AI-Enabled Acute Kidney Injury Prediction and the Challenge of Prevention

By
Charat Thongprayoon
Francesco Pesce
Wisit Cheungpasitporn
July 10, 2026

Jama Network Open

Overview

A randomized clinical trial evaluated the effectiveness of a structured early nephrology consultation, triggered by an AI risk score, in preventing acute kidney injury (AKI) among hospitalized patients. The trial found no significant reduction in AKI incidence or related outcomes, highlighting challenges in translating predictive analytics into effective preventive measures. The study utilized the ESTOP-AKI model to identify patients at increased risk of stage 2 AKI and randomized them to early nephrology consultation or usual care.

Background

Acute kidney injury (AKI) is a prevalent complication in hospitalized patients, associated with increased morbidity and mortality. Despite advancements in predictive analytics and decision support systems, preventing AKI remains a significant challenge due to the timing of intervention and the complexities of clinical management. Understanding the limitations of current predictive models is crucial for improving patient outcomes.

Data Highlights

No significant differences were observed in peak serum creatinine change, incident AKI, severe AKI, kidney replacement therapy, mortality, readmission, or 90-day outcomes between the early consultation and usual care groups.

Key Findings

The early nephrology consultation did not reduce peak serum creatinine change over 7 days.
Consult recommendations in the early consultation arm were often not followed, which may have contributed to the lack of improved outcomes.
Primary teams may perceive preventive recommendations as less urgent, impacting adherence to consult advice.
Alerts and risk scores alone have often failed to improve AKI outcomes without proper implementation.
Future trials may need to consider alternative primary endpoints beyond peak serum creatinine to assess AKI prevention effectively.

Clinical Implications

The findings suggest that while predictive models can identify at-risk patients, translating this information into effective preventive strategies requires systematic implementation and adherence to recommendations.

Conclusion

The trial underscores the importance of identifying at-risk patients and ensuring that actionable recommendations are effectively integrated into clinical workflows.