-------
DISCLAIMER
This report has been reviewed by the Health Effects Research
Laboratory, U.S. Environmental Protection Agency, and approved
for publication. Approval does not signify that the contents
necessarily reflect the views and policies of the U.S. Environmental
Protection Agency, nor does mention of trade names or commercial
products constitute endorsement or recommendation for use.
if
-------
FOREWORD
The many benefits of our modern, developing, industrial society
are accompanied by certain hazards. Careful assessment of the relative
risk of existing and new man-made environmental hazards is necessary
for the establishment of sound regulatory policy. These regulations
serve to enhance the quality of our environment in order to promote the
public health and welfare and the productive capacity of our Nation's
population.
The Health Effects Research Laboratory, Research Triangle Park,
conducts a coordinated environmental health research program in toxicology,
"eprtlemialogy,Hand clinical studies using human volunteer subjects. These
studies address problems in air pollution, non-ionizing radiation,
environmental jcarcinogenesis and the toxicology of pesticides as well as
other chemical pollutants. The Laboratory participates in the development
and revision o'f air quality criteria! documents on pollutants for which
national amb01erft air quality standards exist or are proposed, provides
the data for registration of new pesticides or proposed suspension of
those already 'in use, conducts research on hazardous and toxic materials,
and is primarily responsible for providing the health basis for non-
ionizing radiation standards. Direct support to the regulatory function
of the Agency is provided in the forni of expert testimony and preparation
of affidavits as well as expert advice to the Administrator to assure
the adequacy of health care and surveillance of persons having suffered
imminent and substantial endangerment of their health.
The study described in this report was conducted under our program
to determine tyie health implications of children exposed to low levels
of lead. !
F. G. Hueter, Ph.D.
Director
Health Effects Research Laboratory
ill
-------
,$**'"',.,.-,
i/iiv WING <;.;:'> r..'.-.:'
! 'r if- .!;!'
'""' T- * ABSTRACT
; ; ; . ,' Two separate^studies were conducted with the overall objective of
.;;• examining the impact of lead at low dose on the neuropsychological
v; function of children.
In the first study, a sample of Children identified as having
elevated lead levels in the dentine of shed deciduous teeth (N = 19)
I were compared to children with low dentine lead (N = 22) on electro-
, encephalograms,and a panel of 8 auditory and speech processing tasks.
; Quantitative electroencephalograms were obtained from 20 sites
j ' .
i under 4 conditions in these subjects.; The spectrum from 0.5 - 32 Hz was
i examined; four. I W" . • . .
, shw:x>f the U.5. Environmental Protection Agency.
-------
PREFACE
Concern over the health effects from exposure to lead has shifted in
recent years from those obvious impairments which follow high dose exposure
to those changes in the adaptation of the host which, while they may be
less dramatic and are often missed, may have important, if difficult to
identify, long-term consequences.
A number of biochemical alterations have been recently demonstrated in
humans which appear to have virtually no threshold for lead (1, 2, 3). Con-
siderable controversy has been raised as to whether these biochemical changes
can be considered adverse health effects. Any decrease in the ability of the
host to process psychological stimuli clearly compromises the organism's
ability to deal adaptively with a changing environment. A change of this
nature would, therefore, meet any reasonable definition of an adverse health
effect.
Because young organisms appear to be more susceptible to lead, and
because childrens' exposure to lead is widespread, the study of the neuro-
pathological consequences of lead in children has become an area of intense
study. In 1979, the Lead Exposure Program at the Boston Children's Hospital
Medical Center demonstrated cognitive and behavioral changes in children who
were dose-related to lead. The changes were found in three general areas:
verbal intelligence, speech and auditory processing, and attentional perfor-
mance. It seemed important to determine whether these alterations were ac-
companied by parallel alterations in the electrical activity of the brain,
and whether the auditory and language processing could be better defined.
Another Massachusetts community reported as long as 150 years ago to have a
-------
problem with lead, provided a simultaneous opportunity to examine the class-
room behavior of a Different sample of children in relation to their lead
exposure in more depth than previous studies.
The studies reported here are part of the ongoing activities of the
Lead Exposure Program, Children's Hospital Medical Center, whose goal is to
contribute to the comprehensive understanding of lead's impact on man.
VI
-------
CONTENTS
Foreword iii
Abstracts . iv
Preface . . .; . v
Figures . viii
*& -~ - -• -- . • v
I
Tables .....; , ix
•i ' '•'
Acknowledgments .... x
1. Introduction 1
2. Conclusions 3
3. Recommendations 5
4. Methods of Procedure . . 6
'' . , Study I 6
Study II . 33
5. Discussion 58
References 60
vii
-------
FIGURES
i
Page
Figure One: Placement of electrodes for EEG analysis 11
Figure Two: Distribution of spectral energy (delta) in high
and low lead! children 16
i
Figure Three: bistribution of spectral energy (alpha) in
high and low! lead children 17
ffigufeTdurf Tfeache'rs' ratings in relation to dentine lead
levels, males
35
\
Figure Five: Teachers' ratings in relation to dentine lead
levels, females 36
viii
-------
TABLES
Number
1
i
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
i
EEG Features in Hi ah and Low Lead Subiects .......
1EG and Psychologic Features ..with Greatest Discrim-
inating Power
Separation of High and Low Lead Groups by Discriminant
Function Analysis of EEG and Pyschologic Variables . . .
Stepwise Discriminant Analysis with EEG and Pyschologic .
Variables
Retrospective Classification of Subjects by Stepwise
Discriminant Analysis
Prospective Classification of Subjects by Jackknifing . .
Speech 'Processing Tests in High and Low Lead Children:
Outcqme'iof Speech Processing: F Ratios and P Value . . .
i
Mean audiometric Values of High and Low Lead Children:
Mean Values and Significance Tests ...........
I
Covaris|tes and Categories Scored in Logistic Regression
;
Logisti'c Regression Analysis: Cluster 1 Without Lead . .
Logistic Regression Analysis: Cluster 1 With Lead ....
Logistic Regression Analysis: Cluster 2 Without Lead . .
Logistic Regression Analysis: Cluster 2 With Lead ....
Logistic regression Analysis: Cluster 3 without Lead . .
Logistic Regression Analysis: Cluster 3 With Lead ....
Logistic Regression Analysis:! Frustratable Without Lead .
Logistic Regression Analysis: Frustratable With Lead . .
Logistic Regression Analysis: Daydreaming Without Lead .
Logistic Regression Analysis: Daydreaming With Lead . . .
Pages
19
i •+
20
22
23
25
26
30
\j^f
31
32
42
45
46
47
49
51
52
53
54
55
56
ix
-------
ACKNOWLEDGMENTS
The electroencephalograms in this study were obtained in the Seizure
Clinic, Children's Hospital Medical Center, and analyzed by Drs. Janes
Burchfield, Frank Duffy and Peter Bartels. The speech and hearing battery
was administered and scored by Drs. Martha Lyle and Sylvia Topp. Family
histories and psychological data in the Lowell, Massachusetts, sample were
collected by Elise Martin. Biostatistical analyses were done by Robert
Stiratelli.
The cooperation of the Department of Schools, Chelsea, Somerville,
and Lowell, Massachusetts, the principals, nurses and teachers is gratefully
acknowledged.
Dr. Alan Leviton, Director, Division of Neuroepidemiology, Children's
Hospital Medical Center was an essential contributor to the design of these
studies.
-------
INTRODUCTION
While the toxicity of lead at high dose is widely acknowledged, con-
siderable controversy exists as to whether lesser doses produce adverse
health effects. Lead is known to inhibit certain enzyme systems at extremely
low concentrations. Among those most sensitive are D-aminolevulinic acid
dehydrase (1), adenyl cyclase (2), and ferrochelatase (3). It is not, however,
universally accepted that these biochemical changes and others represent adverse
health effects.
For the purposes of the investigations described below, an adverse
health effect is defined as "an alteration in the functioning of an organism
which diminishes its ability to adapt to a changing environment. As a result,
the organism's longevity, vigor, or reproductive capacity is reduced." In
a changing environment, the ability to register change and then make decisions
based upon the data is essential to survival. The organ in which these executive
functions primarily reside is the central nervous system. The organizing princi-
pal of these investigations is that the child's CNS is a candidate site for the
early expression of adverse effects of lead exposure. The Lead Exposure Pro-
gram at Children's Hospital Medical Center has focused on CNS performance. Earlier
studies by the group (4) had reported that children thought to be undamaged by
lead, but who had elevated dentine levels, were less able on a number of
measures of CNS function when compared to controls. Among the performances
which appeared sensitive to lead were verbal intelligence, auditory and speech
processing, attention and classroom behavior (4).
-------
This study examines the electrophysiology of the CNS of a subsample
of high and low lead subjects, and also compares them on a number of
other measures of auditory and speech processing. A separate sample of
children from another city were evaluated by teacher ratings, family history,
and their dentine lead levels measured. This offered an opportunity, not
»TV . ' .
only to replicate our previous observations, but to evaluate the effects
pf lead on classroom performance while controlling more completely for
other covariates which could be confounding.
-------
CONCLUSIONS
Study I
Alterations in neuropsychologic function were demonstrated in electro-
encephalograms of asymptomatic children with elevated body lead burdens.
These changes tended to be located in nvidline brain structures. The addition
of EEG analysis to a battery of psychologic outcomes sharpened diagnostic pre-
cision remarkably.
