A Bayesian Approach for Handling Covariate Measurement Error When Estimating Population Treatment Effect

• Activity Number: 501 - Innovative Methods for Measurement Error Correction
• Type: Topic Contributed
• Date/Time: Wednesday, July 31, 2019 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Hwanhee Hong* and Juned Siddique and Elizabeth A Stuart
• Companies: and Feinberg School of Medicine, Northwestern University and Johns Hopkins Bloomberg School of Public Health

Abstract

Results of randomized controlled trials (RCTs) could be generalized to a target population for which we want to make decisions regarding treatment implementation after some calibration. However, methods for handling covariate measurement error when estimating population treatment effects have not been developed. Measurement error can be found in either RCT or target population data, and the degree of measurement error could be different between the two data sources because the data collection process could be different (e.g., more stringent measurement practices in the RCT leading to higher reliability in the RCT than in the target population). In this talk, we propose a flexible Bayesian approach for handling this situation. We consider three scenarios where an error-prone covariate exists in (1) both RCT and target population, (2) RCT only, and (3) target population only. We investigate these scenarios via extensive simulation studies. We apply our methods to a real data example to assess the population treatment effect of a program to reduce sodium intake on hypertension using the PREMIER study (RCT) and the INTERMAP study (target population).