Chapter 2Assessing Risks to Human Health

from Chemicals in the Environment

SUSAN P. FELTER, MICHAEL L. DOURSON ANDJACQUELINE PATTERSON

2.1 INTRODUCTION

Humans are exposed to a multitude of potentiallyhazardous chemicals in their indoor and outdoorair, food, soil and ambient and drinking waters.Risk assessment is the process whereby scientistsevaluate the toxicity data for chemicals to whichhumans are, or may be, exposed, and attempt toidentify and quantify potential risks to health. Therisk assessment process is also used to estimatelevels of intake via the various media that areexpected to be 'safe'. These values are then usedin conjunction with information on exposure inorder to determine acceptable levels for concen-trations of hazardous chemicals in environmentalmedia.

The process of human health risk assessmentwas first described as a four-component paradigmby the National Research Council (NRCI of theNational Academy of Sciences (NASI in 1983and was subsequently updated in 1994. Thischapter follows the NAS paradigm and introduceseach component with an excerpt from the NAS(1994) publication, Science and lodgment in RiskAssessment. The focus is primarily on the risk-assessment methods used by national or healthagencies, such as the International Programme onChemical Safety (IPCSI or the US EnvironmentalProtection Agency IUSEPAI. However, scientistsfrom other groups have made contributions to thefield, especially in the area of research to improvethe standard methods.

pose health hazards, quantification of theconcentrations at which they are present inthe environment, a description of thespecific forms of toxicity (neurotoxicity,carcinogenicity, etc.) that can be caused bythe contaminants of concern, and anevaluation of the conditions under whichthese forms of toxicity might be expressedin exposed humans...

NAS, 1994

2.2.1 Hazard identification ofnon-cancer end-points

Hazard identification is generally the first stepof the risk assessment process, in which it isdetermined if there is a potential cause for con-cern over human exposure to an agent. Thisinvolves an evaluation of the appropriateness,nature, quality and relevance of scientific dataon the specific chemical; the characteristics andrelevance of the experimental routes of exposure;and the nature and significance to human healthof the effects observed. The USEPA, for example,has developed hazard identification guidelinesfor developmental and reproductive toxicity thatcarefully address these issues (USEPA 1991, 1994al.Table 2.1 gives a brief list of considerations.

Much of the process of hazard identification fornon-cancer end-points depends on professionaljudgement as to whether or not an observed effect,or collection of effects (or syndrome I, constitutesan adverse response. This is not always easy, andoften requires the views of experts in the subjectarea, because although many effects are clearlyadverse (e.g. fatty infiltration of the liverl, many

2.2 HAZARD IDENTIFICATION

Hazard Identification entails identificationof the contaminants that are suspected to

9

Chapter 210

Table 1.1 Considerations in characterizing hazard. (Information from USEPA 1995b.1

What are the key toxicological studies and of what quality?Are the data from laboratory or field studies! Single or multiple species?Por cancer: Was there a single or multiple tumour sitejsl! Benign or malignant? Was the maximum tolerated doeeachieved?Por other-thaD-cancer: What end-points were observed and what is the basis for the critical effect? Othersupporting stUdies! Conflicting?

Besides for the critical effect, are there other end-points of concern?

What are the significant data gaps!

What are the available epidemiological or clinical data?. What types of studies were used lie. ecological, case-c:ontrol, cohort)?. Describe the degree to which exposures were described adequately, to which confounding factors were accountedfor adequately, and to which other causal factors were excluded .

Were there non-positive animal or human data?

How much is known about the biological mechanism of action and how does this aid in the interpretation of thedata?

Summarize the hazard identification and discuss the confidence in the conclusions, alternative conclusions thatare also supponed by the data, significant data gaps and highlights of any major assumptions

evaluation of the target organ or 'critical' effect;i.e. the first adverse effect or its known precursorthat occurs as the dose rate increases. This isshown hypothetically in Fig. 2.1, where severaleffects are evoked from chemical exposure:

others are of uncertain toxicological consequencele.g. decrease in body weight gain).

Because toxic chemicals often elicit more thanone adverse effect, the process of hazard identi-fication for non-cancer toxicity includes an

Fig. 2.1 The judgement of thecritical effect and its NOAEL (no-observed-adverse-effect levell.along with the appropriateuncertainty factor (UF) andmodifying factor (MFI. leads to theestimation of the RfD (referencedose'. NOEL. no-observed-effectlevel; LOAEL, lowest-observed-adverse-effect level; FEL. frankeffect levrJ.

11Risks to Human Health from Chemicals

enzyme change, slight decrease in body weight,fatty infiltration of the liver and convulsions. Enzymechange and slight decrease in body weight arejudged not to be adverse effects. Fatty infiltrationof the liver is judged to be the critical effect.

The judgement of whether an effect is adverseor critical may change among toxicity studies ofdifferent durations, and may be influenced bytoxicity in other organs or by toxicokinetics. Agood example of this is increased liver weight dueto a proliferation of smooth endoplasmic reticulumwith chemical exposure. Such an effect may bejudged as not adverse if the parent chemical isthe toxic moiety and such an increase is likelyto quicken its metabolism, or may be judged tobe adverse if a metabolite is the toxic moiety(Farland &. Dourson 19921. The distinction ofadverse compared with non-adverse effects and thechoice of critical effects in the hazard identificationcomponent of the paradigm is then used as a basisfor the dose-response assessment.

Group 3: cannot be classified as to its carci-nogenicity to humans.

In 1986, the USEPA published general guidelinesto be used by Agency scientists in developingand evaluating risk assessments for carcinogens(USEPA 1986). Based on the weight-of-evidencefrom epidemiological and laboratory animal bio-assays, chemicals are placed in one of six categories.Supporting data le.g. mutagenicity data, mechan-istic data) may then be used to move a chemicalup or down in the ranking. These categories aremodelled after those used by IARC.Group A: carcinogenic to humans.Group Bt: probably carcinogenic to humans.Group C: possibly carcinogenic to humans.Group D: not classifiable as to human carci-

nogenicity.Group E: evidence of non-carcinogenicity for

humans.In April 1996, the USEPA proposed revisions tothe carcinogen risk assessment guidelines. Incontrast to the concise alpha-numeric classifi-cation system of 1986, the guidelines proposedadvocate the development of a more comprehen-sive characterization of the carcinogenic hazardin the form of a narrative. Within this context,a cancer hazard characterization should includeall information relevant to the weight-of-evidencefor carcinogenicity, not just tumour data inhumans and animals. This means that mechanisticdata can play an integral role in the hazardidentification step for carcinogenicity, and mayalso influence the choice of a dose-response model.Another change is that the hazard characteriz-ation can provide specific information aboutthe conditions under which a chemical is likelyto be carcinogenic; for example, it may be likelyto be carcinogenic by the route of inhalationbut not by ingestion. These proposed changesin the USEPA methods reflect a general move-ment in the field of cancer risk assessment toinclude more chemical-specific data and to moveaway from the use of default positions wherever

possible.

