Abstract
Background: Medical progress and sustainability pressures have made reducing hospital Length Of Stay (LOS) for total joint arthroplasty increasingly feasible and necessary. Monitoring rehabilitation duration and outcomes after surgical ward discharge needs equal attention. The aim of this retrospective, cohort study is to evaluate perioperative predictors of Inpatient Rehabilitation LOS (IRLOS), Discharge Destination (DD) (home versus residential care unit) and Need for Assistance at Discharge (NAD), in patients undergoing inpatient rehabilitation after total hip or knee arthroplasty in a high-volume, specialized research hospital. Methods: Electronic hospital datasets were employed to identify all adults with hip or knee osteoarthritis who received specialistic inpatient rehabilitation after total joint replacement between January and December 2019. Associations between demographic, clinical, surgical and functional variables and postoperative outcomes were examined using binary logistic regression for dichotomous outcomes (DD, NAD) and linear regression for continuous outcomes (IRLOS). Results: Based on a cohort of 1679 patients, we found various patient-related (age, working status, living alone, pre-existing comorbidities, osteoarthritic characteristics), surgical (duration of intervention, LOS, joint approach) and postoperative (hemoglobin levels, functional status) predictors. Overall, the regression models explained a modest but meaningful proportion of the variability in rehabilitation duration and post-discharge outcomes (R(2) ranging from 0.12 to 0.34), resulting in marginal changes compared to a preliminary version of the same study on a smaller dataset. Conclusions: External validation on another cohort from the same hospital could be used to test the model's predictivity at the local level, supporting the continuity of care between an orthopedic hospital hub and outpatient care and rehabilitation. Gains in predictive capacity may follow from including local factors like the operating surgeon and team. Although these factors could significantly improve the model performance at the local level, they would not be generalizable in different settings.