EPA/S&B/81/QQ2
                       Review of
"A General Method for Assessing Health Risks Associated
 With  Primary  National Ambient  Air Quality Standards"
                (Draft dated  April  1981}
A  Report  of  the  Subcommittee on Health Risk Assessment
                     September -1981
                 Science Advisory Board
          U.S.  Environmental Protection Agency
                Washington, D.C.   20460

-------
          ^
                      EPA NOTICE
     This report has been written as" part of the activities of
the Science Advisory Board, a public advisory group providing
extramural scientific • information to the Administrator and
other officials of the Environmental Protection Agency.  The
Board is structured to provide a balanced expert assessment
of the scientific matters related to problems facing the
Agency.  This report has not been reviewed for approval
by the Agency, hence its contents do not necessarily
represent the views'an4 "policies of the Environmental
Protection Agency» nor does mention of trade names or
commercial products constitute, endorsement or recommendation
for use,            '•••••'•'•

-------
                   fable  of Contents
                  '•'• :; ••'..-.            .                         Page

Background	V* *«...,,,*,«*.........„..*..,,**«**       I


Comments *..........,,,,.,.	,..	....»,...„       4


Advisory Statement  ,»»'„»»,.„	*.*..,..„..,....,...,       9


Conclusions ..............,	*...;...<,*.* + ...*.,,....       9


Appendices         •  . ';'•


Appendix A,
     Subcowmittet Roster	.*.*.....*.....*.*«..     A-l
Appendix B,
     Cornnte^ts by Geoffrey  S. Watson  on Appendix  Gr
     Mathematical freatment of Dncertainty  in Pol-
     lutant Concentrations of MA  General Method  for
     Assessing Health 'Risks Associated With Primary
     National Ambient Air  Quality Standards" (Draft
     dated April 1981)  *......*.*........	     B~l

-------
Background         ;   •• •;-.;i-:.-^':

     This is the th*rd;: report of the Subcommittee on Health
Risk Assessment of' he ; Science Advisory Boarderitictuing.
approaches to health risk' assessment for alternative national
ambient air quality standards developed under the auspices of
FPA's Office of Air Quality ^lannina and Standards  CO&QPSK

     In 1979, the Subdomait-fcee reviewed a methodology for
assessing health risks associated with alternative  ambient
air auality standards forVosone*  The Subcommittee  stated
its continentsr conclusions,  and recommendations  in a report
dated September 1§79»  ^/

     Among other comments and suggestions, the  Subcommittee

     - found that the methodology developed by  QAQPf? was
       not ready>for application in the process. of  estab-
       lishing national arijjbi'ient air quality standards;  and

     - recommended that; Q&QPS should formulate, a plan
       outlining how the ';,ftgency will (a) develop the
       proposed methodology?: including standards and
       protocols for application, (fa) consider  alter-
       native apprr .iches, and.Cc) select and establish
       the credibility of life.e best methodology,
                            . $''
     Consistent witr the Subcommittee's recommendations, the
Office of Air Quality Planning and Standards developed  a "Risk
Assessment Program Plan1* and contracted with si* research teams
to develop integrated conceptual risk assessment procedures.
This effort resulted in aix 'documents which the Subcommittee
reviewed at a public meeting in September 1980.  Also at that
weetino, the Subcommittee; lizard update reports  on the F/B Risk
Assessment Method  (named'.after its authors, Thomas  B* Peagans
and William P. Billert and  on contemplated next steps in the
OAOPS "Risk Assessment Program Plan**  The Subcommittee's
comments, findings, and conclusions are contained in a  report
dated Deeeisiber 1980,**/
    Review of  "A Method of  Assessing  the Health  Risks  Associated
    with Alternative  Air  Quality Standards for Ozone*1  (Draft
    dated July 1978)„ A Report  of the Subcommittee on  Health
    Risk Assessment,,  EPA/SAB/79/001,  September 1979.   iSie method
    is referred to as the F/B  Risk Assessment  Wethod or the
    P/B Hethod after  its  authorsf Itiomas B.  Peagans, OAQPS
    Analyst, and Dr»  William F«  Biller,  Consultant.

    Approaches to Health  Risk  Assessment for Alternative
    National Ambient  Air  CMality Standards,  A  Bepott of the
    Subcommittee on Health  Risk Assessment,  JSPA/SAB/8Q/003r
    December 1980.

