Case contamination in electronic health records based case-control studies.

Area of Science:

• Biomedical Informatics
• Epidemiology
• Statistical Genetics

Background:

• Electronic Health Records (EHRs) offer rich clinical data for phenotype derivation.
• Identifying precise case and control groups in EHRs is challenging due to data complexity and lack of a priori hypothesis.
• Candidate case pools in EHR studies often contain non-case patients, leading to potential bias.

Purpose of the Study:

• To develop a bias-correction method for case identification in EHR-based case-control studies.
• To address the challenge of case contamination in large, EHR-derived case pools.
• To improve the efficiency and accuracy of phenotype identification from EHR data.

Main Methods:

• Proposed an estimating equation approach for bias correction.
• Studied the large sample properties of the proposed method.
• Evaluated performance through extensive simulation studies and a pilot study of aortic stenosis in the Penn Medicine EHR.

Main Results:

• The developed method effectively corrects for bias introduced by case contamination in EHR studies.
• The approach allows for the use of enlarged, albeit contaminated, case pools.
• Demonstrated improved estimation of association parameters in the presence of case contamination.

Conclusions:

• The proposed bias-correction method enhances the utility of EHR data for case-control studies.
• This approach facilitates more efficient research by accommodating larger, imperfectly defined case groups.
• Accurate phenotype identification from EHRs is crucial for reliable epidemiological research.