Skip to content
Contact UsStart Building

Healthcare Operations

Hospital readmission

Estimate early readmission after a diabetic hospital encounter to inform follow-up planning.
AI-generated representative illustration of an empty hospital bed, bedside cabinet and a plain discharge-document tray.
AI-generated task illustration. Representative scene; not a customer site or benchmark observation.
From known examples to a prediction

Available inputs

  • Admission details and prior care
  • Recorded procedures and medications

Known examples

71,518 recorded examples

Prediction target

The curated early-readmission outcome after a diabetic encounter
0.70 pphigher ROC AUC

vs. LightGBM (tuned ensemble), the strongest conventional result in this benchmark.

Hospital readmissionROC AUC · Higher is better

71,518 examples · 47 inputs · 9 evaluation splits

01
  1. Causilo
    0.6793
  2. LightGBM
    0.6723
  3. CatBoost
    0.6717
  4. XGBoost
    0.6696
  5. Logistic regression
    0.6586
  6. Random forest
    0.6573
  7. Extra trees
    0.6524
  8. K-nearest neighbors
    0.6512

The task in practice

The curated study contains diabetic hospital encounters with admission details, procedures, medications and prior care. Repeated patient records are removed. The target is early readmission.

  1. Pair encounters with outcomes. Use curated encounter records and their observed readmission labels.
  2. Define the scoring moment. Provide only encounter fields available at the time an intended analysis would run.
  3. Evaluate follow-up priorities. Compare discrimination and calibration on independent diabetic encounters before assessing a follow-up workflow.

Scope. Retrospective benchmark classification does not establish a safe clinical decision or a reduction in readmissions.

Dataset, sources and evaluation

Hospital readmission dataset. Original source and curated OpenML task linked below. Original dataset source.

The TabArena-curated OpenML version (46922) supplies the dataset and recorded target, EarlyReadmission. Comparison scores come from the retained results_per_split_official.csv evaluation snapshot. These pages describe the task and compare the recorded means; source records are not republished here.

ROC AUC measures how well the scores rank the recorded positive outcomes above negative ones. A higher score is better. The change is reported in percentage points, rather than relative error reduction.

Causilo uses its default configuration. Each conventional family uses its strongest reported mean across default, tuned and tuned-ensemble settings. The chart compares reported means, without a claim of statistical significance, customer deployment or operational savings.

ROC AUC difference in percentage points = (Causilo mean − strongest conventional mean) × 100. ROC AUC is a ranking metric, not a percentage of correct predictions.

How to read these comparisons · All configurations and original values.