Online Program Home
My Program

Abstract Details

Activity Number: 179 - Statistical Methods for Measurement Error and Missing Data in Covariates/Exposures
Type: Contributed
Date/Time: Monday, July 29, 2019 : 10:30 AM to 12:20 PM
Sponsor: Section on Statistics in Epidemiology
Abstract #305051
Title: Relative Risk Estimation Using Multiple Imputation with Logistic Regression and Discretization
Author(s): Jay Xu* and Thomas Belin
Companies: University of California, Los Angeles and UCLA
Keywords: Logistic Regression; Missing Data; Multiple Imputation; Relative Risk; Discretization; Propensity Score

In epidemiology and medicine, the relative risk is an important quantity of interest. In epidemiological and medical settings with incomplete data, failure to address the missing data can result in misleading conclusions. Sullivan et al. (2017, BMC Medical Research Methodology) considered a setting with a binary outcome subject to missingness generated from a log-binomial model where the relative risks were known. Considering estimation using multiple imputation with logistic regression to impute for binary variables, they found that using a mis-specified logistic regression model to impute binary variables can lead to substantial bias. We propose a novel multiple imputation method that uses logistic regression with discretization with respect to the propensity score. Using simulations, we demonstrate that our novel imputation method achieves substantial bias reduction relative to ordinary logistic regression. Using simulations, we compare our novel imputation method to ordinary logistic regression in estimating the relative risk and evaluate the methods with respect to coverage, bias, and interval length.

Authors who are presenting talks have a * after their name.

Back to the full JSM 2019 program