Statistically significant decrease in function was found on three
of nineteen speech processing ou-comes. Outcomes tended to favor low lead
subjects generally. The relatively small sample size may be responsible for
the failure of certain outcomes to reach statistical significance.
Study II
Teachers' Behavioral Ratings as a Function of Lead Burden.
The proportion of negative teachers' reports of classroom behavior tended
to increase with increasing dentine lead concentrations. The items most
sensitive to lead were distractibility, disorganization, ability to follow a
sequence of directions, and overall classroom functioning.
After data reduction by cluster analysis, the cluster containing the
items "distractible" and "disorganized" was found most sensitive to lead
exposure. The impact of lead was evaluated by stepwise multiple logistic
regression. A significant (p s 0.013) lead effect was found, controlling for
major covariates. This study replicated in part a previous investigation from
this laboratory which show that teachers' assessment of classroom behavior are
sensitive to lead exposure (4). It goes further than that study by controlling
for a number of covariates that could be related to outcome. The ability of
-------
the child to inhibit irrelevant stimuli is critical to the task of academic
learning. The item "distractible" which evaluates that ability is consis-
tently sensitive to lead exposure. It is reasonable to expect that this defi-
cit is related to the impaired performance on speech processing measured here.
-------
RECOMMENDATIONS
These studies add to the weight of those which report lead effects at
low dose, and demonstrate that the threshold for appearance of adverse be-
havioral effects is a function of the sensitivity of the methods brought to
bear on measuring outcome.
The findings of alterations in EEG, speech processing, and classroom
behavior in samples clinically assumed to be free of frank toxicity supports
the prudence and hygienic utility of reducing lead in the environment avail-
able to children.
Studies of lead effects in children should attend to those behaviors
which are expressions of the organism's ability to focus attention.
-------
METHODS OF PROCEDURE
Study I; EE6 and Speech and Auditory Processing In Children Exposed
i
to Low Level Lead.
f~r •
Study Sample; A subsample of children selected for our previous study was
• *>.
chosen for EEG and speech and hearing evaluation. The 3,329 children attending
first and second grades in the period between 1975 and 1978 1n Chelsea and
Soraervllle, Massachusetts, made up the population sampled. The children at-
tended 19 public schools, and came from 105 classrooms. Children were asked
to submit their shed teeth to the teacher, who then verified and recorded
the presence of an appropriate fresh socket.
Classification of Exposure: Children were classified according to the concen-
tration of lead In dentine. A 1mm sagittal slice was taken from each donor's
tooth, dentine chiseled out, the tooth dried, weighed and then digested 1n
perchloric-nitric add. Lead was measured by anodic stripping voltammetry (5).
The distribution of dentine lead levels (N • 3321 teeth) was log normal
(median 12.0 ppm). Samples were analyzed in replicate, employing either a second
tooth or the opposite slice of the first tooth. Agreement between replicate
samples was required or the candidate subject was excluded from further analy-
sis. To be classified as high lead, the mean of the samples was required to be
) 20 ppm. To be classified as low lead, the mean of the samples was required
to be ( 10 ppm. The extremes of the distribution were selected for sampling
because it was felt that any effect would most clearly be demonstrated by this
contrast.
Selection of Children for Neuropsychological Evaluation; After their lead
status had been determined, children were admitted to the study if they met
-------
met the following criteria:
1. English was the primary home language.
2. They were full term births.
3. They had not suffered noteworthy head
Injury.
4. They were discharged after birth at the
same time as their mothers.
Examination of the distribution of dentine lead levels and teachers'
behavioral ratings Included and excluded groups which demonstrated that no
selection bias was Imposed in this process. The sample of children then
accepted for neuropsychological data analyses numbered 158 (low = 100,
high « 58). A large number of covariates (39) were measured. The subjects
received a 4-hour neuropsychological battery, administered 1n fixed order.
Each child was examined singly; the psychometrldans were blind to the child's
lead status. A summary of the neuropsychological data analysis 1s presented
In the appendix.
Teacher's Behavioral Rating: The teacher of every child who gave a tooth
'** .
was asked to fill out an 11-ltem forced-choice behavioral rating scale scor-
ing the child as "yes" or "no" on the following questions:
1, Is this child easily distracted during
his/her work?
2. Can (s)he persist with a task for a rea-
sonable amount of time?
3. Can this child work independently and
complete assigned tasks with minimal
assistance?
4. Is his/her approach to tasks disorganized
(constantly misplacing pencils, books, etc.)?
5. Do you consider this child hyperactive?
6. Is (s)he over-excitable and impulsive?
-------
7. Is (s)he easily frustrated by difficulties?
8. Is (s)he a daydreamer?
9. Can (s)he follow simple directions?
10. Can (s)he follow a sequence of directions?
11. In general, is this child functioning as well
as other children his/her own age in the class-
room?
These questions were extracted from a questionnaire developed at the
Harvard Graduate School of Education. The instrument's reliability was eval-
uated by Miss Margaret Guild as part of her Ed.O. thesis requirement. Inter-
observer agreement on the Hitems was in the range of .640 to .786. Intra-
observer reliability was in the range of .801 to .905.
The EE6 Speech Sub-sample; From the 158 children evaluated neuropsychologic-
ally, 19 high lead and 22 low lead subjects were selected at random and in-
vited to return to the laboratory. No refusals were encountered. In most
instances, children first were evaluated in the speech and hearing clinic
and then were taken to the EEG clinic.
Electroencephalographic (EEG) Evaluation; Twenty-two low lead and 19 high lead
subjects had electroencephalographic evaluations. All subjects were studied
in a standard EEG laboratory. Twenty Grass gold cup electrodes were applied
to the scalp according to the International 10-20 System (Figure 1). Recording
of brain electrical activity was done in the "monopolar" mode using linked
electrodes on the two earlobes as a reference. Signals were amplified via
a Grass Model 78 polygraph, charted on polygraph paper as for a standard EEG,
and recorded on a 28 channel FM Honeywell 5600E tape recorder for subsequent
off-line analysis. Spectral analysis was performed on the raw EEG from each
electrode via the Fast Fourier Transform (FFT) technique on a PDP-12 computer
(Digital Equipment Corporation). Spectral analysis and file handling were
done under the SIGSYS-12 biomedical data analysis system (Agrippa Data Systems).
8
-------
Epochs of spontaneous EEG activity were recorded from each subject
under 4 standard conditions: (1) relaxed, alert with eyes open (EO); (2)
same, but with eyes closed (ECP);(3) during 3 minutes of hyperventilation (HV);
and (4) post-hyperventilation (PHV).
Raw EEG data recorded in each of the above conditions were bandpass
filtered between 0.5 and 35 Hz and analog-to-digital converted as a series
of 50 2-second segments. FFT analysis was individually performed on each
2-second segment and the ensemble of 50 segments was averaged to yield a final
spectrum representing 110 seconds of EEG. 'The final spectrum covered the
frequency range of 0-64 Hz, but only the 0.5-32 Hz range was subsequently
analyzed. Each average spectrum was divided into the 4 classical EEG frequency
bands: delta (0.5-3.5 Hz), theta (4.0-7.5Hz), alpha (8-12 Hz), and beta (12.5-
31.5 Hz). Spectral energy was summed within each of these bands and expressed as
a percentage of the total spectral energy over the full .frequency range 0.5-32 Hz.
Prior to statistical analysis these percentages were subjected to inverse sine
transformation.
In summary, the EEG data set for each subject consisted of 320 values:
percent spectral energy in 4 frequency bands, recorded from 20 scalp electrode
positions, under 4 conditions.
Psychologic Data: For use in multlvariate statistical analysis, subjects'
scores on psychologic performance tests were obtained from the data of Needleman
(4)
et al. Nine measures were chosen as having potentially high discriminating
value based upon their showing significant differences between high and low
lead populations. These were as follows: (1) Full-scale I.Q., (2) Verbal I.Q.,
(3) Performance I.Q., (4) Seashore Rhythm Test, (5) Sentence Repetition Test,
(6) Reaction time under varying intervals of delay, Block 2 (tested at 12 sec),
(7) Reaction time, Block 3 (12 sec), (8) Reaction time, Block 4 (3 sec), and
(9) Token Test. Parental I.Q., a control variable was also entered Into the
9
-------
model.
Data Analysis; The overall goal of data analysis was twofold. First, we
wanted to determine whether the high and low lead populations differed sig-
nificantly on any of the recorded EEG measures. To examine this we calcu-
lated population means and standard deviations of the high and low groups
respectively for each of the 320 derived spectral values. Significance of
group separation for homologus pairs of variables was tested by Student's
t-test and by the non-parametric W1lcoxin-Mann-Whitney two-sample rank test.
The second goal was to determine the power of EEG and psychologic
variables to discriminate between high and low lead subjects. To this end,
multlvariate analysis and non-parametric pattern recognition techniques were
performed on a PDP-11/45 computer using the TICAS system of statistical pro-
grams. ^ ' This comprehensive system was originally developed for the auto-
matic recognition of cancer cells based upon morphological features of their
microscopically displayed images (TICAS * Taxonomlc Intra£ellular Analytic
System). However, its statistical evaluation programs are not jSroblem specific;
they can be applied to any set of descriptive features, in this case, EEG and
psychologic measures.
