Inference on treatment effect modification by biomarker response in a three-phase sampling design

Area of Science:

• Biostatistics
• Clinical Trial Methodology
• Epidemiology

Background:

• Evaluating treatment effects on clinical endpoints across subgroups defined by intermediate outcomes is crucial in randomized trials.
• Principal surrogates and marginal causal effect predictiveness (mCEP) curves are key estimands.
• Previous methods for mCEP curve estimation are inefficient with complex three-phase sampling designs, as seen in dengue vaccine trials.

Purpose of the Study:

• To develop a novel, efficient, and robust method for estimating the mCEP curve in three-phase sampling designs.
• To avoid restrictive modeling assumptions like "placebo structural risk."
• To provide statistical inference tools, including confidence intervals and hypothesis testing, for the mCEP curve.

Main Methods:

• Proposed a novel approach for mCEP curve estimation in three-phase sampling designs.
• Utilized non-parametric kernel smoothing for biomarker density estimation to enhance robustness.
• Developed bootstrap-based procedures for pointwise and simultaneous confidence intervals and hypothesis testing.
• Investigated finite-sample properties and compared with an alternative method.

Main Results:

• The novel method demonstrated improved efficiency and robustness compared to existing approaches under three-phase sampling.
• The proposed bootstrap procedures provide reliable confidence intervals and hypothesis testing for mCEP curves.
• The methods were successfully applied to Phase 3 dengue vaccine trial data.

Conclusions:

• The developed method offers a significant advancement for analyzing principal surrogates and mCEP curves in complex clinical trial sampling designs.
• This approach enhances the evaluation of treatment effects in settings like dengue vaccine efficacy studies.
• The findings support more accurate and reliable interpretation of intermediate biomarker data in clinical trials.