2.2.2 Hazard identification of carcinogens

Hazard identification of carcinogens refers to theprocess of determining if a compound has the po-tential to elicit a carcinogenic response in humans.Many types of information may be used to deter-mine the overall weight-of-evidence of carcino-genicity: epidemiological information, chronicanimal bioassays, mechanistic data, mutagenicitytests, other short-term tests, structure-activityrelationships, metabolic and pharmacokineticproperties, toxicological effects and physical andchemical properties.

The first organization to develop a classificationscheme for carcinogenicity was the InternationalAgency for Research on Cancer (IARC) in 1978.Based on a strength-of-the-evidence approach (evid-ence coming from human or laboratory animaldata or short-term studies I, chemicals were placedin one of three categories.Group 1: carcinogenic to humans.Group 2*: probably carcinogenic to humans.

. Group 2 includes subgroups 2A (for chemicals havinglimited evidence of carcinogenicity in humans) and 2B(for chemicals having sufficient evidence of carcino-genicity in laboratory animals, and inadequate evidencein humans).

t Group 8 includes subgroups 81 (for chemicals havinglimited evidence of carcinogenicity in humans! andB2 (for chemicals having insufficient human data butsufficient animal data).

.12 Chapter 2

evaluation of the conditions under whichthe toxic properties of a chemical might bemanifested in exposed people, withparticular emphasis on the quantitativerelation between the dose and the toxicresponse. The development of thisrelationship may involve the use ofmathematical models. This step may includean assessment of variations in response, forexample, differences in susceptibilitybetween young and old people.

NAS, 1994

2.3.1 Non-cancer end-points

In addition to what has been described herefor the USEPA, other groups have publishedcarcinogen classification schemes along similarlines. Moolenaar 11994) has provided a summaryand comparison of several international classifi-cation schemes, including eight governmentalagencies and two independent organizations. Theseclassification schemes have anywhere from twoto six distinct categories with varying degrees ofemphasis on mechanistic data. In addition, Ashbyet ai. (19901 have recommended an eight-categorysystem.

Common to many of these groups, the deter-mination of carcinogenic hazard includes a deter-mination of whether the incidence of tumourtypes observed to occw in laboratory animalsis statistically significantly elevated over thatobserved in controls. Two forms of statistical testsare used to answer this question: trend tests, whichlook for an overall trend of increasing tumowincidence with increasing dose; and pairwisecomparison tests, which directly compare thetumour incidence in an individual dose group withthat seen in controls.

Determination of the mechanism by which achemical causes cancer in laboratory animalsalso provides information about the potential forhuman carcinogenicity relevant to the hazardidentification process. The potential for the sameor a related mechanism to be operative in humansprovides the basis for extrapolation from otheranimal species to estimate the risk of cancer tohumans. For some specific tumow types, or mech-anisms of carcinogenicity, there are indicationsthat tumows observed in laboratory animals mayhave no relevance or limited relevance to humancarcinogenicity. Tumour types that are includedin this group include kidney tumours in malerats that are caused by the accumulation of amale-rat-specific protein (alph~u-globulin); livertumours in male B6C3Fl mice; thyroid follicularcell tumours; and bladder tumours related to theformation of silicate-containing precipitate andcrystals (e.g. as seen in saccharin-induced bladdercancer in ratsl.

The 'Safe' dose approach

Dose-response assessment follows hazard iden-tification in the risk assessment process. Dose-response assessment involves the quantitativeevaluation of toxicity data to determine the likelyincidence of the associated effects in humans. Theinformation available for dose-response assessmentranges from well-conducted and well-controlledstudies on human exposures, and epidemiologystudies with large numbers of subjects, well-characterized exposures, and supportive studies inseveral animal species, to a lack of human andanimal toxicity data with only structure-activityrelationships to guide the evaluation. In any case,scientists should consider all pertinent studiesin this process; even a single human case studycan provide useful information. However, onlydata of sufficient quality, as judged by experts,should be used in the dose-response assessmentof a chemical. Table 2.2 lists some questions to beconsidered in characterizing the dose-responserelationship for an agent.

Most non-cancer effects resulting from ex-posure to toxic agents are thought to be associatedwith a threshold; i.e. an exposure exists belowwhich toxicity does not occur. The dose-responsecomponent of risk assessment involves the quan-titative evaluation of toxicity data to determine alevel of exposure for humans that is considered byrisk assessors to be below the threshold for toxicityfor sensitive subgroups.

Health agencies throughout the world supportthe use of a 'safe' dose concept, and define terms

2.3 DOSE-RESPONSE ASSESSMENT

Dose-Response Assessment entails a further

Risks to Human Health from Chemicals 13

Table 2.2 Considerations in characterizing the dose-response assessment. (Information from USEPA 1995b.)

DataWhich data were used to develop the dose-response curve? Would the results have been different if based on adifferent data set?

If animal data were used, which species were used-most sensitive, average of all species, or other? Were anystudies excluded and why?

If epidemiological data were used, were they only the positive, all studies, or a combination? Were any studiesexcluded and why?

Was a meta-analysis performed to combine the studies?If so, what approach was used?

ModelsWhat model was used to develop the dose-response curve? The rationale for this choice? Is chemical-specificinformation available to support this approach?

For non-cancer end-points how was 'safe' dose calculated? What assumptions and/or uncertainty factors wereused? For benchmark doses, what model was used and why?

For cancer end-points, what dose-response model was used and why was it selected? Would other models haveprovided as plausible results?

Discuss the route and level of exposure observed in the data compared with anticipated human exposure. U dataare from a different route, are pharmacokinetic data available to extrapolate across routes? How far is theextrapolation from the observed data to environmental exposures and what is the impact of this extrapolation?

Thxicity valuesSummarize the risk value and discuss the confidence in the value. Can a range of values be provided? What are theresults of different approaches or models?

and conditions for use. This 'safe'or subthresholddose often goes by different names, such as: HealthCanada's Tolerable Daily Intake or Concentration(mI or TDC) IMeek et al. 19941; IPCS's TolerableIntake (ll) (IPCS 19941; US Agency for ToxicSubstances and Disease Registry's IATSDR's)Minimum Risk Level (MRL) (Pohl &.Abadin 19951;USEPA's Reference Dose (RfD) (Barnes &. Dourson1988; Dourson 1994) or Reference ConcentrationIRfC) (Jarabek 1994; USEPA 1994bl; or the WorldHealth Organization's Acceptable Daily Intake(ADI, (Lu 1985, 19881. Many of the underlyingassumptions, judgements of critical effect, andchoices of uncertainty factors (or safety factors)are similar among health agencies in estimatingthese subthreshold doses.

One of the best-known methods is that used bythe USEPA to derive reference doses IRfDs) andreference concentrations (RfCs', which are sub-threshold exposures for non-cancer toxicity. Theyare defined as: '... an estimate (with uncenainty

spanning perhaps an order of magnitude) of a dailyexposure to the human population (includingsensitive subgroups) that is likely to be withoutan appreciable risk of deleterious effects during alifetime' (Barnes & DouISon 1988).