Multlvariate statistical analysis proceeded in a sequential manner as
follows:
Reduction of the Dimensions of the Data Set; Statistical theory specifies a
limit to the number of features per subject which can be used to demonstrate
group differences or to develop classification rules to diagnose a subject
as belonging to one group or another. If this limit is exceeded, the analysis
becomes excessively individualized to the immediate data set and consequently
does not do well on prospective tests of repeated measure on subsequent data
sets. As a rough approximation, this limiting number Is derived by dividing
10
-------
Figure 1: Electrode placements according to the International 10-20 System.
F7
T5
F8
T6
11
-------
the number of subjects In the smallest group by 3. In the present case, this »
19/3 » 6 l'/3. There/ore, the Initial task was to reduce the number of features
from 330 (320 EEG and 10 psychologic) to 6 or less.
Data reduction was done In two steps. First, an Initial screening of
'if-
features was done based upon the results of univariate analysis. Those
variables which did not reach a high level of significance with Student's t-test
or the W1lcox1n-Mann-Whitney two sample rank test ( p j. 0.05) were excluded from
further analysis. Next, the remaining features were subjected to Merit Value
analysis. The Merit Value as formed by TICAS 1s a relative measure of a
feature's ability to distinguish between the two populations. It is derived as
the average of three independent factors. The first two factors are measures of
the degree to which a given variable separates two populations. One is d-prfme,
the measure of "detectablllty" calculated from the receiver operating character-
istic (ROC) curve. '6' The value of this statistic ranges from 0 (no detectabi-
lity of group separation) to 0.5 (complete detectabllity of group separation).
The other factor is the "ambiguity" value based upon the ambiguity function of
Genchi and Morl.^'' A feature which totally separates two populations has an
ambiguity value of 0. On the other hand, exact overlap of the populations
yields the maximum ambiguity value of 1. The final factor contributing to the
Merit Value is a measure of the correlation of a given feature with the other
features under consideration. This was calculated as the average of the correla-
tion coefficients between a given feature and each of the other features. The
average correlation was used to detect the presence of redundant features.
Calculation of the Merit Value for a given variable was as follows:
Merit Value * 2 /d-prime) + H-ambiguity value) -frfl-average correlation coefficient)
«j
Merit Value, therefore, ranges from 0 to 1. The higher a variable's Merit Value,
the better its potential usefulness for demonstrating group separation or for
developing rules to classify subjects into one of two populations.
12
-------
Features with high Merit Value possess a combination of high detectab111ty,
low ambiguity and low correlation with other variables.
tests of Separation of High and Low Lead Populations and Retrospective Classi-
fication of Subjects; The features with the best Merit Value scores were sub-
mitted to direct and stepwlse discriminant analyses to assess the collective
ability of these features to separate high and low lead groups. The signifi-
cance of group separation was tested by Milks' lambda and the Mahalanobis
(8)
distance between group centroids. The former statistic may be considered
as a multlvarlate analog of the variance ratio (F) test; therefore, an approx-
imation of the F-stat1st1c was calculated using the technique of Rao. For step-
wise discriminant analysis, retrospective classification of subjects was carried
out by the Bayes classification procedure using the linear discriminant function.
Prospective Classification of Subjects: Prospective classification was carried
out using the non-parametric, supervised learning programs of the TICAS.
These algorithms develop classification rules for assigning subjects to one
of two populations by the sequential determination of decision boundaries along
the feature axes within multlvarlate feature space. The technique is non-parametric
because subjects are arrayed along each axis by rank, rather than by absolute
value of a given variable. Thus, each subject is represented by a vector whose
elements are the relative rank the subject occupies within the sample population
for each variable.
The classification rules are developed on a "training set" of subjects
whose membership in one of the two populations is known. The procedure involves
examining subjects in rank order along a given feature axis. Sampling continues
as long as all subjects belong to the same population; however, as soon as the
first subject of the other population is encountered, sampling stops and a
13
-------
decision boundary 1s established. The sampled subjects of the first population
are then removed and sampling proceeds 1p the same way with the remaining subjects
on another feature axis. This continues until all subjects of one population
have been removed or until the end ranks of each axis contain subjects of the
other population.
Successive runs of this procedure are carried out and the order of feature
selection 1s varied based upon previous outcomes In terms of classification
success and efficiency. This 1s repeated until the optimum set of classification
rules 1s developed. This procedure differs from parametric discriminant tech-
niques In that It selects features according to their ability to classify sub-
jects correctly from the remaining subset of as yet unclassified subjects. Thus,
1t may select features which are not weighed highly by parametric selection
methods, which always consider a feature's usefulness In classifying subjects
from the full sample.
Where a significant separation exists between the centrolds of the two
populations in multlvariate space, a set of classification rules can usually
be developed which will classify subjects of the training set with nearly 100Z
accuracy. Therefore, the true test of the diagnostic power of the classifica-
tion rules, 1s to prospectlvely evaluate them on a "test set" of subjects who
were not part of the original training set. Ideally, for this purpose one would
like to divide the study population Into separate training and test sets. In
the present case, however, the population was not large enough to allow this.
Instead, we established test sets by the "jackknlffng" or "leaving one out"
procedure.^) Here, diagnostic classification rules were developed on a train-
Ing set of 37 of our 41 subjects and tested on a test set of the remaining 4
subjects who had not been used for development of the classification rules.
14
-------
This was repeated 11 times with, a different set of 4 "left out" each time.
The final time only one subject was left out. To be considered successful,
such rules should correctly classify substantially more than 50% of the test
set subjects. The "jackknlflng" technique 1s generally taken as a method to
appraise the prospective performance of diagnostic classification rules when
the rules are developed on small data sets. In Its original form, "jackknlflng"
referred to the "leaving out" of single subject test sets. The "leaving out"
of larger test sets, 4 in our case, is a more conservative procedure.
RESULTS
Electroencephalographic (EEG) Evaluation: EEG frequency spectra derived from
19 children with high dentine lead concentration were compared with those de-
rived from 22 of their classmates with low dentine lead concentration. The
high lead children differed significantly from their low lead counterparts
for a number of EEG spectral features (Table 1). The distribution of these
differences among the various combinations of recording conditions, electrode
position and frequency band was decidedly non-random.
By far the largest number of differences occurred in spontaneous EEG
recorded in the state of relaxed wakefulness with the eyes clsoed (EC). More-
over, these differences clustered with respect to frequency band and spatial
topography. Two patterns are clearly evident. First, the high lead group
had consistently higher percentages of slow frequency delta (0.5-3.5 Hz)
activity 1n the EEG recorded from a series of adjacent electrodes covering
the central, parietal and occipital regions of the head bilaterally and ex-
tending forward into the right prefrontal area (Figure 2).
The second pattern of differences 1n the eyes closed EEG was a decrease
in the percentage of alpha (8-12 Hz) activity In the occipital lobes and
the nrfdline central parietal region of the high lead subjects (Figure 3).
15
-------
LOW t'EAD
DELTA (0.5-3.5 Hz)
°/0 ENERGY
HIGH LEAD
L: Topography of delta EEG spectral energy in high and low lead child-
ren. EEG recorded while subjects were relaxed, alert with eyes closed. Delta
energy was calculated as the percentage of total EEG spectral energy in the
frequency range 0.5-3.5 Hz. This was done for each electrode derivation from the
mean EEG spectrum of the high and low lead group, respectively, and topographical
maps of delta energy were constructed by linear interpolation based upon the val-
ues of the nearest 3 electrode points (11).
16
-------
ALPHA (8-12 Hz)
LOW LEAD
Figure 3: Topography of alpha (8-12 Hz) EEG spectral energy in high and low
lead children. Same recording condition as in Figure 2. See legend of Figure
2 for further details.
17
-------
Somewhat similar regional patterns of EEG differences were seen 1n
the eyes open (EO) condition, but to a much smaller extent (Table 1); the
high lead group haiTslgnlflcantly larger percentages of delta activity 1n the
nrfdllne, central and parietal regions (CZ and PZ). It 1s Interesting to
note that these were the same two electrode positions which showed the greatest
differences 1n delta activity in the eyes closed EEGs. During the eyes open
state, the high lead group also had decreased percentages of alpha activity;
however, in contrast to the eyes closed condition, this occurred in the pre-
frontal area rather than the occipital area.
No statistically significant differences occurred in any of the compari-
sons between spectral values derived from EEGs recorded during the conditions
of hyperventHatlon or post-hyperventilatlon. Likewise, the high and low lead
groups did not d1ffer anywhere in the theta (4-7.5 Hz) or beta (12.5-31.5 Hz)
frequency ranges.
Discrimination Between High and Low Lead Groups Based Upon EEG and Psychologic
'<".
Features; Merit Value analysis culled from the available features those with
the greatest potential for discriminating between high and low lead groups.
EEG and psychologic features were considered both alone and in combination.
Table 2 lists the 6 features in each category with the highest Merit Value.
The power of these features to discriminate between the high and low
lead exposure was tested with direct and stepwise discriminant analyses. For
!
both types of analysis, the 4 "best" features among, respectively, EEG features
alone, psychologic features alone and combined EEG and psychologic features were
used as Input variables. The most striking result of these discriminant
analyses was the marked difference in significance of group separation using
combined variables as opposed to EEG or psychologic variables alone (Table 3).
