Electronic health record problem lists: accurate enough for risk adjustment?

Area of Science:

• Health Informatics
• Clinical Epidemiology
• Medical Data Analysis

Background:

• Electronic Health Records (EHRs) are increasingly used for clinical research.
• Accurate comorbidity data is essential for patient risk stratification and case-mix adjustment.
• Problem lists within EHRs are a potential source for comorbidity information.

Purpose of the Study:

• To assess the accuracy of comorbidity information extracted from EHR problem lists.
• To compare EHR problem list data with manual review of EHR free-text notes.
• To evaluate the predictive value of EHR-derived comorbidity scores for long-term mortality.

Main Methods:

• Retrospective cohort study of 1596 prostate cancer patients (1998-2004).
• Comparison of EHR problem list comorbidity assessment against manual review of EHR free-text notes for sensitivity and specificity.
• Evaluation of Charlson Comorbidity Index (CCI) scores derived from both methods in predicting long-term mortality.

Main Results:

• EHR problem lists demonstrated poor sensitivity for major comorbidities (e.g., myocardial infarction 8%, diabetes 46%).
• Specificity for identifying comorbidities from EHR problem lists exceeded 94%.
• Free-text-based CCI scores predicted long-term mortality, while EHR problem list-based scores did not.

Conclusions:

• EHR problem list data is often inaccurate for comorbidity assessment.
• Inaccurate comorbidity data can misrepresent patient case mix.
• Validation of EHR problem list data is necessary before its use in risk adjustment.