (1992) National Status and Trends Program approach. In: Sediment classification
methods compendium. EPA 823-R-92-006. Washington. DC: U.S. Environmental Protection Agency.
Long. ER; Morgan. LG. (1990) The potential for biological effects of scdimcnt-sorbcd contaminants tested in the
National Status and Trends Program. NOAA Tech Memo NOS OMA 52. National Oceanic and Atmospheric
Administration. Seattle. WA.
Long. ER; Morgan. LG. (1991) The potential for biological effects of scdimcnt-sorbcd contaminants tested in the
National Status and Trends Program. NOAA Tech Memo NOS OMA 52. National Oceanic and Atmospheric
Administration. Seattle. WA.
Long. ER; MacDonald. DD; Smith. SL; ct al. (1995) Incidence of adverse biological effects within ranges of
chemical concentrations in marine and cstuarinc sediments. Environ Manage 19:81-97.
Long. ER; Field. LJ; MacDonald. DD. (1998) Predicting toxicity in marine sediments with numerical sediment
quality guidelines. Environ Toxicol Chcm 17:714-727.
MacDonald. DD; Carr. RS: Caldcr. FD; ct al. (1996) Development and evaluation of sediment quality guidelines
for Florida coastal waters. Ecotoxicology 5:253-278.
MacDonald. DD; Ingcrsoll. CG: Bcrgcr. TA. (2000) Development and evaluation of consensus-based sediment
quality guidelines for freshwater ecosystems. Arch Environ Contain Toxicol 39:20-31.
MacDonald. DD; Moore. DR; Paw lis/.. A; ct al. (2001) Calcasieu Estuary remedial investigation/feasibility study
(RI/FS): baseline ecological risk assessment (BERA). Vol. 2.
Mearns. AJ; Swart/. RC: Cummins. JM; ct al. (1986) Intcr-laboratory comparison of a sediment toxicity test using
the marine amphipod. Rhcpoxynius ahronius. Mar Environ Res 19:13-37.
Morgan. BJT. (1992) Analysis of quantal response data. London. UK: Chapman and Hall.
Nctcr. J; Kutncr. MH; Nachtschcim. CJ: ct al. (1996) Applied linear statistical models. 4,hcd. Chicago. IL: Irwin.
NOAA (National Oceanic and Atmospheric Administration). (1989) Standard analytical procedures of the NOAA
Analytical Facility. 2'"1 cd. NOAA technical memorandum NMFS F/NWC-92. 1985-86. National Status and Trends
Program. U.S. Department of Commerce. Rockvillc. MD.
NOAA (National Oceanic and Atmospheric Administration). (2004) Predicting sediment toxicity. Available online
at http://www.response.rcstoration.noaa.gov/cpr/scdimcnt/scd_tox.html.
Phillips. BM; Hunt. JW; Anderson. BS: ct al. (2001) Statistical significance of sediment toxicity test results:
threshold values derived by the detectable significance approach. Environ Toxicol Chcm 20:371-373.
156
-------
Redmond. MS: Crocker. PA; McKenna. KM; ct al. (1996) Sediment toxicity testing with the amphipod. /.sew
ohdita in Calcasieu Estuary¦. Arch Environ Contain Toxicol 30:53-61.
Robinson. AM; Lambcrson. JO: Cole. FA; ct al. (1988) Effects of culture conditions on the sensitivity of a
phoxoccphalid amphipod. Rhepoxynius ahronius. to cadmium in sediment. Environ Toxicol Chcm 7:953-959.
SAS Institute. (1990) SAS/STAT user's guide, version 6. 4,h cd. Vols. 1 and 2. Can . NC.
Smith. SL; MacDonald. DD; Kccnlcysidc. KA; ct al. (1996) A preliminary evaluation of sediment quality
assessment values for freshwater ecosystems. J Great Lakes Res 22:624-638.
Smith. E; Robinson. T; Field. LJ; ct al. (2003) Predicting sediment toxicity using logistic regression: a
concentration-addition approach. Environ Toxicol Chcm 22:565-575.
Swart/. RC: Ditsworth. GR: Schults. DW; ct al. (1985) Sediment toxicity to a marine infaunal amphipod:
cadmium and its interactions with sewage sludge. Mar Environ Res 18:133-153.
Swart/. RC: Schults. DW; DcWitt. TH; ct al. (1987) Toxicity of fluoranthene in sediment to marine amphipods: a
test of the equilibrium partitioning approach to sediment quality criteria. 89th Annual Meeting. Society for
Environmental Toxicology and Chemistry¦. Pcnsacola. FL.
Swart/. RC: Kemp. PF; Schults. DW; ct al. (1988) Effects of mixtures of sediment contaminants on the marine
infaunal amphipod. Rhcpoxyinius ahronius. Environ Toxicol Chcm 7:1013-1020.
Swart/. RC: Kemp. PF; Schults. DW; ct al. (1989) Acute toxicity of sediment from Eagle Harbor. Washington, to
the marine infaunal amphipod. Rhcpoxyinius ahronius. Environ Toxicol Chcm 8:215-222.
Swart/. RC: Schults. DW; O/rctich. RJ; ct al. (1995) ^PAH: a model to predict the toxicity of poly nuclear
aromatic hydrocarbon mixtures in ficld-collcctcd sediments. Environ Toxicol Chcm 14:1977-1987.
Thursby. GB; Hcltshc. J: Scott. KJ. (1997) Revised approach to toxicity test acceptability criteria using a statistical
performance assessment. Environ Toxicol Chcm 16:1322-1329.
Tukcy. JW. (1977) Exploratory data analysis. Reading. MA: Addison-Wcslcy.
U.S. EPA (Environmental Protection Agency). (1994a) Methods for measuring the toxicity and bioaccumulation of
scdimcnt-associatcd contaminants with freshwater invertebrates. Is'cd. EPA/600/R-94/024. Duluth. MN.
U.S. EPA. (1994b) Methods for measuring the toxicity of scdimcnt-associatcd contaminants with cstuarinc and
marine amphipods. EPA-600/R-94/025. Narragansctt. RI.
U.S. EPA. (1997) The incidence and severity of sediment contamination in surface waters of the United States.
Vol. 1: National sediment quality survey. EPA 823-R-97-006. Office of Science and Technology. Washington. DC.
U.S. EPA. (2000) Supplementary guidance for conducting health risk assessment of chemical mixtures. EPA 630-
R-00-002. Office of Research and Development. Washington. DC.
U.S. EPA. (2004) The incidence and severity of sediment contamination in surface waters of the United States.
National sediment quality survey. Second edition. EPA-823-R-04-007. Office of Science and Technology.
Washington. DC.
Word. JQ: Ward. JA; Franklin. LM; ct al. (1987) Evaluation of the equilibrium partitioning theory for estimating
the toxicity of the nonpolar organic compound DDT to the sediment dwelling amphipod Rhepoxynius ahronius.
Battcllc Washington Environmental Program Office. Report prepared for Criteria and Standards Division. U.S.
Environmental Protection Agency.
157
-------
APPENDIX Al: REFERENCES FOR MARINE SEDTOX02 DATABASE
Adams. DA; OConner. JS: Wcisbcrg. SB. (1998) Sediment quality of the NY/NJ Harbor system: an investigation
under the Regional Environmental Monitoring and Assessment Program (R-EMAP). Final report. U.S. EPA.
Region 2. Division of Environmental Sciences & Assessment EPA/902-R-98-001.
Analytical. (1988) Volatile organics. pesticides. PCBs. acid, base neutral analyses for PSDDA phase i baseline
monitoring (Hardcopies and Lotus worksheet files). Received from ARI on July 5. 1988.
Anderson. JW; Bay. SM; Thompson. BE. (1988) Characteristics and effects of contaminated sediments from
Southern California. California State Water Resources Control Board. Sacramento. CA.
Anderson. B; Hunt. J: Tudor. S: ct al. (1997) Chemistry, toxicity and bcnthic community conditions in sediments
of the Southern California bays and estuaries. California State Water Resources Control Board. Sacramento. CA.
Anderson. B; Hunt. J: Phillips. B; ct al. (1998) Sediment chcmistn. toxicity, and bcnthic community conditions in
selected water bodies of the Los Angeles Region. Final Report. California State Water Resources Control Board.
Sacramento. CA.
Bay. S: Grccntcin. D; Brown. J; ct al. (1994) Investigation of toxicity in Palos Vcrdcs sediments. Final report
1994. Prepared by Southern California Coastal Water Resources Project. Westminster CA. Prepared for Santa
Monica Bay Restoration Project.
Carr. RS. (1992) Survey of Galveston Bay bottom sediments and bcnthic communities. Draft final report.
Prepared by U.S. Fish and Wildlife Sen ice. Corpus Christi State University. Prepared for Galveston Bay National
Estuary Program and U.S. Environmental Protection Agency (USEPA). Region 6.
Casillas. E; Weber. D; Haley. C: ct al. (1992) Comparison of growth and mortality in juvenile sand dollars
(Dem/raster excentricus) as indicators of contaminated marine sediments. Environ Toxicol Chcm 11:559-569
Chapman. PM; Dexter. RN; Long. ER. (1987) Synoptic measures of sediment contamination, toxicity and infaunal
community composition (the Sediment Quality Triad) in San Francisco Bay. Marine Ecol Progr Scr 37.
Chapman. PM; Power. EA; Dexter. RN; ct al. (1991) Evaluation of effects associated with an oil platform, using
Sediment Quality Triad. Environ Toxicol Chcm. Vol. 10.
Crocker; ct al. (1991) Water quality, ambient toxicity and biological investigations in the Houston Ship Channel
and tidal San Jacinto River. U.S. EPA.
