Statistical inference for association studies using electronic health records: handling both selection bias and outcome misclassification

Abstract

Health research using electronic health records (EHR) has gained popularity, but misclassification of EHR-derived disease status and lack of representativeness of the study sample can result in substantial bias in effect estimates and can impact power and type I error. In this paper, we develop new strategies for handling disease status misclassification and selection bias in EHR-based association studies. We first focus on each type of bias separately. For misclassification, we propose three novel likelihood-based bias correction strategies. A distinguishing feature of the EHR setting is that misclassification may be related to patientspecific factors, and the proposed methods leverage data in the EHR to estimate misclassification rates without gold standard labels. For addressing selection bias, we describe how calibration and inverse probability weighting methods from the survey sampling literature can be extended and applied to the EHR setting.Addressing misclassification and selection biases simultaneously is a more challenging problem than dealing with each on its own, and we propose several new strategies to address this situation. For all methods proposed, we derive valid standard errors and provide software for implementation. We provide a new suite of statistical estimation and inference strategies for addressing misclassification and selection bias simultaneously that is tailored to problems arising in EHR data analysis. We apply these methods to data from The Michigan Genomics Initiative (MGI), a longitudinal EHR-linked biorepository.

EXTENDING THE SUSCEPTIBLE-EXPOSED-INFECTED-REMOVED (SEIR) MODEL TO HANDLE THE HIGHfaLSEnEGATIVE RATE ANDsyMPTOM-BASED ADMINISTRATION OF COVID-19 DIAGNOSTIC TESTS:SEIR-fansy

“…The sample tested is, therefore, not representative of the population intended to be analyzed, and predicted case-counts and estimated parameters can deviate from the truth. While it may be possible to estimate the underlying selection model or run sensitivity analyses, it is difficult to prove that bias has been reduced or eliminated [10,5]. In the context of COVID-19 testing, such biases will continue to impact estimates of disease prevalence and the effective reproduction number until a random sample of the population is tested and/or testing becomes abundantly available.…”

Bias reduction and inference for electronic health record data under selection and phenotype misclassification: three case studies

“…External summary statistics or data are then used to address selection bias through weighting. Beesley and Mukherjee (2020) demonstrates good bias reduction and inferential performance of these methods when variables related to selection (collectively, denoted W ) and phenotype misclassification (denoted X) are known and observed. In reality, drivers of selection and misclassification may not be known, and known drivers may not always be observed (e.g., income, residential information, access to health care).…”

“…However, these works do not address how to account for both sources of bias in a single data analysis. Recently, Beesley and Mukherjee (2020) proposed novel strategies for addressing these two sources of bias simultaneously. The work by Beesley and Mukherjee has several new features.…”

“…In this paper, we explore how the methods proposed in Beesley and Mukherjee (2020) can be applied in practical EHR data analysis through three case studies. In doing so, we demonstrate the potential for bias reduction in practice and highlight some limitations.…”

“…In Section 2, we introduce the two observational databases with individual-level EHR data used for our analysis. Section 3 provides an overview of the bias-correction strategies proposed in Beesley and Mukherjee (2020). In Sections 4-6, we apply these methods to obtain corrected point estimates and standard errors for the three examples.…”