Characterizing Uncertainty in Human Health Risk Assessment:

An Agency Perspective

Lynn Flowers, PhD, DABT National Center for Environmental Assessment Office of Research and Development U.S. Environmental Protection Agency Washington DC This presentation does not necessarily represent the views or policies of the U.S. Environmental Protection Agency.

2

o IRIS Program Background

o IRIS Users

o Characterizing Uncertainty: NRC Recommendations

o IRIS Program’s Workshop on NRC Recommendations (2014)

o Examples: Recent IRIS Assessments

o Summary and Next Steps

Overview

3

IRIS assessments systematically review publicly-available peer-reviewed studies to:

Identify adverse health outcomes

Characterize exposure-response relationships

IRIS Background

HAZARD IDENTIFICATION

Which health outcomes are caused by the agent?

DOSE-RESPONSE ASSESSMENT

Characterize exposure-response relationships

Account for high-to-low-dose, animal-to-human, route-to-route, and other differences

EXPOSURE ASSESSMENT

How do people come in contact with the agent?

How much are they exposed to?

RISK CHARACTERIZATION

Integrate HAZARD, DOSE-RESPONSE, and EXPOSURE

RISK MANAGEMENT

Analyze and compare options

Select an appropriate action

LEGAL

POLITICAL

SOCIAL

ECONOMIC

TECHNICAL CONSIDERATIONS

IRIS assessments

Risk assessment – other steps

Risk management

4

Results from epidemiological and animal studies generally need to be extrapolated to inform risk management:

Clean Air Act specifies “an ample margin of safety to protect public health.”

Safe Drinking Water Act specifies “no adverse effects on the health of persons may reasonably be anticipated to occur, allowing an adequate margin of safety.”

Cancer decisions often consider a range of risks between 1/10,000 and 1/million.

It is not feasible to always wait for new studies.

General Principles - The characterization of uncertainty should promote assessments that: provides useful information to risk managers; are completed in a reasonable time; use a reasonable level of resources; can use the data at hand.

EPA’s Programs and Regions Make Decisions About Potential Risks

5

1. Systematic identification of relevant evidence

2. Criteria for evaluating the strength of the evidence

Standardized to avoid ambiguity

3. Unify dose-response framework

Cancer assessments should reflect variability and uncertainty

Noncancer assessments should reflect probability of response

4. Combine information from multiple studies

Should be unusual to use only one study

Consideration of meta-analyses, Bayesian analyses

5. Characterization and communication of uncertainty

NRC Recommendations for Hazard Identification and Dose-Response Assessment

6

“Overall, the committee finds that substantial improvements in the IRIS process have been made, and it is clear that EPA has embraced and is acting on the recommendations in the NRC formaldehyde report.”

“The NRC formaldehyde committee recognized that its suggested changes would take several years and an extensive effort by EPA staff to implement.”

“Substantial progress, however, has been made in a short time, and the present committee’s recommendations should be seen as building on the progress that EPA has already made.”

IRIS was Recently Reviewed by the NRC (May 2014)

“Overall the committee expects that EPA will complete its planned revisions in a timely way and that the revisions will transform the IRIS Program.”

7

1. Improve documentation and presentation of dose-response information.

Recommendation: EPA should clearly present two dose-response estimates: a central estimate (such as a maximum likelihood estimate or a posterior mean) and a lower-bound estimate for a POD from which a toxicity value is derived.

2. Advanced analytic methods, such as Bayesian methods, for integrating data for dose-response assessments and deriving toxicity estimates are underused by the IRIS Program.

Recommendation: As the IRIS program evolves, EPA should develop and expand its use of formal quantitative methods in data integration for dose-response assessment and derivation of toxicity values.

3. IRIS-specific guidelines for consistent, coherent, and transparent assessment and communication of uncertainty remain incompletely developed.

Recommendation: Uncertainty analysis should be conducted systematically and coherently in IRIS assessments.