Downing. J; Fairey. R; Roberts. C; ct al. (1998) Chemical and biological measures of sediment quality in the
Central Coast Region. Final report. California State Water Resources Control Board. Sacramento. CA.
EA Mid-Atlantic. EA Engineering. Science, and Technology. Inc. (1994) Results of dredged material testing of
samples from the proposed Bayou Casottc Turning Basin. Pascagoula. Mississippi. U.S. Army Engineer District.
Mobile. Alabama.
Englc. V; Harwell. L; Hcitmullcr. T. (1999) Catalog documentation EMAP-cstuarics program level database
Louisianian Province 1991-1994 sediment chcmistn data & sediment toxicity data. U.S. EPA and U.S. Geological
Survey.
Fairey. R; Bret/. C; Lamcrdin. S; ct al. (1996) Chemistry. toxicity and bcnthic community conditions in sediments
of the San Diego bay region. California State Water Resources Control Board. Sacramento. CA.
Fairey. R; Downing. J; Roberts. C; ct al. (1998) Chemistry. toxicity and bcnthic community conditions in
sediments of the San Diego bay region: final addendum report. California State Water Resources Control Board.
Sacramento. CA.
158
-------
Fcrraro. SP: Swart/. RC: Cole. FA; ct al. (1991) Temporal changes in the benthos along a pollution gradient:
Discriminating the effects of natural phenomena from sewage-industrial wastewater effects. Estuar Coast Shelf Sci
33 (4): 3 8 3 —107.
Flcgal. AR; Riscbrough. RW; Anderson. B; ct al. (1996) San Francisco estuary pilot regional monitoring program:
Sediment studies. San Francisco Bay Regional Water Quality Control Board. State Water Resources Control Board.
Hall. LW. Jr; Zicgcnfuss. MC: Fischer. SA; ct al. (1992) A pilot study for ambient toxicity testing in Chesapeake
Bay. Year 2 report. Prepared for U.S. Environmental Protection Agency. CBPATRS 82/92. Contract No. 68-WO-
00-43. Printed by U.S. EPA.
Hart Crowscr & Associates Inc. (1986) Results of chemical and biological analyses of sediment samples. Everett
Marina, and Jetty Island Pier. Everett. Washington. Prepared by Hart Crowscr and Associates. Inc.
Hunt. JB; Anderson. B; Phillips. B; ct al. (1997) Evaluation of sediment toxicity tests and reference sites in San
Francisco Bay. California State Water Resources Control Board. Sacramento. CA.
Hunt. JB; Anderson. B; Phillips. B; ct al. (1998) Sediment quality and biological effects in San Francisco Bay.
Bay Protection and Toxic Cleanup Program. Final technical report. California State Water Resources Control
Board. Sacramento. CA.
Hyland. JL; Hcrrlingcr. TJ; Snoots. TR; ct al. (1996) Environmental quality of estuaries of the Carolinian Province:
1994. Annual statistical summary for the 1994 EMAP-Estuarics Demonstration Project in the Carol. NOAA
Technical Memorandum NOS ORCA 97. National Oceanic and Atmospheric Administration. Silver Spring. MD.
Hyland. JL; Balthis. L; Hackney. CT; ct al. (1998) Environmental quality of estuaries of the Carolinian Province:
1995. Annual statistical summary for the 1995 EMAP - Estuaries demonstration project in the Car. NOAA
Technical Memorandum NOS ORCA 12.3. National Oceanic and Atmospheric Administration. Silver Spring. MD.
Jacobi. M; Faircy. R; Roberts. C; ct al. (1998) Chemical and biological measures of sediment quality and tissue
bioaccumulation in the North Coast Region. Final Report. California State Water Resources Control Board.
Sacramento. CA.
Lambcrson. JO: Redmond. MS: Jones. JK.P; ct al. (199.3) Development of a 10-day amphipod sediment toxicity test
for Hawaii. USEPA. Environmental Research Lab - Narragansctt. Contract No 68-CO-0051.
Long. ER; Buchman. MF; Bay. SM; ct al. (1990) Comparative evaluation of five toxicity tests with sediments form
San Francisco Bay and Tomalcs Bay. California. Environ Toxicol Chcm 9:119.3-1214.
Long. ER; Wolfe. DA; Carr. RS: ct al. (1994) Magnitude and extent of sediment toxicity in Tampa Bay. Florida.
NOAA Technical Memorandum NOS ORCA 78. National Oceanic and Atmospheric Administration. Silver Spring.
MD.
Long. ER; Wolfe. DA; Scott. KJ; ct al. (1995) Magnitude and extent of sediment toxicity in the Hudson-Raritan
Estuary¦. NOAA Technical Memorandum NOS ORCA 88. National Oceanic and Atmospheric Administration.
Silver Spring. MD.
Long. ER; Sloanc. GM; Carr. RS: ct al. (1996) Sediment toxicity in Boston Harbor: Magnitude, extent, and
relationships with chemical toxicants. NOAA Technical Memorandum NOS ORCA 96. National Oceanic and
Atmospheric Administration. Silver Spring. MD.
Long. ER; Sloanc. GM; Carr. RS: ct al. (1997) Magnitude and extent of sediment toxicity in four bays for the
Florida panhandle: Pcnsacola. Choctawatchcc. St. Andrew and Apalachicola. National Oceanic and Atmospheric
Administration. Silver Spring. MD.
159
-------
Long. E; Scott. G: Kucklick. J; ct al. (1998) Magnitude and extent of sediment toxicity in selected estuaries of
South Carolina and Georgia. NOAA Technical Memorandum NOS ORCA 128. National Oceanic and Atmospheric
Administration. Silver Spring. MD.
Long. E; Sloanc. G: Scott. G: ct al. (1999) Magnitude and extent of chemical contamination and toxicity in
sediments of Biscay nc Bay and vicinity-final draft. NOAA Technical Memorandum NOS ORCA 141. National
Oceanic and Atmospheric Administration. Silver Spring. MD.
McGcc. BL: Schlckat. CE: Rcinharz. E. (1993) Assessing sublethal levels of sediment contamination using the
cstuarinc amphipod l.epiocheirusplumulosus. Environ Toxicol Chcm 12:577-587.
McLcay. D: Ycc. S: Doc. K: ct al. (1991) Final report: Phase II and phase III studies by EC Laboratories of 10-day
tests for sediment toxicity using marine or cstuarinc infaunal amphipods. Prepared by McLcay Associates Ltd. for
Environment Canada.
Murdoch. MH; Chapman. PM; Johns. DM; ct al. (1996) Chronic effects of organochlorinc exposure in sediment to
the marine polvchactc Xeanthes arenaceodentata. Prepared bv EVS Environmental Consultants. North Vancouver.
BC.
Nelson. WG: Bergen. BJ; Bcnyi. SJ: ct al. (1996) New Bedford Harbor long-term monitoring assessment report:
Baseline sampling. Research Report United States Environmental Protection Agency.
PTI Environmental Sen ices (1989) PSDDA baseline survey of phase ii disposal sites. Bcllcvuc. WA.
PTI Environmental Sen ices. (1990) Reconnaissance survey of reference area sediments in shallow waters of Carr
inlet. Technical memorandum. Prepared for U.S. Environmental Protection Agency. Region X. Seattle. WA.
PTI Environmental Sen ices. (1991) Reconnaissance survey of chemical contamination and biological effects in
southern Pugct Sound. Prepared for U.S. Environmental Protection Agency. Region X. Seattle. WA.
PTI Environmental Sen ices: Tctra Tech. Inc. (1988) Everett Harbor Action Program: analysis of toxic problem
areas. Final report. Prepared for U.S. Environmental Protection Agency. Region X. Office of Pugct Sound. Seattle.
WA.
PTI Environmental Sen ices: Tctra Tech. Inc. (1988) Pugct Sound Estuary Program Elliott Bay Action Program:
Analysis of toxic problem areas. Final report. Prepared for U.S. Environmental Protection Agency. Region X.
Office of Pugct Sound. Seattle. WA.
Pacific Northwest Laboratory. Battcllc Marine Research Laboratory. (1986) Reconnaissance survey of eight bays
in Pugct Sound. Final report - Volumes I and II. Prepared for U.S. Environmental Protection Agency. Region X.
Seattle. WA.
Pastorok. RA; Becker. DS. (1990) Comparative sensitivity of sediment toxicity bioassays at three Supcrfund sites
in Pugct Sound. ASTM STP 1096.
Phillips. B; Anderson. B; Hunt. J: ct al. (1998) Sediment chemistry, toxicity, and bcnthic community conditions in
selected water bodies of the Santa Ana Region. Final report. California State Water Resources Control Board.
Sacramento. CA.
Rice. CA; Plcsha. PD; Casillas. E; ct al. (1995) Growth and survival of three marine invertebrate species in
sediments from the Hudson-Raritan Estuary¦. New York. Environ Toxicol Chcm 14(11): 1931-1940
Striplin. PL; Sparks-McConkcy. PJ; Davis. DA; ct al. (1992) Pugct Sound ambient monitoring program, marine
sediment monitoring program, annual report 1990. Washington State Department of Ecology. Olympia. WA.
Strobcl. CJ: Buffum. HW; Bcnyi. SJ: ct al. (1995) Statistical summary. EMAP-Estuarics. Virginian Province -
1990 to 1993. U.S. Environmental Protection Agency. Narragansctt. RI. EPA/620-R-94/-26.
160
-------
Swart/. RC: Schults. DW; Ditsworth. GR: ct al. (1985) Sediment toxicity, contamination, and macrobcnthic
communities near a large sewage outfall. Validation and predictability of laboratory methods for assessing the fate
and effects of contaminants in aquatic ecosystems. ASTM STP 865.
