From Heterogeneous Treatment Effects to Treatment Policies: A Decision Focused Framework for Treatment Personalization in Drug Development

Clinical trials often report evidence that treatment effects vary across patients, but such evidence does not show whether an individualized treatment policy will improve decisions. We propose a two-stage framework that separates treatment effect assessment from policy evaluation. Stage 1 addresses a prespecified treatment effect claim on a defined causal scale, using an analysis and multiplicity strategy suited to that claim. Stage 2 separately estimates treatment effect scores and evaluates candidate policies against a defined comparator through ranking, policy value, incremental value, validation, and empirical surrogate benefit and harm metrics. A Stage 1 result may motivate Stage 2, but it does not validate a policy or determine the inferential status of Stage 2. Simulations covering no effect, constant benefit without HTE, quantitative HTE, and qualitative HTE show that heterogeneity supports personalization mainly when it changes the preferred treatment. Additional simulations examine sample size, candidate covariate dimension, and learner complexity. An AIDS Clinical Trials Group (ACTG) 175 secondary analysis illustrates that population level HTE evidence can coexist with little improvement in held out policy value. The framework provides setting-specific guidance for matching methods, validation, and reporting to distinct treatment effect and policy claims.