0. A Decision Analytic Framework for Considering the Economic Value of Improved Risk Assessment Data
Eric Ruder
Henry Roman Industrial Economics
October 12, 2011

1. Overview of Presentation
Objective: Explore the potential value of improved risk assessment data in the context of environmental regulation and research priority setting (both public and private sector)
Nature of the problem
Analytical framework.
Examples
Results of Case Studies: Lead regulation and mobile source related air toxics requirements
Hypothetical scenario of research to support product development
Implications

2. Nature of the Problem – Evaluating Improvements in Risk Assessment
There are numerous sources of uncertainty in risk assessment that may be addressed (at least partially) by improved risk assessment methods and/or data:
Improving methods for evaluating the potential risks of chemicals.
Identifying susceptible populations and characterizing factors that may place them at higher risk.
Understanding how chemicals move and change along pathways from sources to potentially exposed populations.
Advances in these areas seems of obvious value in the field of risk assessment and the protection of human health and the environment, but are there quantifiable net benefits for society or economic benefits for the private sector?

3. Value of Information Analysis
Value of information (VOI) represents the improvement in the expected value of a decision outcome that would result from collecting additional information about one or more factors affecting a decision.
Decision analysis approaches can quantify the value of collecting additional information before making a specific decision.
Involves comparison of the expected outcomes of a suite of alternative regulatory choices made with and without information believed critical for the issue.

4. Why is VOI Important?
Structuring decision helps decision-makers make better decisions
identifies key decision inputs
helps minimize expected loss / maximize expected gain
increases chances of good outcome
makes explicit the "costs" of uncertainty
Results provide a measure of social willingness-to-pay for new research
Helps prioritize competing research opportunities to allocate resources efficiently

5. Evaluations of VOI as a Tool
NRC’s Science and Decisions: Advancing Risk Assessment (2009)
Discusses VOI as a tool for understanding the tradeoffs between timely decision-making and the desire to refine the underlying science.
Barriers to applying formal VOI broadly
Underlying concepts and structure still valuable: linkages between information, decision-maker behavior, and decision making objectives
Recommends informal VOI or value of methods analyses to inform risk assessment design.

6. Evaluations of VOI as a Tool (cont’d)
2010 Report of EPA’s Board of Scientific Counselors (BOSC) decision analysis workshop with ORD/NRMRL
VOI can be valuable, but is challenging to implement
Need extensive data, including probabilities for decision options & how p’s change conditional on new information
Nonetheless, group mentions VOI as possible method to inform several case studies – e.g., regulations of chemicals with biomarker data, prioritizing IRIS evaluations
Key to consider benefits as the net improvements in decision outcomes minus the costs of obtaining improved information.

7. VOI Analytic Framework
Decision analysis approaches can quantify the value of collecting additional information before making a specific decision.
Involves comparison of the expected outcomes of a suite of alternative regulatory choices made with and without information believed critical for the issue.
Framework in a regulatory context can be illustrated using two diagrams:
Influence diagram - showing the role of exposure data and other inputs in estimating NSB.
Decision tree - depicting a choice among:
Taking no further regulatory action.
Instituting additional control measures based on current information.
Collecting additional exposure information prior to deciding whether to institute new controls.

8. VOI Framework Influence Diagram Dose-Response Information
Effectiveness of Controls
Value of
Health Effects
Change in Exposure
Change in Health Effects
Monetized Benefits (Avoided Health Effects)
Implement Controls?
Baseline Exposure
Control Costs
Net Benefits

9. VOI Framework (continued)
Hypothetical Decision Tree: VOI = $500 - $437.5 = $62.5
Policy Decision
Exposure Distribution
Control Decision
Net Social Benefits
No Controls $0
High
.25
$1,000
Controls
[$437.5]
Medium .5
$500
Low
.25
$-250
No Controls $0
High
.25
Controls
$1,000
Collect Exposure
Information
No Controls
[$500] $0
Medium .5
Controls
$500
No Controls $0
Low
.25
All values in millions of $
Controls
$-250

10. VOI Framework (continued)
Variations in complexity of analysis:
Value of perfect information -- assumes that collection of information will eliminate all uncertainty in exposure.
Value of imperfect information -- examines the potential that addition of information will resolve some, but not all of the uncertainty in exposure.
Value of partial information -- considers the impact of uncertainty in other inputs to the decision (e.g., toxicity).

