Anirban Basu1, David Meltzer2. 1. The Comparative Health Outcome, Policy, and Economics (CHOICE) Institute, Department of Pharmacy and the Departments of Health Services and Economics, University of Washington, Seattle, WA, USA. 2. Section of Hospital Medicine, Department of Medicine, Harris School of Public Policy Studies and the Department of Economics, The University of Chicago, Chicago, IL, USA.
Abstract
BACKGROUND: In value of information (VOI) calculations, such as the expected value of perfect information (EVPI), partial perfect information (EVPPI), sample information (EVSI) or implementation (EVIM), the maximum expected value criterion defines the decision making criterion for the adoption of decisions for treatments. However, because decision makers are often risk averse, the uncertainty that accompanies a decision problem may influence adoption decisions. METHODS: VOI estimates were studied based on 2 alternate decision making criteria: 1) maximum expected value and 2) 95% credible intervals. These criteria were applied to a probabilistic minimal lifetime model of incident cardiovascular events and mortality among target patients comparing 2 daily doses of aspirin (81 mg and 325 mg). Model parameters were based on literature reviews and data analyses. RESULTS: Expected life-years under 81 v. 325 mg of aspirin were estimated to be 14.86 (SE, 0.10) and 14.72 (0.31) respectively, with a difference of 0.14 (0.29). The probability that 81 mg was optimal was estimated to be 67%. Under Decision Criterion 1, EVIM and EVPI were about 233-thousand and 411-thousand years, respectively. Under Criterion 2, EVIM was undefined, as there remains ambiguity about what to implement. Consequently, EVPI becomes the entire 644-thousand years. Also, under Criterion 1, EVSI reaches an asymptote at a sample size of 10,000 per arm, with minimal gains in value beyond a 5,000 person per arm trial. With Criterion 2, a sample size of 10,000 per arm or higher is substantially more valuable than lower sample sizes. CONCLUSION: Alternative decision criteria for treatment adoption change the VOI. Decision criteria should be justified for VOI analyses. If multiple criteria may be relevant, analysts should complete VOI estimates using multiple criteria.
BACKGROUND: In value of information (VOI) calculations, such as the expected value of perfect information (EVPI), partial perfect information (EVPPI), sample information (EVSI) or implementation (EVIM), the maximum expected value criterion defines the decision making criterion for the adoption of decisions for treatments. However, because decision makers are often risk averse, the uncertainty that accompanies a decision problem may influence adoption decisions. METHODS: VOI estimates were studied based on 2 alternate decision making criteria: 1) maximum expected value and 2) 95% credible intervals. These criteria were applied to a probabilistic minimal lifetime model of incident cardiovascular events and mortality among target patients comparing 2 daily doses of aspirin (81 mg and 325 mg). Model parameters were based on literature reviews and data analyses. RESULTS: Expected life-years under 81 v. 325 mg of aspirin were estimated to be 14.86 (SE, 0.10) and 14.72 (0.31) respectively, with a difference of 0.14 (0.29). The probability that 81 mg was optimal was estimated to be 67%. Under Decision Criterion 1, EVIM and EVPI were about 233-thousand and 411-thousand years, respectively. Under Criterion 2, EVIM was undefined, as there remains ambiguity about what to implement. Consequently, EVPI becomes the entire 644-thousand years. Also, under Criterion 1, EVSI reaches an asymptote at a sample size of 10,000 per arm, with minimal gains in value beyond a 5,000 person per arm trial. With Criterion 2, a sample size of 10,000 per arm or higher is substantially more valuable than lower sample sizes. CONCLUSION: Alternative decision criteria for treatment adoption change the VOI. Decision criteria should be justified for VOI analyses. If multiple criteria may be relevant, analysts should complete VOI estimates using multiple criteria.
Entities:
Keywords:
aspirin; cardiovascular; decision criterion; value of information
Authors: David D Kim; Gregory F Guzauskas; Caroline S Bennette; Anirban Basu; David L Veenstra; Scott D Ramsey; Josh J Carlson Journal: Pharmacoeconomics Date: 2020-02 Impact factor: 4.981