Chapter 20

Compliance and Survival Analysis

Els Goetghebeur

20.1 Compliance: Cause and Effect

Today, new treatments must prove their worth in comparative (double blind) randomized clinical trials, the gold standard design for causal inference. With noninformatively right-censored survival outcomes, a typical robust intention-to-treat analysis compares groups as randomized using the popular (weighted) logrank test. Accompanying Kaplan–Meier, curves describe nonparametrically how survival chances differ between arms. A one-parameter summary of the contrast follows from a semiparametric Cox proportional hazards (PH) model Accelerated Failure-Time Model [6].

In general, and especially with long-term treatments, patients tend to deviate from their prescribed treatment regime. Varying patterns of observed exposure relative to the assigned are called “compliance (levels)” and recognized as a likely source of variation in treatment effect. Because deviations from prescribed regimes occur naturally in clinical practice, it is wise to learn about them within the trial context rather than restrict the study population to perfect compilers, an atypical and sometimes small and unobtainable subset of the future patient horizon [12].

Treatments that are stopped or switched or are less dramatic lapses in dosing happen in response to a given assignment. Different exposure patterns between treatment arms therefore point to (perceived) differences following alternative assignments. Studying compliance ...

Get Methods and Applications of Statistics in Clinical Trials, Volume 1: Concepts, Principles, Trials, and Designs now with the O’Reilly learning platform.

O’Reilly members experience books, live events, courses curated by job role, and more from O’Reilly and nearly 200 top publishers.