Validation of the Immobility Harm Risk Score for Predicting Inpatient Immobility-Associated Outcomes: A Retrospective Cohort Study

Background Hospitalized patients are at high risk for immobility-related complications such as deconditioning, venous thromboembolism (VTE), falls, pressure injuries, prolonged length of stay (LOS), discharge to post-acute facilities, and 30-day readmissions. Current prediction tools often focus on single outcomes and demonstrate limited calibration or generalizability. The Immobility Harm Risk Score was developed to provide a unified, parsimonious model to predict multiple immobility-associated outcomes using structured data available at admission. To validate the predictive performance of the Immobility Harm Risk Score in a large, diverse inpatient population. Methods We conducted a retrospective cohort study of adult inpatients (≥18 years) admitted to Indiana University Health between 2013 and 2023. Exclusion criteria included elective procedures discharged within 24 hours, direct hospice admissions, or records with >20% missing data. Predictors included demographic, clinical, and functional variables. Outcomes were defined as deconditioning, VTE, falls, pressure injuries, prolonged LOS (>7 days), discharge to post-acute facility, and 30-day readmission. Model performance was assessed using AUCs, Brier scores, calibration slopes and intercepts, Hosmer–Lemeshow tests, and net reclassification improvement (NRI) against established comparator tools. Internal validation was performed via 10-fold cross-validation. We fit seven regularized logistic models sharing a common predictor set, used multiple imputation (m=10), and validated via stratified 10-fold cross-validation with bootstrap optimism correction; outcomes and comparators were defined from EHR assessments and validated code sets. Results The cohort comprised 9,842 admissions (mean age 64 ± 14 years; 52% female; 28% non-White; 20% ICU admissions). Outcome frequencies were: deconditioning 19%, VTE 1.1%, falls 2.0%, pressure injuries 3.2%, prolonged LOS 33%, discharge to post-acute 29%, and 30-day readmission 15%. The Immobility Harm Risk Score achieved AUCs of 0.82 for deconditioning, 0.78 for VTE, 0.75 for falls, 0.80 for pressure injuries, 0.77 for prolonged LOS, 0.81 for discharge disposition, and 0.73 for readmissions. Calibration slopes ranged from 0.92 to 1.04 with intercepts near zero; Brier scores ranged from 0.08–0.19. Comparative performance exceeded Braden (pressure injuries, AUC 0.64), Morse (falls, 0.61), Padua (VTE, 0.66), Charlson (LOS, 0.70), and LACE (readmission, 0.68) with meaningful NRIs. Decision curve analysis demonstrated clinical net benefit across plausible thresholds. Conclusions The Immobility Harm Risk Score demonstrates strong discrimination, calibration, and superiority over standard instruments in predicting multiple immobility-associated outcomes. Its single, multi-outcome structure offers clinical efficiency and potential for integration into electronic health records to inform therapy prioritization, patient counseling, and discharge planning.