Swart/. RC: Cole. FA; Schults. DW; ct al. (1986) Ecological changes in the Southern California Bight near a large
sewage outfall: benthic conditions in 1980 and 1983. Marine Ecol Progr Scr. U.S. Environmental Protection
Agency.
Swart/. RC: Kemp. PF; Schults. DW; ct al. (1988) Acute toxicity of sediment from Eagle Harbor. Washington, to
the infaunal amphipod Rhcpoxynius ahronius. Environ Toxicol Chcm. Vol. 8.
Swart/. RC; Schults. DW; Lambcrson. JO; ct al. (1991) Vertical profiles of toxicity, organic carbon, and chemical
contaminants in sediment cores from the Palos Vcrdcs shelf and Santa Monica Bay. California. Marine Environ Res
31(3):215-225
Tay. Kok-Lcng; Doc. KG; Wade. SJ; ct al. (1990) Sediment assessment in Halifax Harbour. In: Aquatic toxicity
workshop proceedings: 17th 1991. Vol. 2.
Tctra Tech. Inc. (1990) Pugct Sound ambient monitoring program 1989 marine sediment monitoring final report.
Bcllcvuc. WA.
Tctra Tech. Inc. (1988a) Commencement Bay ncarshorc/tidcflats feasibility study, appendices, public review draft.
Bcllcvuc. WA.
Tctra Tech. Inc. (1988b) Commencement Bay ncarshorc/tidcflats feasibility study. Vol. 1. public review draft.
Bcllcvuc. WA.
Ward. J A; Pin/a. MR; Barrows. ME; ct al. (1992) Ecological evaluation of proposed dredged material from
Wilmington Harbor. North Carolina. Draft report. Prepared by Pacific Northwest Laboratory. Prepared for the
U.S. Army Corps of Engineers with the U.S. Department of Energy. Contract DE-AC06-76RLO 1830.
Washington State Department of Ecology. (1994) Pugct Sound ambient monitoring program, marine sediment
monitoring program, annual report 1991. Olympia. WA. Publication #94-93.
Weyerhaeuser Analytical Laboratory. (1984) Blair/Milwaukee waterways |URS001 ]: metals and conventional
analyses. Tacoma. WA.
Windom. H. (1995) An evaluation of ambient sediment quality in the Brunswick Harbor and Savannah River
entrance channels. Georgia. Excerpts from an unpublished report. Chemical data. Skidway Institute of
Oceanography. Savannah. GA.
Wolfe. DA; Brickcr. SB; Long. ER; ct al. (1994) Biological effects of toxic contaminants in sediments from Long
Island Sound and environs. NOAA Technical Memorandum NOS ORCA 80. National Oceanic and Atmospheric
Administration. Silver Spring. MD.
Word. JQ; Ward. J A; Squires. AL. (1990) Results of chemical, toxicological. and bioaccumulation evaluations of
dioxins. furans. and guiaicol/organic acids in sediments from the Grays Harbor/Chahalis R. Prepared by Battcllc
Memorial Institute. Battcllc/Marinc Sciences Laboratory. Prepared for the U.S. Army Corps of Engineers and the
U.S. Department of Energy.
161
-------
APPENDIX A2: REFERENCES FOR FRESHWATER SEDTOX02 DATABASE
Abraham. WJ; Prcddicc. T; Kuzia. EJ; ct al. (1993) Lower Gcncscc River Project phase I report: summary 1992
results. New York State Department of Environmental Conservation.
Anderson. B; Hunt. J: Phillips. B; ct al. (1998) Sediment chcmistn. toxicity, and bcnthic community conditions in
selected water bodies of the Los Angeles Region. Final report. California State Water Resources Control Board.
Anklcy. GT; Leonard. E; Glass. G: ct al. (1990) Chemical and biological assessment of sediments from Oswego
Harbor. 1990: final report. New York State Department of Environmental Conservation.
Anklcy. GT; Collyard. SA; Monson. PD; ct al. (1994) Influence of ultraviolet light on the toxicity of sediments
contaminated with polycyclic aromatic hydrocarbons. Environ Toxicol Chcm 13( 11).
Applied Biology. Inc. (1982) Chemical, physical and bioassay analysis of sediment samples. Eric Harbor. Eric.
Pennsylvania.
Aqua Tech Environmental Consultants. Inc. (1985) The analyses of sediments from Rochester Harbor. Prepared
for U.S. Army Corps of Engineers. Buffalo.
Aqua Tech Environmental Consultants. Inc. (1986) The analyses of sediments from Cleveland Harbor. Cleveland.
Ohio. Prepared for U.S. Corps of Engineers. Buffalo.
Aquatcc Biological Sciences. (1997) Results of Chironomus teutons survival and growth toxicity tests for the 1997
Lake Ontario sediment and biota survey BTRs 1275& 1279. RECRA LabNct. New York State Department of
Environmental Conservation. Aquatcc Biological Sciences.
AScI Corporation/AScI-Duluth Environmental Testing Division. (1996) Results of chronic toxicity testing with
whole sediment samples from IT Corporation-Intcrlakc Project. International Technology Corp.
Bcdard. D; Pctro. S. (1995) Laboratory sediment bioassay report on Whcatlcy Harbour sediments 1992. Ontario
Ministry of Environment and Energy.
Bcdard. D; Pctro. S. (1997) Laboratory sediment bioassay report on St. Man 's River sediments 1992 & 1995.
ISBN 0-7778-6674-9; Standards Development Branch. Ontario Ministry of Environment and Energy.
Bcdard. D; Pctro. S. (1997) Laboratory sediment bioassay report on Upper St. Clair River sediments in the vicinity
of industrial point sources 1994 & 1995. St. Clair River remedial action plan.
Bcdard. D; Pctro. S. (1998) Laboratory sediment bioassay report on Lyons Creek sediments. ISBN -7778-7195-5;
Ontario. Ministry of the Environment.
Bennett. J: Cubbagc. J. (1992) Copper in sediments from Stcilacoom Lake. Picrcc County. Washington.
Washington State Department of Ecology. Environmental Investigations and Laboratory Sen ices Program.
Olympia. WA.
Bennett. J: Cubbagc. J. (1992) Effects of polycyclic aromatic hydrocarbons (PAHs) in sediments from Lake
Washington on freshwater bioassay organisms and bcnthic macroiin crtcbratcs. New York State Department of
Environmental Conservation.
Borgmann. U: Norwood. WP. (1997) Identification of the toxic agent in metal-contaminated sediments from
Manitouwadgc Lake. Ontario, using toxicity-accumulation relationships in Ilyolcllo aztcco. Can J Fish Aqua
Science 54.
Bortlcson. G. (1994) Sediment-quality assessment of Franklin D. Roosevelt Lake and the upstream reach of the
Columbia River. Washington. 1992. Personal Comm. Gil Bortlcson. USGS. Tacoma. WA.
162
-------
Bowser. M. (1992) Sediment sampling and testing. Eric Harbor. Pennsylvania. Prepared for U.S. Army Corps of
Engineers. Buffalo District.
Brady. MD. (1994) Baseline sediment characterization Columbia Aluminum Company. Goldcndalc. Washington.
ENSR Consulting and Engineering.
Burton. D. (1994) Sediment sampling and analysis plan - West Branch Grand Calumet River. U.S. EPA.
Department of Justice.
Call. DJ; Balccr. MD; Brooke. LT; ct al. (1991) Sediment quality evaluation in the Lower Fox River and Southern
Green Bay of Lake Michigan. Center for Lake Superior Environmental Studies. University of Wisconsin-Superior.
Superior. WI.
Carter. JA; Mroz. RE; Tay. K.L; ct al. (1998) An evaluation of the use of soil and sediment bioassays in the
assessment of three contaminated sites in Atlantic Canada. Water Qual Res J Canada.
Crane. JL; Schubaucr-Bcrigan. M; Schmudc. K. (1997) Sediment assessment of hotspot areas in the
Duluth/Supcrior Harbor. Minnesota Pollution Control Agency.
Cubbagc. J. (1992) Survey of contaminants in sediments in Lake Union and Adjoining Waters (Salmon Bay. Lake
Washington Ship Canal, and Portage Bay). Urban Bay Action Team. Northwest Regional Office. Washington State
Department of Ecology.
Dickson. K.L; Waller. WT; Kennedy. JH; ct al. (1989) A water quality and ecological survey of the Trinity River.
University of North Texas and University of Texas. Prepared for City of Dallas Water Utilities.
Downing. J; Fairey. R; Roberts. C; ct al. (1998) Chemical and biological measures of sediment quality in the
Central Coast Region: final report. California State Water Resources Control Board.
Earth Tech Inc. (1998) Phase II. Flat River sediment investigation. Washington State Department of Ecology.
Water Body No. WA-08-9350 (Segment No. 04-08-001).
EG & G. Bionomics. (1983) An assessment of the chemical and toxicological properties of dredged sediments
collected from Cleveland Harbor. Ohio. Prepared for U.S. Army Engineering District. Buffalo.
ENSR Consulting and Engineering. (1994) Aluminum Company of America: Vancouver Works baseline sediment
characterization. ENSR Consulting and Engineering. Document # 0225-006-405.
EVS Environment Consultants. (1997) Risk assessment of Sheboygan River. Wisconsin. 1997. Prepared for
National Oceanic and Atmospheric Administration.
FishPro Engineering and Environmental Consultants. (1991) biological report on sediment and water bioassays and
bcnthic community determination at UnimarYard 1 dry dock facility.
Garabcdian. B; Estabrook. F; Ncudcrfcr. G: ct al. (1997) Oswego River sediment study: Final report - Summary of
May 23. June 28 and September 13-15. 1994 Results. Prepared for Department of the Army. Buffalo District.
