Online Program

Return to main conference page
Wednesday, September 25
Wed, Sep 25, 10:45 AM - 12:00 PM
Thurgood Marshall East
Advances in Statistical Methodology Motivated by the Estimand Framework

A General Framework for Treatment Effect Estimators Considering Patient Adherence (300939)

*Yongming Qu, Eli Lilly and Company 
Junxiang Luo, Sanofi 
Haoda Fu, Eli Lilly and Company 
Stephen J. Ruberg, Analytix Thinking, LCC 

Keywords: Counterfactual effect, adherence causal estimator, inverse probability weighting, marginal structure models, mixed-effect model with repeated measures, tripartite estimands.

Randomized controlled trials remain a gold standard in evaluating the efficacy and safety of a new treatment. Ideally, patients adhere to their treatments for the duration of the study, and the resulting data can be analyzed unambiguously for efficacy and safety outcomes. However, some patients may discontinue the study treatment due to intercurrent events, which leaves missing observations or observations that do not reflect the randomly assigned treatment. Frequently, an intent-to-treat analysis (or a modification thereof) is done to estimate the treatment effect for all randomized patients regardless of the occurrence of intercurrent events. Alternatively, clinicians may be more interested in understanding the efficacy and safety for those who can adhere to the study treatment. The naive per-protocol analysis may provide a biased estimate for the treatment difference because the observed adherence populations may not be comparable between two treatments. In this article, we propose two methods for estimation of the treatment difference for those who can adhere to one or both treatments based on the counterfactual framework. Theoretic derivations and a simulation study show the proposed methods provide consistent estimators for the treatment difference for the adherent population of interest. A real data example comparing two basal insulins for patients with type-1 diabetes is provided using the proposed methods.