18
-------
Table 1
COMPARISON OF EEG FEATURES FROM HIGH AND LOW LEAD SUBJECTS BY
WILCOXIN-MANN-WHITNEY TWO-SAMPLE RANK TEST
Condition
Eyes Open (EO)
'
Eyes Closed (EC)
Electrode
F4
• FZ
CZ
PZ
F4
C4
C3
P4
P3
02
01
FZ
CZ
v PZ
oz
Frequency
Band
Alpha
Alpha
Delta
Delta
Delta
Delta
Delta
Delta
Delta
Delta
Alpha
Delta
Alpha
Delta
Delta
Alpha
Delta
Alpha
Delta
Alpha
Mean Percentage of Total
Spectral Energy
Low Lead
25.6
25.4
31.2
31.0
29.9
28.1
28.1
27.7
27.5
27.1
33.8
26.8
34.0
31.2
30.4
27.5
29.5
30.0
26.1
33.9
High Lead
24.5
24.3
32.3
32.1
31.0
29.4
29.3
28.9
28.7
29.1
31.4
28.5
31.6
32.6
32.0
26.0
31.1
28.2
27.7
31.6
P- Value
.05
.05
.05
.02
.05
.02
.05
.05
.02
.02
.05
.02
.05.
.05
.01
.05
.01
.05
.02
.05
HyperventHatlon (HV)
Post-hype rventHa-
tfon (PHV)
NO SIGNIFICANT DIFFERENCES
NO SIGNIFICANT DIFFERENCES
19
-------
. Table 2
FEATURES WITH THE GREATEST POTENTIAL POWER FOR DISCRIMINATING BETWEEN HIGH AND
LOU LEAD GROUPS AD DETERMINED BY MERIT VALUE ANALYSIS. THE SIX BEST WERE SEL-
ECTED FROM, RESPECTIVELY, EEG FEATURES ALONE, PSYCHOLOGIC FEATURES ALONE, AND
COMBINED EEG AND PSYCHOLOGIC FEATURES.
Feature
EEG*
Delta PI EC
Alpha FZ EO
Delta OZ EC
Alpha OZ EC
Alpha PZ EC
Alpha CZ EC
Ambiguity
Value
0.64274
0.76054
0.69427
0.77362
0.68875
0.75213
d-Prfme
0.18900
0.18780
0.16268
0.18660
0.17703
0.18660
Average
Correlation
Coefficient
0.61419
0.51113
0.55674
0.61946
0.68980
0.66728
Merit
Value
0.3736?
0.3679C
0.35811
0.32671
0.32517
0.31793
PSYCHOLOGIC
Seashore Rhythm Test 0.68725
Sentence-Repetition Test 0.79442
Performance I.Q. 0.67626
Token Test 0.71324
Full-Scale I.Q. 0.69818
Verbal I.Q. 0.74706
EEG AND PSYCHOLOGIC
Delta PZ EC
Alpha PZ EC
Seashore Rhythm Test
Verbal I.Q.
Full-Scale I.Q.
Performance I.Q. 0.67626
0.64274
0.68875
0.68725
.74706
.69818
0.
0.
0.20215
0.20335
0.19737
0.15311
0.25000
0.23086
0.18900
0.17703
0.20215
0.23076
0.25000
0.19737
0.23922
0.38112
0.48870
0.37508
0.61164
0.54674
0.25568
0.21053
0.26510
0.39184
0.47989
0.42667
0.49261
0.41039
0.40993
0.40597
0.39673
0.38931
0.49319
0.48493
.48399
.44094
0.44064
0.43060
0.
0.
*0es1gnation of EES features is by frequency band, electrode location and recording
condition; thus, DELTA PZ EC designates percentage of total spectral energy in the
delta (0;5 - 3.5 Hz) frequency band for EEG recorded from the PZ electrode during
the state of relaxed wakefulness with eyes closed.
20
-------
The separation of high and low lead groups using combined variables was
highly significant (P a 0.005). By contrast, the P-value using psychological
features alone was an order of magnitude less (P a 0.041); however, this value
still exceeded conventional standards for statistical significance. The P-value
for EEG features alone was P * 0.079.
Significant separation of the high and low lead groups was also demonstrated
by stepwise discriminant analysis (Table 4). Here again the most significant
discrimination resulted from the use of a combination of EEG and psychologic
features (Milks' lambda » 0.557, p = 0.0002). Note that for EEG features alone,
the total discriminatory power resided in a single feature; therefore, WUk's
lambda has little meaning. In this case, the significance of the Mahalanobis
distance is a more meaningful statistic (p J.0.01). For comparison, the signi-
ficance of the Mahalanobis distance for psychologic features alone is p <_ 0.025
and that for combined EEG and psychologic features is p ^ 0.001.
Discriminant analysis defines a boundary in multivariate space which optim-
ally separates the two populations. Multivariate space, therefore, is divided
into two population domains and any subject whose feature vector falls into one
of these domains would be classified as a member of that population. Obviously,
the more significant the group separation, the fewer the errors that will be
made in such classification. Somplete separation would yield 1002 accuracy of
classification, whereas two identical populations would result in classification
at the chance level of 502. Hence, the success rate of retrospective classifi-
cation based upon the discriminant function is another test of the significance
of group separation for a set of features.
Retrospective classification of high and low lead children was significantly
better than chance (Table 5). This was true for all categories of features.
Based upon EEG features alone, 70.7? of subjects were correctly classified. For
psychologic features alone, accuracy was 68.3%. Combined psychologic and EEG
21
-------
Table 3
SEPARATION OF HIGH AND LOW LEAD
GROUPS BY DISCRIMINANT ANALYSIS
EEC Features Alone
Feature
Delta PZ EC
Delta OZ EC
Alpha OZ EC
Alpha FZ EC
Performance I.Q.
Seashore Rhythm Test
Sentence-Repetition
Test
Token Test
u/^f ^ , "f^ F-Stat1st1c
High Lead Low Lead
31.14 29.53 8.32
27.71 26.11 5.96
31.61 33.86 4.43
24.27 25.39 4.54
Group centroids: high a -11.61, low
Milks' Lambda: 0.798
Significance of Group Separation by
Approximation: F (4,36) » 2.273, P
Psychologic Features Alone
103.63 111.50 3.98
19.68 22.77 4.44
10.90 13.81 7.09
22.90 25.23 3.53
Group centroids: high * 37.00, low
Milks' Lambda: 0.764
Significance of Group Separation:
F (4,36) » 2.785,
P- Value
0.006
0.018
0.040
0.037
» 9.24
Rao ' s
= 0.079
0.050
0.039
0.011
0.065
« 42.58
P = 0.041
Discriminant
Coefficient
-0.863
-0.050
0.1 GO
0.478
0.626
0.807
0.807
0.592
Combined EEC and Psychologic features
Delta PZ EC
Verbal I.Q.
Seashore Rhythm Test
Alpha PZ EC
31.14 29.53 8.32
97.26 106.55 6.22
19.68 22.77 4.44
28.20 30.00 4.78
0.006
0.016
0.039
0.033
-0.721
0.637
0.549
0.568
Group centroids: high » -14.20, low » -10.91
W11ks' Lambda: 0.661
Significance of Group Separation:
F (4,36) * 4.61, P » 0.005
22
-------
Table 4
SEPARATION OF HIGH AND LOW LEAD
GROUPS BY STEPWISE DISCRIMINANT ANALYSIS
EEG Features Alone
Step
Feature Entered
F to Enter
Delta PZ EC
Significance of Mahalanobis Distance
F (1,39) - 8.33, p 10.01
8.33
Psychologic Features Alone
Step
1
2
3
Feature Entered
Full-Scale I.Q.
Sentence-Repetition Test
Seashore Rhythm Test
F to Enter
7.13
2.85
1.47
WUks1 Lambda » 0.756, P = 0.015
Significance of Mahalanobls Distance
F (3,37) = 3.97, p 10.025
Combined EEG and Psychologic Features
Step
Feature Entered
F to Enter
1
2
3
4
Delta PZ EC
Full-Scale I.Q.
Verbal I.Q.
Performance I.Q.
Milks' Lambda » 0.557, P * 0.0002
Significance of Mahalanobls Distance
F (4,36) * 7.15, P 10.001
8.33
14.70
1.27
1.12
23
-------
features yielded the most successful classification with 75.6% correct.
Prospective Classification of Subjects Into the High or Low Lead Group Based
Upon EEG and Psychologic Features: A better test of the discriminating power
of a set of features 1s prospective classification. In retrospective classi-
fication one defines and tests a discriminant function using the same subjects.
In prospective classification, on the other hand, classification rules are de-
veloped on one set of subjects (the "training" set) and tested on a separate
set (the "test" set). Clearly, 1n order for prospective classification to
succeed a greater than a chance level, the subjects must fall into two distinct
clusters in multivariate space.
Based upon the 6 best combined EEG and psychologic features we were able
to classify test sets of children as high or low lead with an overall accuracy
of 65.9% well above chance (Table 6). Moreover, the additive discriminating
power of EEG and psychologic features was clearly demonstrated by this pro-
cedure; prospective classification using either EEG or psychologic features
alone yielded only chance levels of classification (48.82 and 41.52 respec-
tively). This Indicates that the separation of high and low lead subjects
into two distinct clusters is much more prominent in a multivariate space
consisting of both EEG and psychologic features than it is In one consisting
of either type of feature alone.