-------
     Briefly, the Subcommittee concluded that
                        ' ' : " ''•/. '                    \
     - the program has produced two and possibly three
       promising approaches,to health risk assessment for
       use in developing" alternative national ambient air
       quality standards; and

     - information presented'on the F/B Risk Assessment
       Method did not refljjsct^'significant improvements since
       the Subcommittee filttte;". re viewed it and discmssed its
       deficiencies in public*session in April 1979*

     Following the Subcommittee's recommendations, GAQPS
selected two approaches as•alternatives for further
development;  the approach presented by M.W, Merkhofer and
the one by Bobert L+ Winkler and Rakesh K. Sarin* As regards the
F/B Method, QAQPS advised the Subcommittee that the method will
be presented in a comprehensive report and will be subject to
extensive in-house and external reviews including another review
by this Subcommittee,  OAQPS then planned to proceed as follows.
If the reviews indicate that  further development of the F/B
Method is warranted? OAQPS, will hold it in abeyance until
alternative methodologies can be brought  into a parallel state of
development.  If the reviews indicate that further OAQPS support
of the F/B approach is not.warranted, all work, on that method
will be terminated except "for the generally-applicable exposure
analysis module. Resources currently designated for further
development of the F/B Method would be applied to the alternative
approach{e s}.  V1

      A comprehensive report, describing the F/B approach was issued
as a draft ancf became available for 'Subcommittee review in
mid-April*   **/
*7   Letter from' Joseph Padgettr  Director,  Strategies and  Mr
~    standards Division, OAQPS, to Dr. Anita  S.  Curran,
     Subcommittee Chairman, dated December  17, 1980.

**/  "A General Method for Assessing Health Risks  Associated with
     Primary National Ambient  Mr Quality Standards"  (Draft  dated
     April 1981) by  fhoroas B.  Ftagans  and William  F,  Biller
                              .-  2  -

-------
      OH.QPS also arranged  for  six  experts to provide in-depth
technical reviews of  the P/i report which became available
shortly before or at  the'Meeting,
   ~1he six experts selected:.'by OAQPS to provide technical
   reviews were*    ;   •  ;v,';>
   Dr. DaviS S. Bell,  Har1iit|f«5 -University Business' School
   Or, C*S« Burton, .Syst$»4" ^applications, Smn Rafael,
      California    '    - .'^\i
   Dr. Richard C. Jeffrey^'lDepartment of Linguistics and
      Philosophy, Maasachiii^tts Institute of f^chnolofy? after
      June 15, 1931s  Department of Philosophy, Princeton
      University       . • .;!. -
   Dr. Isaac Levi, Departaiertt of Philosophy, Columbia
      University        •  :'.'.' :
   Dr. John H. Seinftld, .Department of Chemical Engineering,
      California  Institute'of Technology
   Or. Andrew J.  Van  Horn,  Energy an<3 Environmental Analysis
      Division, Teknetron  itesearch, Qrinda, California
                               — 3 v

-------
Comments       _        ..,•.-.•••

     We again wish to conmend and compliment the Office of Air
Quality Planning and Standards  (OAQFS) on their « Cforts in
developing health risk assessment and evaluation 'methodologies
and recommend that they continue these efforts.

     The contracts comX$$$.oned over this past year and a half
have produced 'valuable -t^iarch on the application of decision
analysis principles and'-t^cfeiiques in the development "of-health
risk assessment and evaluation methodologies to aid in setting
ambient air quality standards.  But we can see little or no  impact
from this research or fraa other sources in remedying the
deficiencies this Subcommittee has perceived with the F/B
approach.  It remains unnecessarily complex, continues to be
poorly communicated despite-repeated attempts at clarification,
contains concepts that have not been fully worked out* and in
some areas contains unclear mathematics. {See Appendix Br Dr.
Watson's critique of Appendix G of the F/B Report,}

     Our perception is tits'fc . EPA * s goal is to have an impiemen table
methodology within the ine'act several years for assessing health
risks to aid the Administrator  in setting ambient air quality
standards.  While the F/B approach may have commendable aspects
as a research effort, it is not, in its present ^orm, an
implementable tool for'pifoiic policy decision ma. ing, an<3» we
believe, will not be accepted by a large part of., the scientific
community or the public* .- J^s we have repeatedly 'uggested,
Peagans and Biller should publish their work in peer-reviewed
journals* and the professional  community shoulcl 3udge the
merits of their viewpoint*  It  is our belief -that even  if the
F/B viewpoint were to become widely accepted, this would not
occur in the foreseeable future,

