TOPLINE:

AI models identified patients with end-stage kidney disease (ESKD) receiving hemodialysis who faced an imminent risk for hospital admission due to infections or fluid status abnormalities. When paired with nurse-led case reviews and targeted interventions, this strategy helped avert short-term admissions, demonstrating AI’s potential to guide timely, focused care.

METHODOLOGY:For patients with ESKD treated in outpatient dialysis clinics, advanced AI-driven machine learning approaches can improve care by more accurately predicting hospital and emergency department readmissions, but it has been applied less often in nephrology than in other medical specialties.Researchers conducted a retrospective observational matched cohort study of 10,294 adult Medicare patients with ESKD who received value-based hemodialysis at integrated kidney care clinics across the US in 2023.Two validated machine learning models — one trained on inputs for fluid overload and one on inputs for infection — calculated daily risk scores ranging from 0 to 1 and identified patients with scores ≥ 0.64 at a risk for hospitalization within the next 7 days due to fluid overload or infection.Remote nurses systematically reviewed high-risk patients in order of high- to low-risk score and conducted case reviews, with interventions managed either by remote nurses or clinic staff; observations linked to an intervention were matched to noninterventional observations.The study assessed the association between AI-triggered case reviews and interventions and all-cause hospital admission within 7 days of predicting the risk score, using linked electronic medical records and Medicare claims data.TAKEAWAY:Among 83,928 risk score observations, 13,988 prompted a case review and intervention; of these, 60% were managed remotely by nurses, and 40% were managed by clinic staff.Patients with chronic high-risk scores had 53% higher odds of hospitalization than those with occasional spikes in risk scores (odds ratio [OR], 1.53; P < .001).Patients who received AI-driven interventions had an 8% reduction in the odds of hospitalization within 7 days compared with those who did not receive interventions (OR, 0.92; P = .025); outcomes were comparable for remote nurse-managed and clinic-managed patients.When analyzed by risk scores, these interventions were associated with a 12% reduction in the risk for hospitalization for those with scores above 0.75-0.85 (OR, 0.88; P = .02); patients with final risk scores above 0.85 had higher hospitalization rates than those with lower risk scores, and although this group received more interventions, the interventions did not significantly reduce the 7‑day risk for hospital admissions.IN PRACTICE:

“These findings underscore the potential of AI-driven machine learning models to assist healthcare providers in targeted risk interventions for patients with ESKD,” the authors of the study wrote.

SOURCE:

The study was led by Sheetal Chaudhuri, PhD, Renal Research Institute, Waltham, Massachusetts. It was published online in NEJM Catalyst Innovations in Care Delivery.

LIMITATIONS:

The retrospective design introduced inherent bias. The lack of detailed documentation on specific interventions limited generalizability to other clinical settings. Potential confounding exists because cases were not matched on the basis of demographic factors. The primary outcome focused on 7-day hospitalization consistent with the model’s prediction horizon, and longer-term utilization outcomes such as 30-day hospitalizations were not assessed.

DISCLOSURES:

The study received funding from Fresenius Medical Care. Some authors reported being employed at healthcare companies, including the funding agency; having consulting arrangements; holding stock or stock options; owning patent ownership or having pending patent applications related to diagnostic or analytics tools; and serving in institutional research or leadership roles.

This article was created using several editorial tools, including AI, as part of the process. Human editors reviewed this content before publication.