Speech and Auditory Evaluation:
. Each subject received the following tests, administered singly by examiners
in the blind to the child's lead status:
1. Pure Tone Audiometry: Tones were delivered to each ear separately
at 250 Hz, 500 Hz, 1 K Hz, 2K Hz, 3K Hz, 4K Hz, 6K Hz, and 8K Hz.
24
-------
Table 5
RETROSPECTIVE CLASSIFICATION OF SUBJECTS
BASED UPON STEPWISE DISCRIMINANT ANALYSIS
Percentage of Subjects Correctly Classified
Features • . i _.......
Low Lead Sigh Lead All
EE6 Alone 68.2 73.7 70.7
Psychologic Alone 72.7 63.2 68.3
Combined EEC and Psychologic 77.3 73.7 75.6
25
-------
Table 6
PROSPECTIVE CLASSIFICATION OF SUBJECTS
USING "JACKKNIFING" PROCEDURE
Percentage of Subjects Correctly Claaai-fled
Features
Low Lead Sigh Lead All
EE6 Alone 36.4 63.2 48.8
Psychologic Alone 50.0 31.5 41.5
Combined EEG and Psychologic 59.1 73.7 65.9
26
-------
2. Tone Decay Test: In order to determine if auditory adapta-
tion 1s present, a Modified Tone Decay Test was administered
at 1000 Hz and 4000 Hz.
3. Speech Discrimination: (a) Speech intelligibility was esti-
mated by using phonetic balanced word lists. Stimuli were
presented at 40 dB SL re: the hearing level for speech using
the open response format, (b) Speech in words function was
estimated using phonetic balanced word lists at a signal to
noise ratio of +10.
4. Speech Reception Threshold: Two syllable words (Spondees) are
uttered to the subject in 2 dB decreasing steps until the sub-
ject can no longer repeat words correctly. The SRT 1s defined at
2 dB above the failure level.
5. Staggered Spondaic Word Test was administered. This test is
reported to be sensitive to central auditory dysfunction. Over-
. V .
lapping spondaic words were fed into each ear separately,, and the
individual then asked to report on what was heard. This may be
diagrammed as follows:
Left Ear Both Ears Right Ear
Input: up stairs
down town
Correct Report: "upstairs", and "downtown"
Incorrect Report: "updown" and "up town"
Errors are reported for both ears in the competing and noncompetlng
circumstance.
6. Goldman-Frfstoe-Woodcock Auditory Selective Attention Test: The
subject is asked to identify pictures corresponding to words uttered
27
-------
i'.O
on a tape under three conditions: (a) against a background of
fan-like noise, (b) against a background of cafeteria noise,
(c)against a background of a verbal story. The signal-to-noise
ratio varies from a positive to negative ratio in each situation.
Percentite scores for each age are calculated.
These tests were administered in fixed order to conform with the assessment
1'-
protocol of the speech clinic. While order effects are a possible risk, these
'would be experienced by both high and low, lead subjects.
Data Analysis:
The speech and hearing outcomes were compared using ANACOVA. Distribu-
tions of outcomes were first evaluated and appropriate transformations (either
log or z-score) were applied to non-Gaussian distributions of data. Several
outcomes were severely non-normal in distribution (e.g. bimodal). These were
dichotomized as close to the median as possible. Tone Decay and Pure Tone
Audiometry Variables did not require transforming. The following variables
were transformed:
Variable Hone a Score log Dichotomize
SRT Right /
SRT Left /
SSW Right Non-competing /
SSW Left Non-competing /
SSW Right Competing / /
' SSW Left Competing / /
GFW - fan /
cafeteria /
voice /
28
-------
(transformed variables, continued)
Variables None z Score log Dichotomize
Speech Discrimination:
quiet - right /
quiet - left /
noisy - right /
noisy - left /
Five covariates identified as important predictors of outcome in the
earlier neuropsychological study (4) were employed in the ANACOVA to con-
trol for socioeconomic confounding. These covariates were: mother's edu-
cation, father's social class, family size (number of pregnancies), mother's I.Q.
(Peabody Picture Vocabulary Test), and mother's age at subject's birth.
Pure tone audiogram results are shown in Table 7.
The F-ratios and p values on all tests with both raw and transformed
data are displayed in Tables 8 and 9. While no significant differences at the
p * 0.05 level were found on the pure tone audiometric examinations, low lead
subjects had lower mean values (indicating greater accuracy) on 13 of 16 out-
comes. The 4 tests of speech processing were comprised of 17 outcome measures.
Four outcomes differed at p _£ 0.05 (SSW left non-competing, GFd voice and total
score, and speech descriminations, quiet, left). Of the 17 outcomes, 15 tended
to favor one low lead group.
The speech data suggest low lead affects speech processing, but replication,
with a smaller battery of tests is indicated.
29
-------
Table 7
MEAN AUDIOMETRIC VALUES (+ STANDARD DEVIATION)
OF HIGH AND LOW LEAD SUBJECTS
Purs Tones High Lead Lou Lead
Right 250
500
1000
2000
3000
4000
6000
8000
Left 250
500
1000
2000
3000
4000
6000
8000
12.50 ± 10.1
11.07 ± 8.1
8.21 ± 9.9
5.00 ± 6.5.
8.61 t 8.9
7.50 ±8.7
15.00 ± 11.1
11.43 t 13.2
11.07 ± 6.8
9.00 t 4.4
5.71 ± 5.5
2.14 t 3.2
7.14 t 4.7
5.36 t 4.6
13.57 ± 7.4
15.35 ± 16.1
11.09 ±
9.78 t
6.52 ±
2.83 ±
8.47 t
6.95 ±
14.13 t
12.17 ±
8.48 ±
7.17 t
4.34 ±
2.17 t
6.74 ±
5.87 t
11.74 ±
10.22 ±
7.2
5.3
4.9
3.9
7.0
9.6
12.5
10.1
4.6
4.5
7.0
5.6
5.8
5.6
7.5
7.0
30
-------
Table 8
MEAN VALUES AND SIGNIFICANCE TESTS
(AFTER ANALYSIS OF COVARIANCE) FOR TESTS OF SPEECH PROCESSING
X X
Sigh Lead Lou Lead *Signif£eance
Right SRT
Left SRT
SSW - right non -competing
SSW - right competing
SSW - left competing
SSW - left non-competing
SSW - right
SSW - left
SSW - total
GFW - fan
GFW - cafeteria
GFW - voice
GFW - total
Speech Disc. : Quiet right
left
Noisy right
left
5.28
4.86
24.57
38.14
41.14
22.14
31.65
33.21
31.57
50.77
36.62
53.23
35.92
93.71
91.86
72.57
74.00
6.08
5.13
25.31
39.60
41.09
24.65
32.70
33.69
33.31
66.78
47.48
77.35
61.39
96.17
96.25
74.44
72.00
p » .048
p ' .05
p » .035
p = .036
*after appropriate transformation and adjustment for covariates
31
-------
Table 9
OUTCOME OF SPEECH PROCESSING: VALUES GIVEN ARE F RATIOS AND P VALUES AFTER ANALYSIS OF COVARIANCE
OF RAM AND TRANSFORMED DATA WITH ALL COVARIATES, AND SELECTED COVARIATES FOR EACH ITEM
All Covariateg Selected Covariatee All Cavariatea Selected Covariatee
Rou Raa Raw Transformed
1.09
.7
.02
.07
.08
4.2
.05
.00
•3
1.65
t.5
4.2
4.9
1.9
6.2
.7
.14
NS
NS
NS
NS
NS
.048
NS
NS
NS
NS
NS
.05
.035
NS
.019
NS
NS
-
^
.09
.07
.2
3.7
.14
.08
.5
.97
1.2
8.2
8.4
.97
9.2
.04
.14
-
_
NS
NS
NS
.06
NS
NS
NS
NS
NS
.007
.006
NS
.004
NS
NS
1.2
,05
.02
.01
.085
4.2
.05
.00
.31
1.5
1.57
-
-
.2
3.5
1.05
.089
NS
NS
1
NS
NS
NS
.048
NS
NS
NS
NS
NS
-
-
NS
.069
NS
NS
1.5
?.49
.01
.01
.00
2.5
.16
.09
.6
.45
2.3
-
-
1.1
4.7
1.2
0
NS
NS
NS
NS
NS
NS
NS
NS
NS
NS
NS
-
-
NS
.036
NS
NS
-------
Study II; Teachers' Behavioral Ratings 1n Lowell School Children 1n
Relation to Dentine Lead Levels
Study Sample: Children attending first grade 1n Lowell, Massachusetts In
academic years 1977 and 1978 comprised the cohort sampled. After two months'
experience with each class, teachers were released from class to evaluate the
subjects on an 11-item teachers' rating questionnaire (see appendix 1 for
questionnaire format).
In evaluating each Item, a good outcome (I.e. Hyperactive? "No" or
Able to follow directions? "Yes") was scored as 1, a bad outcome scored as
0. Thus, a score of 11 would be perfect, a score of 0 very poor, and a score
of 5 mldrange. This sum score was only used to select patients for follow-up.
Item analyses were also done for each group of subjects.
Shed decisuous teeth were collected by the teachers who were asked to
note whether an appropriate vacant socket was found in the donor's mouth.
Teeth were collected from 1,273 children (624 males, 652 females) who were
evaluated by the questionnaire.