     The scientific data and  reasoning should be emphasizedf not
the mathematical complexities of processing probabilities. There
is a lack of emphasis in 'the  F/B approach on what constitutes  an
adverse health effect, what constitutes the most sensitive
group, and which health effects should be most important  in
establishing ambient standards.  The discussion of  responses of
concern, sensitive populations, seriousness of effect*
uncertainty about causality  (pages 7-2 to 7-4 of the  F/B  Report)
is unsatisfactory and superficial.  Agreement on basic
definitions and measurement-criteria  is essential  if  comparisons
between experts1 probabilities  are to have any meaning.
                              — 4 —

-------
     Another area of difficulty is that of working through the
mathematics of a complex'.:;model using upper and lower
probabilities,  in this situation, a probability density function
does not exist, and integration over the range of an uncertain
quantity becomes much more:'difficult.  In complex parts of their
analysis, such as the dose—response discussion in Appendix R,
Feaqans and Biller go back to usinq singIs-valued probabilities
and density functions,  Peagans and Biller state (p. E-2)»
"With care, it is relatively straightforward to recast the
results and the supporting, treatment in the form of upper and
lower probabilities." ;We are not convinced that the extensive
development of pages E~'2 to S-29 and Appendix F can be carried  -
through with upper and lower probabilities without encountering
some intractable or severely restrictive mathematical difficulties.
Only the most diligent and determined probability expert  could
understand the mathematical development of Appendices fi and F,
given that it is carried;out using upper and lower probabilities.

     A third area of di.£;|4calty was pointed out by Professor Levi.
It is not clear, how the filler and lower probabilities are to be
used to arrive at decisicmp,  *?fte risk ribbon approach advocated
by Feaqans and Biller isT«3;,hoe and unsatisfactory.  Professor Levi
notes that approaches to^tf ision theory using tapper and  lower
probabilities have been '«l$e;v;eloped by himself and others and published
in the literature, :but..th*t«* approaches have not been incorporated
into the P/B approach,
                            i
     The Subcoiroittte sucrqwvts that all material relating to upper
and lower probability,'the "theoretical foundations of probability?
and risk ribbons should be:considered-basic research and
separated from the approaeto.  For an implementable tool,>we
suggest the analysis shooltf. be based on standard decision analysis
or Bayesian methods, using single-valued probability
distributions*  Bach such  distribution would represent the
judgment of one expert or a consensus among a group of experts-for
each uncertain input to the' risk analysis/ such as the percentage
of a well-defined sensitive group suffering a health effect from a
given pollutant concentration-  "Hiese distributions would be
combined with meteorology and human activity models to produce
probability distributions on threshold exceedances anc! the number
of adverse health effects for alternative standards.  Where
experts disagree on a probability distribution that is used as
input, EP& should carry out a sensitivity analysis to determine
the impact of the disagreement. Where the  disagreement is
significant to the decision on choice of standard, each of the
disagreeing experts should be asked to communicate the rationale
for his or her probability assessments in such a form that this
rationale can be reviewed by scientific peers and presented to
EPA's senior management •.

-------
     Any analysis which'-ls^to be successfully osed and defended
by EPA in the standardHSH&'tlting process must be as simple as-is
consistent with a sattsfftci^pry solution.  Its developers and  users
must b€ able to describe;;*t!i
-------
     Rereading the comments of the six reviewers, as Regards  the
merits of the P/i approach, it seems to us that  they'were  mainly
addressing the questions Is the F/B approach sufficiently
interesting research thatpli;deserves continued  support?   This,
in our opinion, is very different front the question of  whether
the P/B approach could or should be implemented  for an  ambient
air quality standard in th'^;:,»ext several years.  'We read most of
the reviews to state or iiJply that the 'approach  and the
communication of it had.'Significant problems yet to be  resolved*
We find Bell's comments o?i'; Efficiency and the  caveat  in Burton's
review particularly important in relation to the feasibility  of
successfully implementing the P/B approach.

