Causal models for randomized trials with two active treatments and continuous compliance.

Area of Science:

• Biostatistics
• Clinical Trial Methodology
• Causal Inference

Background:

• Clinical trials often involve continuous measures of treatment compliance.
• Estimating causal effects with continuous compliance, especially with two active treatments, presents significant challenges.
• Existing methods may struggle to accurately assess treatment efficacy under varying levels of patient adherence.

Purpose of the Study:

• To propose a novel structural model for principal causal effects with continuous compliance.
• To develop identifiable marginal compliance models within study arms.
• To provide a framework for sensitivity analysis regarding unidentifiable parameters in compliance modeling.

Main Methods:

• A structural model for principal causal effects was developed.
• Marginal compliance models were specified and shown to be identifiable.
• A Gaussian copula model was employed for the joint distribution of observed and counterfactual compliance.
• A sensitivity analysis was conducted by varying the dependence parameter of the copula model.

Main Results:

• The proposed methodology allows for the estimation of causal effects in the presence of continuous compliance measures.
• The analysis of a smoking cessation trial demonstrated the practical application of the model.
• Causal effects were estimated at specific compliance levels and within subpopulations exhibiting similar compliance behaviors.

Conclusions:

• The developed statistical framework effectively addresses the challenges of estimating causal effects with continuous compliance in clinical trials.
• The methodology provides valuable insights into treatment effectiveness by accounting for varying patient adherence.
• The approach is applicable to real-world clinical trial data, as shown in the smoking cessation trial example.