Children of both sexes were evaluated with respect to teachers' questionnaire
and dentine lead level. Only male children were candidates for follow-up with
family interviews and psychological testing. Within each class, all male
children who were designated as "Hyperactive - yes" on the teachers' rating
scale were selected for follow-up. In the same class, we attempted to select another
male identified as "Hyperactive - no" with a sum score of 10 or IT (referred to
as "control"). We also attempted to select from the same class a child identified
as "Hyperactive - no" with a sum score of 5-7 (referred to as "midrange").
Follow-up was attempted on a total of 447 children.
33
-------
Essentially complete follow-up was obtained on a total of 215 children. In-
complete but substantial follow-up data was collected on an additional 18
children. The distribution of the 447 children selected for follow-up and the
215 children with complete follow-up data are displayed In Table I.
Selected for Pb
Follow-up Avail.
151 53
160 78
136 60
Follou-up
Completed
Yes
Yes
Yes
Pb '
Avail.
36
53
41
"hyperactive" males
" control" males
"nridrange" males
In addition, a number of performance tests were utilized to evaluate those
children who were followed up.
Data Analysis: The classroom behavior of students from Lowell, Massachusetts
In relation to their dentine lead levels was evaluated in two ways. In the
first evaluation, subjects were classified Into five groups according to their
dentine lead level. Group 1 _<. 6.4 ppm; Group 2 * 6.5 to 8.7 ppm; Group'3 »
8.71 to 12 ppm; Group 4 * 12.01 to 18.1 ppm;. Group 5 }_ 18.2 ppm. These bounda-
ries were selected to be similar to those used 1n the study by Needleman et al.
(1979), but were modified slightly by examination of discontinuities in the
dentine lead distributions. Males and females were then separated, and the
proportion of negative reports on each item of the teachers' rating scale calcu-
lated for each group. Figures 4 and 5 display the Increased incidence of negative
reports in relation to increased dentine lead level.
In the second evaluation, male subjects were followed up and socioeconomic
data obtained from parental questionnaires. These data were treated as covariates
and the relationship of lead to teachers' ratings evaluated by multiple logistic
regression analysis. This allowed us to more clearly estimate the role of lead
in classroom behavior.
34
-------
Figure 4
60
50
40
f
20
MALES-LOWELL
Dentine Lead
I. <6.4ppm
2.6.5-87
3.8.71 -12
4.I£OI -18.1
5. > 18.2
has difficulty
The .relationship between negative teachers' ratings and dentine lead level
in males. Each sample was classified into 5 groups according to dentine
lead level.. Each item was then scored. Within each item, Group 1, lowest
lead level, is at the left; Group 5, highest lead level, is at the right.
35
-------
Figure 5
60?
50
40
30
20
10
FEMALES-LOWELL
Dentine Lead
I. <6.4ppm
2. 6.5-8.7
3.8.71 -12
4.12.01-18.1
5.
has difficulty
The relationship between negative teachers' ratings and dentine lead levels
in females. Each sample was classified into 5 groups according to dentine
lead level. Each item was then scored. Within each item, Group 1, lowest
lead level, is at the left; Group 5, highest lead level, is at the right.
36
-------
Evaluation of Teacher Ratings; .A cluster analysis of the 11 Items In the
teachers rating scale was performed In order to reduce the outcome space
and thereby alleviate the potential problems of multiple comparisons. Two
separate cluster analyses were performed for each sex, utilizing tetrachorlc
and Pearson correlation coefficients. In addition, factor analyses of the
teachers rating scale were performed as a validation procedure. The various
procedures all yielded similar results for each sex. The resulting clusters
for males are defined below. (Because follow-up data was only obtained for
males, final data analysis was restricted to males.) Two items (frustrated
and day dreams) were not segregated, and were treated as separate outcomes.
Cluster 2
hyperactive
Impulsive
Cluster 3
distract! ble
disorganized
Cluster 4
Cluster 5
frustrated daydreams
Cluster 1
not persistent
dependent
poor overal1
functioning
less able to
follow simple
directions
less able to
follow sequences
of directions
Cluster scores were assigned on a monothetlc basis. The distributions of
scores within the clusters (males) were evaluated 1n order to determine
the most sensitive method foe scoring the clusters dlchotomously. The
distributions are displayed below: (Entries are the number scoring yes on
the variable given their total score.)
37
-------
Cluster 1
Total 1
Yes
0
1
2
3
4
5
Persist Independent Directions 6 Seq°W
-*>.
16 4 132 2
59 16 122 41
91 79 125 48
147 154 175 113
Overall Total
126
2 156
26 134
50 134
175 18q
1802
Cluster 2
Yes Hyperactive Disorganized
o
1 16 191
2
Total
1458
207
J31
1896
Cluster 3
T°yf] # D1stract1b1e Disorganized
0
1 422 40
2
Total
953
462
378
1793
38
-------
It Is clear from the results for Cluster 1 that a great majority of
the cases wich received one yes score for the entire cluster obtained it
from the item "follow simple directions". It was, therefore, decided to
eliminate this Item from the cluster. Thus, Cluster 1 was evaluated as a 4-
ttem cluster in the following manner. All cases who had not received a "yes"
response to any of the four items were given a cluster score of 0. After
eliminating ineligible cases (lead poisoning, etc.), a total of 243 cases
received a score of 0. All cases who had received a "yes" to every one of
the four Items were given a cluster score of 1. A total of 1063 cases received
a score of 1. All other cases (i.e., all cases which had not received the
same score on every item) were eliminated from the subsequent analyses.
The results for Cluster 2 indicated that most cases which received a
single "yes" within the cluster, received it for the impulsivity item.
Since 1t was felt that this was reflective of the hesitancy of teachers to
label children as hyperactive, it was decided to pool cases which had re-
'<*"
celved at least one "yes" score in the cluster. Thus, there were 1,446 cases
who had been scored "no" for both items and received a 0 for the cluster.
A total of 332 cases had received a "yes" for at least one item and, there-
fore, were scored as 1 for the cluster.
Cluster 3 exhibited a similar pattern to Cluster 2 in that a large
proportion of the cases with discrepant results for the two items followed
a particular pattern—In this case, a "yes" score for distractlbility.
However, since no underlying reason could be identified for this pattern,
the:strategy utilized for Cluster 1 was applied to Cluster 3. Thus, all
cases with discrepant results for the two items were eliminated. Cases with
a "no" response to both items (375 cases) were given a cluster score of 0.
Cases with a "yes" response to both items (941 cases) were given a cluster
score of 1.
39
-------
Multiple Logistic/Model : The effect of dentine lead level upon each of the
3 clusters was evaluated utilizing a multivariate logistic model:
log (*i/l-p ) « n B1 x 1,
P , •
j * probability of a score of 1 on a cluster for subjects
X1, » 'predictor variable i, for subjects
BJ a effect of predictor variable 1
1 <.
The model was formulated so that the maximum number of available subjects was
utilized and had the following features:
1. a main effect Indicator (0-1) variable was Included
which had the value 1 If follow-up Information was
available; 0 otherwise.
2. a main effect indicator variable was Included which had
*
the value 1 if dentine lead data was available; 0 other-
wise.
3. the follow-up variables were transformed into sets of
discrete indicator variables (Table 10 lists all follow-
up variables utilized in the analysis and their categori-
zations. All follow-up variables were scored as 0 if
follow-up data had not been obtained.
Thus, the model took on the following form:
log (P/l-p) = I7 * I, Cc BI Xj) + I2+ I2 (3-al - Xj Lj
Ij a 1 if follow-up data available
0 if otherwise
x * set of indicator variables for follow-up data
Ig « 1 if dentine lead data available
0 otherwise
L » set of indicators for lead data
40
-------
Representation of continuous predictor variables by sets of Indicator
variables was deemed desirable^ This approach was utilized since 1t was likely
that the relationship between the covarlates and the logit of the probabilities
was not linear In form. Therefore, utilization of a continuous variable
would be Inappropriate without employing complex functions such as polynomials.
The generation of discrete classes was based upon both theoretical and true
distribution of each variable. A number of cases had two to four dentine
lead determinations. Multiple determinations were examined for concordance.
If all determinations were determined to be concordant, an average lead value
was determined and used in the analysis. If two of three (1 case) or three
of four (3 cases) readings were concordant and markedly different from the
remaining result, the discordant result was excluded before averaging. If two
or more results were discordant, the case was treated as if no dentine lead
data was available (10 cases).
Method of Determining the Effect of Lead Upon Behavioral Clusters: The goal
of the analysis was to evaluate the effect of dentine lead level on the be-
havioral cluster scores after controlling for socioeconomic - environmental
effects. In order to achieve this goal, the following sequence of analyses
was performed for each cluster:
1. Determination of the follow-up variables which had a signi-
ficant ( p ( .05) readership with the cluster by using step-
wise multiple logistic regression. This model automatically
Included the main effect Indicator variables for the presence
of a dentine lead assay result. The dentine lead level indi-
cator variables were not included at this stage of the analysis.