11. Summary of Case Study Results
Lead Case Study
There are significant benefits of improving data ($13.2 billion to $24.3 billion).
This value is robust when we consider potential delay in obtaining exposure data and the magnitude of uncertainties in other inputs.
Conclusion depends on the ability to resolve the uncertainty regarding lead uptake, especially among the highly exposed (i.e., children).
Mobile Source Air Toxics Case Study
There are potential benefits to improving exposure data, up to $45 million annually.
This value decreases substantially when we consider the larger uncertainty in toxicity for these compounds.

12. Background
Decision we analyzed: regulating lead in residences under TSCA Section 403.
Rule sets three standards: floor dust loading (µg/ft2); window sill dust loading (µg/ft2); and soil concentration (ppm). Violation triggers abatement.
Economic and risk analysis supporting the rule looked at 1000 combinations of these three standards.
Risk and economic estimates based on two alternative models for relationship between environmental lead and children’s blood lead:
IEUBK: more sensitive, says regulate stringently
The “Empirical” model: much less sensitive, says do not regulate
November 1998 proposed standards reflect EPA’s balancing of these two exposure outcomes.

13. VOI Case Study - Lead Exposure

14. Lead Rule RIA Influence Diagram

15. Decision Tree for Lead Regulatory Decision
VOI = $43.3b - $19b = $24.3b
No_Controls
[0]
IEUBK
$173.2
.250
Empir
$-34.9
.750
HighControls
[$17.1]
$107.2
$-10.4
Medium_Controls [$19]
$77.9
$ -8.9
Low_Controls
[$12.8]
High_Controls
Medium_Controls $0
[$173.2]
$-8.9
[$0]
Exposure_Data [$43.3]

16. Effect of Delay on VOI Delay scenarios assume: Results:
Benefits are delayed if “collect exposure data” option is chosen.
Discounted over delay period at 3 percent annually.
Results:
Perfect VOI, no delay: $24.3 billion
Perfect VOI, five-year delay: $18.4 billion
Perfect VOI, ten-year delay: $13.2 billion.

17. Value of Imperfect Information
Value of imperfect information analysis asks: What if improved data cannot completely resolve uptake uncertainty?
Looked at best and worst case scenarios for power of NHEXAS-like data to resolve IEUBK/Empirical uncertainty (from 50 percent predictive accuracy to 90 percent predictive accuracy).
Results show VOI remains substantial when improved data achieves a high level of predictive accuracy and the prior likelihood that IEUBK is correct is less than 50 percent.

18. Expected Value of Imperfect Lead Exposure Information

19. Conclusions of Lead Case Study
Value of exposure information could be very high; under best case conditions, a national exposure survey could have value in the tens of billions of dollars.
Effect of delay in provision of information is to reduce VOI, but value of perfect information remains substantial even with 10-year delay.
Improved exposure information must achieve a predictive accuracy of 80 to 90 percent certainty in forecasting the "correct" lead uptake model for VOI to remain substantial (e.g., no more than a 20 percent chance of predicting IEUBK if the empirical model is correct); lower levels of certainty quickly erode VOI.

20. VOI Case Study: Mobile Air Toxics
Retrospective analysis of VOI of exposure data for benzene and 1,3-butadiene, two air toxics regulated under the 1990 CAAA.
Developed estimates of exposure and risk in 2000 both with and without control programs -- based on information available as rules were promulgated.
Other inputs: control costs (over $2 billion annually); other benefits of air toxics reductions (about $850 million annually).
Estimate value of improved information for mobile source air toxics relative to improvements in other key uncertain inputs (e.g., toxicity).

21. Mobile Source Air Toxics - Exposure Uncertainty
Alternative Exposure/Distributions for 1,3-Butadiene

22. Calculation of Net Social Benefits

23. Decision Tree for Motor Vehicle Air Toxics

24. Conclusions of Mobile Sources Case Study
Expected value of perfect information = $93 million.
Explored value of partial information by considering impact of uncertainty in toxicity of 1,3-butadiene and benzene.
Value of exposure information decreases to $9 million.
Value of toxicity information is greater, $118 million.
Explored impact of varying dollar value of a statistical life (VSL).
Exposure information has value across a wide range of VSL (approximately $3 million to $7 million).