Gcraghty & Miller. (1997) Confirmation sampling results. Green Bay Paint sludge site. Menominee. Michigan.
Gcraghty & Miller Inc. Environmental Sen ices. Michigan Department of Environmental Quality.
Hall. LW. Jr; Zicgcnfuss. MC: Fischer. SA; ct al. (1992) A pilot study for ambient toxicity testing in Chesapeake
Bay. Year 2 report. Prepared for the U.S. Environmental Protection Agency for the Chesapeake Bay Program.
Prepared by the University of Man land System. Queensland. Man land.
Hefty. LN. (1998) Toxicity testing using Ihale/la azteca. Department of the Army. Buffalo District. Corps of
Engineers.
163
-------
Hunt. J: Anderson. B; Phillips. B; ct al. (1998) Sediment quality and biological effects in San Francisco Bay: final
report. California State Water Resources Control Board.
Ingcrsoll. CG: Buckler. DR; Crccclius. EA; ct al. (1992) Draft final report : Biological assessment of contaminated
Great Lakes sediment. Prepared for the US EPA GLNPO Assessment and Remediation of Contaminated Sediments
(ARCS) Project.
Ingcrsoll. CG: Brumbaugh. WG: Farag. AM; ct al. (1993) Effects of mctal-contaminatcd sediment, water, and diet
on aquatic organisms. Final report for the USEPA Milltown cndangcrmcnt assessment project. NTIS. PB93-21529.
NFCRC-U.W. Springfield. VA.
Ingcrsoll. CG: Havcrland. P: Brunson. E; ct al. (1996) Calculation and evaluation of sediment effect concentrations
for the amphipod Ilyalella aztcca and the midge ('hironomus riparius. Prepared for the US EPA GLNPO
Assessment and Remediation of Contaminated Sediments (ARCS) Project.
Ingcrsoll. CG: Brunson. EL; Dwycr. FJ; ct al. (1998) Sublethal endpoints in sediment tests with Ilyalella aztcca.
Environ Toxicol Chcm 17:1508-1593.
Jaagumagi. R: Pctro S. (1992) An in-placc pollutants study of the Otonabcc River and Rice Lake. Queens Printer
forOntario. ISBN 0-7729-8010-1.
Jaagumagi. R: Bcdard. D. (1997) Sediment and biological assessment of Canagagiguc Creek at the Uniroyal
Chemical Ltd. Plant. Elmira. Ontario 1995-96. ISBN 0-7778-6938-1: Environmental Monitoring and Reporting
Branch and Standards Development Branch. Ontario Ministry of Environment and Energy.
Johnson. A. (1992) Review of metals, bioassay. and macroim crtcbratc data from Lake Roosevelt bcnthic samples
collected in 1989. Department of Ecology. State of Washington.
Johnson. A; Norton. D. (1988) Screening survey for chemical contaminants and toxicity in sediments at five lower
Columbia River ports. September 22-24. 1987. Washington State Department of Ecology.
Johnson. A; Norton. D. (1989) Screening survey for chemical contaminants and toxicity in drainage basins at Paine
Field. August 10-12. 1987. Washington State Department of Ecology.
Johnson. A; Norton D: Yakc. B. (1989) An assessment of metals contamination in Lake Roosevelt. Washington
State Department of Ecology. Toxics Invcstigations/Groundwatcr Monitoring Section. Olympia. WA.
Kcmblc. NE: Burton. EL; Canficld. TJ; ct al. (1997) An assessment of sediments from the Upper Mississippi
River. Final report - June 1997. EPA 823-R-97-005.
Kcmblc. NE; Hardcsty. DK; Ingcrsoll. CG: ct al. (2000) An evaluation of the toxicity of contaminated sediments
from Waukcgan Harbor. Illinois following remediation. Arch Environ Contain Toxicol 39:451-462.
Kolok. AS: Plaisancc. EP; Abdclghani. A. (1998) Individual variation in the swimming performance of fishes: an
overlooked source of variation in toxicity studies. Environ Toxicol Chcm 17 (2).
Kosmond. LD. (1996) Lower Fox River system sediment characterization: Sediment quality triad assessment and
application of sediment quality guidelines. Wisconsin Department of Natural Resources. Sediment Management
and Remedial Techniques Team.
Krant/bcrg. G. (1994) Spatial and temporal variability in metal bioavailability and toxicity of sediment from
Hamilton Harbour. Lake Ontario. Environ Toxicol Chcm 13(10): 1685-1698.
Landau Associates. (1993) Mill Creek and East Drain sediment sampling and analysis report, western processing,
phase II (Draft Report). Landau Associates.
164
-------
Lcppancn. CJ. (1997) Effects of North Hollywood dump leaching on Wolf River sediment toxicity, sediment
quality and bcnthic macroinvcrtcbratc community. University of Memphis.
Malucg. KW; Schuytcma. GS: Krawc/yk. DF: et al. (1984) Laboratory sediment toxicity tests, sediment chemistry
and distribution of henthic macroinvertehrates in sediments from the Keweenaw Waterway. Michigan. Environ
Toxicol Chcm 3:233-242.
Malucg. KW; Schuytcma. GS: Gakstattcr. JH; ct al. (1984) Toxicity of sediments from three metal-contaminated
areas. Environ Toxicol Chcm 3:279-291.
McGcc. BL; Schlckat. CE; Rcinharz. E. (1993) Assessing sublethal levels of sediment contamination using the
cstuarinc amphipod Leptocheirusplumulosus. Environ Toxicol Chcm 12:577-587 (E).
Minnesota Pollution Control Agency. (1993) Toxicity test results of sediment samples collected in 1993 from the
Intcrlakc/Duluth Tar and USX Supcrfund sites.
New York State Department of Environmental Conservation (NYSDEC). (1995) Phase II - final report: Lower
Genesee River study: Summary of 1992. 1993 and 1994 results. U. S. Army Corps of Engineers Buffalo District.
U. S. EPA. University of Wisconsin-Superior. AScI Corporation.
Pcthybridgc. AH. (1981) Sediment quality testing at Ogdcnsburgh Harbor. New York. Technical Report #10175-
04. Prepared for U.S. Army Engineer District. Buffalo.
Phelps. HL. (1993) Sediment toxicity of the Anacostia River estuary. Washington. DC. Bull Environ Contain
Toxicol 51.
Pope. RJ. (1993) Analysis of bcnthic invertebrates from Algoma Slip Core Samples. Water Resource Branch. The
Ontario Ministry of the Environment and Energy.
Prairie. R: Mckcc. P. (1992) Sediment quality assessment near two Ontario mine sites: How relevant arc the
provincial guidelines? Paper presented at the International Land Reclamation and Mine Drainage Conference and
the Third International Conference on the Abatement of Acidic Drainage.
PTI Environmental Sen ices. (1992) McCormick & Baxter crcosoting company. Remedial investigation report.
Volume IV of IV. Prepared for Oregon Department of Environmental Quality. Portland. Oregon.
Rcdsikc. R: Fahncnsticl. G: Meier. P: ct al. (1997) Preliminary investigation of the extent and effects of sediment
contamination in White Lake near the Whitehall leather tannery. U.S. Environmental Protection Agency. National
Oceanic and Atmospheric Administration.
REMAP 95. (1997) Ten-day toxicity tests with llyalella azteca and ('hironomus teutons on sediments from the St.
Louis River area of concern 1995 R-EMAP test results batch #1-7. Minnesota Pollution Control Agency.
Monitoring and Assessment Section.
REMAP 96. (1997) Ten-day toxicity tests with llyalella azteca and (hironomus tentans on sediments from the St.
Louis River area of concern 1996 R-EMAP test results batch #1-7. Minnesota Pollution Control Agency.
Monitoring and Assessment Section.
Simonin. HA; Paul. E: Ncudcrfcr. G: ct al. (1996) The effects of long-term copper sulfate use on the toxicity of
lake sediments. Presented at Socictv of Environmental Toxicologv and Chcmistrv Annual Meeting. Washington.
DC.
Smith. JA: Glowacky. RS: Crcrar. PJ: ct al. (1984) Analysis of sediment from Toledo Harbor - Maumcc River.
Aqua Tech Environmental Consultants. Inc. Department of the Army. Buffalo District. Corps of Engineers.
165
-------
Smith. JA; Glowacky. RS: Crcrar. PJ; ct al. (1984) Analysis of sediment. Oak Orchard Harbor. Oak Orchard. New
York. Aqua Tech Environmental Consultants. Inc. Prepared for Department of the Army. Buffalo District. Corps of
Engineers.
Smith. JA; Glowacky. RS: Crcrar. PJ; ct al. (1985) Analyses of Sediments from Huron Harbor. Huron. Ohio.
Prepared for Department of the Army. Buffalo District. Corps of Engineers. Prepared by Aqua Tech Environmental
Consultants. Inc.
Smith. JA; Crcrar. PJ; Glowacky. RS; ct al. (1985) Analysis of sediment from Sandusky Harbor. Sandusky. Ohio.
Aqua Tech Environmental Consultants. Inc. Prepared for Department of the Army. Buffalo District. Corps of
Engineers.
Smith. JA; Crcrar. PJ; Glowacky. RS; ct al. (1985) The analyses of sediments from Conncaut Harbor. Conncaut.
Ohio. Aqua Tcch Environmental Consultants. Inc. U.S. Army Corps of Engineers. Buffalo District.
Smith. JA; Glowacky. RS; Crcrar. PJ; ct al. (1986) The analyses of sediments from West Harbor. Ottawa County.
