SYSTEM FEATURES
Supporting Better Patient Monitoring
The system combines patient information, risk prediction and dashboard visualisation in one platform.
Patient Assessment
Enter patient information through a structured assessment form.
Risk Prediction
Estimates the likelihood of hospital readmission within 30 days.
Dashboard Monitoring
View patient and prediction information through an integrated dashboard.
READING THE RESULT
What the risk levels mean
Each patient receives a predicted probability of readmission within 30 days, grouped into three levels.
Lower estimated risk. Standard discharge planning is likely appropriate.
Moderate estimated risk. The care team may want to review follow-up arrangements.
Higher estimated risk. The patient may benefit from closer follow-up after discharge.
ABOUT THE MODEL
Calibrated XGBoost
Predictions come from an XGBoost classifier, a model made up of many gradient-boosted decision trees that performs well on structured patient records.
The model's output has been calibrated using Platt scaling (sigmoid calibration), so the predicted probability can be read as a percentage. A predicted risk of 20% represents an estimated 20% probability of readmission within 30 days based on the model.
What information the model uses
Have these details ready before starting an assessment.
- Age
- Gender
- Length of stay
- Admission type
- Admission location
- Discharge location
- Insurance
- Marital status
- Race
RESPONSIBLE USE
Limitations
- The model was trained on historical data from MIMIC-IV, collected at a single US hospital (Beth Israel Deaconess Medical Center in Boston), which may not reflect every hospital, region or patient population.
- The model's ability to separate patients who were readmitted from those who were not is modest (ROC AUC 0.60), so results should be read as a rough guide.
- The model uses admission and demographic details only. It does not use lab results, medications or diagnoses.
- Results depend on the quality and completeness of the information entered.
- A predicted risk describes what happened to similar patients. It is not a certainty about an individual patient.
- The system supports clinical judgement and should not be the only basis for discharge or treatment decisions.