Appropriate patient selection for outpatient shoulder arthroplasty: a risk prediction tool.

BACKGROUND: The transition from inpatient to outpatient shoulder arthroplasty critically depends on appropriate patient selection, both to ensure safety and to counsel patients preoperatively regarding individualized risk. Cost and patient demand for same-day discharge have encouraged this transition, and a validated predictive tool may help decrease surgeon liability for complications and help select patients appropriate for same-day discharge. We hypothesized that an accurate predictive model could be created for short inpatient length of stay (discharge at least by postoperative day 1), potentially serving as a useful proxy for identifying patients appropriate for true outpatient shoulder arthroplasty. METHODS: A multicenter cohort of 5410 shoulder arthroplasties (2805 anatomic and 2605 reverse shoulder arthroplasties) from 2 geographically diverse, high-volume health systems was reviewed. Short inpatient stay was the primary outcome, defined as discharge on either postoperative day 0 or 1, and 49 patient outcomes and factors including the Elixhauser Comorbidity Index, sociodemographic factors, and intraoperative parameters were examined as candidate predictors for a short stay. Factors surviving parameter selection were incorporated into a multivariable logistic regression model, which underwent internal validation using 10,000 bootstrapped samples. RESULTS: In total, 2238 patients (41.4%) were discharged at least by postoperative day 1, with no difference in rates of 90-day readmission (3.5% vs. 3.3%, P = .774) between cohorts with a short length of stay and an extended length of stay (discharge after postoperative day 1). A multivariable logistic regression model demonstrated high accuracy (area under the receiver operator characteristic curve, 0.762) for discharge by postoperative day 1 and was composed of 13 variables: surgery duration, age, sex, electrolyte disorder, marital status, American Society of Anesthesiologists score, paralysis, diabetes, neurologic disease, peripheral vascular disease, pulmonary circulation disease, cardiac arrhythmia, and coagulation deficiency. The percentage cutoff maximizing sensitivity and specificity was calculated to be 47%. Internal validation showed minimal loss of accuracy after bias correction for overfitting, and the predictive model was incorporated into a freely available online tool to facilitate easy clinical use. CONCLUSIONS: A risk prediction tool for short inpatient length of stay after shoulder arthroplasty reaches very good accuracy despite requiring only 13 variables and was derived from an underlying database with broad geographic diversity in the largest institutional shoulder arthroplasty cohort published to date. Short inpatient length of stay may serve as a proxy for identifying patients appropriate for same-day discharge, although perioperative care decisions should always be made on an individualized and holistic basis.

Duke Scholars

Author Cynthia Lea Green Biostatistics & Bioinformatics, Division of Biostatistics

Author Tally Edward Lassiter Orthopaedic Surgery

Author Oke Adrian Anakwenze Orthopaedic Surgery

Author Christopher Scott Klifto Orthopaedic Surgery