Comparison of EHR-based diagnosis documentation locations to a gold standard for risk stratification in patients with multiple chronic conditions.

Objective: To measure variation among four different Electronic Health Record (EHR) system documentation locations versus 'gold standard' manual chart review for risk stratification in patients with multiple chronic illnesses.

Methods: Adults seen in primary care with EHR evidence of at least one of 13 conditions were included. EHRs were manually reviewed to determine presence of active diagnoses, and risk scores were calculated using three different methodologies and five EHR documentation locations. Claims data were used to assess cost and utilization for the following year. Descriptive and diagnostic statistics were calculated for each EHR location. Criterion validity testing compared the gold standard verified diagnoses versus other EHR locations and risk scores in predicting future cost and utilization.

Results: Nine hundred patients had 2,179 probable diagnoses. About 70% of the diagnoses from the EHR were verified by gold standard. For a subset of patients having baseline and prediction year data (n=750), modeling showed that the gold standard was the best predictor of outcomes on average for a subset of patients that had these data. However, combining all data sources together had nearly equivalent performance for prediction as the gold standard.

Conclusions: EHR data locations were inaccurate 30% of the time, leading to improvement in overall modeling from a gold standard from chart review for individual diagnoses. However, the impact on identification of the highest risk patients was minor, and combining data from different EHR locations was equivalent to gold standard performance. The reviewer's ability to identify a diagnosis as correct was influenced by a variety of factors, including completeness, temporality, and perceived accuracy of chart data.