Specific NRC Recommendations Regarding Uncertainty (NRC, 2014)

8

Uncertainty: Lack or incompleteness of information. Quantitative uncertainty analysis attempts to analyze and describe the degree to which a calculated value may differ from the true value; it sometimes uses probability distributions. Uncertainty depends on the quality, quantity, and relevance of data and on the reliability and relevance of models and assumptions.

Variability: Variability refers to true differences in attributes due to heterogeneity or diversity. Variability is usually not reducible by further measurement or study, although it can be better characterized.

Vulnerability: The intrinsic predisposition of an exposed element (person, community, population, or ecologic entity) to suffer harm from external stresses and perturbations; it is based on variations in disease susceptibility, psychological and social factors, exposures, and adaptive measures to anticipate and reduce future harm, and to recover from an insult.

Sensitivity: The degree to which the outputs of a quantitative assessment are affected by changes in selected input parameters or assumptions.

What is Uncertainty? (Silver Book, 2009)

9

The level of uncertainty analysis might be tiered according to quantification level:

from a single default (no variation);

to qualitative and systematic characterization;

to quantitative characterization with bounds, ranges, and sensitivity;

to a probabilistic distribution.

Framework for Uncertainty Analysis and Communication (NRC, 2014)

10

Goal: Evaluating Uncertainty by Incorporating Probabilistic Approaches

Deterministic

RfD

… a daily oral exposure to the human population

(including sensitive subgroups) that is likely to

be without an appreciable risk of deleterious

effects during a lifetime.

… a daily oral exposure where, with 95%

coverage (confidence), 1% of the human

population shows more than 5% decrease in

red blood cell counts during a lifetime.

Probabilistic

RfD

(with 95%

coverage)

Target human

dose HDMI

(e.g., HD0501)

95% coverage

uncertainty

11

“EPA has developed standard descriptors to characterize the level of confidence in each reference value on the basis of the likelihood that the value would change with further testing. Development of the descriptors is consistent with guidelines for deriving recommendations from systematic reviews that evaluate the quality of evidence.” (NRC, 2014)

Standard Descriptors to Characterize Level of Confidence:

High confidence: The reference value is not likely to change with further testing, except for mechanistic studies that might affect the interpretation of prior test results.

Medium confidence: This is a matter of judgment, between high and low confidence.

Low confidence: The reference value is especially vulnerable to change with further testing.

Descriptors to Characterize Level of Confidence

12

NRC (2014) Support for Data Representation in IRIS PCE Assessment

13

o EPA hosted a public workshop to discuss several specific recommendations from the NRC May 2014 report.

o Topics included:

Session 1: Systematic Integration of Evidence Streams for IRIS

Session 2: Adapting Systematic Review Methodologies for IRIS

Advancing Dose-Response Analysis – Combining Multiple Studies

Advancing Dose-Response Analysis – Uncertainty Analysis

o Not a consensus workshop; beginning discussion; varied opinions

o Date: October 15-16, 2014, Arlington, VA

http://www.epa.gov/iris/irisworkshops/NRC_workshop/index.htm

EPA IRIS Workshop on the 2014 NRC Recommendations

14

Session 3a

Advancing Dose-Response Analysis:

Combining Multiple Studies

IRIS already uses multiple studies

Candidate values from different studies for each health outcome

Different databases for different parameters in complex models

How to combine human studies in a systematic and replicable manner?

How to derive confidence bounds from different experimental systems with divergent results?

IRIS Workshop on the NRC Recommendations

15

Session 3b

Advancing Dose-Response Analysis:

Uncertainty Analysis

For users of IRIS assessments:

How do you use estimates of uncertainty and variability?

What information would be most useful in your decisions?

For analysts:

What practical approaches would meet these user needs?

How to derive appropriate and defensible confidence bounds?

