OBJECTIVES: For a payer’s perspective, entering an outcomes-based contract (OBC) can reduce risk in the face of uncertainty about treatment effectiveness. However, implementing OBC is often costly, and sustaining it may be even costlier. Little is known about the optimal duration of OBC.

METHODS: A central premise for terminating an OBC is whether uncertainty about the treatment effectiveness is resolved. The optimal length of an OBC can be reduced if payers are willing to learn from their own data during the execution of an OBC contract. We combine concepts from Value of Information methods and Bayesian designs to compare alternative designs of learning within the real-world system. Specifically, we compare 1) static approaches, where sample size are determined apriori, either through traditional sample size calculation or through Expected Value of Sample Information (EVSI) criterion, and 2) dynamic approaches, that follow adaptive allocation of patients to treatments following evolving posterior means or posterior variance. Using simulated scenarios, we compare the time to resolution of uncertainty under each of these mechanisms with 1000 Monte-Carlo runs.

RESULTS: The average time to uncertainty resolution was 1/20 to 1/70 for dynamic approaches compared to static approaches. Among the static approaches, EVSI approach required shorter time. Under the dynamic approaches, adaptive allocation based on posterior variance had the shortest time to uncertainty resolution.

CONCLUSIONS: Payers can shorten time to termination of OBC by engaging in active learning during the OBC period. Dynamic allocation of patients to alternative treatments are better approaches as it helps resolve decisional uncertainty faster than the uncertainty in the true comparative effect. We found that payers may want to steer patients to use the treatment that has the greatest uncertainty associated with it’s effectiveness during the OBC period. Further work, requiring application of methods to specific clinical scenarios will be most useful.