     Bell statess

         "It is well known that there are decreasing
     returns to scale with the complexity of a model  to
     the point where you can -end up worse off  than no
     model at all.  I thiiiJt: the models here are  overly'
     complex for the current state of applied  art.   I
     agree with the approach' but I believe it  is a
     little too much all at ones.  I would be  happier
     seeing more modest-goals at this point.   If the
     report is only intended'-to be a look at the future
     or as a research document as opposed to a draft  of
     an EPA manual then I'fn content.  I don't  believe
     it's realistic to expert? a methodology such as  this
     to be performed witn'':iitoch creeden.ce  Esic] given  to
     it in the next 10 yewta."  (p.3)

     Burton adds*          :.:;;

         "The concepts and iiethodology embodied  in the
     approach appear; sound/^4 merit  (immediate) further
     development towards ;tpi$,%oal of  incorporating the
     techniques into, the-''si&&dar3 setting process*
     However, while I support'completely the approach,  •'
     it is apparent that, .neither the  style nor the  tone
     of the report will facilitate acceptance  of the
     approach by the technical, regulating,•or regulated-
     community.  To taker as,the authors do, such an
     unyielding position regarding the approach  of
     eliciting probability-'.estimates, and to use lan-
     guage that to many, ii.-not most, will be  perceived
     as offensive, appears to be both unnecessary and
     counterproductive.  This statement  should not  be
     construed to mean that  the authors  should not be
     bold and clear  in stating their  views?  it {the
     statement) is meant to  suggest,  however,  that  there
                               . 7 -

-------
     exist less polarislitq iseans of conmainicating their
     views,  'fie authors,*, should be satisfied,  that both
     their peers and time 'will be the final 1udge.   In
     short why make the .Introduction and use  of  this new
     methodology into thi (f«cision-€isking process more
     cliff ieult .than
     The comments of th'e^ir^iewers reflect  in general many of  the
sentiments expressed . hy-;..^jiljj, Subcommittee 2% years  ago  in that
they endorse the use of''.'ji^p€ntal probabilities "as a 'useful
additional tool in the ;.flifi|ard-settin« process r while  raising a
variety of specific conc^ferii^ with the methodology.  Ihough the
overall cciranents • appear. fc£;Se  favorable, it must  be recognized
that this is a first view
-------
Advisory Statement ^J  vv.•':%";•••

      •Hie Subcommittee again .wishes  to  commend and compliment the
Office of Air Quality Plafwting and standards  (QAQPS)  on their
efforts in- developing health risk assessment  and evaluation
methodologies and recommends that these efforts be continued.
But we see no significant progress.in the  P/B Beport.  The
comments of the reviewers^reflect, in general,  many of  the
sentiments expressed by.-,tfcls Subcommittee  2 1/2 years ago in- that
they endorse the use of \J:t3-pE|mental probabilities as a useful
additional .tool in ; the:'sl'iti«3ar«3-setting process while raising a
variety of specific coocetriii.

      We recognize that this is  a first view  of the P/B approach
by these commentators,, but  we have had  the advantage  of watching
the development of the process over  time,  and our opinion is that
there has been a lack of' significant progress toward  developing
an implementable tool.  We'-see no reason to believe this
situation will change* •; -'_,;  ••'

      Despite our conc«rtt&.''about this specific piece  of analysis,
we believe that the overall program  of  health risk assessment in
which Q&QFS is now ' enga$&& is well conceived  and making excellent
progress.  We believe that ••'•'the OAQPS is the appropriate unit
within EFA to pursue this work,  and  we  recommend that they
continue to develop the alternative  approaches now under
consideration,              :
Conclusions

      In the view  of  the .Sufocojpiittee,  the P/B method is not a
promising approach  for  developing  an iinplementable tool in the
foreseeable future.

      The Subcommittee  recommends  that  OAQPS continue to develop
the two alternative approaches.
VThis  is  a  slightly edited version of a statement, reflecting
    a consensus  of  Subcommittee members, read into the record by
    the Subcommittee Chairman, Dr.  Curran, at,the May 13, 1§81
    meeting  of the  Subcommittee*


                                 - 9 -

-------
                                                  Appendix A
                     SCIENCE ADVISORY BOARD

             SUBCOMMITTEE'::;0W HEALTH RISK ASSESSMENT
Chairman:
Dr. Anita S. Curran, Commissioner, Department of Health,
    Westchester County, County Office Building 2,
    150 Grand Street, white Plains, New York  10601
Members:
Dr. Bernard Altshuier, Director of Biomathematics  laboratory,
    Institute of Envirowueiital Medicine, New fork  University
    Medical Center, 550 First Avenae, New York, New York   10016

Dr. Judy A. Bean, Associate 'Professor, College of  Medicine,
    Department of Preventive ledieine ana Environmental Health,
    University of Iowa, ^ Iowa i, City, Iowa  52242