25. Estimating the Economic Value of Research in the Private Sector – Product Decision
Studies are routinely performed to better understand exposure and toxicity prior to making a product development decision in the private sector, but these are usually still uncertain elements, and there may be concerns about study assumptions or study quality that can compound this uncertainty.
What is the value of research that improves both our fundamental understanding of exposure and toxicity and the methods used to assess these important decision elements? Improved methods can lead to increased confidence and reduced uncertainty in results that will help inform firm decision-making and regulators.

26. Influence Diagram for Estimating the Economic Value of Research
Price
Demand/
Volume
Sold
Revenue
Produce
Chemical?
Profit
Research
Exposure
Study
Tox
Exposure
Study
Tox
Regulatory
Action
Cost
Exposure
Cancer
Potency
Regulatory
Action

27. Value of Perfect Toxicity Information
MOA__1__High
Yes
[50]
.50
-100
MOA__2__Low
.50
200 No
[0]
Produce_Chemical_
[100]
Produce_Chemical2
Yes
MOA__1__High
[0]
-100
Toxicity
.500
.50 No
Research
[100]
Produce_Chemical2
Yes
MOA__2__Low
[200]
200
.50 No
Values in millions of dollars. Bracketed numbers represent expected values for the outcome measure.
Example: Bioassays for compound X in two different animal species suggest two possible modes of action (MOAs) with very different implications for cancer potency, and hence for regulatory action and/or consumer demand and, ultimately, potential profit. Each MOA appears equally plausible at present. The firm can proceed with production or abandon the chemical. However, additioanl research may help to resolve which MOA is relevant for humans at expected ambient levels and lead to a better informed decision. (Note: This example assumes exposure is well known.)
In the ideal case, research fully resolves the uncertainty. The value of this “perfect” information is the difference between the expected value of the Research branch ($100 million) and the next best alternative, the Yes branch ($50 million) = $50 million. This represents an upper bound estimate on the value of research on toxicity.

28. Value of Imperfect Toxicity Information
MOA__1__High
Yes
[50]
.50
-100
MOA__2__Low
.50
200 No
[0]
Toxicity_Given_High_Result
MOA__1__High
Produce_Chemical_
Produce_Chemical2
Yes
[-40]
.80
-100
[70]
MOA__1__High
[0]
MOA__2__Low
.50
.20
200
Study_Prediction No
[0]
LRI_Study
[70]
Toxicity_Given_Low_Result
MOA__1__High
Values in millions of dollars.
Bracketed numbers represent expected
values for the outcome measure.
Produce_Chemical2
Yes
[140]
.20
-100
MOA__2__Low
[140]
MOA__2__Low
.50
.80
200 No
[0]
Research is unlikely to fully resolve uncertainty, yet It may still be valuable if it reduces the level of uncertainty. In the bottom branch following the study prediction, there is still some residual uncertainty about the MOA. However, if the study (e.g., PBPK modeling for MeCl2) is viewed as highly credible, the uncertainty in the MOA, given the study prediction, could be significantly reduced (80/20 in favor of the study prediction in this hypothetical example).
In this case of “imperfect” information, the research would still be worth pursuing, though the value would be less ($70 million - $50 million = $20 million). When using this type of VOI tool it is possible to explore ranges of values associated with different levels of confidence in the predictive power of LRI studies (or their likelihood of influencing regulatory decisions and customer demand).

29. Applications / Requirements
VOI framework can be applied to variety of issues (pollutant regulation, global warming) and focus on different inputs (exposure, toxicity, economic valuation).
Key constraint is data availability.
Need to be able to estimate outcomes (ideally, net social benefits) in dollars
Need to quantitatively characterize decision inputs (e.g., costs, population exposure levels, toxicity, non-health effects, benefit values)
Need to quantify uncertain elements either discretely or using distributions (e.g., lognormal) pre- and post-information -- may require expert elicitation

30. Identifying Decisions Where VOI May be High
Stakes are high; divergent outcomes
Decision expected to be sensitive to health outcomes and not dictated by non health-related factors
Uncertainty can be represented using small number of scenarios (e.g., alternative MOAs) and associated probabilities
Additional basic research can change beliefs about these states (e.g., improved PBPK modeling, interpretation of biomarker data, MOA research)
Uncertainty in other decision inputs not significantly greater than uncertainty in input of interest.

31. IEc INDUSTRIAL ECONOMICS, INCORPORATED
Eric Ruder, Principal
Henry Roman, Principal