Development of a Machine Learning Algorithm to Predict Nonroutine Discharge Following Unicompartmental Knee Arthroplasty.

Background: Reliable and effective prediction of discharge destination following unicompartmental knee arthroplasty (UKA) can optimize patient outcomes and system expenditure. The purpose of this study is to develop a machine learning algorithm that can predict nonhome discharge in patients undergoing UKA.

Methods: A retrospective review of a prospectively collected national surgical outcomes database was performed to identify adult patients who underwent UKA from 2015 to 2019. Nonroutine discharge was defined as discharge to a location other than home. Five machine learning algorithms were developed to predict this outcome. Performance of the algorithms was assessed through discrimination, calibration, and decision curve analysis.

Results: Overall, of the 7275 patients included, 263 (3.6) patients were unable to return home upon discharge following UKA. The factors determined most important for identification of candidates for nonroutine discharge were total hospital length of stay, preoperative hematocrit, body mass index, preoperative sodium, American Society of Anesthesiologists classification, gender, and functional status. The extreme boosted model achieved the best performance based on discrimination (area under the curve = 0.875), calibration, and decision curve analysis. This model was integrated into a web-based open access application able to provide both predictions and explanations.

Conclusion: The present model can, following appropriate external validation, be used to augment clinician decision-making in patients undergoing elective UKA. Patients with high preoperative probabilities of nonroutine discharge based on nonmodifiable risk factors should be counseled to start the insurance authorization process with case management to avoid unnecessary inpatient stay, and those with modifiable risk can attempt prehabilitation to optimize these parameters before surgery.