Related Experiment Videos
Predictive Factors of Inpatient Rehabilitation Outcomes and Stay: A Machine Learning Study with Temporal Validation
Andrea Campagner1,2, Claudio Cordani3, Catia Pelosi4
1Laboratory of Mechanics of Biological Structures, IRCCS Galeazzi-Sant'Ambrogio Hospital, Via Belgioioso 157, 20157 Milan, Italy.
Healthcare (Basel, Switzerland)
|July 28, 2026
Summary
Machine learning models show moderate predictive performance for inpatient rehabilitation length of stay and daily living function after joint replacement. Models identified key predictors like perioperative complexity and social factors, demonstrating stable performance over time.
Area of Science:
- Orthopedic surgery outcomes
- Rehabilitation medicine
- Health informatics
Background:
- Optimizing post-discharge rehabilitation is crucial for patient safety and care sustainability, especially with rising rates of osteoarthritis and related surgeries.
- Real-world data from tertiary orthopedic hospitals offers a valuable resource for understanding rehabilitation trajectories.
- Machine learning (ML) presents an opportunity to predict key rehabilitation outcomes.
Purpose of the Study:
- To evaluate the predictive performance of ML models for Inpatient Rehabilitation Length Of Stay (IRLOS), Function in the Activities of Daily Living (FADL), and discharge destination (DD).
- To identify key predictors influencing rehabilitation outcomes in patients undergoing total joint replacement for hip and knee osteoarthritis.
- To assess the stability and generalizability of ML models using temporal validation.
Main Methods:
- Utilized routinely collected perioperative data from 2103 patients undergoing hip or knee total joint replacement.
- Developed and validated ML models using a temporal split (2019 development, 2018 validation).
- Modeled IRLOS and FADL as regression tasks, and DD as a binary classification task, employing rigorous feature selection and hyperparameter tuning.
Main Results:
- ML models achieved modest predictive performance for IRLOS (R²=0.17) and FADL (R²=0.25), with stable performance over time.
- DD prediction demonstrated good discrimination (AUC=0.85) despite class imbalance, showing high sensitivity and negative predictive value.
- Interpretability analysis highlighted perioperative complexity, baseline function, and social factors as significant predictors of rehabilitation outcomes.
Conclusions:
- ML models can identify relevant predictors of rehabilitation outcomes, including medical and non-medical determinants.
- The developed models exhibit stable performance across joint types and temporal validation, suggesting their utility in capturing rehabilitation pathway patterns.
- While predictive accuracy was moderate, the insights gained from ML models can inform strategies for optimizing patient rehabilitation after joint replacement surgery.