An Empirical Study for Impacts of Measurement Errors on EHR based Association Studies

Area of Science:

• Biomedical Informatics
• Statistical Genetics
• Health Services Research

Background:

• Electronic Health Records (EHR) systems are widely adopted in US hospitals, offering vast potential for research.
• However, challenges in clinical documentation and data quality, including information bias from measurement errors, hinder EHR data analysis.
• Understanding the impact of these biases is crucial for reliable EHR-based association studies.

Purpose of the Study:

• To empirically quantify the impact of information bias on association studies utilizing EHR data.
• To assess the loss of statistical power resulting from misclassifications in case ascertainment and covariate extraction within EHR data.

Main Methods:

• Conducted simulation studies designed using data characteristics from the Electronic Medical Records and Genomics (eMERGE) Network.
• Quantified power loss based on varying levels of misclassification rates, disease prevalence, and covariate frequencies.
• Evaluated the effects of measurement errors in outcome and covariate data extraction.

Main Results:

• Simulation results demonstrated a quantifiable loss of power in EHR-based association studies due to misclassification.
• The extent of power loss was shown to be dependent on misclassification rates, disease prevalence, and covariate frequencies.
• Measurement errors in covariate status extraction significantly impacted study power.

Conclusions:

• Information bias, stemming from misclassification and measurement errors in EHR data, can substantially reduce statistical power in association studies.
• These findings highlight the need for careful consideration of data quality and potential biases when conducting EHR-based research.
• Investigators can use these empirical findings to better anticipate and account for power loss under various conditions.