Measuring Exposure to Incarceration Using the Electronic Health Record

Area of Science:

• Health Informatics
• Public Health
• Social Determinants of Health

Background:

• Electronic health records (EHRs) contain valuable health data, but extracting social determinants of health like incarceration is difficult.
• Incarceration history significantly impacts health outcomes and contributes to health disparities.

Purpose of the Study:

• To compare the accuracy of patient self-report with various EHR-based methods for identifying incarceration exposure.
• To evaluate the sensitivity and specificity of different data sources and techniques for detecting incarceration history within EHRs.

Main Methods:

• A validation study was conducted using data from the Veterans Aging Cohort Study (VACS).
• Methods included linking EHR data to state Department of Correction (DOC) and Centers for Medicare and Medicaid Services (CMS) administrative data.
• Additional methods involved using EHR-specific reentry service identifiers and natural language processing (NLP) on unstructured EHR text.

Main Results:

• Linking EHR to DOC data yielded a sensitivity of 2.5% and specificity of 100%.
• Linking EHR to CMS data showed a sensitivity of 7.9% and specificity of 99.3%.
• NLP demonstrated the highest sensitivity (63.5%) with a specificity of 95.9% for identifying incarceration exposure.

Conclusions:

• Natural language processing (NLP) tools offer a feasible and valid approach for identifying individuals with incarceration exposure within EHRs.
• Further research involving larger datasets and refined NLP methods could enhance the accuracy of detecting incarceration history in electronic health records.