Causal Inference
Toward a Clearer Definition of Per-protocol Estimand When Emulating a Target Trial Haidong Lu* Haidong Lu
The adherence-adjusted per-protocol effect, which represents the effect of adhering to the treatment protocol, has received increasing attention among both trialists and observationalists employing observational data to emulate hypothetical target trials, as this estimand is considered more patient-centered compared to the intention-to-treat effect (i.e., the effect of the assigned treatment, regardless of adherence status). However, the definition of the per-protocol effect remains ambiguous, especially in observational studies designed to emulate a target trial. To enhance clarity in defining this estimand, we propose the following guidelines: 1) Distinguish the causal adherence-adjusted per-protocol effect from the non-causal per-protocol difference; 2) Utilize the ICH E9 (R1) addendum to provide a more precise definition of the per-protocol estimand, which is often mistakenly equated with the as-treated effect; 3) Clarify the feasibility of identifying or approximating treatment assignment in observational settings; 4) Establish a comprehensive study protocol that explicitly defines protocol deviations (e.g., including specifying those allowable deviations) before initiating a target trial emulation. Implementing these suggestions will sharpen the definition of the per-protocol estimand and improve transparency in reporting studies.