Calibrating Validation Samples When Correcting for Measurement Error in Intervention Study Outcomes

• 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): Benjamin Ackerman* and Elizabeth A Stuart and Juned Siddique
• Companies: Johns Hopkins Bloomberg School of Public Health and Johns Hopkins Bloomberg School of Public Health and Feinberg School of Medicine, Northwestern University
• Keywords: measurement error; causal inference

Abstract

Many lifestyle intervention trials depend on collecting self-reported outcomes, like dietary intake, to assess the intervention’s effectiveness. Self-reported outcome measures are subject to measurement error, impacting treatment effect estimation. Methods exist to correct for measurement error by using external validation studies to model the error structure, but they require a transportability assumption that the error structure of the outcome in the validation sample matches that of the outcome in the intervention study. There is growing concern over the performance of these methods when the validation and intervention studies differ on pre-treatment covariates related to treatment effect. We formalize cases where such covariate imbalance introduces bias when modeling the measurement error using validation data, and evaluate the relationship between the two through simulation. Using data from PREMIER, a sodium-reduction intervention study, and OPEN, a validation study measuring both self-reported diet and urinary biomarkers, we implement propensity score-type methods to account for study covariate differences and improve upon the outcome measurement error correction.