The subthreshold dose approach starts with anidentification of the critical effectjsl, as describedin Hazard Identification j2.2.2). The critical doseis then chosen. All groups rely on the experimentaldose that represents the highest level tested atwhich the critical effects were not demonstratedas this critical dose. This dose is often called theno-observed-adverse-effect leveljNOAEL), or theno-observed-effect level (NOEL). If a NOAEL is notavailable, the use of a lowest-observed-adverse-effect level (LOAEL) is often used as the criticaldose.

Human data are preferred in the determinationof an RfD or RfC. However, in the absence of thesedata, animal data are closely scrutinized. Riskassessment scientists seek to identify the animal

14 Chapter 2

model that is most relevant to humans, based onthe most defensible biological rationale. In theabsence of a clearly most relevant species, thecritical study and species that shows an adverseeffect at the lowest administered dose are generallyselected. This is based on the assumption that, inthe absence of data to the contrary, humans maybe as sensitive as the most sensitive experimentalanimal species. Uncertainty factors IUFs) arethen used as divisors to this critical dose INOAELor LOAEL) to determine the subthreshold dose.These factors are considered as reductions inthe dose rate to account for several areas ofscientific uncertainty inherent in most toxicitydatabases. As shown in Table 2.3, these areasinclude interhuman variability (designated as HI;extrapolation from experimental animals tohumans (designated as AI; extrapolation fromsubchronic to chronic exposure jdesignated as 5);extrapolation from an experimental LOAEL toNOAEL (designated as L); and how to account forthe lack of a complete database. In addition tothese UFs, several groups also use a modifyingfactor that can be used to account for uncertaintiesnot explicitly dealt with by the standard factors.

All groups occasionally use a factor less than 10or even a factor of I, if the existing data reduceor obviate the need to account for a particular areaof uncertainty.. For example, the use of a I-yearrat study as the basis of an R£D may reduce theneed for a tenfold factor for the area of subchronic-to-chronic extrapolation to threefold, becauseit can be demonstrated empirically that I-yearNOAELs for rat are generally closer in magnitudeto chronic values than are 3-month NOAELs.Lewis et al. (19901 investigate this concept ofvariable uncertainty factors more fully through ananalysis of expected values.

The choice of appropriate uncertainty andmodifying factors reflects a case-by-case judgementby experts and should account for each of the

applicable areas of uncertainty (described in Table2.31 and any nuances in the available data thatmight change the magnitude of any factor. Severalreports describe the underlying basis of uncertaintyfactors (Zielhuis & van der Kreek 1979; Dourson& Stara 1983) and research into this area (Calabrese1985; Hattis et aI. 1987; Hartley & Ohanian 1988;Lewis et aI. 1990; Renwick 1991, 1993; Calabreseet aI. 1992; Dourson et aI. 1992; Calabrese &Gilbert 1993; Kroes et aI. 1993; Abdel-Rahman &Dourson 1995).

The scientific strengths and limitations of thisapproach have been discussed in the literature(Munro & Krewski 1981; Lu 1983, 1985, 1988;Krewski et al. 1984; Crump 1984, 1986; Doursonet al. 1985, 1986; Barnes & Dourson 1988; Kimmel& Gaylor 1988). The scientific strengths, in brief,are that all toxicity data are reviewed in thechoice of the NOAEL for the critical effects,and that uncertainties in the entire data base,can be factored into the t:esulting value of thesubthreshold dose through the use of professionaljudgement as to the appropriate uncertainty andmodifying factors.

The limitations, in brief, are that the NOAEL isrestricted by the choice of dose-spacing and thenumber of animals, as well as factors that influ-ence the quality of the study. Studies with widedose-spacing and a low number of animals perdose group can lead to a more poorly characterizedsubthreshold dose as compared to studies withtighter dose-spacing and more animals per dosegroup (see, e.g., Hattis et aI. 1987; Leisenring &Ryan 1992). The NOAEL is also not generallyinfluenced by the nature of the dose-responsecurve. Uncertainty factors, although considerednecessary and perhaps to reflect accurately thepotential underlying areas of uncertainty, arequite imprecise. Nor does the subthreshold-doseapproach enable an estimate of risks at exposuresgreater than the subthreshold dose.

Scientists are developing methods that addresssome of these latter limitations le.g. DeRosaet al. 1985; Dourson et aI. 1985; Kimmel &Gaylor 1988; Kimmel et aI. 1988; Hertzberg1989; Hertzberg & Dourson 1993; Renwick &Walker 1993; Faustman et aI. 1994; Allen et aI.1994a,b). TWo of these methods are describedbriefly here.

. The usual intermediate factor used is 3 because it is

the approximate logarithmic mean of 1 and 10. Thechoice of 3, instead of 5 for example, reflects both theexpected precision of the UFs labout 1 digit, log base 10)and the view that it is not generally possible to be moreprecise in considering the nuances of these areas ofuncertainty than about half-way.

