Automated algorithms to detect and report complex notifiable diseases from electronic medical record data

Background:  Electronic medical records (EMR) can improve the completeness and timeliness of notifiable disease surveillance.  Detection of complex diseases such as tuberculosis or acute viral hepatitis, however, is elusive since these diagnoses require subjective clinical assessments and integration of multiple laboratory tests.  We report on the development of novel case definitions for automated notifiable disease surveillance from EMRs.

Methods:  The Electronic medical record Support for Public Health (ESP) system scans EMR data to detect and report notifiable conditions to the state health department (MMWR 2008;57:376).  Available data include vital signs, diagnostic codes, laboratory orders and results, and medication prescriptions.  Algorithm development involves:  1) translating existing CDC definitions into electronic terms, 2) applying the candidate algorithm to retrospective EMR data, 3) chart review of identified cases to determine positive predictive value (PPV) relative to existing CDC definition as reference standard, 4) comparison of electronically identified cases with health department case records to determine algorithm sensitivity, 5) algorithm refinement to maximize sensitivity and positive predictive value, 6) implementation of final algorithm in the live ESP system, 7) routine review of algorithm performance.

Results:  We have developed definitions for acute hepatitis A, B, and C, active tuberculosis, and pelvic inflammatory disease.  Application of case definitions within the ESP system spanning June 2006 to present has led to detection and reporting of 6 cases of acute hepatitis A (PPV 83%), 9 cases of acute hepatitis B (PPV 100%), 6 cases of acute hepatitis C (PPV 100%), 13 cases of active tuberculosis (PPV 62%), and 34 cases of pelvic inflammatory disease (PPV 97%).  Comparison with health department records of manually reported cases showed no cases missed by the electronic algorithms.

Conclusions: Electronic algorithms can detect complex conditions with high accuracy.  Standardization of electronic definitions will facilitate more complete, comparable, and meaningful national notifiable disease surveillance.