Ohio. Prepared for Department of the Army. Buffalo District Corps of Engineers. Prepared by Aqua Tcch
Environmental Consultants. Inc.
Smith. DL; Talbot. LM; Campbell. CK. (1992) Contaminated sediment bioassay (toxicity tests) study of Wisconsin
Great Lakes coastal harbors and tributaries, draft.
Smith. JA; Glowacky. RS; Crcrar. PJ; ct al. (1996) The analyses of sediments from Fairport Harbor. Prepared for
Department of the Army. Buffalo District. Corps of Engineers. Prepared by Aqua Tcch Environmental Consultants.
Inc.
Sprcngcr. MD; Kracko. K. (1992) Final report for supplemental environmental investigation. Cannclton Industries
site. Sault Stc. Marie. MI. U.S. Environmental Protection Agency.
Tctra Tcch. (1994) Lower Columbia River backwater reconnaissance survey Lower Columbia River bi-statc
program.
T.P. Associates International. Inc. (1987) The analyses of sediments from Oswego Harbor. Minnesota Pollution
Control Agency.
T.P. Associates International. Inc. (1987) The analyses of sediments from Oak Orchard Harbor. Prepared for
Water Quality Section. U.S. Army Engineer District. Buffalo.
T.P. Associates International. Inc. (1987) The analyses of sediments from Olcott Harbor. Prepared for Department
of the Army. Buffalo District. Corps of Engineers.
T.P. Associates International. Inc. (1988) The analyses of sediments from Wilson Harbor. Prepared for U.S. Army
Engineer District. Buffalo.
U. S. Army Engineer District. Chicago. (1994) Clean Water Act Section 404 (b) (1) Contaminant Determination
for Bums Waterway Harbor. U. S. Army Engineer District. Chicago.
U.S. EPA (Environmental Protection Agency). (1996) Assessment of sediment in the Indiana Harbor area of
concern.
U.S. EPA. (1997) Field sampling methodologies: Clinton River. MaComb County. Michigan (Stations CL9716-
CL9727). U.S. EPA. Great Lakes National Program Office.
Wade. TL; Vclinsky. DJ; Rcinharz. E; ct al. (1994) Tidal River sediments in the Washington. D.C. area. II.
Distribution and sources of organic contaminants. Estuarinc Research Federation.
166
-------
Webster. CD; McLaughlin. JC: Wan. L; et al. (1995) A report of the whole sediment toxicity of 12 sites in the
Maumcc area of concern to Ihale/la ozteca. Ohio EPA.
WENCK. Associates. Inc. (1995) Harbor sediment sampling documentation report. Prepared for Lakchcad Pipe
Line Company.
167
-------
APPENDIX B: SEDIMENT TOXICITY DATABASE (SEDTOX02) STRUCTURE
The following tables describe the structure for the database management system
developed for the SEDTOX02 database (NOAA, 2004). Above each table is the name of the
database table and a brief description of the purpose. Following the database table name is a
"key" that describes how unique records in the table are defined. The SEDTOX02 database is
divided into separate databases with identical structure for marine/estuarine and freshwater data.
Site: Table to define general location
Key: Siteid
FIELD NAME
TYPE
WIDTH, DEC
DESCRIPTION
SITEID
Char
4
Site identifier
SITENAME
Char
40
Descriptive name for site
EPAREGION
Num
2,0
Region for site location; 11 for Canada
Study: Provides basic information regarding the study (e.g., name, contact, etc.).
FIELD NAME
TYPE
WIDTH, DEC
DESCRIPTION
SITEID
Char
4
Site identifier
STUDYID
Char
2
Study identifier
STUDYNAME
Char
40
Short name of study
CONTACT
Char
40
Contact source for data
SEDCHEM
Logical
1
Sediment chemistry data, Y or N?
SEDTOX
Logical
1
Sediment toxicity data, Y or N?
Studynot: Primarily a table for descriptive notes regarding the study design and method for
recording data in the database. Information may include how replicates are recorded and any
chemical sums calculated.
FIELD NAME
TYPE
WIDTH, DEC
DESCRIPTION
SITEID
Char
4
Site identifier
STUDYID
Char
2
Study identifier
NOTES
Memo
10
Memo field data processing notes
Studyref: Contains information regarding the document that describes the study and data.
FIELD NAME
TYPE
WIDTH, DEC
DESCRIPTION
SITEID
Char
4
Site identifier
STUDYID
Char
2
Study identifier
YEAR
Char
4
Year of publication
AUTHORS
Char
160
Authors, if published
TITLE
Char
160
Title, if published
SOURCE
Char
160
Citation or agency, if published
168
-------
Station: Listing of stations for sediment chemistry, tissue chemistry, and bioassay sample
collections.
Key: Siteid+Studyid+Stationid
FIELD NAME
TYPE
WIDTH, DEC
DESCRIPTION
SITEID
Char
4
Site identifier
STUDYID
Char
2
Study identifier
STATION ID
Char
6
Station identifier
LATITUDE
Num
12, 8
Latitude in decimal degrees, NAD83
LONGITUDE
Num
13, 8
Longitude in decimal degrees, NAD83
EST STN
Char
8
How coordinates were established
Sample: Sediment sample collection informat
depth (in centimeters). Field sample lab replic
Key: Siteid+Studyid+Stationid+Sampleid+La
on, including station id, sample date, sample
ates are treated as separate samples.
? rep
FIELD NAME
TYPE
WIDTH, DEC
DESCRIPTION
SITEID
Char
4
Site identifier
STUDYID
Char
2
Study identifier
STATION ID
Char
6
Station identifier
SAMPLEID
Char
2
Sample identifier
LABREP
Char
2
Lab replicate number
SAMPDATE
Char
8
Date sample collected as YYYYMMDD
SAMPTIME
Char
5
Time sample collected
UDEPTH
Num
8,2
Top depth of sample from sed/water
interface in cm
LDEPTH
Num
8,2
Bottom depth of sample from sed/water
interface in cm
TOC
Num
6,2
Total organic carbon as percent
PCTFINES
Num
6,2
Percent fines
UAN PW
Num
10, 4
Unionized ammonia in porewater
H2S PW
Num
10, 4
Hydrogen sulfide in porewater
EXSAMPID
Char
12
Investigator's sample identifier
169
-------
Chem: Chemistry data associated with surface sediment samples.
Key: Siteid+Studyid+Stationid+Sampleid+La
?rep+Chemcode
FIELD NAME
TYPE
WIDTH, DEC
DESCRIPTION
SITEID
Char
4
Site identifier
STUDYID
Char
2
Study identifier
STATION ID
Char
6
Station identifier
SAMPLEID
Char
2
Sample identifier
FIELDREP
Char
2
Field replicate number
LABREP
Char
2
Lab replicate number
CHEMCODE
Char
10
Code for parameter name
CONC
Num
12,5
Measured concentration
QUALCODE
Char
5
Assigned qualifier for concentration
UNITS
Char
6
Units of concentration for parameter
MEASBASIS
Char
2
Wet (WW) or dry weight (DW) indication
MISSINGVAL
Logical
1
Data missing, Y or N?
Biosumm: Mean results (of replicate data) for sediment bioassay.
Key: Siteid+Studyid+Stationid+Sampleid+Testid
FIELD NAME
TYPE
WIDTH, DEC
DESCRIPTION
SITEID
Char
4
Site identifier
STUDYID
Char
2
Study identifier
STATION ID
Char
6
Station identifier
SAMPLEID
Char
2
Sample identifier
TESTID
Char
12
Bioassay test code
FIELDREP
Char
2
Field replicate number
GROUP
Char
2
Sample grouping
SERIES
Char
2
Bioassay test series number
EFFECTVAL
Num
7,2
Measured effect value
SIGEFFECT
Logical
1
Was effect significant, Y or N?
NEG
Logical
1
Negative control sample, Y or N?
REF
Logical
1
Reference sample, Y or N?
STAT
Logical
1
Used for statistical comparison, Y or N?
CTRLADJ
Num
6,2
Control-adjusted effect value
SIG ORIGIN
Char
2
Code for toxic sample used by original study
170
-------
Bmaster: Descriptive notes associated with bioassay tests where available.
Key: Siteid+Studyid+Testid
FIELD NAME
TYPE
WIDTH, DEC
DESCRIPTION
SITEID
Char
4
Site identifier
STUDYID
Char
2
Study identifier
GROUP
Char
2
Sample grouping based on spatial or
temporal
TESTID
Char
12
Bioassay test code
SPIKED
Logical
1
Sediment spiked with contaminant, Y or N?
TESTCOMM
Memo
10
Bioassay test comments
Qualify: Defines qualifiers used with chemical data
Key: Siteid+Studyid+Qualcode
FIELD NAME
TYPE
WIDTH, DEC
DESCRIPTION
SITEID
Char
4
Site identifier
STUDYID
Char
2
Study identifier
QUALCODE
Char
5
Assigned qualifier code for concentration
QUALIFIERS
Char
30
Qualifiers used in original study
DESCRIPT
Char
80
Description of qualifiers used in study
Chemdict: Provides a unique list of chemical names and associated chemical codes that are used
in the Chem table.
Key: Chemcode
FIELD NAME
TYPE
WIDTH, DEC
DESCRIPTION
CHEMCODE
Char
10
Code for parameter name
CHEMNAME
Char
40
Full chemical name
CHEMCLASS
Char
8
Chemical classification
CATEGORY
Char
8
Alternate chemical classification
SUBCATGY
Char
10
Subclassification for alternate chemical
class
CHEMTOTAL
Char
10
Classification used for totaling chemicals
MOLWT
Num
7, 3
Molecular weight of chemical
CASNUM
Char
24
CAS number
UNITS
Char
6
Units of concentration for chemical
(sed/tiss)
WAJJNITS
Char
6
Units of concentration for chemical in water
media
171
-------
Testdict: Provides a unique list of bioassay tests and associated test codes that are used in the
Biosumm and Bmaster tables.