~Iti.1

-..~~.BJ-0:8~:;:i'05ego

~~~..,9~U~l-00e-0;t;-,9~u<

.I3--;;<

.I

'~..0§,tjQ,

'5~u0'"Nu=a~

0)

i!:;

c="2 ~

:t:.u

0 U

)

"'..~~

U;:J

::>~

-

~ ~~~Q~~.nU~\h)

-s~~

-;~5

XU

~

..~.9"'ii-0S0

0-0...

g 0...8I

;0--...;X\0-a-0~

; .i

~ ~

-Ed

~

.~..

~.E

.£

Risks to H

uman H

ealth from C

hemicals

c0;ojXInd0 0- 0-'35 0

e ..

.~]

~~

;.d

~ Q-va8-I

- 8...I Q-u'a u

] '~

2.-5:j

01/)...

0 s~~

0..~~

~~

g~

~ ~ 0..~u] u

Q,U

)e

..

8~.E

,g

-<z ~ 00"-vV

l0

-0- u

~

.. ..

.d° ...,

~"""

"'... 0

U°d

c"'~"

..~

"UU

;..

~6'"'O

~

...~~

~

~

~

5~~

.8

:au-O

cs§ ..

~

~

8"" ..

= ~

~

~.d:3

8.~

~

.2i. '"

~'" .Q

.t:

;B ~

-0

.. 0

. -0'"

".,..~

=

u".c U

., ~

~

~

'"'O

~

c.~

~..

~

: ~

..

.~

u ~

.. 0

0<

. M

~

.

U~

u88

- ~ ...

.,.5 'S

"" 0

ii! ~

U

~

0 $

13~

~

'"'O:a ~

~ ~

0:J

Z

<

.~

.5 'is ~

5 >

-~

:a

.~

~ ;

~

s ~

~

§ u 0

8' :J"

~

:E- F

-0

~

~

~

:3 u

u 0

'a z "U

~

.. ~

~

u~

,,;~

~'"'O

:a~§ a~

~

..I,. ~

~~

'"'O

;g,"=

~~

...,J-O

..u.c.Q

o~

~

..H-0

.. =

.. -0"

,.. ..

u ..

d -0

-0 =

U

:.:j e .. =

~

oj ~ §

~

2 a ~

.~

~

..

U

..

'~1:~

§ :~~

~~

]~

.s ~

'(;-~ :~

-t ~

~j

.Jf. i f j ~ f.s 1 ] ~

] ~ ~

"I ~ j j I

~ ~

-~

5 8 u

~

.~ ~

0 ~

,,; 8 ~

~

="

~ ~

-o -;; ..-0

.~.d

...~

.~ ~

~ ~

.. oS

~ 0"

'S

p ~

.~

8 .. =

.a Y

o

~u'(;-~

..

~...~

~

~s~

o '-'3~

..82

.9-ou8. ~

:~ ~

1° ~

~ ~

8.:3 ~

~ 1 d

... "8.:.5:ti

~ a ;;;

.. U

P

-g .-

U 8.'P

...d.S

<

.. U

U

.. E

~

-;;

ti ; .~.~

ti 1 ~

D]

@ ~

i ~ ij.~

':E, ij'~

: ~

] .~

5u=2

5e;.. 0

5e

~u

5eo~

5-O"bO

=u

.d ~

-0 .d.~

~ .5.u

.d. H

'8 .d;J

~

.d u

.~

8. ~ 0

~~

§~

~uo>

-~

~~

u~

~~

o ~

~>

- =

'"'O~

u eo'P

u

u eo =

e"

~

.S.Q

~

.. ~

u

=

~

~

~.c

~

;J.:5'~

~;J

~

'.. '"'0

3] '(;- ~

=

~

.. ~

~

- 3.~

~

>--o~

e >

-d ~

-o >

-"5" >

-'""8 >

-~o.,

>-1;=

~

i ~

;w

~

~ ~

g ~

=S

.8 0~:J

=s ~

] =

s ~

.~ 8

~'5

~~

0.t:8 ~

:a~=

~

u5t~

uPu

U§~

8 ~

~..

~ 1]'~

~

~ ~

..g ~ ~ ~

§ 9 ~ ~

.g ~.~

1 g ~ § ~

0-I Q.....I~

bO

:~ ~

;:gU~

~

... ~

... ~

,

U'"

&I

&I

~

~~

~

~

]'"

;.. -ou.d

~

-; ~

5~i

.~

~ ~

~i

0 ~

~.g

:.. =

...

w~

&

I ..."'~

~

~

~~

-o'9

~

.g~~

... ~

~

0

"'... -;

ro!~

~

""...

~~

...

0 a

W0

;C

&lU

~-

... "'s-~

' .6

.E

=

- 8.w

~

-0 ~

o'<

~

.6.~

t

9 ~

-;a~

. ...~

§'"

.. .A

...

, &

I 0

v, 0

.- :S

.;3 2. =

8:9 ~

z<

~=

->-

&I

~~

.g~

e ~

.3~'!i

~!

s j~

~~

=

'"

5 '

~

.~

~

Z .bIt

bO]

~

~

.bIt &

I ~ -g :s

oWw >-

.d ...c

a ~

&I

,0~

s ..

-~~

=~

w

~

~""".~

~~

~

i5~~

-0;~

~

_-o~

&I ~

&I

0 -

'"

:0 ...

t- "08.

.-8:i ~

~

~at1

0 - w

Uo

u w

(/)

~

u=

- ;..

~

l1 ...

's,S

..

u"'=o

~:s

... .~

.c ~

..

S

w

~:s-o~

~ . 9

.~

&I

(/) -0

uP

o A

U

-0~

. e

U

.R ,!!

CuS

~

o~

"~

u

~

~ F

&

.u-::~

&

I u~

~8

~~

~

;~

~~

8: 'I:! u

~

u=

-~

,=

5-.c(/)1

"iJ.~

~~

~

~~

tij o<

~=

~(/)~

"u

u-::ou-o

w...~

~

a:u:~o

~~

"iJu~

w

~

...

vi0

;".c v,""

'-0 .""

..=

->.u

&1.=

..(/)-

="-t!..U

CL.oW

OoW

.~

iJ.S

5 ij '< .2

.;3

15i~~

g~S

i_u-o~

...~

(/)..:::g~,d

u' &

lU

d ~

~

.- :s ~

= A

a. .0 .~ H

0\~

.. U

...

~

;3 .~

..

0\ ~

ro!>-.;ou~

..-

.c~.c~

U~

o-

~o">

-&lo~

""g.

8i]~~

]'E~

.- u

d -; u

u~

S

~--;

~

0 ...

;.. J!0

.. ~

=

co ';3 d ~

>-

O.d,<

o&l

...u

~

0 .~

.1=

.>

8 ~

.~

u"Z

",-o&ldO

;..&

1 &

I£..o(/)~

gz°"iJ-e3~.~

!:::8(/)

~~

&I

c""C. +

~*

:;\Q)~

,,:j 15

16 Chapter 2

Categorical regression

Another method that has been proposed forquantitative dose-response analysis for non-cancereffects is that of categorical regression. Thisinvolves statistical regression on severity cate-gories of overall toxicity (Hertzberg & Miller 1985jHertzberg 1991j Hertzberg & Wymer 1991). Byassigning severity categories, all adverse effectsmay be taken into account rather than just thecritical effect. Categorical regression also allowsuse of group data (ie. at the dose-group level, notindividual animals) as well as toxicity data frommultiple studies. The results of the regression canthen be used to develop a subthreshold dose muchas the BMD is used.

Categorical regression can also provide informa-tion about relative risks from exposures exceedingthe RfD. The NOAEL and BMD approaches arelimited in that they are focused on the deter-mination of 'safe' and 'acceptable' intake levels(i.e. they are point estimates designed to be belowthe population threshold for toxicity). In situationswhere exposures may exceed these levels, however,information is needed to help determine theurgency of a situation. Herein lies one of theadvantages of categorical regression because itprovides information about increasing toxicitywith increasing dose rate. If human data areavailable, categorical regression can be used toactually estimate potential risk above the RED orRfC. With only animal data, categorical regressioncan help prioritize risks based on how quickly thetoxicity severity changes with dose.

Benchmark dose

Another form of quantitative risk assessmentof non-cancer end-points is the benchmark doseIBMD) method. The USEPA 11995a) has definedthe BMD as: 'a statistical lower confidence limitfor a dose that produces a predetermined changein response rate of an adverse effect... comparedto background.'

This method, which was first described byCrump (1984) and Dourson et ai. (19851, wasdeveloped in an attempt to remedy some not-able shortcomings of the use of a NOAEL in thesubthreshold-dose approach described above. Forexample, the NOAEL is limited by the experi-mental doses chosen by the investigators in thetoxicity studies. The larger the dose spacing, theless accurate the experimental NOAEL lor LOAELIis apt to be. Also, the slope of the dose-responsecurve provides valuable information that is notused explicitly in this approach (although itmay influence the choice of uncertainty factorsl.The BMD method attempts to use more of theavailable dose-response information by fittinga mathematical model to the data and thendetermining the dose associated with a specifiedincidence of adverse effect. In this way, the BMDis not limited to the experimental doses chosenby the investigators.

Although the BMD method offers some ad-vantages over the NOAEL, it can be used only incases where data are available that are suitablefor modelling. It is not, therefore, a replacementfor the NOAEL, but should be considered as anadditional method that may offer advantages forsome risk assessments.

There are a number of decisions to be made inapplying the BMD method, for example: whichmathematical model to use; what degree of con-fidence limit to use; what incidence rate to pre-determine as the benchmark response (e.g. a 1 %,5% or 10% incidence of an effect). For moreinformation, the reader is referred to a guidancedocument on the use of the benchmark-doseapproach in risk assessment that was issuedby USEPA's Risk Assessment Forum (USEPA1995al. .

2.3.2 Cancer end-points

