Cumulative Risk Assessment for Pesticide Regulation: A Risk Characterization Challenge

Mary A. Fox, PhD, MPH

Linda C. Abbott, PhD

USDA Office of Risk Assessment and Cost-Benefit Analysis

Cumulative Risk Assessment for Pesticide Regulation

• Debut of multi-chemical assessment of pesticide exposure through food, water, and residential uses

• Highly refined dose-response and exposure assessment

• Nationally representative dietary assessment

• What do we know about risk characterization for such complex assessments?

Risk Characterization DefinedNAS 1996

• From Understanding Risk:– A synthesis and summary of information about

a potentially hazardous situation that addresses the needs and interests of decision makers and interested and affected parties

– Analytic-deliberative process– The process of organizing, evaluating, and

communicating …

Outline

• Identify key elements of risk characterization for probabilistic assessments

• Evaluate the risk characterization chapter of the revised organophosphate (OP) assessment

• Review example highlighting importance of uncertainty and sensitivity analyses

Resources

• Presidential/Congressional Commission on Risk Assessment/Management, 1997

• US EPA Guidance– Principles for Monte-Carlo Analysis, 1997– Risk Characterization Handbook, 2000

• US EPA Revised OP Cumulative Risk Assessment, 2002

• DEEM™ and DEEM-FCID ™• Data files for methamidophos

Presidential Commission, 1997

• Quantitative and qualitative descriptions of risk• Summarize weight of evidence • Include information on the assessment itself• Describe uncertainty and variability• Use probability distributions as appropriate • Use sensitivity analyses to identify key uncertainties • Discuss costs and value of acquiring additional information

Did not recommend:• Use of formal quantitative analysis of uncertainties for

routine decision-making (i.e. local, low-stakes)

Excerpts from Guiding Principles of Monte Carlo Analysis, US EPA 1997

• Selecting Input Data and Distributions– Conduct preliminary sensitivity analyses

• Evaluating Variability and Uncertainty– Separate variability and uncertainty to provide greater

accountability and transparency.

• Presenting the Results– Provide a complete and thorough description of the model. The

objectives are transparency and reproducibility.

Risk Characterization Handbook, 2000

• Transparency– Explicitness

• Clarity– Easy to understand

• Consistency– Consistent with other EPA actions

• Reasonableness– Based on sound judgment

Transparency Criteria

• Describe assessment approach, assumptions

• Describe plausible alternative assumptions

• Identify data gaps

• Distinguish science from policy

• Describe uncertainty

• Describe relative strengths of assessment

Key Elements of Risk Characterization

• Separately track and describe uncertainty and variability

• Conduct sensitivity analyses

• Conduct formal uncertainty analyses

• Transparency and reproducibility– Model components – Basic operational details

Evaluation of the Revised OP Cumulative Assessment

• Track and describe uncertainty and variability

• Sensitivity analyses

• Uncertainty analyses– Yes, but …spotty, qualitative, not comprehensive

• Transparency/reproducibility – No– Significance of many inputs unknown

– No mention of random seed, # iterations used

Recipes – essential to dietary model

• Break down foods reported in dietary recall records to commodities that can be matched with pesticide residue data

• Recipes are ‘representative’ with nutritional basis– May not accurately reflect commodities eaten– E.g. beef stew with vegetables – recipe includes carrots

but could be broccoli or leafy greens

• DEEM ™ – proprietary recipes• DEEM-FCID ™ – EPA & USDA collaboration• Policy relevant

foodcode DESCR Food Commodity

(FC)

com_amt g/100g

foodform

74602030 ~Tomato soup, canned, undiluted~ Tomato 85.82 240 (cooked, canned)

74602030 ~Tomato soup, canned, undiluted~ Wheat 8.05 240 (cooked, canned)

74602030 ~Tomato soup, canned, undiluted~ Sugarcane 1.932 240 (cooked, canned)

74602030 ~Tomato soup, canned, undiluted~ Beet 1.518 240 (cooked, canned)

74602030 ~Tomato soup, canned, undiluted~ Soybean 1.108 240 (cooked, canned)

74602030 ~Tomato soup, canned, undiluted~ Cottonseed 0.099 240 (cooked, canned)

74602030 ~Tomato soup, canned, undiluted~ Corn 0.065 240 (cooked, canned)

74602030 ~Tomato soup, canned, undiluted~ Rapeseed 0.052 240 (cooked, canned)

74602030 ~Tomato soup, canned, undiluted~ Sunflower 0.002 240 (cooked, canned)

74602030 ~Tomato soup, canned, undiluted~ Safflower 0.001 240 (cooked, canned)

Tomato Soup Recipe

Experiment to examine importance of recipes

• Focus on one chemical- methamidophos• Look at dietary exposure using DEEM ™

and DEEM-FCID ™

• Forty 1000 iteration replicates with different random number seeds

• 1-6 year olds, 99.9th %ile, exposures in mg/kg-day

Between Model Exposure Variability Forty 1000-Iteration Replicates, Different Random Number Seeds

DEEM ™ Estimate

DEEM-FCID ™

Estimate

Difference % Difference

Minimum 7.43 x 10-4 8.54 x 10-4 1.02 x 10-4 13.56 %

Maximum 7.63 x 10-4 8.80 x 10-4 1.32 x 10-4 17.74 %

Mean 7.53 x 10-4 8.69 x 10-4 1.17 x 10-4 15.48 %

Within Model Exposure Variability Forty 1000-Iteration Replicates, Different Random Number Seeds

DEEM ™ DEEM-FCID ™

Within model exposure variability

2.69 % 3.04 %

On par with US EPA findings for 1000-iteration runs

Exposure variability findings in contextPreliminary data files, Children 1-2, Single 1000 iteration runs

Commodities DEEM-FCID ™

Estimate

Difference % Difference from complete

model

Exclude grapes

0.00157 0.00024 15.29 %

Exclude apples

0.00163 0.00018 11.00 %

All included 0.00181

Average DEEM vs. FCID difference is 15%

Risk Metric Comparison – 15% Difference

Margin of Exposure (MOE) = Toxicological Benchmark

Exposure Estimate

Revised OPCRA Tox. Benchmark for dietary = 0.08 mg/kg-d

MOE average exposure DEEM = 0.08 / 0.000753 = 106

MOE average exposure FCID = 0.08 / 0.000869 = 92

Conclusions

• Risk characterization is incomplete

• Good guidance on risk characterization for complex models

• Continue to work and share findings