41
-------
Variable
Race
Age
Table 10
COVAJUATES AND CATEGORIES SCORED IN LOGISTIC
REGRESSION ANALYSIS OF TEACHER'S RATINGS
Beferrent Category
White
1968
Blrthwelght
Delivery
Pregnancy Length
Mother and child left
hospital together
Mother's education
Father's education
Mother HolHngshead:
occupation
Father HolUngshead:
occupation
111-120 ounces
Vaginal
40 months
Infant's time In hospital 5 days
Yes
12 years
12 years
1-5
Scored Categories
Nonwhlte
1969
1970
1971
1972-1973
< 88
89-110
121-134
> 135
Caesarian
( 37
38-39
) 41
5-7
> 7
No
< 8
9-11
13-15
> 16
( 8
9-11
13-15
> 16
1-4
42
-------
Table 10 (continued)
Variable Sgfervent Category
Mother head of household No
Number of hospitallzatlons 0
Number of head injuries
Number of seizures
Number of Illnesses
Immunizations
Other health problems
Medications
Enures1s
Number of pregnancies
Miscarriages
Birth order of child
Mother's IQ
Mother's marital status
Mother's age at birth of
child
Attended preschool
Attended daycare
Dentine lead
0
0
0
Yes
No
No
No
1
0
first
101-115
married
20-25 years
Yes
Yes
0-6.37 ug
Scored Categories
Yes
1-2
> 3
> 1
> 1
> 1
No
Yes
Yes
Yes
2-3
4-5
> 6
> 1
second
third
fourth
> fourth
< 35
86-100
116-125
> 126
unmarried
< 19
26-30
> 31
No
No
6.37-8.70 ug
8.71-12.0 ug
12.01-18.2 ug
) 18.2 ug
43
-------
2. Determination of the significance and dose-response pattern of
the lea'd effect. This was done by adding the set of dentine
lead Indicator variables to the liiultlvarlate model which was
derived 1n step 1.
•; RESULTS
jt
Glister 1; Effect of Follow-up Variables; Table 11 displays the results for
the variables which were selected by the stepwlse logistic regression. The
relative risks for the blrthwelght variable Indicate an Increase for both the
lowest and higher blrthwelght groups. Although the risk for those not
attending daycare appears quite large, only 12 children in the sample did
attend daycare; thus, the estimate 1s probably unstable.
Cluster 1; Lead Effect; The addition of the set of lead level variables to
the model derived 1n step 1 did not yield evidence of a significant lead
effect for this cluster. The results are displayed In Table 12. The risk
In the highest lead level group 1s essentially equal to the risk In the
lowest lead level group (referrent).
Cluster 2: Effect of Follow-up Variables: The results of the stepwise logis-
tic regression utilizing the follow-up variables alone are shown in Table 13.
The most notable effect is that of mother's education. Children whose mothers
had less than 9 years of education had approximately 17 times the risk of
being perceived to exhibit hyperactive/impulsive behavior than children whose
mothers graduated from high school (12 years) but went no further. An Inverse
relationship between yejars of school and a problematic score on this cluster
can be seen very clearly upon examining the relative risks:
Years of Education Relative Risk
8 17.07
9-11 1.93
12 1.00
13-15 .42
16 .16
44
-------
Table 11
CLUSTER 1: N - 1306
Variable
Intercept
V
I2
Daycare
Birth-
weight
s
136
465
12
12
34
36
21
Value
Follow-up Done
Lead Done
Yes
^ 88 02.
89-110
121-134
) 135.
4*
95 Q
0 Jr»
£fc CJ
1.364
-0.206
0.331
5.092
0.749
-0.888
0.842
1.447
]o A
Cj Co
15.62
-0.44
2.12
1.25
0.65
-1.45
1.10
1.28
Relative
Risk
.81
T.39
162.67
2.11
.41
2.32
4.25
£&*.(§
ST
-------
Table 12
ADDITION OF DENTINE LEAD LEVEL VARIABLES
TO THE MODEL DESCRIBED IN TABLE 11 (CLUSTER I)
Dentine Laag_ _ Regression Coefficient/ Relative
Level (ppm) Coefficient Standard Error Riak
0.00-6.36 146 Referrent • 1.0
6.37 - 8.70 103 .404 1.10 1.50
8.71 -12.0 90 .565 1.41 1.76
12.01 -18.2 83 .196 0.53 1.22
) 18.2 43 -.117 -0.27 0.89
46
-------
TABLE 13
CLUSTER 2
Variable
Intercept
IT
V
B1rthwe1oht
(referrent
= N 1-120
oz.)
Mother's
education
(referrent
= 12 yrs.)
Mother's IQ
(referrent =
101-115 pts)
N
195
618
N«50
17
57
48
23
N=87
19
44
28
17
N=44
19
5D
38
29
Value
Follow-up Done
Lead Done
](. 88 ozs .
89-110
121-134
I 135
J. 8 years
9-T1 "years
13-15 years
>. 16 years
j.85
86-100
116-125
> 126
•w
c c
.2-3
l'<
QJ Cj
-1.561
1.004
-0.244
-1.499
-1.104
-1.154
-0.439
2.838
0.659
-0.859
-1.845
-0.678
0.980
0.490
0.900
1s-
Q •+•*
G to
-19.85
"1.89
- 1.79
- 2.15
- 2.30
- 2.42
- .70
2.86
1.43
- 1.31
- 1.53
- .83
2.05
0.89
1.48
1
t*» jo •» "9
2.73
.78
1
.22
.33
.32
.64
2
17.07
1.93
0.42
0.16
"3
0.51
2.66
1.63
2.46
J
0
.014
.018
.018
47
-------
TABLE 13
CLUSTER 2
(continued)
•» -w
0 « V Q
Variable N Value wo o t,
M |t* 41 weeks -1.233 -2.17
1
vS Qi
* J5 •**
•t* W
-W 3>
OrX 8. 0
1*4^03 4i -W
-------
Table U
ADDITION OF DENTINE LEAD LEVEL VARIABLES
TO THE MODEL DESCRIBED IN TABLE 13 (CLUSTER II)
Dentine Lead
Level (ppm)
0.00 -
6.37 -
8.71 -
12.01 -
)
6.36
8.70
12.0
18.2
18.2
.. Regression
Coefficient
179
125
123
123
65
Referrent
-.433
0.113
0.177
0.286
Coefficient/
Standard Error
-1.27
0.36
-0.54
0.78
Relative
Risk
1.0
0.65
1.12
0.84
1.33
Overall P to enter » 0.414
49
-------
It should be noted that these are estimates controlling for the other vari-
ables in the model. It should also be noted that a significantly elevated
nil
risk is found in children of mothers whose IQ is between 86 and 100 as com-
pared to those whose mothers had IQs of 101-115. Also, children who were
born after very long pregnancies had a low risk on this cluster.
It is more difficult to explain the patterns in birthweight and father's
education, but the interaction of the 8 variables in the model should be
kept in mind.
Cluster 2; Lead Effect: The set of lead level indicator variables was added
to the model derived in step 1. The results are displayed in Table 14. Al-
though the highest risk is in the group with the highest lead level, the ef-
fect is not significant nor is the overall effect for the set of lead vari-
ables. There is little evidence of a dose-response effect in these results
for this cluster.
Cluster 3: Effect of Follow-up Variables; The model resulting from the step-
wise regression on the follow-upn/ariables is shown in Table 15. The results
for mother's education show a clear inverse relationship between years of
education and the child's risk on this cluster as shown below:
Years of Education Relative Risk
8 3.62
9-11 3.78
12 1.00
13-15 .54
16 .19
50
-------
Table 15
CLUSTER 3 - N = 1316
— — — — — —
Variable
^^^^"""••^••^•••^•••^•M
Intercept
II
r2
Glrthwefght
(referrent
» 111-120
oz., N-33)
Mother' s
education
(referrent
• 12 yrs.,
N » 67}
— • . —
* Value
•^
^ —————— .
135 Follow-up Done
452 Lead Done
12 ^88 oz.
39 89-110
33 121-134
18 >_135
11 j. 8 years
32 9-11 years
15 13-15 years
10 >. 16 years
Regression
Coefficient
1
-0.898
1.362
-.303
-1.983
-0.597
-1.950
-1.944
1.287
1.329
-0.622
-1.680
+A
K S.
-------
Table 16
ADDITION OF DENTINE LEAD LEVEL VARIABLES
TO THE MODEL DESCRIBED IN TABLE 15 (CLUSTER 3)
Dentine Load
Level (ppml
0.00 - 6.36
6.37 - 8.70
8.71 -12.00
12.01 -18.20
H8.20
ft
131
94
92
83
52
Regress-urn
Coefficient
Referrent
-0.242
0.028
-0.456
1.030
Coefficient/
Standard Error
-0.72
0.01
-0.14
2.93
Relative
Risk
1.00
0.79
1.00
0.96
2.30
overall p to enter « .013
52
-------
Table 17
Frustratable
Variable
Intercept
IT
!
