AI-Triggered Nephrology Consults Fail to Prevent AKI

Early nephrology consultations, initiated by real-time machine-learning risk scores, did not successfully prevent increases in peak serum creatinine among hospitalized patients identified as being at risk for severe acute kidney injury (AKI). This finding comes from a randomized trial published this week in JAMA Network Open.

The trial involved 1,134 patients across 12 hospitals. The intervention group received alerts for nephrology consultation when their calculated AKI risk score exceeded a predefined threshold. The control group received standard care. The primary outcome measured was the difference in peak serum creatinine levels between the two groups within 7 days of hospital admission.

Analysis of the trial data revealed no significant difference in peak serum creatinine levels between the AI-alert group and the standard care group. The mean peak serum creatinine was 1.7 mg/dL in the intervention arm and 1.8 mg/dL in the control arm, with a mean difference of -0.1 mg/dL (95% CI, -0.3 to 0.1; P=.30). Furthermore, the incidence of AKI Stage 2 or 3 did not differ significantly between the groups, occurring in 20.3% of the intervention group versus 21.7% of the control group.

Secondary outcomes also showed no significant benefits from the AI-triggered consultations. Rates of kidney failure requiring dialysis, in-hospital mortality, and length of hospital stay were comparable across both groups. The study authors concluded that while the AI system successfully identified patients at high risk, the early nephrology consultations prompted by these alerts did not translate into improved clinical outcomes regarding acute kidney injury prevention or management in this specific hospital setting.