|
Activity Number:
|
73
- Alternative Designs and Related Topics
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Sunday, July 28, 2019 : 4:00 PM to 5:50 PM
|
|
Sponsor:
|
Biopharmaceutical Section
|
|
Abstract #304123
|
|
|
Title:
|
Two Novel Non-Parametric Methods for the Analysis of Stepped-Wedge Cluster Randomized Trials
|
|
Author(s):
|
Lee Kennedy-Shaffer* and Victor De Gruttola and Marc Lipsitch
|
|
Companies:
|
Harvard University and Harvard T.H. Chan School of Public Health and Harvard T.H. Chan School of Public Health
|
|
Keywords:
|
stepped wedge;
synthetic control;
cluster randomized trials;
non-parametric;
econometrics
|
|
Abstract:
|
Stepped-wedge cluster randomized trials (SWCRTs) have become increasingly popular as a method to evaluate interventions in cases where individual randomized trials and parallel-arm cluster randomized trials are not feasible or not acceptable to participants. As the settings of such trials grow more diverse, new methods for analyzing the results of SWCRTs are necessary to ensure unbiased effect estimates and adequate power. We propose two non-parametric estimators and inferential methods for SWCRTs, one based on the synthetic control method of causal inference commonly used in econometrics. This method avoids the need for assumptions about the data-generating process and improves power by exploiting similarities among subsets of clusters. We present properties of these two methods derived from simulations drawn from infectious disease settings and compare their performance to commonly-used parametric methods (mixed effects models) and non-parametric methods. Using these properties, we discuss settings for which each method may have the most desirable operating characteristics. Finally, we propose extensions that may account for treatment effects that vary over time.
|
Authors who are presenting talks have a * after their name.
