El-Nahal W, Grader-Beck T, Gebo K, Holmes E, Herne K, Moore R, Thompson D, Berry S. Designing an electronic medical record alert to identify hospitalised patients with HIV: successes and challenges. BMJ Health Care Inform 2022;29:e100521. doi:10.1136/bmjhci-2021-100521
An electronic medical record (EMR) alert system was developed to use readily available data elements to accurately identify hospitalized people with HIV. Authors described the design and implementation of the EMR alert and methods to evaluate its accuracy for identifying people with HIV. Over 24 months, the EMR alert was used to notify an intervention team and data abstraction team in real time about admissions of people with HIV. Sensitivity was assessed by comparing the machine-learning alert system to manual chart reviews. Positive predictive value (probability that a patient with a positive test result actually has the disease), was assessed by false positives identified in chart review (not having HIV despite alert triggering). Results demonstrated high sensitivity (sensitivity=100%, 95% CI 82-100%) and good predictive value (84%, 95% CI 82-86%). A combination of data (diagnosis, prescriptions, and lab orders) in the EMR alert system achieved high sensitivity and positive predictive value in identifying people with HIV. ICD Code diagnoses were the strongest contributors to predictive value, compared to the other criteria. Use of data-driven alerts in electronic health record systems can facilitate the deployment of multidisciplinary teams for medication review, education, case management, and outpatient linkage to follow-up.