Mining Deviations from Patient Care Pathways via Electronic Medical Record System Audits

  • Authors:

  • He Zhang;Sanjay Mehotra;David Liebovitz;Carl A. Gunter;Bradley Malin

  • Affiliations:

  • Northwestern University;Northwestern University;Northwestern University;University of Illinois at Urbana-Champaign;Vanderbilt University

  • Venue:
  • ACM Transactions on Management Information Systems (TMIS) - Special Issue on Informatics for Smart Health and Wellbeing
  • Year:
  • 2013

Quantified Score

Hi-index 0.00

Visualization

Abstract

In electronic medical record (EMR) systems, administrators often provide EMR users with broad access privileges, which may leave the system vulnerable to misuse and abuse. Given that patient care is based on a coordinated workflow, we hypothesize that care pathways can be represented as the progression of a patient through a system and introduce a strategy to model the patient’s flow as a sequence of accesses defined over a graph. Elements in the sequence correspond to features associated with the access transaction (e.g., reason for access). Based on this motivation, we model patterns of patient record usage, which may indicate deviations from care workflows. We evaluate our approach using several months of data from a large academic medical center. Empirical results show that this framework finds a small portion of accesses constitute outliers from such flows. We also observe that the violation patterns deviate for different types of medical services. Analysis of our results suggests greater deviation from normal access patterns by nonclinical users. We simulate anomalies in the context of real accesses to illustrate the efficiency of the proposed method for different medical services. As an illustration of the capabilities of our method, it was observed that the area under the receiver operating characteristic (ROC) curve for the Pediatrics service was found to be 0.9166. The results suggest that our approach is competitive with, and often better than, the existing state-of-the-art in its outlier detection performance. At the same time, our method is more efficient, by orders of magnitude, than previous approaches, allowing for detection of thousands of accesses in seconds.