Incomplete comparisons between the predictive power of data from administrative claims and electronic health records

The recent paper by Kharrazi and colleagues suggested administrative claims data predicted future acute care utilization and costs better than the diagnostic and pharmacy codes found in the corresponding electronic health record (EHR) encounters. While these results are stimulating given the growing enthusiasm for EHR-based research and diagnostic algorithms, we are concerned that the authors’ conclusion that “…EHR based models may underperform compared with administrative claims” is founded on an incomplete comparison.

The authors rightly acknowledge that future EHR work should examine other data sources within the EHR, including laboratory values, vital signs, and free-text sources, among others. This latter group, including the text of clinical encounter notes, represents a major source of rich and nuanced data that has already been demonstrated to vastly improve case identification across a range of diagnostic categories. While such data sources may not have been readily available to the investigators, their exclusion prevents a complete or definitive comparison between administrative and EHR data for purposes of risk stratification and prediction of utilization patterns.

National Heart Lung and Blood Institue (NHLBI)