The elicitation of a carcinogenic response tradi-tionally has been presumed to occur withouta threshold. Because of this, it has often beenassumed by several regulatory agencies (e.g.USEPA 1986; Rademaker &. Linders 1994) thatany dose of a carcinogen is associated with someincreased risk. As a result, dose-response assess-ment for carcinogens often focused on deterD1ininga de minimis risk level, frequently expressedas the risk of one-in-a-million, by using a linearmodel to extrapolate risks down to low-doselevels. Other groups, such as Health Canada IMeek

Risks to Humtm Health from Chemicals 17

et al. 19941, do not advocate extrapolation beyondthe range of observable data, but rather use amargin-of-exposure approach (described more fullybelow). In EPA's 1996 proposed revisions to thecarcinogen risk assessment guidelines, an optionof using a margin-of-exposure (MaE) approach isalso described.

Exposure to a carcinogenic agent often causesmore than one tumour type. Similar to the processused for the evaluation of non-cancer toxicity, therisk assessor must evaluate the data to determinewhich end-pointls) occurls) at the lowest dose.Unless there are data to support otherwise, itis generally assumed that humans may be assensitive as the most sensitive animal model.After identifying the study(ies) that is(are! mostappropriate for developing a quantitative riskestimate, the next step is to transform the dosesto which the animals were exposed into humanequivalent doses. For example, in the absenceof a chemical-specific model, the USEPA 11996)recommends the use of a cross-species scalingfactor of lbody weight)3/4 for oral exposures, anddefault methodology (USEPA, 1994al for estimatingrespiratory deposition and absorption of particlesand gases for inhalation exposures. Finally, thedose-response data are modelled to determine thecarcinogenic potency of the chemical at low doses.

Estimating risk with mathematical models

Dose-response assessment for carcinogens isconcerned with estimating the central estimateand/or the upper confidence bound for carcinogenicrisk associated with environmental exposures.Alternatively, risk managers may be interested insetting standards for exposures by various media(e.g. air, drinking water) based on a carcinogenicrisk level that is considered to be de minimis le.g.one-in-a-million excess risk). The cancer bioassaysgenerally used in the dose-response assessment,however, are performed in laboratory animalsat very high doses relative to levels at whichhumans may actually be exposed. These high dosesare necessary in order to produce a statisticallymeasurable effect given the relatively small num-ber of animals used. The nature of the curve atlevels of exposure below the lowest experimentaldose is not known. Many models have been

developed to estimate cancer risk in this low-dose

region.A common model used to perform this extra-

polation is adapted from the multistage model.This model assumes that cancer is the result ofa sequence of changes in a cell or organ andthat exposure to a carcinogen can increase thetransition rate between these stages, resultingin malignancy (Armitage & Doll 1954, 1961;Crump et aI. 19761. The 95% upper confidencelimit of the linear component of this model(often referred to as the q;) has been used bythe USEPA (1986) as an upper bound estimate ofcancer potency because it is numerically morestable than a central estimate and also is in keepingwith the low-dose linear approach adopted forcancer-risk assessments. Although there are nodata to demonstrate that the linearized multistagemodel is more appropriate than any other model,it has been used as the default because it providesa plausible and stable upper bound estimate thatis not likely to underestimate the cancer risk. TheUSEPA and others recognize that at very low dosesthe response could be as low as zero.

In the USEPA's 1996 revised guidelines, it isproposed that the dose-response assessment beeonsidered as a two-step process. The first step isto fit a model to data in the observed range only.If sufficient data are available, a biologicallybased model is the preferred approach. Also, theremay be cases in which data other than tumourincidence le.g. information on ONA adductslcan be used to extend dose-response below theobservable range. The outcome of the first stepis the estimation of an EOlo or LEOlo' The EOlo(effective dose at the 10% level) is the doseassociated with a 10% increase over backgroundin the end-point being measured le.g. tumourincidence or other). The LEOlo is the lower 95%confidence limit on this dose.

The second step under the USEP A's 1996 revisedguidelines is to use an extrapolation procedure toestimate risk in the low-dose region ji.e the rangeof human exposure) if it is appropriate to do so. Forcases where data do not support the developmentof a biologically based model, and for which thedose-response relationship is thought to be linear,the proposed guidelines suggest terminating themodel in the range of experimental data, and

Chapter 218

and The Netherlands use a subthreshold doseapproach: they set ADIs using the method describedabove for non-cancer toxicity. For genotoxiccarcinogens thought to have no threshold, TheNetherlands extrapolates linearly from the lowestexperimental dose having an increased incidenceof tumours. Clearly, there are many variationsin the ways that dose-response assessment forcarcinogens can be, and are, performed. A commontheme among all of these groups is that themechanism by which the agent is believed to causecancer is playing a greater role in the way in whichthe dose-response assessment is approached.

drawing a straight line to the origin. Whereas thelinearized multistage model may still be used inmodelling the data in the experimental range,justification for using this (or any other! modelneeds to be provided.

For chemicals that have a non-linear dose-response relationship, the proposed guidelinesadvocate the use of an MOE analysis. The MOEis the LEDtO (or another predetermined startingpoint within the range of observation) divided bythe exposure of concern. The risk manager thendecides whether the margin of exposure is largeenough to satisfy management policy criteria. Theproposed guidelines suggest that a factor of 100 beused as a science policy default position to reflectallowances for intra- and interspecies variability.Chemical-specific data can then be used to adjustthis factor upward or downward as appropriate.

This type of MOE analysis is similar to that usedcurrently by Health Canada jMeek et aI. 19941.Main differences between the MOE approachesof the agencies are twofold: Health Canadadetermines a mas (the dose associated with a 5%increase over background of tumour incidencel,whereas the USEPA is proposing determination ofan EDto' Secondly, the USEPA uses the 95% lowerconfidence bound on the EDtO whereas HealthCanada uses the central estimate.

Approaches with uncertainty factors

A number of organizations have developed othermethods for quantitative dose-response assess-ment for carcinogens. Moolenaar 11994) has de-scribed the similarities and differences betweenapproaches used by the USEPA, the UK, Denmark,the European Union IEUI, The Netherlands, andNorway. He points out that the USEPA is theonly organization to have described carcinogenicrisk in terms of an 'upper bound' li.e. the 95thpercentile of the slope of the dose-response curvein the low-dose region).