Dr. Granger Morgan, Head, • Etepartraent of  Engineering ani3 Poblic
    Policy, Carnegie-Mellon University,  5000 Forbes Avenue,
    Pittsburgh f Pennsylvejtia  15213

Dr, D, Warner North, Principal, Decision Focus Inc*,  5 Palo Alto
    Square, Suite 410, Palo'Mto, California  S4304

Dr. Geoffrey Watson, Professor and Director of Graduate Studies?
    Department of statistics, Fine Hall, Princeton University:,
    Princeton, Slew Jersey 08544
SAB Staff Officer?
Mr, Ernst Linde? Scientist Administrator,  Science' Advisory
    Board, A-lOlKjr U,S, ^'Environmental  Protection Agency,
    Washington, B.C.- 204'«o:"'"Phones  (202)  472-0444
                                A-l

-------
                                                        Appendix B
Comments by Geoffrey S»..Wat.son on Appendix G, Mathematical
Treatment of Un,eertaintyVi*C;:;P611utanit Concentrations of  BA
General Method  for Assess-J^f Health Risks Associated with  •
Primary National Ambient'Air Quality Standards*  (Craft dated
April 1981}             " "'t'^i '
Appendix G seeks  the  probability distribution of the
n  largest, in  a series of  la  values,  ISiree years  ago  I
pointed out the need  to consider dependent values,  ^he authors
now recognize  this bat 'laalie.': totally incorrect statements,  The 1
lines giving the  theory for/- the independence case are marred by
five typos i one leading to 'nonsense.  Hie discussion of this
topic is at best  opaque.. ;.jf.

If the  15  values or<3'eri&C:|fem largest to 'smallest  are
                If  G(x) :;».^i;|t:' i>x)  , for any
 (as in  (Ol)  }  t  the c&aiieelithat exactly  v  of the   m  values
will be greater than "C^IisVr from the binomial distribution,

  ,                ™™ipnnmiN^^^^HtHHnHiiMWM ^  ^  %,J ^ ^rf £   ^ ^
Now x,
-------
         particular case,,of  the Weibull distribution  (first
paragraph on p. CS-3) 'co^feetly sets
 ;     • .            G(C) :*;''ejcp - (C/5)k        .   (G-6)
fhe expected number .ofAeitceeclaiices  E  of a standard '$S£$  in
iS            '  .     v''i^';^ •      ."• • '       - :.^::.-    /"':•: '" •  '   ''
 ;    To proceed-one iBu^t$pii6w what: is meant by, .the P^
This is apparently 'defin%'d'r;in Section 8.2, but'."this.section  is
absolutely baffling.   '."•':.-;:--. .
     From Section'8.1, ••ifc;:^«ems clear that if the standard
requires  E  in  (i) to  he less than unity., then cst£ could be
determined from  (i) if  --'^-•and   ic  are known,
     Returning to Section'8.2 the notation and definition of P^
seems to change  from  line to line.  Hence I cannot say whether  (G-7)
and the paragraph below it. 'are right or wrong,
     The remainder  of Appendix G deals with the problems caused  by
hour*by-ho«r correlations and non-'stationarity.  The latter  means
daily and seasonal  variations.  For daily masisBSBD the authors refer
to a paper by Sorowits  and Sarakatz (which I have not been able  to
see) who use the representation   ,  •
                    log CCtj) m o(ti) -f ab1i...,ib_                   (ii)
I put a finite number of  b*s  because that is presumably intended,
                               supposed to be •uneorrel&ted and
that the t^  are  equally spaced,  Sie condition  {ii} no  doubt  is
                                B-2

-------
supposed to be necessary, and sufficient to guarantee stationarity
 i

It is NEITHER as the trivial example


 !                  Mt) »   *(t-l) +   a(t-2) + a0(t)
 j

should make clear to anyone knowing this theory*


 |    Further, Peagans and 'Biller assert that Horowitsj 'and Barakatz


have derived the distribution of the maximum (presumably over a


long period like a year) from their model*  Uiis is a major achievement


because mathematicians have been trying fruitlessly to solve this


problem for at least 30 years (see, e.g., Watson/ Ann. Math. Stat. 1§54)


     Finally go one can guess these results.  I have done many


simulations an<3 yet am totally unable to*guess the results when I
 i

vary the parameters.
                                B-3

-------