Truace
Subscribe

Predicting the Risk of Unplanned Hospital Encounters in Hemodialysis Patients: A Comparison Between Clinicians and a Machine Learning Prediction Algorithm

Rationale & objective: Dialysis patients have a higher risk of hospitalization than the age-matched general population. Unplanned hospital encounters (UHEs), including emergency department visits and hospitalizations, may be preventable but are difficult to anticipate. We aimed to stratify dialysis patients' risk of UHE, using a machine learning (ML)-based prediction model. Study design: We created an ML-based prediction model to stratify hemodialysis (HD) patients' risk of UHE in the subsequent 7 and 30 day…

By the Truvace desk2 min read1 sourcePeer-reviewedTRV-2026-1891· cite & verify
Predicting the Risk of Unplanned Hospital Encounters in Hemodialysis Patients: A Comparison Between Clinicians and a Machine Learning Prediction Algorithm
Kidney dialysis patients : a population at undue risk? : hearing before the Special Committee on Aging, United States Senate, One Hundred Sixth Congress, second session, Washington, DC, June 26, 2000 by United States. Congress. Senate. Special Committee on Aging. Public domain

In brief

Researchers built a machine learning model that uses demographic, clinical, and dialysis session data, updated weekly, to estimate which hemodialysis patients will have an emergency visit or hospitalization within 7 or 30 days. They then compared its predictions with those made by clinicians for a separate group of patients.

The model caught many more future events than clinicians, particularly over 30 days, which could help target preventive attention. However, its 7-day precision was lower, and the authors say it remains unknown whether using the tool in care would actually reduce unplanned hospital encounters.

Main points

  1. The model was developed using 627 maintenance hemodialysis patients treated at an academic hospital between October 2015 and April 2022.
  2. A separate group of 285 patients was used to compare model predictions with clinician predictions.
  3. At 7 days, sensitivity was 0.63 for the model versus 0.25 for clinicians, while PPV was 0.07 versus 0.14.
  4. At 30 days, PPV was comparable (0.21 model, 0.25 clinicians; P = 0.54), but sensitivity was 0.73 versus 0.24.

The problem

For 7-day predictions the model was less precise than clinicians, so it produced proportionally more false alarms, and whether it reduces hospital encounters is not yet shown.

The rundown

Dialysis patients are hospitalized more often than the age-matched general population, and unplanned hospital encounters, which include emergency department visits and hospitalizations, may be preventable but are hard to anticipate. The study aimed to stratify risk using predictors updated weekly.

The comparison showed a trade-off. Clinicians were more precise for 7-day events, meaning a higher share of their flagged patients had an event, but they missed most events. The model found a much larger share of events at both horizons.

The authors conclude that the model anticipates events with greater sensitivity than clinicians at comparable PPV, and that further research must determine whether integrating the tool into clinical care leads to fewer unplanned encounters.

Sources

  1. Peer-reviewedKidney Medicine2026-08-13

The debate