IRIS Workshop on the NRC Recommendations

16

o Libby Amphibole Asbestos (2014)

o Dioxin (noncancer) (2012)

o Tetrachloroethylene (2012)

o Trichloroethylene (2011)

o Inorganic arsenic (ongoing) (http://www.epa.gov/iris/publicmeeting/iris_bimonthly-

jun2014/mtg_docs.htm#ia)

o Ethylene oxide (ongoing) (http://cfpub.epa.gov/ncea/iris_drafts/recordisplay.cfm?deid=282012)

Characterization of Uncertainty in IRIS Assessments

17

II.B. QUANTITATIVE ESTIMATE OF CARCINOGENIC RISK FROM ORAL EXPOSURE

II.B.1. SUMMARY OF RISK ESTIMATES

II.B.1.1. Upper Bound Oral Slope Factor = 8.2 x 10–3 per mg/kg-day

Calculation of the upper bound oral slope factor: The oral slope factor is derived from the LED10 (12.2 mg/kg-day), the 95% lower bound on the exposure associated with an 10% extra cancer risk, by dividing the risk (as a fraction) by the LED10, and represents an upper bound, continuous lifetime exposure risk estimate.

Calculation of the slope of the linear extrapolation from the central estimate: The slope of the linear extrapolation from the central estimate is derived from the ED10 (18.7 mg/kg-day), the central estimate of exposure at 10% extra cancer risk, by dividing the risk (as a fraction) by the ED10 and equals 5.3 x 10–3 per mg/kg-day.

Presentation of the Upper Bound and Central Estimate of Cancer Risk: Biphenyl (2013)

18

o MOA analysis:

Induction of urinary bladder tumors in male rats is likely to be a high-dose phenomenon related to the formation of urinary bladder calculi.

Urinary bladder tumors in male F344 rats likely will not occur without the development of calculi.

MOA is assumed to be relevant to humans, but humans would likely be less susceptible to the tumors than rats.

o Characterization of uncertainty:

What if the formation of urinary calculi was the critical effect? A candidate RfD for bladder calculi of 0.9 mg/kg-day was derived that is approximately two-fold higher than the final RfD of 0.5 mg/kg-day based on papillary mineralization in the kidney.

What if there is an alternative MOA that has a linear low-dose component? A linear extrapolation approach for urinary bladder tumors was performed. A slope factor of 2 x 10-3 per mg/kg-day was derived that is lower than the slope factor derived from mouse liver tumors.

Characterization of Uncertainty Related to MOA and Formation of Bladder Tumors in Rats (Biphenyl, 2013)

19

Exposure-Response Array for Developmental Effects Following Oral Exposure to Benzo[a]pyrene (External Review Draft)

20

Candidate Reference Values for Benzo[a]pyrene with Corresponding PODs and Composite UFs

21

Organ/System-Specific RfDs and Proposed Overall RfD for Benzo[a]pyrene

Effect

Point of

Departure

(mg/kg-d) UF

Chronic RfD

(mg/kg-d) Confidence

Developmental:

Neurobehavioral changes

Chen et al. (2012)

Neurodevelopmental study in rats

BMDL: 0.086 Total UF = 300

UFA = 10

UFH = 10

UFDB = 3

3 x 10 -4 Medium

Reproductive:

Decreased ovary weight

Xu et al. (2010)

60 day reproductive study in adult rats

BMDL: 0.37 Total UF = 1000

UFA = 3

UFH = 10

UFS = 10

UFDB = 3

4 x 10 -4 Medium

BMDL: 1.9 Total UF = 1000

UFA = 3

UFH = 10

UFS = 10

UFDB = 3

2 x 10 -3 Low

Proposed Overall Reference Dose (RfD) - Developmental 3 x 10 -4 Medium

Immunological:

Decreased thymus weight and IgM

De Jong et al. (1999)

35 day study in adult rats

22

Tumor

Species/

sex

POD = BMD

(mg/kg-d)

CENTRAL TENDENCY

POD = BMDL

(mg/kg-d)

UPPER BOUND

Forestomach, oral cavity: squamous cell tumors

Kroese et al. (2001)

Male Wistar rats 0.453 0.281

Hepatocellular adenomas or carcinomas

Kroese et al. (2001)

Male Wistar rats 0.651 0.449

Jejunum/duodenum adenocarcinomas

Kroese et al. (2001)

