This study aimed to identify predictors of functional independence at discharge and at 3- and 6-month follow-up after stroke using machine learning techniques. Data were obtained from the Florence centre of the prospective observational STRATEGY study, including 244 patients admitted to intensive inpatient rehabilitation within 30 days of stroke. Demographic, premorbid, neurological, functional and clinical variables collected at rehabilitation admission were analysed to predict functional outcomes measured by the modified Barthel Index (mBI). Three machine learning algorithms were trained and validated using nested cross-validation, and predictor contributions were assessed through SHAP analysis. Among the 135 patients with complete follow-up data, XGBoost achieved the best predictive performance, with Median Absolute Errors of 7.83 at discharge, 9.15 at 3 months and 9.52 at 6 months. Baseline mBI was consistently the strongest predictor, while the importance of other variables changed over time. Clinical and functional measures were particularly relevant at discharge, whereas premorbid disability, cognitive function, neurological severity and comorbidity burden became increasingly important at longer follow-up. These findings highlight the potential of machine learning models based on routinely collected clinical variables to support personalised prognostic assessment and rehabilitation planning after stroke.

Time-varying relevance of functional independence predictors after stroke: a machine learning analysis from the STRATEGY study / Ester Marra, A.S.. - In: JOURNAL OF NEUROENGINEERING AND REHABILITATION. - ISSN 1743-0003. - ELETTRONICO. - (2026), pp. 0-0.

Time-varying relevance of functional independence predictors after stroke: a machine learning analysis from the STRATEGY study

Alessandro Sodero;Andrea Mannini
Project Administration
;
Marco Baccini;Francesca Cecchi
2026

Abstract

This study aimed to identify predictors of functional independence at discharge and at 3- and 6-month follow-up after stroke using machine learning techniques. Data were obtained from the Florence centre of the prospective observational STRATEGY study, including 244 patients admitted to intensive inpatient rehabilitation within 30 days of stroke. Demographic, premorbid, neurological, functional and clinical variables collected at rehabilitation admission were analysed to predict functional outcomes measured by the modified Barthel Index (mBI). Three machine learning algorithms were trained and validated using nested cross-validation, and predictor contributions were assessed through SHAP analysis. Among the 135 patients with complete follow-up data, XGBoost achieved the best predictive performance, with Median Absolute Errors of 7.83 at discharge, 9.15 at 3 months and 9.52 at 6 months. Baseline mBI was consistently the strongest predictor, while the importance of other variables changed over time. Clinical and functional measures were particularly relevant at discharge, whereas premorbid disability, cognitive function, neurological severity and comorbidity burden became increasingly important at longer follow-up. These findings highlight the potential of machine learning models based on routinely collected clinical variables to support personalised prognostic assessment and rehabilitation planning after stroke.
2026
0
0
Ester Marra, Alessandro Sodero, Andrea Mannini, Marco Baccini, Chiara Castagnoli, Angela Maria Politi, Julieta Giacani, Chiara Pedrini, Ilaria Pellegr...espandi
File in questo prodotto:
Non ci sono file associati a questo prodotto.

I documenti in FLORE sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificatore per citare o creare un link a questa risorsa: https://hdl.handle.net/2158/1501472
Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus ND
  • ???jsp.display-item.citation.isi??? ND
social impact