|
Activity Number:
|
29
- Statistical Issues Specific to Therapeutic Areas, Power and Sample Size Calculations, and Trial Monitoring
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Sunday, August 8, 2021 : 1:30 PM to 3:20 PM
|
|
Sponsor:
|
Biopharmaceutical Section
|
|
Abstract #318171
|
|
|
Title:
|
Optimizing Precision and Power in COVID-19 Trials by Covariate Adjustment
|
|
Author(s):
|
Nicholas Williams* and Ivan Diaz and Michael Rosenblum
|
|
Companies:
|
Weill Cornell Medicine and Weill Cornell Medical College and Johns Hopkins University
|
|
Keywords:
|
Covariate adjustment;
Doubly robust;
Variable selection;
Randomized trial;
Coronavirus
|
|
Abstract:
|
The rapid finding of effective therapeutics for COVID-19 requires the efficient use of available resources in clinical trials. The use of covariate adjustment may yield statistical estimates with smaller error and may result in a reduction in the number of participants and resources required to draw futility or efficacy conclusions. A key question for covariate adjustment in randomized studies is how to choose an adjustment set that yields the largest efficiency gains. We conducted a simulation study to evaluate the performance of different variable selection methods (L1 penalization, Random Forests, XGBoost, and Multivariate Adaptive Regression Splines) for time-to-event and ordinal outcomes in COVID-19 trials using longitudinal data from over 1,500 patients hospitalized with COVID-19 at Weill Cornell Medicine New York Presbyterian Hospital. We show that a fully adjusted estimator with L1 penalization maintains type-1 error control and is more efficient than an unadjusted estimator across the sample sizes tested. We also show that when covariates are not prognostic of the outcome, L1 penalization remains as efficient as the unadjusted estimator.
|
Authors who are presenting talks have a * after their name.