For example, each of the aforementioned groupshas developed separate methods for dealing withgenotoxic versus non-genotoxic carcinogens.Norway does not perform low-dose extrapolationfor any carcinogens, but rather uses the roso todetermine a potency classification for a carcinogen.For non-genotoxic carcinogens, the UK, the EU

2.4 EXPOSURE ASSESSMENT

Exposure assessment involves specifyingthe population that might be exposed to theagent of concern, identifying the routesthrough which exposures can occur, andestimating the magnitude, duration, andtiming of the doses that people mightreceive as a result of their exposure.

NAS, 1994Environmentally relevant routes of exposure forhumans are inhalation, oral, and dermal. Anexposure assessment may include a componentfor each, such as an assessor would conduct wheninvestigating the potential impact of a point sourceof pollution. In such a multimedia investigation,an exposure assessment is initiated by estimatingthe amount and rate at which a toxic agent isreleased from a given source. Fate and transportmodels are then used to estimate the movementof the agent through environmental media towhich humans may be exposed. A number ofmodels is available for use in estimating trans-port and fate; many of these are described inthe USEPA's exposure assessment guidelinesIUSEPA 1992). Table 2.4 provides guidance incharacterizing the exposure assessment step.

An exposure assessment may also be focused onone particular medium and one route of exposure,for example, the oral intake of a chemical fromdrinking water. This type of exposure assessmentmay be used, for example, to detennine whetherthere is sufficient exposure of humans to a chemi-cal in a given medium to warrant regulation.

Exposure can be determined directly, through

Risks to Human Health from Chemicals 19

Table 2.4 Considerations in characterizing exposure. (Information from USEPA 1995b.I

What are the most significant sources and pathways of environmental exposure, presently and in the future (if

appropriate!?Are there data on other sources of exposure and what is the relative contribution of different sources of exposure?

Describe the populations that were assessed, including highly exposed groups and highly susceptible groups

Describe the basis for the exposure assessment, including any monitoring, modelling, or other analyses of exposuredistributions such as Monte Carlo.

Describe the range of exposures to 'average' and 'high-end' individuals, the general population, high exposuregroup(sJ, children, susceptible populations

How was the central tendency estimate developed? What factors and/or methods were used in developing thisestimate?

Are there highly exposed subgroups and how are they accounted for?

Is there reason to be concerned about cumulative or multiple exposures?

What are the results of different approaches, i.e. modelling, monitoring, probability distributions?

What are the limitations of each and the range of most reasonable values?

What is the confidence in the results obtained and the limitations to the results?

also include a full discussion of theuncertainties associated with the estimatesof risk.

personal monitoring devices, or indirectly, throughenvironmental monitoring. If environmentalmonitoring is used, then the assessor mustestimate the extent to which individuals may beexposed to the media for which monitoring dataare available. Risk assessment scientists often usedefault values for these assessments le.g. assumingan inhalation rate of 20 m3 per day or consumptionof 21 of water daily!.

A need exists to estin1ate the distribution ofexposures that may result to individuals andpopulations. For example, the USEPA !USEPA1992) recommends assessing exposure to the totalpopulation, and also for assessing the upper end ofthe exposure distribution; i.e. a 'high-end expo-sure estimate' and a 'theoretical upper boundingestimate'.

NAS, 1994Risk characterization is the final step of the riskassessment process, in which information from thehazard identification, dose-response and exposuresteps are considered together to determine andcommunicate the actual likelihood of risk toexposed populations. The risk characterizationdiscussion includes an evaluation of the overallquality of the data, the specific assumptions anduncertainties associated with each step, and thelevel of confidence in the resulting estimates.

Specific key qualities, or attributes, of riskcharacteriZations have been identified (AIHC1992; USEPA 1995bl. These attributes includetransparency in decision making, clarity in com-munication, consistency and reasonableness.Exercising transparency and clarity result inscientific conclusions being identified separatelyfrom policy judgements. In addition, default values,assumptions and uncertainties are disclosed so thatthe end-user can better identify what is based ondata and what is assumed. Greater consistency inthe terminology used, along with definitions andassumptions, will provide for better comparabilityacross assessments. In addition, use of standard

2.5 RISK CHARACTERIZATION

Risk characterization involves integrationof information from the first three steps todevelop a qualitative or quantitativeestimate of the likelihood that any of thehazards associated with the agent of concernwill be realized in exposed people. This isthe step in which risk-assessment resultsare expressed. Risk characterization should

Chapter 220

descriptors, such as those outlined in the USEPA'sExposure Assessment Guidelines IUSEPA 1992},reduce confusion and lead to greater under-standing. Lastly, the risk characterization shouldbe reasonable and balanced in its presentation. Theinformation and conclusions should be presentedin such a fashion that they are clearly understoodby the intended audience. The ultimate goal ofrisk characterization is to provide the decisionmakers with enough information, presented ina comprehensible fashion, that they understandwhat is known and unknown about the risk tohuman health from the situation being evaluated,thereby leading to the best possible risk-baseddecisions.

In order for risk assessors to meet this goal, theymust understand the need for the risk assessmentand its intended end-use. Risk assessors shouldmeet with the decision makers and engage themin the process throughout. Involving the riskmanagers and decision makers will help the riskassessor to meet a level of detail and analysisappropriate for the situation le.g. initial screeningversus national regulation}. By communicatingwith the end-user, risk assessors can ensure thatthe risk manager will comprehend the results ofthe analysis.

Involvement of the end-user supports takingan iterative approach to the risk assessment.For example, if the risk assessment is for acontaminated site, it would be very useful to havethe exposure assessment scientists involved indeveloping a monitoring plan and in reviewinginitial results so that the monitoring could berefined to collect the most useful data. Only if afirst, conservative screening indicates that somelevel for concern is warranted would a more in-depth analysis be pursued. The iterative approachto risk assessment assures better use of limitedresources to address problems.

