A validated preoperative risk prediction tool for discharge to skilled nursing or rehabilitation facility following anatomic or reverse shoulder arthroplasty

As bundled payment models continue to spread, understanding the primary drivers of cost excess helps providers avoid penalties and ensure equal health care access. Recent work has shown discharge to rehabilitation and skilled nursing facilities (SNFs) to be a primary cost driver in total joint arthroplasty, and an accurate preoperative risk calculator for shoulder arthroplasty would not only help counsel patients in clinic during shared decision-making conversations but also identify high-risk individuals who may benefit from preoperative optimization and discharge planning. Anatomic and reverse total shoulder arthroplasty cohorts from 2 geographically diverse, high-volume centers were reviewed, including 1773 cases from institution 1 (56% anatomic) and 3637 from institution 2 (50% anatomic). The predictive ability of a variety of candidate variables for discharge to SNF/rehabilitation was tested, including case type, sociodemographic factors, and the 30 Elixhauser comorbidities. Variables surviving parameter selection were incorporated into a multivariable logistic regression model built from institution 1’s cohort, with accuracy then validated using institution 2’s cohort. A total of 485 (9%) shoulder arthroplasties overall were discharged to post-acute care (anatomic: 6%, reverse: 14%, P < .0001), and these patients had significantly higher rates of unplanned 90-day readmission (5% vs. 3%, P = .0492). Cases performed for preoperative fracture were more likely to require post-acute care (13% vs. 3%, P < .0001), whereas revision cases were not (10% vs. 10%, P = .8015). A multivariable logistic regression model derived from the institution 1 cohort demonstrated excellent preliminary accuracy (area under the receiver operating characteristic curve [AUC]: 0.87), requiring only 11 preoperative variables (in order of importance): age, marital status, fracture, neurologic disease, paralysis, American Society of Anesthesiologists physical status, gender, electrolyte disorder, chronic pulmonary disease, diabetes, and coagulation deficiency. This model performed exceptionally well during external validation using the institution 2 cohort (AUC: 0.84), and to facilitate convenient use was incorporated into a freely available, online prediction tool. A model built using the combined cohort demonstrated even higher accuracy (AUC: 0.89). This validated preoperative clinical decision tool reaches excellent predictive accuracy for discharge to SNF/rehabilitation followin