Kidney Week

Abstract: TH-PO1289

Multihospital Electronic Decision Support for Drug-Associated AKI (MEnD-AKI): A Prospective Cluster-Randomized Clinical Trial

Session Information

• Late-Breaking Research Posters
  October 22, 2026 | Location: Exhibit Hall A, Convention Center
  Abstract Time: 10:00 AM - 12:00 PM

Category: Acute Kidney Injury

• 102 AKI: Clinical, Outcomes, and Trials

Authors

• Kane-Gill, Sandra, University of Pittsburgh School of Pharmacy, Pittsburgh, Pennsylvania, United States

Sandra Kane-Gill, PharmD, MSc, RPh

• Kazarov, Christina L., University of Pittsburgh School of Pharmacy, Pittsburgh, Pennsylvania, United States

Christina L. Kazarov, PharmD

• Lukan, Caiden, University of Pittsburgh School of Pharmacy, Pittsburgh, Pennsylvania, United States

Caiden Lukan, PharmD

• Crouse, Britney A S, University of Pittsburgh School of Pharmacy, Pittsburgh, Pennsylvania, United States

Britney A S Crouse, PharmD, PhD

• Kennedy, Jason N., University of Pittsburgh School of Medicine, Pittsburgh, Pennsylvania, United States

Jason N. Kennedy, MS

• Ozrazgat-Baslanti, Tezcan, University of Florida College of Medicine, Gainesville, Florida, United States

Tezcan Ozrazgat-Baslanti, PhD

• Adiyeke, Esra, University of Florida College of Medicine, Gainesville, Florida, United States

Esra Adiyeke, PhD

• Murugan, Raghavan, University of Pittsburgh School of Medicine, Pittsburgh, Pennsylvania, United States

Raghavan Murugan, MD, MS

• Chang, Chung-Chou H., University of Pittsburgh, Pittsburgh, Pennsylvania, United States

Chung-Chou H. Chang, PhD

• Bihorac, Azra, University of Florida College of Medicine, Gainesville, Florida, United States

Azra Bihorac, MD, MS, FASN

• Kellum, John A., University of Pittsburgh School of Medicine, Pittsburgh, Pennsylvania, United States

John A. Kellum, MD

Group or Team Name

• MEnD-AKI Study Team

Background

Clinical decision support systems (CDSS) have improved AKI detection, but passive electronic alerts alone show only modest impact on outcomes. Enhancing CDSS with a predictive deep learning model and an active, pharmacist-led intervention may improve prevention and management of drug-associated AKI (D-AKI). Our aim was to evaluate the effect of a protocolized, tiered (Level A & B alerts) pharmacist-led intervention prompted by a predictive, deep learning-based CDSS delivered to physicians via telemedicine on reducing complications and progression of D-AKI in hospitalized non-ICU patients.

Methods

38 hospital service clusters in 8 hospitals were randomized to a pharmacist-led intervention prompted by machine learning-driven CDSS or an EMR-based passive alert (usual care) for patients with or at risk of AKI. Intervention occurred at the physician level; outcomes were assessed at the patient level using de-identified data. The primary endpoint was major adverse kidney events within 30 days (MAKE30). Secondary endpoints included AKI progression, AKI intensity, and nephrotoxin burden.

Results

Of 698 enrolled physicians, 480 (40.0% female) cared for a patient with a qualifying CDSS alert. Intention to treat (ITT) analysis included 5939 patients (3841 intervention, 2098 usual care). MAKE30 occurred in 50.8% of intervention patients vs. 42.9% of usual care patients (adjusted OR 1.05; 95% CI 0.82–1.35). Intervention patients were older and had more chronic kidney disease, congestive heart failure, diabetes, hypercholesterolemia, and hypertension, while usual care patients had more anemia, malignancy, and trauma (all p<0.001). No secondary outcome differed significantly between groups. Of the ITT alerts, 52.3% were non-actionable, and only 12% of Level B recommendations were delivered by phone (vs. email), despite phone being the intended primary route.

Conclusion

A pharmacist-led intervention prompted by a predictive, deep learning-based CDSS did not significantly reduce MAKE30 or any secondary outcome compared with usual care. Key limitations include reliance on asynchronous email over real-time phone delivery, a high proportion of non-actionable alerts, and significant baseline comorbidity imbalances. Real-time delivery and improved alert actionability are needed before this approach can be considered effective.

Acknowledgment

We would like to acknowledge the following additional members of the MEnD-AKI Study Team:
Nabihah Amatullah
Tiffany L Tran
Michele M Elder
Yuanfang Ren
Dan Ricketts
Beth Emanuele
Parisa Rashidi

Funding

• NIDDK Support