Using K-Means Clustering to Identify Physician Clusters by Electronic Health Record Burden and Efficiency

Abstract

Objectives: Electronic health records (EHRs) have transformed the way modern medicine is practiced, but they remain a major source of documentation burden among physicians. This study aims to use data from Signal, a tool provided by the Epic EHR, to analyze physician metadata in the Montefiore Health System via cluster analysis to assess EHR burden and efficiency. Methods: Data were obtained for a one-month period (July 2020) representing a return to normal operation post-telemedicine implementation. Six metrics from Signal were used to phenotype physicians: time on unscheduled days, pajama time, time outside of 7 AM to 7 PM, turnaround time, proficiency score, and visits closed the same day. k-Means clustering was employed to group physicians, and the clusters were assessed overall and by sex and specialty. Results: Our results demonstrate the partitioning of physicians into a higher-efficiency, lower-time outside of scheduled hours (TOSH) cluster and a lower-efficiency, higher-TOSH cluster even when stratified by sex and specialty. Intra-cluster comparisons showed general homogeneity of physician metrics with the exception of the higher-efficiency, lower-TOSH cluster when stratified by sex. Conclusions: Taken together, the clusters uniquely reflect the EHR efficiency-burden of the Montefiore Health System. Applying k-means clustering to readily available EHR data allows for a scalable, efficient, and adaptable approach of assessing physician EHR burden and efficiency, allowing health systems to examine documentation trends and target wellness interventions.

Keywords: COVID-19; cluster analysis; documentation burden; electronic health records and systems; machine learning; telemedicine.