Predicting Unplanned Hospital Readmissions in People with Multiple Long-Term Conditions

The prevalence of multiple long-term conditions (MLTC) is associated with increased healthcare utilisation and higher risk of unplanned 30-day hospital readmission. Researchers developed Self-HR, a two-stage machine learning framework tested on data from 99,207 UK Biobank participants with multiple long-term conditions. The system achieved AUROC of 0.92 and AUPRC of 0.75 in the main dataset and outperformed all compared methods including XGBoost and Random Forest. In external validation on 79,224 patients from the Clinical Practice Research Datalink, it achieved AUROC of 0.86 and F1 score of 0.67. Self-HR demonstrated superior robustness to incomplete data and class imbalance, maintaining an F1 score of 0.62 even when trained on only 50% labelled data. Analyses identified admission diagnoses as the strongest predictor, followed by medications and long-term condition history. The system proved to be an effective and generalisable tool for predicting readmission risk in patients with multiple conditions.