BACKGROUND: A reliable prediction tool for 90-day adverse events not only would provide patients with valuable estimates of their individual risk perioperatively, but would also give health-care systems a method to enable them to anticipate and potentially mitigate postoperative complications. Predictive accuracy, however, has been challenging to achieve. We hypothesized that a broad range of patient and procedure characteristics could adequately predict 90-day readmission after total joint arthroplasty (TJA). METHODS: The electronic medical records on 10,155 primary unilateral total hip (4,585, 45%) and knee (5,570, 55%) arthroplasties performed at a single institution from June 2013 to January 2018 were retrospectively reviewed. In addition to 90-day readmission status, >50 candidate predictor variables were extracted from these records with use of structured query language (SQL). These variables included a wide variety of preoperative demographic/social factors, intraoperative metrics, postoperative laboratory results, and the 30 standardized Elixhauser comorbidity variables. The patient cohort was randomly divided into derivation (80%) and validation (20%) cohorts, and backward stepwise elimination identified important factors for subsequent inclusion in a multivariable logistic regression model. RESULTS: Overall, subsequent 90-day readmission was recorded for 503 cases (5.0%), and parameter selection identified 17 variables for inclusion in a multivariable logistic regression model on the basis of their predictive ability. These included 5 preoperative parameters (American Society of Anesthesiologists [ASA] score, age, operatively treated joint, insurance type, and smoking status), duration of surgery, 2 postoperative laboratory results (hemoglobin and blood-urea-nitrogen [BUN] level), and 9 Elixhauser comorbidities. The regression model demonstrated adequate predictive discrimination for 90-day readmission after TJA (area under the curve [AUC]: 0.7047) and was incorporated into static and dynamic nomograms for interactive visualization of patient risk in a clinical or administrative setting. CONCLUSIONS: A novel risk calculator incorporating a broad range of patient factors adequately predicts the likelihood of 90-day readmission following TJA. Identifying at-risk patients will allow providers to anticipate adverse outcomes and modulate postoperative care accordingly prior to discharge. LEVEL OF EVIDENCE: Prognostic Level IV. See Instructions for Authors for a complete description of levels of evidence.
BACKGROUND: A reliable prediction tool for 90-day adverse events not only would provide patients with valuable estimates of their individual risk perioperatively, but would also give health-care systems a method to enable them to anticipate and potentially mitigate postoperative complications. Predictive accuracy, however, has been challenging to achieve. We hypothesized that a broad range of patient and procedure characteristics could adequately predict 90-day readmission after total joint arthroplasty (TJA). METHODS: The electronic medical records on 10,155 primary unilateral total hip (4,585, 45%) and knee (5,570, 55%) arthroplasties performed at a single institution from June 2013 to January 2018 were retrospectively reviewed. In addition to 90-day readmission status, >50 candidate predictor variables were extracted from these records with use of structured query language (SQL). These variables included a wide variety of preoperative demographic/social factors, intraoperative metrics, postoperative laboratory results, and the 30 standardized Elixhauser comorbidity variables. The patient cohort was randomly divided into derivation (80%) and validation (20%) cohorts, and backward stepwise elimination identified important factors for subsequent inclusion in a multivariable logistic regression model. RESULTS: Overall, subsequent 90-day readmission was recorded for 503 cases (5.0%), and parameter selection identified 17 variables for inclusion in a multivariable logistic regression model on the basis of their predictive ability. These included 5 preoperative parameters (American Society of Anesthesiologists [ASA] score, age, operatively treated joint, insurance type, and smoking status), duration of surgery, 2 postoperative laboratory results (hemoglobin and blood-urea-nitrogen [BUN] level), and 9 Elixhauser comorbidities. The regression model demonstrated adequate predictive discrimination for 90-day readmission after TJA (area under the curve [AUC]: 0.7047) and was incorporated into static and dynamic nomograms for interactive visualization of patient risk in a clinical or administrative setting. CONCLUSIONS: A novel risk calculator incorporating a broad range of patient factors adequately predicts the likelihood of 90-day readmission following TJA. Identifying at-risk patients will allow providers to anticipate adverse outcomes and modulate postoperative care accordingly prior to discharge. LEVEL OF EVIDENCE: Prognostic Level IV. See Instructions for Authors for a complete description of levels of evidence.