Key points are not available for this paper at this time.
PURPOSE: To inform graduate medical education (GME) outcomes at the individual resident level, this study sought a method for attributing care for individual patients to individual interns based on "footprints" in the electronic health record (EHR). METHOD: Primary interns caring for patients on an internal medicine inpatient service were recorded daily by five attending physicians of record at University of Cincinnati Medical Center in August 2017 and January 2018. These records were considered gold standard identification of primary interns. The following EHR variables were explored to determine representation of primary intern involvement in care: postgraduate year, progress note author, discharge summary author, physician order placement, and logging clicks in the patient record. These variables were turned into quantitative attributes (e.g., progress note author: yes/no), and informative attributes were selected and modeled using a decision tree algorithm. RESULTS: A total of 1,511 access records were generated; 116 were marked as having a primary intern assigned. All variables except discharge summary author displayed at least some level of importance in the models. The best model achieved 78.95% sensitivity, 97.61% specificity, and an area under the receiver-operator curve of approximately 91%. CONCLUSIONS: This study successfully predicted primary interns caring for patients on inpatient teams using EHR data with excellent model performance. This provides a foundation for attributing patients to primary interns for the purposes of determining patient diagnoses and complexity the interns see as well as supporting continuous quality improvement efforts in GME.
Schumacher et al. (Thu,) studied this question.