S Value
215 Follow-up Done
616 Lead Done
el:
*<•> '!*
aj o
l£
w ^%
ty^ Qt
-1.2T5
0.346
-0.017
t: s, « s.
os. a Os
«b "** 2* a
-17.04
1.42 1.41
-0.14 0.98
Mother's
education
(referrent
* 12 yrs.,
N » 107)
13
55
26
14
1 8 years
9-11 years
13-15 years
)_ 16 years
-5.232
0.607
-0.958
-0.248
- 1.31
1.69
- 1.46
- 0.36
0.01
1.83
0.38
0.78
Preschool
(referrent
attended)
Not attended
.696
- 1.79 0.50
.030
.034
53
-------
Table 18
ADDITION OF DENTINE LEAD LEVEL VARIABLES
TO THE MODEL DESCRIBED IN TABLE 17 (FRUSTRATABILITY)
Dentine Lead „ '
Level IPPm) *
0.00
5.37
8.71
12.01
- 6.36
- 8.70
- 12.00
- 18.20
> 18.20
17S
125
121
123
69
Regression Coefficient/
Coefficient Standard Error
Referrent
.063
.046
.211
.631
0.22
0.16
0.74
1.98
Relative
Risk
1.00
1.06
1.05
1.23
1.88
overall p to enter » .3524
54
-------
Table 19
OAYOREAMNfi • N « 1778
si
Variable 3 'Satu* • '£ *a
||
Intercept -0.959
If 195 Follow-up Done -0.238
I. 679 Lead Done -0.088
Oaycare (re-
« - 181)*
Not attended 14 -1.399
Aae of sottier
(referrsnt ••
20-25 yrs.f
N » 107)
44 , j19 years -0.344
30 26-30 years -1.743
14 I 31 years -0.017
Otter heal til
no, N • 183)
Problem 12 res 1.517
Patter's
education
\iGim i BIIU
12 yrs.)
13 J. 3 years 0.212
439 9-11 years 1.159
31 13-15 years -0.053
T7 >.16 years -1.345
Patter's
occupation
Holl Ings-
head (refer-
ent « 5)
77 1-4 0.949
52 6-7 0.330
Reprodu
best ava
B «. IE.
So "3 «
•••5 a t, «
_o > . .2 ° •"
S" '£.* ? «
^*» • Q MC *• Q
01 ^7 "•* * ** •M
S ** "8 •?* C
Cj &i as fir 4i a*
-14.30
- 0.56 0.79
- 0.76 0.92
1 .017
- 1.90 0.15
2 .016
- 0.62 0.71
- 3.31 0.18
- 0.22 0.90
3 .024
2.23 4.56
4 .017
0.32 1.24
2.40 3.19
- 0.11 0.95
- 1.63 0.26
5 .042
1.96 2.58
1.33 2.29
ccd from RS>P*^
ilable copy. fyP
55
-------
Table 20
ADDITION OF DENTINE LEAD LEVEL VARIABLES
TO THE MODEL DESCRIBED IN TABLE 19 (DAYDREAM)
-
Dentine Lead
fr"*Z(ppmj "
f^m ^^••^^^••WBiMa^^^MH^^^^^K
0.00 -
6.37 -
8.71 -
12.01 -
>
' i
6.36
8.70
12.00
18.20
18.20
177
126
123
123
70
_
teffresaion Coefficient/
Coefficient Standard Error
"^^^^^^^^^•^•••^•^•^^^^^^^^^__^^__
^^^^^^^^""^•""^^•••••^••^••^^__^.
Referrent
-0.446
0.047
-0.025
0.577
-1.55
0.17
-0.09
1.89
• • „
Relative
Risk
^
^^"•••^•^'^"^^"••^•^•^^^
1.0
0.64
1.05
0.98
1.78
overall p to enter » .059
56
-------
The results of birthwelght indicate those children in the referrent group
(111-120 ozs.) were most at risk on this cluster.
Cluster 3; Lead Effect: Addition of the set of lead level indicator vari-
ables into the model resulted in a highly significant lead effect (Table 16)
(p - .013). The group of children with the highest dentine lead level 0 18.2
ppra) had a significantly higher risk of exhibiting distractible/disorganized
behavior than the children with lower lead levels. The ratio of the coefficient
to its standard error is asymptotically normally distributed and the coefficient
for the highest lead level group is 2.93 standard deviations above 0.0 (p < .01).
Cluster 4; Easily Frustrated.
Effect of Cdvan'ates: Children who attended preschool or whose mothers
were less educated were more likely to b.e scored as easily frustrated.
Effect of Lead; Those children with dentine lead levels greater than 18.2 were
at highest risk for this item, and the coefficient for this group is signifi-
cantly different from zero. The lead effect estimates display a clear dose-
response relationship, although the overall p to enter was not significant.
Cluster 5: Daydreaming
Effect of Cdvan'ates: Infants who did not attend daycare, and whose
mothers were between 26 and 30 years when they gave birth were less likely to
5e seen to daydream. Infants of better educated fathers were also at lower
risk, whereas children with non-specific health problems were at higher risk..
Effect of Lead: Dentine lead levels over 18.2 were associated with increased
relative risk for daydreaming, and the lead effect for this group was nearly
significant (p * 0.0591.
57
-------
DISCUSSION
These studies show that children with elevated lead In their bodies,
but who are considered not lead poisoned, differ In their electroencephalo-
grams, In certain areas of speech perception, and 1n their classroom behavior,
from control subjects with lesser amounts of lead.
In setting standards for permissible exposure to lead, Intense consider-
ation has been given to determining what alteration constitutes a health ef-
fect. The data presented 1n this report bear on that question, and demonstrate
that employing sensitive outcome measures and sophisticated b1ostat1st1cal
techniques enhances the possibility of discovering biologically significant
alterations at lower exposure levels.
i
Smith (10) reported changes 1n EEG patterns, but only in children who
had had encephalopathy. Children entered Into this study clearly had no
. i
episode of Intoxication severe enough to be diagnosed as encephalopathy.
(Any child whose mother was told he or she had a lead problem was excluded
from consideration.) High lead children tended to have midline slowing (In-
creased delta) and decreased mldline alpha activity on EEG. The results
obtained here demonstrate the Increased sensitivity of quantitative EEG
analysis compared to clinical readings.
\
While only 4 of 17 possible outcomes in the speech processing bat-
tery were significantly different at the p < .05 level, most outcomes (14/
17) evaluated tended to favor the low lead group. The data reported here are
in'agreement with the findings previously reported by our group (4). In
that study, among the areas most sensitive to lead effects were sentence
repetition, the token test, and attention.
In our previous study (4), we demonstrated that negative teachers'
ratings on most Items tended to increase in frequency in direct relationship
58
-------
to dentine lead levels. The relationship between classroom behavior and
dentine lead level is generally similar in this study. The overall incidences
of negative reports of each item also resemble those reported earlier in
two different cities.
While the finding of a dose response relationship in our previous ex-
amination of teachers' ratings was intriguing, we were unable to control for
nonlead covariates in that sample. In this study, the follow-up group of
males provided an opportunity to evaluate the contribution of lead separated
out from other possibly confounding variables. After cluster analysis had
reduced the variables to three, the cluster characterized as "distractible-
disorganized" was found to be mos-: sensitive to lead, controlling for signi-
ficant covariates (birthweight and mother's education). A dose-effect rela-
tionship between dentine lead level and relative-risk of exhibiting the
characteristic of cluster 3 was found.
Those characteristic behaviors which cause children to be identified
by their teachers as distractible or disorganized seem to be among the more
prominent expressions of low level lead exposure. The significant differences
on the 6.P.M. selective attention test support this possibility. Longer
reaction time under varying intervals of delay, another measure of disturbed
attention, has also been shown (4). It is tempting to speculate that the
midline slowing on the EEG reported here is an electrophysiological expression
of the same process.
The lead-associated alterations reported here span the range from CNS
electrophysiology, through speech processing to classroom behavior. The
breadth of these effects and their intimate relation to functions critical
to the child's ability to attend to stimuli and learn argue that at the ex-
posure encountered here, lead has produced adverse health effects. Follow-
up studies of these children will provide important information about the con-
sequences of these alterations upon their academic and social adaptations.
59
-------
REFERENCES
1. Hernberg, S., Nikkannen, J., Mellis, 6., and L1l1us, H. D-aminolevulinic
acid dehydrase as a measure of lead exposure. Archives of Environmental
; Health 21:140, 170.
2. Nathanspn, J,Atjjnd Bloom, F.E. Lead Induced inhibition of brain adenyl
cyclase, Nature^2S5:419. 1975.
3. Gibson, S.M., Goldberg, A. Defects in heme synthesis in mammalian tis-
sues in experimental lead poisoning and experimental porphyria.
Clinica.1 Science 38;63. 1970.
4. Needleman, H.C., Gunnoe, C., Leviton, A., Reed, R., Peresie, H., Maher,
C., and Barrett, P. Deficits in psychologic and classroom performance
of children with elevated dentine lead levels. New England Journal of
Medicine 300:689. 1979. '
5. Bartels, P.H. and Wied, G.L. Computer analysis and interpretation of
microscope images: Current problems and future directions. Proc. I.E.E.E.
65:252, 1977.
6. Sherwood, E.M., Bartels, P.H., and Wied, G.L. Feature selection in cell
image analysis: Use of the ROC curve. Acta Cytol. 20:254, 1976.
7. Genchi, H. and Mori, K. Evaluation and feature extractionon automated
pattern recognition systems. Denki-Tsuchin Gakkaj.. Part 1, 1965 (Japanese)
8. Rao, C.R. Advanced Statistical Methods in Biometric Research. Hafner
Press, Mew York, 1974.
9. Lachenbach, P.A. Discriminant Analysis. Hafner Press, New York, 1975.
10. Smith, H.D., Backner, R.L., Carney, T., and Majors, W.J. The sequelae
of pica with and without lead poisoning. American Journal of Disability
1n Children 105:609.
60
-------