Key: Testid
FIELD NAME
TYPE
WIDTH, DEC
DESCRIPTION
TESTID
Char
12
Bioassay test code
MEDIUM
Char
15
Medium tested
MEDCODE
Char
2
Code for medium tested
GROUP
Char
20
Bioassay species grouping
ALTGROUP
Char
20
Alternate bioassay species grouping
SPECIES
Char
40
Bioassay organism
SPPCODE
Char
3
Code for bioassay organism
LHS
Char
10
Life history stage of bioassay organism
LHSCODE
Char
1
Code for life history stage
ENDPOINT
Char
30
Bioassay test endpoint
ENDCODE
Char
2
Code for test endpoint
DURATION
Char
10
Duration of test
DURCODE
Char
4
Code for test duration
172
-------
APPENDIX C: DATA ACQUISITION SCREENING METHODS
The SEDTOX02 database was developed to support the development and assessment of
numerical sediment quality guidelines. The database is composed of matching (i.e., synoptically
collected) sediment chemistry and laboratory toxicity data from freshwater, estuarine, and
marine sites. The following screening criteria provided a means of evaluating candidate data sets
and ensuring general consistency in the information included in the database. The evaluation
criteria were based on ASTM (2002a, b), Environment Canada (1998a, b), and U.S. EPA (2000).
However, the screening criteria are not necessarily recommended for applications beyond their
intended purpose. Data from spiked-sediment bioassays were not included in the database.
A. Approach for Evaluating Data Set Acceptability
1. Data sets must contain synoptically collected sediment chemistry and biological effects
data. That is, the sediment samples for biological and chemical testing must have been
collected from the same location and at the same time.
2. Data sets may contain any number of sediment samples, provided that there is at least one
nontoxic sample. Preference should be given to data sets that contain >5 samples.
3. Data sets that include toxicity and chemistry data generated on sediment samples from
any sediment horizon (i.e., surficial sediments, cored sediments, etc.) should be
considered to be acceptable provided that the sediment chemistry and toxicity data are
matching (i.e., for the same sediment samples).
4. Data sets that include toxicity and chemistry data generated on composite sediment
samples should be considered to be acceptable. Preference should be given to data sets
that composite sediments over limited geographic areas.
5. Data sets that include data generated from dilution series of bulk sediments and/or
porewater are not acceptable for incorporation into the database (however, data from the
100% dilution are acceptable).
6. It is not essential that data sets include the coordinates (i.e., longitude and latitude) of the
sampling site along with the chemical and biological data. However, these data will be
included as available.
7. It is not essential that data sets have a minimum range of chemical concentrations in the
sediment samples (i.e., the 10-fold criteria that was used previously to assess data
acceptability is no longer required).
173
-------
8. Data sets should be preferentially included if full sediment chemistry has been conducted
(i.e., metals, PAHs, PCBs, and pesticides). Data sets with incomplete sediment chemistry
may also be included, particularly if prior knowledge indicates that certain chemicals are
unlikely to occur at the site.
9. Analytical detection limits should be below the respective ERLs or TELs for each
chemical analyte.
10. The chemical analytical methods used in the study must be reported and should meet
minimum data quality requirements/objectives (i.e., the precision, accuracy, and detection
limits must be reported; data quality may be evaluated using various protocols and best
professional judgement; and the rationale for decisions regarding data acceptability must
be documented).
11. Concentrations of SEM metals (e.g., Cd, Cu, Pb, Ni, Zn) may be included in the database,
provided that data on the concentrations of total metals (i.e., strong acid digestion) are
also available.
12. Information on the environmental conditions in the bioassay chambers should be
captured in the database, including data on DO, pH, salinity, water hardness, temperature,
NH3, and H2S.
13. Chemistry data that were generated using atypical methods (e.g., X-ray fluorescence for
metals, screening methods for PAHs, etc.) are not acceptable for inclusion in the
database.
14. Data from elutriate tests must not be included in the database because there is too little
connection between the chemistry and the laboratory toxicity data.
15. Data sets generated using organic extracts may be included in the database (e.g.,
Microtox), provided they are available with other toxicity data. Microtox and Mutatox
data are not being targeted for inclusion in the database because the linkage between the
toxicity data and effects on sediment-dwelling organisms is tenuous.
16. Acceptable environmental conditions must be maintained throughout the toxicity test (as
defined in the protocol for the toxicity test). Consequently, the temperature, pH,
hardness, conductivity, salinity, and DO of the overlying water should have been
measured during the test. If these variables have been measured but not reported, it is
reasonable to assume that the conditions during the test were not acceptable and
additional information should be obtained from the investigators.
17. The responses of the test organisms exposed to negative controls must be reported and
must be within acceptable limits (i.e., as defined in ASTM standard methods). For
toxicity tests for which a negative control sediment is not available, the selected field
reference sediment must be shown to be functionally equivalent to a negative control
sediment, as indicated by nontoxicity (as defined above); concentrations of measured
contaminants should not exceed their respective TELs or ERLs; and the levels of particle
174
-------
size distribution, pH, Eh, salinity, and TOC must be similar to those in the basin area
under investigation.
B. Considerations for Prioritizing Data Sets
1. The procedures used for collecting, handling, and storing sediments should be consistent
with the protocols that have been established by ASTM. Generally, higher priority for
inclusion in SEDTOX should be assigned if
(i) surficial sediments were collected and tested;
(ii) the sediments were tested within 8 weeks of collection (some flexibility in
applying this criterion is warranted, as similar bioassay responses have been
observed up to 1 to 2 years after sediment collection; and,
(iii) the sediments were not frozen prior to biological testing.
2. Data sets should be preferentially included if full sediment chemistry has been conducted
(i.e., metals, PAHs, PCBs, and pesticides). Data sets with incomplete sediment chemistry
may also be included, particularly if prior knowledge indicates that certain chemicals are
unlikely to occur at the site.
3. Data on all test organisms and endpoints for which standard toxicity testing methods are
available should be captured in the database, as available. However, higher priority
should be given to data sets that include one or more of the following tests/endpoints:
• marine amphipod (Ampclisca and Rhepoxynius) survival;
• marine sea urchin (Arhacia and Slrongy/occnlroliis) fertilization;
• freshwater amphipod (Hyalella) survival, growth, and reproduction; and,
• freshwater midge (Chironomus riparius and C. Icnlans) survival and growth.
The data on other species and associated endpoints should be captured in the database
when available, along with data on the high-priority toxicity tests.
4. Priority should be given to data sets that test surficial sediments and do not composite
samples over large geographic areas.
C. Considerations Related to Preparing Data for Import
1. Sediment samples from the same study for which inconsistent chemistry data are
available (i.e., metals only for some samples and complete chemistry for other samples)
should be grouped separately. The portion of the data set with complete sediment
chemistry should be preferentially included in the database. The portion of the data set
with incomplete chemistry should be considered as a separate data set.
175
-------
2. Detection limits (DLs) should be reported for all measured chemicals (i.e., for all analytes
for which the concentration in one or more samples is reported to be < DL). Below-DL
values must be treated as missing data if the DL has not been reported.
3. Calculations of tPAH, tHMW-PAH, or tLMW-PAH will be conducted using subroutines
in the database. Previously calculated values will not be incorporated into the database.
LMW-PAHs are considered to include the following two- and three-ringed substances:
naphthalene, acenaphthene, acenaphthylene, fluorene, phenanthrene, and anthracene.
HMW-PAHs are considered to include the following four- and five-ringed substances:
fluoranthene, pyrene, benz(a)anthracene, chrysene, benzo(a)pyrene, benzofluoranthene,
dibenzo (a,h)anthracene, and benzo(g,h,i)perylene.
4. The reported concentrations of tPCBs should be treated as equivalent, regardless of
which method was used to determine the levels. If tPCB concentrations were not
reported, the value should be calculated using an appropriate method (e.g., sum of
detected Aroclors or sum of detected congeners, etc.). The method that was used in the
determination should be recorded in the database.
5. The reported concentrations of tDDTs should be treated as equivalent, regardless of
which method was used to determine the levels, provided that the p,p'-isomers of DDT,
DDE, and DDD were measured. The method that was used in the determination should
be recorded in the database.
6. The reported concentrations of tPCDDs/PCDFs should be treated as equivalent,
regardless of which method was used to determine the levels, provided that the most
toxic substances were measured (i.e., TCDD, HCDD, TCDF, HCDF). The method that
was used in the determination should be recorded in the database.
7. The data on the levels of H2S and NH3 in the replicate bioassay chambers will be
summarized on a per-sample basis and included in the database. For these variables, all
of the measurements should be treated as equivalent, regardless of the analytical methods
that were used.
8. Data from tests conducted with different types of media (i.e., porewater vs. organic
extracts for Microtox) should be treated separately in the data analyses.
9. Sediment samples should be grouped in a consistent manner to facilitate data analyses,
specifically:
(i) Samples that were collected from the same area within the same year should
be grouped together.
(ii) Samples that were collected from the same area in different years should be
separated into groups, based on the year that the samples were collected.
(iii) In general, samples that were collected within a single study that was
conducted during one year should be grouped together; however, it may be
176
-------
necessary to create separate groupings for the samples by basin area. For
example, EMAP data will be grouped by basin area using maps of
appropriate scale.
10. The toxicity of sediment samples from the basin area under investigation should be
determined on the basis of statistical comparisons with the negative control. For bulk
sediments and porewater, negative control sediments may be obtained from a suitable
reference site(s), as specified in the ASTM (2002b) standard methods (e.g., >80%
survival and full chemistry) and other relevant information (i.e., contaminant
concentrations < ERLs).