discussions are important because they form thebasis for the overall judgement as to the adequacyof the data and conclusions drawn from it. Inaddition, hi,ghlighting of uncertainties can identifyareas where the collection of additional data mayreduce the uncertainty and strengthen the riskassessment. An uncertainty discussion includesthe quality and quantity of data available (toxicityand exposure), identification of data gaps, use ofdefault assumptions and parameter values, andthe uncertainties in the models used.

Crucial to a discussion of uncertainty is main-taining a clear distinction between uncertaintyand variability within each step of the process.The USEPA distinguishes between these twoconcepts in its risk characterization guidancejUSEPA I 995b). Variability describes inter-individual, spatial or temporal differences withinan animal or human population or within moni-toring data. It reflects the heterogeneity of thedata. Uncertainty, on the other hand, applies toareas for which data are unknown. There areuncertainties associated with both dose-responseor fate and transport models; an uncertaintyanalysis would evaluate the basis for the modeland validation of the model.

Given the extensive use of modelling to estimateexposure in panicular, the uncertainty related tothe chosen parameters can have a great impacton the resulting risk estimates. Risk assessorsmust be careful to identify the parameter valuesand their sources so that others can evaluate theirappropriateness and impact on the final results.Risk assessors can use probability density func-tions and/or likelihood distributions to charac-terize uncertainty quantitatively. Monte Carloanalysis is one statistical procedure used.

To summarize a risk characterization, the riskassessor should consider questions such as thoselisted in Table 2.5. These questions can be usedto help outline a discussion of risk conclusionsand comparisons. These questions build uponthose in Tables 2.2-2.4 relating to hazard identi-fication, dose-response assessment, and exposureassessment, respectively.

2.5.1 Uncertainty and variability

The field of risk assessment is increasinglyutilizing uncertainty and sensitivity analyses tobetter assess risks to human health. Critical to acomplete risk characterization is a full discussionof the uncertainty within each analysis and thatrelated to the overall assessment. Uncertainty

2.6 SUMMARY

This chapter has outlined a process to assess the

Risks to Human Health from Chemicals 21

Table 2.5 Questions to assist in developing a risk characterization summary. (Information from USEPA 1995b.)

Risk conclusionsWhat is the overall picture of risk and the specific risk estimates and/o~ rangeslFor the hazard identification, dose-response and exposure assessment steps:What are the major conclusions and strengthslWhat are the major limitations and uncertaintieslWhat are the science policy options, and what other alternatives were consideredl

Risk contextWhat are the qualitative characteristics of the hazard (e.g. voluntary versus involuntary, one population segmentversus another)? Comment on any risk perception studies related to this type of hazard

What are the alternatives to this hazard and how do the risks compare?

How does this risk compare with other risks?

Are there significant community concerns which influence public perception of risk? Are there perceived or actualinequities in distribution of risks and benefits?

Other informationHave other risk assessments been done on this chemical and were there significantly diHerent conclusions?

human health risks from exposure to chemicalsin the environment. The methods used to assessand characterize these risks are being improvedand expanded upon. Recently, the area ofquantifying uncertainty has greatly expanded,along with developing better procedures to use theavailable data more fully. Improved methods tocharacterize and communicate the results of therisk assessment process 'are also being explored.Although significant research has been done todevelop methods to characterize exposure andhealth effects, much more is needed to providerisk decision makers with the accurate estimatesthey need to make reasonable and cost-effectivedecisions. Risk assessors will always be faced withgaps in data and scientific understanding. Howthey deal with these uncertainties will determinehow useful risk assessment will be to decisionmakers and the public.

2.7 REFERENCES

E.M. (1994al Dose-response assessment for develop-mental toxicity. II. Comparison of generic benchmarkdose estimates with no observed adverse effect levels.Fundamental and Applied Thxicology, 13(4), 487-495.

Allen, B.C., Kavlock, R.J., Kimmel, C.A. &. Faustman,E.M. (1994bl Dose-response assessment for develop-mental toxicity. m. Statistical Models. Fundamentaland Applied Thxicology, 13(4), 496-509.

Armitage, P. &. Doll, R. (1954) The age distribution ofcancer and a multistage theory of carcinogenesis.British TournaI of Cancer, 8, 1-12.

Armitage, P. &. Doll, R. (1961) Stochastic modelsfor carcinogenesis. In: Proceedings of the FourthBerkeley Symposium on Mathematical Statistics andProbability, Vol. 4, pp. 19-38. University of CaliforniaPress, Berkeley.

Ashby, J., Doerrer, N.G., Flamm, F.G. et al. (1990)A scheme for classifying carcinogens. RegulatoryToxicology and Pharmacology, 12,270-295.

Barnes, D.G. &. Dourson M.L. (1988) Reference dose(RfD/: description and use in health risk assessments.Regulatory Thxicology and Pharmacology, 8, 471-486.

Calabrese, E.J. (1985) Uncertainty factors and inter-individual variation. Regulatory Toxicology andPharmacology, 5, 190-196.

Calabrese, E.J. &. Gilbert, C.E. (1993) Lack of totalindependence of uncertainty factors (UPs): implicationsfor the size of the total uncertainty factor. RegulatoryToxicology and Pharmacology, 17,44-51.

Calabrese, E.J., Beck, B.D. &. Chappell, W.R. (1992) Doesthe animal to human uncertainty factor incorporateinterspecies differences in surface areal RegulatoryThxicology and Pharmacology, 15(2), 172-179.

Abdel-Rahman, M.S. &. Dourson, M.L.jGuest Eds} (1995)Proceedings of EPA Uncertainty Factor Workshop.Journal of Human and Ecological Risk Assessment,15(5).

AlHC (1992' Improving Ri~k Characterization. Summary

of Workshop held September 26 and 27, 1991.American Industrial Health Council.

Allen, B.C., Kavlock, R.J., Kimmel, C.A. &. Faustman,

Chapter 222

Crump, K.S. (1984) A new method for determiningallowable daily intakes. Fundamental and Applied

Thxicology,4,854-871.Crump, K.S. (19861 Letter to the Editor: reply.

Fundamental and Applied Thxicology, 6, 183-184.Crump, K.S., Hoel, D.G., Langley, C. &. Peto R.

(19761 Fundamental carcinogenic processes and theirimplications for low dose risk assessment. CancerResearch, 36, 2973-2979.

DeRosa, C.T., Stara, J.F. &. Durkin, P.R. (1985) Rankingof chemicals based upon chronic toxicity data.Thxicology and Industrial Health, 1(41, 177-192.

Dourson, M.L. &. Stara, J.F. (19831 Regulatory history andexperimental suPIK>rt of uncertainty (safety 1 factors.Regulatory Thxicology and Pharmacology, 3, 224-238.

Dourson, M.L., Hertzberg, R.C., Hartung, R. &.Blackburn, K. (1985) Novel approaches for theestimation of acceptable daily intake. Thxicology andIndustrial Health 1(4~, 23-41.

Dourson, M.L., Hertzberg, R.C. &.Stara, J.F.(19861 Letterto the Editor. Fundamental and Applied Thxicology,6, 182-183.

