2020/06/12 by Eric J. Oh, Oh, Eric J., Bryan E. Shepherd +5 · 1 citation
Mathematics · #Statistical Methods and Inference #Statistical Methods and Bayesian Inference #Advanced Causal Inference Techniques
paper · pdf · doi:10.48550/arxiv.2006.07480
Biomedical studies that use electronic health records (EHR) data for\ninference are often subject to bias due to measurement error. The measurement\nerror present in EHR data is typically complex, consisting of errors of unknown\nfunctional form in covariates and the outcome, which can be dependent. To\naddress the bias resulting from such errors, generalized raking has recently\nbeen proposed as a robust method that yields consistent estimates without the\nneed to model the error structure. We provide rationale for why these\npreviously proposed raking estimators can be expected to be inefficient in\nfailure-time outcome settings involving misclassification of the event\nindicator. We propose raking estimators that utilize multiple imputation, to\nimpute either the target variables or auxiliary variables, to improve the\nefficiency. We also consider outcome-dependent sampling designs and investigate\ntheir impact on the efficiency of the raking estimators, either with or without\nmultiple imputation. We present an extensive numerical study to examine the\nperformance of the proposed estimators across various measurement error\nsettings. We then apply the proposed methods to our motivating setting, in\nwhich we seek to analyze HIV outcomes in an observational cohort with\nelectronic health records data from the Vanderbilt Comprehensive Care Clinic.\n