11. Sediment samples must be designated as toxic or nontoxic using the results of statistical
analyses. Negative control data should also be provided for each batch of samples to
facilitate the determination of toxicity on the basis of minimum significant differences
from the negative control responses.
12. In tests that are designed to evaluate effects on growth and/or reproduction, samples will
be treated as though toxic for these endpoints if a significant effect on survival was
determined for that sample (i.e., even if it was not possible to measure growth or
reproduction directly or even if it was possible to measure effects on survivors and no
effects were observed).
References Cited
ASTM (American Society for Testing and Materials). (2002a) Standard guide for conducting
10-day, static sediment toxicity tests with marine and estuarine amphipods. E 1367-99. In:
Annual book of standards. Vol. 11.05. West Conshohocken, PA, pp. 693-719.
ASTM. (2002b) Standard test method for measuring the toxicity of sediment-associated
contaminants with freshwater invertebrates. E1706-00B. In Annual book of standards. Vol.
11.05. West Conshohocken, PA, pp. 1125-1241.
Environment Canada. (1998a) Biological Test Method: Test for growth and survival in
sediment using the freshwater amphipod Hyalclla azlcca. Environment Canada, Ottawa,
Ontario. Technical report number EPSl/RM/33.
Environment Canada. (1998b) Biological Test Method: Test for growth and survival in
sediment using larvae of freshwater midges (Chironomus teutons or Chironomus riparius).
Environment Canada, Ottawa, Ontario, Technical report number EPS1/RM/32.
U.S. EPA (Environmental Protection Agency). (2000) Methods for measuring the toxicity and
bioaccumulation of sediment-associated contaminants with freshwater invertebrates, second
edition. EPA 600/R-99/064, Washington, DC.
177
-------
APPENDIX D: DATA EVALUATION METHODS
SCREENING CRITERIA FOR BEDS/SEDTOX
CO-OCCURRENCE DATA
*Must be Present
Reference: Reference Number:
1 .* Does data set contain matching sediment chemistry and biological effects (i.e., biological
and chemical data collected from the same location at the same time)?
NO UNACCEPTABLE
YES Page Reference(s):
2.
What is the location of sampling site(s)?
Collection Date? Page reference for site
description?
->
J) .
Freshwater Estuarine
Marine
Salinity
I. SEDIMENT CHEMISTRY
4.* Is there at least one nontoxic sample?
NO UNACCEPTABLE
YES Number of nontoxic: Number of toxic:
5. Was bioassay conducted on unique or composite samples.
Number of replicates? Size of composite area?
6. What chemistry data have been collected? (i.e., metals, PAHs, pesticides, pH, DO, TOC)
Metals PCBs pH TOC
PAHs Pesticides DO AYS
7. Are detection limits below the respective ERLs or TELs?
NO UNACCEPTABLE YES
8. Are total metal concentrations measured?
NO SEM metals may not be included.
YES SEM metals may be included.
178
-------
9.
Collection instrument
Sediment depth
10.* What type of sediment was used:
Bulk Sediment Porewater
(Sediment with pore water)
(Kxtract porewater from
sediments and expose
water column species)
OTHER TOXICITY DATA
Organic Extract
(Sediment extracted with
organic solvent and expose
liquid form)
ALSO NEEDED
Elutriate
(Sediments with water, mixed,
settled and exposed water
column species)
UNACCEPTABLE
11. What type of toxicity test was conducted? Length of test?
Static (Water, sed-no change)
Static Renewal (Water, sed-some water change)
Flow-Through (Water, sed-water flowing through)
12.* Are appropriate analytical procedures used to determine total concentrations of the
analytes in bulk sediment samples? What method(s) was used?
(Metals: partial digestion, analysis of elutriates or extracts are unacceptable.)
13. 1 s a di 1 ution series used?
NO YES UNACCEPTABLE
14.* Are measured dry weight contaminant concentrations reported? Conversion from wet
weight to dry weight concentration may occur ONLY if data on moisture or TOC are
provided. Nominal concentrations are unacceptable.
NO UNACCEPTABLE YES Page reference(s):
II. BIOEFFECTS
15.*
a Do toxicity tests employ appropriate laboratory procedures? (ASTM: El367, E1611,
El 706)
NO UNACCEPTABLE YES
b Have the following been recorded during testing?
Temperature ; pH ; Hardness ; Conductivity ;
Salinity ; DO ; Alkalinity ; Ammonia .
c Does DO Remain above 60% Needed for Marine
40% Needed for Fresh Water
NO UNACCEPTABLE
d Temperature
179
-------
e Is temperature within natural range, fluctuate less then 3 C, and have a time-weighted
average within 1 C of selected temp?
YES NO UNACCEPTABLE Range
f Do hardness, alkalinity, pH, or ammonia vary by more than 50% (for freshwater
samples)?
NO YES UNACCEPTABLE
Range: DO ; Alk ; pH ; NH^
g Have salinity levels in porewater been adjusted (for marine samples)?
NO YES UNACCEPTABLE Range
h List procedure reference(s) or brief details:
16.* Were biological responses compared to the control or reference sites?
List the Control and Reference sites? Positive Control .
reference = uncontaminated site within the same waterbody or watershed; control =
uncontaminated site outside the tested water body
17.*
a Have sediment samples used for biological testing been frozen?
NO YES If yes, both biological and chemical testing must be performed after
thawing sediments.
b Have sediment samples been stored for more than eight (8) weeks prior to biological
testing?
NO YES UNACCEPTABLE
What was the holding time? .
c Are appropriate procedures used for collecting, handling, and storage of sediments?
NO YES List procedures reference(s) or brief details:
18. Identify species used in toxicity testing. Identify organism sources.
19. What life stage were the test species at the start of the test?
(Hyalella azteca 7-14 day old; Chironomus tentans third-instar larvae; Chironomus
riparius second instar or younger; Daphnia magna 5 days old; Ceriodaphnia dubia <24h
old; Hexagenia spp. 3-4 months old; Tubifex tubifex adult; Diporeia spp. juveniles)
20. Organism acclimation time
180
-------
21. What percentage of the control survived?
Mean range 70% for (Chironomus riparius, Chironomus tentans )
80% for (Hexagenia spp., Daphnia magna, Ceriodaphnia dubia,
Hyalella azteca)
90% for (Diporeia spp., Tubifex tubifex, Polychaetous annelids, marine amphipods, others)
NO UNACCEPTABLE
22. Reference Samples
Survival %
Cone, less than TEL and ERLs? YES NO
Grain size % sand % silt % clay
23. Benthic Community Analysis
a Is there a benthic community abundance analysis?
NO YES List taxa (e.g., amphipod, sponges,...) upon which the analysis
focuses:
b* Do all of the sites within a sampling area have the same general characteristics (i.e., same
depth of overlying water, same salinity in overlying water, etc)?
NO UNACCEPTABLE YES Briefly list details:
III. STATISTICAL ANALYSIS
24. Are appropriate statistical procedures reported?
NO YES List procedure reference(s):
Additional Notes/Comments:
181
-------
APPENDIX E: SENSITIVITY OF P MAX AND PAVG MODELS TO
DIFFERENT BINNING SCENARIOS
E.l Introduction
Chapter 5 in the main text discusses the development of multiple chemicals to predict the
frequency of toxicity observed in the SEDTOX02 marine amphipod database. The PMax and
PAvg models are used to calibrate the probability of toxicity calculated from the individual
chemical models with the observed incidence of toxicity. This approach uses a nonlinear
regression equation fit to data summarized by binning individual data points according to their
P Max or P Avg values calculated from the individual chemical models. After the data were
placed into bins, the median P Max or P Avg value of each bin was used as the explanatory
variable and the corresponding proportion of toxic samples was used as the dependent variable.
Each bin was selected to contain 50 unique P Max or P Avg values.
This sensitivity analysis evaluated two questions:
1. How does bin size and binning approach affect the regression coefficients and R-squared
value for the P Max and P Avg models
2. What is the effect of the median (vs. the minimum, mean, or maximum) of the P Max
value in the bin for the x-axis values? This latter analysis was conducted only for the
P Max model. Results for the P Avg model would be expected to be similar.
E.2 Methods
Three binning scenarios were identified as reasonable approaches for setting the bin endpoints.
Within each binning scenario, variable bin widths were targeted. The complete set of binning
scenarios plus bin widths or sample sizes are as follows:
Scenario A
This binning approach takes the set of ordered unique P Max or P Avg values in the data
set and sorts them so that there are n unique P Max or P Avg values in each bin. The
number of samples in each bin will be greater than or equal to n, as there may be samples
with duplicate P Max or P Avg values. This scenario was investigated for// = 5, 10, 15,
25, 50, 75, 85, and 100. (Note: this method with n = 50 was used in Chapter 5).
Scenario B
This binning approach takes the set of ordered P Max or P Avg values (duplicates
included) in the data set and sorts them so that there are at least n P Max or P Avg
values in each bin. If the value for the last sample of the bin is the same as the next
value, then those duplicate values are included in the first bin. For example, if we are
using a target bin size of // = 5 and the ordered P Max values start 0.2, 0.233, 0.234,
0.236, 0.239, 0.239, 0.239, 0.34, then the first bin would start with 0.2 and would end
with 0.239, including all three of the 0.239 values for a sample size of seven. The next
bin would start with 0.34. This scenario was investigated for // = 5, 10, 15, 25, 50, 75,
85, and 100.
