658 – Statistical Learning in Various Areas of Clinical Trials
A Versatile Multi-Domain Test with Adaptive Weighting
Yang Zhao
Sanofi R&D
Stephen Lake
Sanofi R&D
The design of a clinical trial to investigate a treatment for a rare disease is often complicated by the multisystemic nature of the disease; no single endpoint can capture the spectrum of potential therapeutic benefits. Multi-domain outcomes which take into account patient heterogeneity of disease presentation through measurements of multiple symptom/functional domains are an attractive alternative to a single endpoint. To obtain the totality of evidence for treatment efficacy over endpoints from various disease progression domains, an extension of a test for equality of two survival distributions based on weighted differences of Kaplan-Meier curves [Uno et al (2015)] is proposed. The test is a weighted sum of domain-specific test statistics with weights selected adaptively via a data-driven algorithm. The null distribution of the test is constructed empirically through resampling. We used data from a clinical trial in a rare lysosomal storage disorder to illustrate the advantage of the combined testing procedure over the conventional methods. Simulations were conducted to demonstrate the statistical properties of the test and to compare to alternative methods.