OBJECTIVES: A previous trial-based Cost-Minimization Analysis (CMA) of office or clinic-based methadone (MO and MC) and office-based buprenorphine (BO) to treat opioid use disorder showed MC as optimal. Our cost-effectiveness analysis (CEA) used the CMA data, incorporating effectiveness measures that were not statistically different. Optimal treatment changed to MO and decision uncertainty (based on a probabilistic sensitivity analysis - PSA) was substantial compared to that in the CMA. In a value of information (VOI) analysis, we quantified expected value of partial perfect information (EVPPI) and identified an appropriate sample size for a trial to reduce uncertainty in the EVPPI parameters.

METHODS: We developed an Excel® model and explored effectiveness and cost parameters in groups. This required sampling values for our grouped parameters of interest (POI), in which one set of POI were held constant while we performed a PSA with the remaining parameters. Then a new set of POI were sampled with another PSA done conditional on those values. The process was repeated for many draws of the POI. We then explored the expected value of sample information (EVSI) for POI with significant EVPPI. Expected Net Gain from Sampling (ENGS) was estimated by setting marginal cost of an additional patient equal to the expected marginal benefit from a study of that respective size.

RESULTS: The only parameters with substantial EVPPI related to effectiveness. We estimated that the study sample size per treatment arm that would efficiently reduce the uncertainty generated from the effectiveness inputs to be an n of 225.

CONCLUSIONS: The literature provides several examples for estimating EVPPI, EVSI, and ENGS, making them fairly straightforward to implement. A potential barrier is computing time (substantial using Excel). Implementation of these VOI methods more generally can provide important information to clarify the potential for resolution of uncertainties in health economic models.