Dourson, M.L., Knauf, L.A. &. Swartout, J.C.(I992J OnReference Dose (RfD) and its underlying toxicity database. Thxicology and Industrial Health, 8(3),171-189.

Dourson, M.L. (19941 Methods for establishing oralreference doses (RfDs).In: Risk Assessment of EssentialElements (eds W. Mertz, C.O. Abernathy &. S.S. Olin I,pp. 51-61. ILSI Press, Washington DC.

Farland, W. &. Dourson, M.L. (19921 Noncancer healthendIK>ints: approaches to quantitative risk assessment.In: Comparative Environmental Risk Assessment (ed.R. Cothern, pp. 87-106. Lewis Publishers, Boca Raton.

FL.Faustman, E.M. Allen, B.C., Kavlock, R.J. &.

Kimmel, C.A. (19941 Dose-response assessment fordevelopmental toxicity. I. Characterization of databaseand determination of no observed adverse effectlevels. Fundamental and Applied Thxicology, 23(4),478-486.

Hartley, W.R. &. Ohanian, E.V. 11988) The use ofshort-term toxicity data for prediction of long-termhealth effects. In: 7race Substances in EnvironmentalHealth-XXII, University of Missouri, May 23-26 (ed.D.D. Hemphil), pp. 3-12.

Hattis, D., Erdreich, L. &. Ballew, M. (1987) Humanvariability in susceptibility to toxic chemicals-apreliminary analysis of pharrnacokinetic data fromnormal volunteers. Risk Analysis, 7(41, 415-426.

Hertzberg, R.C. (1989) Fitting a model to categoricalresponse data with application to species extraIK>lationof toxicity. Health Physics, 57, 405-409.

Hertzberg, R.C. (1991) Quantitative extrapolation oftoxicological findings. In: Statistics in Toxicology (edsD. Krewski &. C. Franklin/. Gordon and Breach Science

Publishers, New York.

Hertzberg, R.C. &. Miller, M. (1985) A statisti:al modelfor species extrapolation using categoric£ responsedata. Toxicology and Industrial Health, 1,43-57.

Hertzberg, R.C. &. Wymer, L. (1991) Modeling theseverity of toxic effects. In: Proceedings' Papersfrom the 84th Annual Meeting and Exhibition, June16-21, British Columbia. Air and Waste ManagementAssociation.

Hertzberg, R.H. &. Dourson, M.L. (19931 Using categoricalregression instead on a NOAEL to characterize atoxicologist's judgment in noncancer risk assessment.In: Proceedings, Second International Symposium onUncertainty Modeling and Analysis (ed. B.M. Ayyub),pp. 254-261. IEEE Computer Society Press, LosAlamitos, CA.

IARC (1978) Monographs on the Evaluation of theCarcinogenic Risk of Chemicals to Humans, Vol. 17.International Agency for Research on Cancer, Lyon.

International Programme on Chemical Safety (IPCS)(19941 Environmental Health Criteria No. 170:Assessing Human Health Risks of Chemicals.Derivation of Guidance Values for Health-basedExposure Limits. World Health Organization, Geneva.

Jarabek, A.M. (1994) Inhalation RfC methodology:dosimetric adjustments and dose-response estimationof noncancer toxicity in the upper respiratory tract.Inhalation Toxicology, 6(Suppl.), 301-325.

Kimmel, C.A. &. Gaylor, D. W. (19881 Issues in qualitativeand quantitative risk analysis for developmentaltoxicity. Risk Analysis, 8(1),15-20.

Kimmel, C.A. Wellington, D.G., Farland, W. et al.(1988) Overview of a workshop on quantitativemodels for developmental toxicity risk assessment.Environmental Health Perspectives, 79, 209-215.

Kroes, R., Munro, I. Poulsen, E. (edsl (19931 ScientificEvaluation of the Safety Factor for the AcceptableDaily Intake. Food Additives and Contaminants, 10(3),269-373.

Krewski, D., Brown, C. &. Mwdock, D. (1984)Determining 'safe' levels of exposure: safety factorsor mathematical models? Fundamental and AppliedToxicology, 4, 5383-5394.

Leisenring, W. &. Ryan, L. (1992) Statistical properties ofthe NOAEL. Regulatory Toxicology and Pharmacology,15(21, 161-171.

Lewis, S.C., Lynch J.R. &. Nikiforov, A.I. (1990) A new

approach to deriving community exposwe guidelinesfrom no-observed-adverse-effect levels. RegulatoryToxicology and Pharmacology, 11,314-330.

Lu, F. 1983. Toxicological evaluations of carcinogens andnoncarcinogens: pros and cons of different approaches.Regulatory Toxicology and Pharmacology, 3, 121-132.

Lu, F. 1985. Safety assessments of chemicals withthresholded effects. Regulatory Toxicology andPhaImQcology, 5, 460-464.

Lu, F. (19881 Acceptable Daily Intake: inception,

Risks to Human Health from Chemicals 23

USEPA 11986) Guidelines for Carcinogen RiskAssessment. US Environmental Protection Agency, 51Federal Register 33992, pp. 1-17.

USEPA 119911 Guidelines for Developmental ThxicityRisk Assessment. Federal Register, Vol. 56, No. 234,pp.63798-63826.

USEPA (1992) Guidelines for Exposure Assessment. USEnvironmental Protection Agency, 57 Federal Register,22888-22938.

USEPA (1994al Methods for Derivation of InhalationReference Concentrations and Application ofInhalation Dosimetry. US Environmental ProtectionAgency, Office of Health and EnvironmentalAssessment, Washington, DC, EPA/600/8-90-066F.

USEPA (1994b) Draft Revisions to the Guidelinesfor Carcinogen Risk Assessment. US EnvironmentalProtection Agency, Office of Research and Develop-ment, Washington, DC, External Review Draft.

USEPA 11995al The Use of the Benchmark DoseApproach in Health Risk Assessment. Risk AssessmentForum, February, US Environmental ProtectionAgency, Washington, DC, EPA/630/R-94/007.

USEPA (1995bl Policy and Guidance for RiskCharacterization. Memo from Carol M. Browner,Administrator to Assistant Administrators, March 21.US Environmental Protection Agency, Washington,DC.

USEPA (1996) Proposed Guidelines for Carcinogen RiskAssessment. US Environmental Protection Agency,Office of Research and Development, Washington, DC,EPA/600/P-92/003C.

Zielhuis, R.L. & van der Kreek, F.W. (19791 The use ofa safety factor in setting health based permissiblelevels for occupational exposure. InternationalArchives of Occupational Environmental Health, 42,191-201.

evolution, and application. Regulatory Thxicology andPharmacology, 8, 45-60.

Meek, M.E., Newhook, R., Liteplo, R.G. &. Armstrong,V.C. (19941 Approach to assessment of risk to humanhealth for priority substances under the CanadianEnvironmental Protection Act. EnvironmentalCarcinogenesis and Ecotoxicology Reviews, CI1(2),105-134.

Moolenaar, R.J. (1994) Carcinogen risk assessment:international comparison. Regulatory ThxicologyandPharmacology, 10, 302-336.

Munro, I.C. &. Krewski, D.R. (1981) Risk assessment andregulatory dec*sion making. Food and Cosmetics1bxicology, 19,549-560.

NAS (National Academy of Sciences) (1983) RiskAssessment in the Federal Government: Managingthe Process. National Academy Press, Washington,DC.

NAS (National Academy of Sciences) (19941 Science andJudgment in Risk Assessment. National AcademyPress, Washington, DC.

Pohl, H.R. &. Abadin, H.G. (1995) Utilizing uncertaintyfactors in minimal risk levels derivation. Regulatory1bxicology and Pharmacology, 11, 180-188.

Rademaker, B.C. &.Linders, J.B.H.J. (1994) Progress Report3: Estimated-concentrations-of-no-concern of pollutingAgents in Drinking Water and Air for Humans.Nationallnstitute of Public Health and EnvironmentalProtection, Bilthoven, The Netherlands.

Renwick, A.G. (19911 Safety factors and establishmentof acceptable daily intake. Food Additives andContaminants, 8(2), 135-lSO.

Renwick, A.G. (19931 Data derived safety factors forthe evaluation of food additives and environmentalcontaminants. Food Additives and Contaminants10(3},275-305.