Flexible Group Sequential Monitoring Methods That Incorporate Covariate Adjustment in Observational Studies

• Activity Number: 183
• Type: Contributed
• Date/Time: Monday, August 10, 2015 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Lu Bai* and Daniel Gillen
• Companies: UC Irvine and UC Irvine
• Keywords: Group sequential ; Monitoring ; Constrained boundaries ; Observational studies ; Alzheimer

Abstract

Group sequential designs are commonly used in clinical trials. Similar approaches could also be used in observational studies. Specifically, group sequential designs could be used to re-evaluate trial design parameters intermittently as new data are observed. While a variety of formula and simulation-based approaches that account for confounder adjustment have been proposed to assess observational design operating characteristics, information on the joint distribution of model covariates is generally scarce. To account for misspecification of design assumptions, we propose a sequential method to monitor observational studies within a generalized linear model framework. The method incorporates common stopping rules that allows interim decisions for or against specified hypotheses. A constrained boundaries approach is introduced to adjust the sample size at each interim analysis based on available population structure information in order to maintain both type I and type II error rates. Procedures incorporating nuisance parameter estimation are also developed and discussed. The method is assessed via simulation and illustrated using a study of Alzheimer's Disease risk factors.