EMR Alert Gives Reliable AKI Detection, Leads to Multicenter Clinical Study Grant

Investigators at the Center for Critical Care Nephrology have shown that an improved specificity for an electronic alert for acute kidney injury (AKI)—generated by the clinical decision support system (CDSS) imbedded in electronic health records of UPMC hospitals in western Pennsylvania—led to more accurate detecting of AKI. Their findings, “Creating a High-Specificity Acute Kidney Injury Detection System for Clinical and Research Applications,” are reported in the American Journal of Kidney Diseases.

“This highly specific AKI alert greatly reduces false positives thus helping to avoid EMR ‘alert-fatigue’ often faced by clinical staff who become desensitized to electronic alerts,” said John Kellum, MD, who is senior author of the paper and professor of Critical Care Medicine, Bioengineering, and Clinical & Translational Science. “For the last eight years, the AKI alert has been embedded in the UPMC electronic medical record and we’ve finessed the alert to give us very reliable AKI detection. We now look forward to a wider clinical study, together with the University of Florida.”

The AKI alert system forms the foundation of a new multicenter research grant that will develop and test more sophisticated alerts using artificial intelligence and pair them to pharmacist-led interventions. The study, “Multicenter Implementation of an Electronic Decision Support System for Drug-associated AKI (MEnD-AKI),” will be led by principal investigator Sandra Kane-Gill, DPharm, and co-investigator Ragavan Murugan, MD, MS, at the Center for Critical Care Nephrology, in collaboration with co-principal investigator Azra Bihorac, MD, MS, at the University of Florida. The clinical study is funded by the National Institute of Diabetes and Digestive and Kidney Diseases/National Institutes of Health.

AKI Alert Development


In a multicenter, sequential period analysis of 528,108 patients, Pitt investigators found that a CDSS resulted in small but sustained decreases in hospital mortality (0.8% absolute decrease), length of stay (0.3 days), and dialysis rates (2.7% absolute decrease) for patients with AKI without affecting outcomes for patients without AKI.


A follow-up study by the same team found sustained effects of the alert through March of 2018. In the latest report, the authors show that the alert can be improved to increase specificity to 92.7% while still maintaining 100% sensitivity. These improvements provide better agreement with clinically adjudicated AKI and should therefore improve performance of CDSS as well as epidemiologic studies involving electronic health record data. The AKI alert logic can be found here as a PDF.