Male Wistar rats 3.03 2.38

Kidney: urothelial carcinomas

Kroese et al. (2001)

Male Wistar rats 4.65 2.50

Skin, mammary:

Basal cell tumors

Squamous cell tumors

Kroese et al. (2001)

Male Wistar rats

2.86

2.64

2.35

1.77

Forestomach, oral cavity: squamous cell tumors

Kroese et al. (2001)

Female Wistar rats 0.539 0.328

Hepatocellular adenomas or carcinomas

Kroese et al. (2001)

Female Wistar rats 0.575 0.507

Jejunum/duodenum adenocarcinomas

Kroese et al. (2001)

Female Wistar rats 3.43 1.95

Forestomach, esophagus, tongue, larynx

(alimentary tract): squamous cell tumors

Beland and Culp (1998)

Female B6C3F1

Mice

0.127 0.071

Summary of Benzo[a]pyrene Oral Slope Factor Derivations

23

Summary of the Dose Response Analysis for Benzo[a]pyrene Oral Cancer Data

Principal Study Elevated tumor types

Selected

model

Oral Slope

factorHED

(mg/kg-d)-1

Kroese et al. (2001)

male rats

Forestomach and oral cavity squamous cell

tumors; hepatocellular adenomas or carcinomas;

small intestine adenocarcinomas;

Kidney urothelial carcinomas; skin/mammary

basal cell and squamous cell tumors

Multistage

Weibull

0.5

Kroese et al. (2001)

female rats

Forestomach and oral cavity squamous cell

tumors; hepatocellular adenomas or carcinomas;

small intestine adenocarcinomas;

Multistage

Weibull

0.3

Beland and Culp

(1998)

female mice

Esophagus, tongue, and larynx squamous cell

tumors

Multistage

Weibull

1

• Tumor types modeled individually and then compiled to estimate overall risk of developing

any tumor type.

• Most potent slope factor was used to represent overall risk.

Proposed Oral Slope Factor = 1 per mg/kg-day

24

Consideration and

impact on cancer risk value Decision Justification and discussion

Selection of target organ

↓ oral slope factor, up to

fivefold, if alimentary tract

tumors not selected

Alimentary tract

tumors

(forestomach,

esophagus, tongue,

larynx)

Tumor site is concordant across rats and mice,

increasing support for its relevance to humans.

As there are no data to support any one result

as most relevant for extrapolating to humans,

the most sensitive result for alimentary tract

tumors was used to derive the oral slope factor.

Selection of data set

↓ oral slope factor ~threefold

if rat bioassay were selected

for oral slope factor derivation

Beland and Culp

(1998)

Beland and Culp (1998) was a well-conducted

study and had the lowest HEDs of the available

cancer bioassays, reducing low-dose

extrapolation uncertainty.

Interspecies extrapolation

Alternatives could ↓ or ↑ slope

factor (e.g., 3.5-fold ↓ [scaling

by body weight] or ↑ 2-fold

[scaling by BW2/3])

BW3/4 scaling

(default approach)

There are no data to support alternatives.

Because the dose metric was not an area

under the curve, BW3/4 scaling was used to

calculate equivalent cumulative exposures for

estimating equivalent human risks. While the

true human correspondence is unknown, this

overall approach is expected to neither over-

nor underestimate human equivalent risks.

Summary of Uncertainties in the Derivation of Cancer Risk Values for Benzo[a]pyrene Oral Slope Factor

25

o Uncertainty in IRIS assessments needs to be “fit-for-multiple-purposes” across EPA.

o Extensive uncertainty analyses can be resource and time-intensive. The level of detail in the characterization of uncertainty should align with what is needed to inform risk-management decisions.

o Advancements in systematic review, evidence integration, and unified dose-response approaches will reduce uncertainty.

o As risk assessment evolves , so will the characterization of uncertainty: High-throughput technologies, computational toxicology,

systems biology and bioinformatics Epigenetics Cumulative risk

Summary and Next Steps: Characterization of Uncertainty (and Variability) Not a Stagnant Process