Evaluation of Documentation Patterns of Trainees and Supervising Physicians Using Data Mining

摘要

Introduction The use of electronic health records (EHRs) in the delivery of patient care is steadily increasing; institutions are making significant investments in EHRs, and payers are providing incentives to promote the use of health information technology.1 This is leading to the broad implementation of EHR systems and the phasing out of paper records in private practice settings, urgent care centers, and hospitals. The EHR has been shown to improve delivery of patient care, increase productivity, and boost billing efficiency.2–4 In addition, EHR data have been processed using data mining and natural language methods to answer patient care–related questions5,6 and to identify system pitfalls in health care delivery and graduate medical education.7,8 In our study, we used a new EHR data mining methodology to objectively measure resident and attending note-entry practices at a large academic medical center.;Materials and Methods To collect the data, an automatic system was used to feed inpatient clinical data (patient encounter notes) from an urban academic hospital's EHR system onto a secure server belonging to a 600-physician university-based multispecialty practice group involved in training medical and surgical residents. An EHR has been used at the institution for more than a decade, and paper patient records have been phased out. We conducted a retrospective observational study of deidentified patient notes during 7 continuous months in 2011 and early 2012. The deidentification included excluding and scrambling protected health information, which was not accessible to any of the investigators or other personnel involved in the study. The study received approval from Wayne State University's Internal Review Board. Notes were automatically analyzed by a commercially available data mining and modeling tool, Healthcare Smartgrid (ProcessProxy Corp). This is a patented method of process mining known as Process Arbitrage, which is normally used by the university physician practice for quality improvement purposes.9,10 The data mining tools allowed for the automatic extraction of fields of interest, which included the month of the encounters, the time of resident note entry, the time of required attending attestations (in the form of cosignatures and any needed corrections), and the medical specialty involved. We divided the 24-hour day into six 4-hour time periods and calculated the frequency of resident notes in each period. We also categorized the time interval between note completion and attending physician completion of the attestation into 5 categories—within 24?hours, 24?hours to 1?week, 1?week to 1?month, 1?month to 1?year, and more than 1?year—and calculated the number of attestations in each category. Finally, we recorded the month of the academic year during which the note was created to assess for any change in pattern due to familiarity and efficiency that may occur later on in the year. We used the Pearson χ2 test to perform comparisons between categorical variables using the statistical package SAS version 9.3 (SAS Institute Inc).;Results The automated system identified 26?802 patient encounter notes in the form of progress notes and consultation reports entered between August 2011 and February 2012. We excluded 2015 notes with missing data, resulting in 24?787 notes that were analyzed. Residents created notes throughout the 24-hour day. However, most notes (33%, 8178 of 24?787) were entered between noon and 4 pm, and 31% (7718 of 24?787) were entered between 8 am and noon (figure?1). The time residents placed notes into the EHR did not significantly change over the academic year for the 9 specialties with the largest number of EHR entries (figure?2). View larger version (27K) FIGURE 1Total Number of Notes Entered by Residents During the 6 Designated Time Periods;Discussion We used objective measures to assess resident and attending documentation practices in an inpatient setting using data mining methodology. We