Inpatient Recovery of Function After Knee Replacement: A Longitudinal Evaluation Across Four Sequential Care Pathway Redesigns

Abstract

Introduction: Inpatient recovery of activities (IRoA) is a key outcome after total knee replacement (TKR), reflecting the patient's ability to regain independence. Preoperative prediction can support shared decision-making and care planning, but its accuracy may decline as healthcare contexts evolve. This study evaluated: Stability of the predictors used for estimating IRoA across sequential care pathway redesigns. Temporal transportability of prediction models for IRoA over different time periods. Methods: A retrospective longitudinal cohort study was conducted at Nij Smellinghe Hospital (the Netherlands) using routine clinical data from 1853 patients opting for TKR surgery between 2009 and 2020. Four consecutive care pathway periods were studied: Joint-Care, Function Tailored, Fast Track, and Prehabilitation. IRoA was assessed using the modified Iowa Levels of Assistance Scale (mILAS). Multinomial regression and minimum Akaike Information Criterion (AIC) guided selection of preoperative predictors (e.g., age, ASA score, TUG, DEMMI). Model performance and temporal transportability were evaluated across periods. Results: Predictor stability was limited across care pathways. Age was consistently retained in all periods for both slow and regular recovery groups. DEMMI was retained in all but the Function Tailored period, while BMI was not retained in any model. ASA and ISAR were variably retained, mainly in later periods and predominantly for slow recovery. TUG was inconsistently retained and primarily selected in earlier periods. Model performance declined when models were applied to later care pathways, with predicted correct proportions decreasing from 0.70 to 0.59, 0.66 to 0.43, and 0.51 to 0.36 across successive transitions. The Prehabilitation model achieved a predicted correct proportion of 0.65 within its own period. Conclusions: Prediction of IRoA after TKR is sensitive to changes in care pathways. Predictor relevance and model performance varied over time, limiting temporal transportability. Ongoing model updating and local recalibration are required to maintain clinical usefulness in evolving care contexts.