Related Experiment Videos
Machine learning using PROFUND components for 30-day readmission prediction in multimorbid patients: a prospective
Amaia Pikatza-Huerga1,2, Aitor Almeida3,4, Raúl Quirós4,5
1Faculty of Engineering, University of Deusto, Av. de las Universidades, 24, Deusto, Bilbao, Bizkaia, E-48007, Spain. a.pikatza@deusto.es.
Scientific Reports
|May 13, 2026
Summary
Predicting 30-day hospital readmissions for multimorbid patients is challenging. The PROFUND index, assessing frailty and social factors, showed only modest improvements in prediction, even with machine learning models.
Area of Science:
- Geriatric Medicine
- Health Services Research
- Clinical Epidemiology
Background:
- Early hospital readmission for multimorbid patients presents a significant clinical challenge.
- Existing risk stratification tools often exhibit limited predictive performance.
- The PROFUND index, encompassing frailty, functional dependence, and social vulnerability, has an unclear role in predicting 30-day readmissions.
Purpose of the Study:
- To evaluate the predictive performance of the PROFUND index and its components for unplanned 30-day hospital readmissions in multimorbid patients.
- To compare machine learning models incorporating PROFUND components against established scores like LACE and HOSPITAL.
- To identify key predictors of 30-day readmission within the PROFUND framework.
Main Methods:
- Prospective, multicentre cohort study of multimorbid patients admitted to Internal Medicine and Geriatrics.
- Development of logistic regression and gradient boosting models (including ensemble) using PROFUND components.
- External validation of developed models and comparison with LACE and HOSPITAL scores.
- Utilized SHAP analysis to identify key predictive factors.
Main Results:
- 14% of 435 patients experienced unplanned 30-day readmission.
- External validation showed modest discrimination (AUC 0.52-0.59).
- The ensemble XGBoost model achieved the highest AUC (0.59), outperforming LACE and HOSPITAL scores incrementally.
- Key contributors included cognitive impairment, anemia, advanced age, heart failure severity, functional dependence, and limited caregiver support.
Conclusions:
- Incorporating frailty, functional, and social vulnerability domains via PROFUND components yielded only modest improvements in predicting 30-day readmissions.
- Even advanced machine learning techniques demonstrated limited predictive discrimination.
- The complexity of short-term readmission and potential constraints from sample size and predictors may limit performance.