Validation of a Predictive Tool for Discharge to Rehabilitation or a Skilled Nursing Facility After TJA.

<bold>Background: </bold>Cost excess in bundled payment models for total joint arthroplasty (TJA) is driven by discharge to rehabilitation or a skilled nursing facility (SNF). A recently published preoperative risk prediction tool showed very good internal accuracy in stratifying patients on the basis of likelihood of discharge to an SNF or rehabilitation. The purpose of the present study was to test the accuracy of this predictive tool through external validation with use of a large cohort from an outside institution.<bold>Methods: </bold>A total of 20,294 primary unilateral total hip (48%) and knee (52%) arthroplasty cases at a tertiary health system were extracted from the institutional electronic medical record. Discharge location and the 9 preoperative variables required by the predictive model were collected. All cases were run through the model to generate risk scores for those patients, which were compared with the actual discharge locations to evaluate the cutoff originally proposed in the derivation paper. The proportion of correct classifications at this threshold was evaluated, as well as the sensitivity, specificity, positive and negative predictive values, number needed to screen, and area under the receiver operating characteristic curve (AUC), in order to determine the predictive accuracy of the model.<bold>Results: </bold>A total of 3,147 (15.5%) of the patients who underwent primary, unilateral total hip or knee arthroplasty were discharged to rehabilitation or an SNF. Despite considerable differences between the present and original model derivation cohorts, predicted scores demonstrated very good accuracy (AUC, 0.734; 95% confidence interval, 0.725 to 0.744). The threshold simultaneously maximizing sensitivity and specificity was 0.1745 (sensitivity, 0.672; specificity, 0.679), essentially identical to the proposed cutoff of the original paper (0.178). The proportion of correct classifications was 0.679. Positive and negative predictive values (0.277 and 0.919, respectively) were substantially better than those of random selection based only on event prevalence (0.155 and 0.845), and the number needed to screen was 3.6 (random selection, 6.4).<bold>Conclusions: </bold>A previously published online predictive tool for discharge to rehabilitation or an SNF performed well under external validation, demonstrating a positive predictive value 79% higher and number needed to screen 56% lower than simple random selection. This tool consists of exclusively preoperative parameters that are easily collected. Based on a successful external validation, this tool merits consideration for clinical implementation because of its value for patient counseling, preoperative optimization, and discharge planning.<bold>Level Of Evidence: </bold>Prognostic Level III . See Instructions for Authors for a complete description of levels of evidence. [ABSTRACT FROM AUTHOR]