182
-------
Scenario C
This binning approach takes the set of ordered PMax or PAvg values (duplicates
included) and divides them into bins of equal width on the P Max or P Avg probability
scale. In this scenario, the number of samples per bin may be highly variable, and some
bins may contain no samples. This scenario was investigated for bin widths of 0.002,
0.004, 0.01, and 0.02.
The three binning scenarios described above were run on the complete data set consisting of
samples that had detected values for 10 or more of the modeled chemicals (n = 2856). All
analyses were conducted using the IX screening approach and the Sig Only classification of
toxicity. For each sample size within a binning scenario, the data set was binned and
summarized, and the nonlinear least squares regression equation was fit using S-PLUS 2000.
E.3. Results
The regression coefficients and R-squared value are shown in the Tables E-l through E-3 for
P Max and E-4 through E-6 for the P Avg models. Each scenario exhibited the same patterns in
the data and goodness of fit as bin size decreased and the regression sample size increased:
variability around the best fit line increased (a smaller R-squared) with increasing regression
sample size (i.e., the more bins and the fewer the number of data points summarized per bin).
For example, Figure E-l shows increasing variability with increasing regression sample size for
the Scenario A bin approach using the P Max values.
All iterations within each scenario resulted in very similar regression coefficients. The models
tend to deviate towards the tails of the P Max or P Avg range; the biggest differences in the
predicted probabilities of toxicity from using a different binning scenario/method will be below
P Max values of 0.2 or above 0.8 (Figure E-2).
The same patterns were apparent across binning scenarios and bin sizes, regardless of
whether the median or the maximum P Max values were used on the x-axis (Tables E-l and E-2,
Figure E-3). However, the differences between the best-fit lines for the different bin sizes varied
more when the maximum P Max values were used than when the median P Max values were
used on the x-axis. This is because the differences between median and maximum values in a
bin increase as the bin sizes increase, making the regression data set more different across bin
size scenarios. In the end, however, all regression lines were very similar.
Conclusions
The binning scenario and bin sample size have a small effect on the outcome of the P Max and
P Avg models. The R-squared values vary considerably as a function of regression sample size
(as would any goodness-of-fit metric) and is not a reliable measure of the accuracy of toxicity
predictions. When compared across the same binning approach, the P Max model produced
slightly higher R-squared values than did the P Avg model for most sample sizes and binning
scenarios.
183
-------
Table E-l. Sensitivity analysis results for P IMax models; Scenario A: unique P IMax values, vary the
minimum unique sample size
Unique
sample
count
Number of
bins'"
Minimum
sample
count per
bin1'
Maximum
sample
count per
bin1'
Coefficients for x
= median (P Max)
Coefficients for x =
max (P Max)
intercept
X
x~
R:
intercept
X
X"
5
440
5
20
0.09 0
394
0.356
0.5417
0.089
0.395
0.354
10
220
10
25
0.098 0
371
0.372
0.6977
0.097
0.374
0.367
15
146
15
-> ->
0.096 0
379
0.369
0.763
0.094
0.384
0.362
25
88
25
46
0.108 0
339
0.398
0.8479
0.105
0.344
0.386
50c
44
52
79
0.112 0
33 1
0.405
0.9299
0.106
0.337
0.387
75
29
80
117
0.112 0
329
0.406
0.9344
0.101
0.352
0.362
85
25
98
181
0.109 0
349
0.382
0.9382
0.095
0.387
0.319
100
22
108
147
0.111 0
-> ->
0.41
0.9481
0.099
0.35
0.362
''Number of data points used in the regression.
''Number of samples summarized within each data point.
cApproach used in Chapter 5.
-------
Table E-2. Sensitivity analysis results for P Max models; Scenario B: nonunique P Max values, vary the
minimum sample size
Unique
sample
count
Number of
bins'"
Minimum
sample
count
per bin1'
Maximum
sample
count
per bin1'
Coefficients for x =
median (P Max)
Coefficients for x =
max (P
Max)
intercept
X
X"
R:
intercept
X
X"
R:
5
533
5
13
0.097
0.386
0.362
0.464
0.096
0.387
0.359
0.464
10
274
10
19
0.107
0.346
0.394
0.609
0.106
0.347
0.391
0.6091
15
185
15
25
0.111
0.335
0.401
0.6909
0.11
0.338
0.395
0.6908
25
113
25
29
0.108
l\ "> 1 ">
0.397
0.8011
0.106
0.348
0.387
0.8012
50
56
50
79
0.111
0.332
0.405
0.8987
0.106
0.347
0.378
0.8994
75
37
75
145
0.108
0.347
0.39
0.9106
0.096
0.388
0.329
0.9109
85
-> ->
85
116
0.105
0.359
0.38
0.9255
0.093
0.397
0.322
0.9257
100
28
100
149
0.105
0.364
0.373
0.9173
0.092
0.404
0.308
0.918
''Number of data points used in the regression.
''Number of samples summarized within each data point.
-------
Table E-3. Sensitivity analysis results for P IMax models; Scenario C:
equal widths on P IMax scale
Width of bin
on PMax
scale
Number of
bins'1
Minimum
sample
count
per bin1'
Maximum
sample
count
per bin1'
Coefficients for x =
median (P Max)
intercept
X
X"
R:
0.002
491
0
21
0.105
0.374
0.342
0.4586
0.004
246
1
28
0.096
0.383
0.368
0.6827
0.01
99
1
47
0.101
0.349
0.404
0.8462
0.02
50
1
92
0.117
0.276
0.477
0.9198
'Number of data points used in the regression.
''Number of samples summarized within each data point.
186
-------
Table E-4. Sensitivity analysis results for PAvg models; Scenario A: unique PAvg
values, vary the minimum unique sample size
Unique
sample
count
Numbcrof
bins'1
Minimum
sample
count
per bin1'
Maximum
sample
count
per bin1'
Coefficients for x =
median (P Avj>)
intercept
X
X"
R"
5
430
5
12
0.11
1.428
-0.525
0.5368
10
215
10
21
0.112
1.421
-0.519
0.7096
15
143
15
32
0.11
1.443
-0.551
0.7765
25
86
25
51
0.109
1.439
-0.538
0.8245
50c
43
52
90
0.109
1.417
-0.488
0.8922
75
28
79
136
0.101
1.517
-0.69
0.9068
85
25
91
150
0.105
1.468
-0.59
0.9086
100
21
107
174
0.103
1.496
-0.644
0.912
'Number of data points used in the regression.
''Numbcrof samples summarized within each data point.
c Approach used in Chapter 5
187
-------
Table E-5. Sensitivity analysis results for PAvg models; Scenario B:
nonunique P Avg values, vary the minimum sample size
Target
sample
count
Number of
bins'1
Minimum
sample
count
per bin1'
Maximum
sample
count
per bin1'
Coefficients for x =
median (P Avg)
intercept
X
X"
R:
5
541
5
9
0.104
1.444
-0.533
0.5017
10
276
10
19
0.104
1.457
-0.558
0.6607
15
186
15
28
0.103
1.466
-0.563
0.723 1
25
112
25
47
0.103
1.466
-0.571
0.7951
50
56
50
92
0.1
1.513
-0.672
0.854
75
37
75
144
0.098
1.54
-0.723
0.8796
85
-> ->
85
124
0.099
1.515
-0.662
0.9019
100
28
100
147
0.1
1.513
-0.661
0.9114
'Number of data points used in the regression.
''Number of samples summarized within each data point.
188
-------
Table E-6. Sensitivity analysis results for PAvg models. Scenario C: equal
widths on P Avg scale
Width of bin
on PMax
scale
Number of
bins'1
Minimum
sample
count
per bin1'
Maximum
sample
count
per bin1'
Coefficients for x =
median (P Avj>)
intercept
X
X"
R:
0.002
455
0
26
0.125
1.266
-0.273
0.5358
0.004
228
0
45
0.122
1.274
-0.235
0.7724
0.01
91
0
104
0.117
1.338
-0.351
0.8477
0.02
46
0
204
0.116
1.343
-0.378
0.8993
'Number of data points used in the regression.
''Number of samples summarized within each data point.
189
-------
bin size = 5
o oooaosar
iHBPBH) P.O. . I
0 0 0 2 0 4 0 6 0 8 10
Median P Max value
bin size = 25
06
o °
0 2
00
0 2
0 4
0 6
0 8
10
Median P Max value
bin size = 85
8
06
04
oo
0 2
Oo,
0 2
0 4
0 6
0 8
Median P Max value
oo ooypo o
Median P Max value
Median P Max value
bin size = 50
0 2 0 4 0 6 0 8 10
Median P Max value
bin size = 75
o
y
X 0.6
o
Jso
c
io
o
O J/rO
r
o
2 0.2
Q_
0 2 0 4 0 6 0 8 10
Median P Max value
bin size = 100
0 2 0 4 0 6 0 8 10
Median P Max value
Figure E-l. Proportion of samples observed to be toxic versus the median P IMax value for
various bin sizes, where the bin size is the number of unique P IMax values in each bin
(Scenario A).
-------
0.8
0.6
a
x
o
c
o
t
o
g" 0.4
CL
0.2
0.2
0.4
0.6
0.8
1.0
Median P Max value
Figure E-2. Best-fit regression lines for all P IMax models. Points shown are for bin size of
50 unique P Max values (Scenario A).
-------
Median P_Max
Maximum P Max
Median-based model
Max-based model
P Max value
Median P Max value
Figure E-3. Comparison of maximum and median P IMax values and resulting models. Points shown
are for bin sizes of 50 unique P Max values (Scenario A).
-------
v/EPA
United States
Environmental Protection Agency/ORD
National Center for
Environmental Assessment
Washington, DC 20460
PB2005-106460
EPA/600/R-04/030
March 2005
------- |