OBJECTIVES: Expected value of sample information (EVSI) quantifies the expected benefits to decision makers of reducing parameter uncertainty through a proposed study. These measures can be used for research prioritisation and optimising study designs. Nevertheless, standard EVSI methods fail to account for heterogeneous treatment effects across subgroups, and the uncertainty in subgroup-specific effects cannot be leveraged to optimise the recruitment for the future trial. This work proposes a novel EVSI measure that incorporates conditional average treatment effects (CATE) and explores its application in optimising recruitment for future trials.

METHODS: We propose a new measure EVSI-CATE, that incorporates subgroup-specific effects, and demonstrate a potential data generation strategy. Given the varying precision in estimating subgroup-specific effects under existing data, we treat the subgroup composition of future trials as a simplex of parameters to be optimised for maximal information gain. We implement and optimise this approach using simulated data and further explore extensions to population-level EVSI-CATE and the expected net benefits of sampling.

RESULTS: In a simplified simulation study involving two subgroups and a fixed sample size for the future trial, the design informed by EVSI-CATE demonstrated a higher value of information. Future work will extend this analysis to more realistic scenarios with additional subgroups and employ Bayesian optimization to identify optimal subgroup composition.

CONCLUSIONS: We introduce EVSI-CATE to account for potential heterogeneous effects in EVSI and compare the expected net benefits of sampling with and without considering CATE. When exploring CATE does offer more value to the decision-maker, EVSI-CATE can further inform future trial design by determining the optimum subgroup composition.