Doctoral Meeting: 'Interpretable Stratification of Functional Outcome and Rehabilitation Length of Stay at Admission'
Accurate estimation of rehabilitation outcomes at admission is important for clinical planning and resource management in spinal cord injury (SCI) care. This study examines admission-time prediction of two routinely documented outcomes: functional independence at rehabilitation discharge, measured using the Spinal Cord Independence Measure (SCIM), and rehabilitation length of stay (LOS). A retrospective cohort of 798 distinct SCI patients admitted to the Swiss Paraplegic Centre (SPZ) Nottwil between 2018 and 2020 was analyzed. Only variables available up to rehabilitation admission were included, covering acute-phase functional status, rehabilitation admission SCIM subscores, demographic information, diagnostic and comorbidity indicators, intensive care exposure, and temporal admission characteristics. Patient-level group splitting was applied to prevent information leakage. Both outcomes were formulated as three-class classification tasks using percentile-based thresholds derived exclusively from the training set. SCIM discharge outcomes were stratified using the 25th and 75th percentiles, while LOS was stratified using asymmetric percentiles to account for right-skewed distributions. Several tree-based ensemble models were evaluated within a unified preprocessing pipeline. Results indicate that discharge functional outcome can be predicted with strong out-of-sample performance using admission-time functional measures and acute-to-rehabilitation transition indicators. In contrast, LOS stratification achieved moderate performance and was more strongly influenced by temporal and organizational factors. The proposed framework provides a transparent and leakage-controlled approach for admission-time outcome stratification in SCI rehabilitation using routinely collected clinical data.
- Supervisor: Jose María Alonso Moral
- Moderator: Antonio Rodríguez Sánchez
Accurate estimation of rehabilitation outcomes at admission is important for clinical planning and resource management in spinal cord injury (SCI) care. This study examines admission-time prediction of two routinely documented outcomes: functional independence at rehabilitation discharge, measured using the Spinal Cord Independence Measure (SCIM), and rehabilitation length of stay (LOS). A retrospective cohort of 798 distinct SCI patients admitted to the Swiss Paraplegic Centre (SPZ) Nottwil between 2018 and 2020 was analyzed. Only variables available up to rehabilitation admission were included, covering acute-phase functional status, rehabilitation admission SCIM subscores, demographic information, diagnostic and comorbidity indicators, intensive care exposure, and temporal admission characteristics. Patient-level group splitting was applied to prevent information leakage. Both outcomes were formulated as three-class classification tasks using percentile-based thresholds derived exclusively from the training set. SCIM discharge outcomes were stratified using the 25th and 75th percentiles, while LOS was stratified using asymmetric percentiles to account for right-skewed distributions. Several tree-based ensemble models were evaluated within a unified preprocessing pipeline. Results indicate that discharge functional outcome can be predicted with strong out-of-sample performance using admission-time functional measures and acute-to-rehabilitation transition indicators. In contrast, LOS stratification achieved moderate performance and was more strongly influenced by temporal and organizational factors. The proposed framework provides a transparent and leakage-controlled approach for admission-time outcome stratification in SCI rehabilitation using routinely collected clinical data.
- Supervisor: Jose María Alonso Moral
- Moderator: Antonio Rodríguez Sánchez
On-site event
Wednesday, March 25, 2026
1774396800000
/events/doctoral-meeting-interpretable-stratification-of-functional-outcome-and-rehabilitation-length-of-stay-at-admission
events_en