Prediction of postoperative respiratory depression and respiratory complications in patients on preoperative methadone

Purpose We developed prediction models for postoperative respiratory depression and respiratory complications for 958 patients who were on methadone preoperatively. Methods The primary outcome was postoperative respiratory depression as defined by respiratory rate < 10/min, oxygen saturation (SpO(2)) < 90%, or requirement of naloxone for 48 h postoperatively. Secondary outcome was the composite of postoperative respiratory complications. Prediction models for postoperative respiratory depression and respiratory complications were constructed using multivariate logistic regression with preoperative and intraoperative characteristics as the predictors. Results For the multivariate logistic regression model for postoperative respiratory depression, surgery duration (P = 0.005), body mass index (BMI) (P = 0.008), surgery involving digestive system (P = 0.031), and American Society of Anesthesiologists (ASA) physical status >= 4 (P = 0.038) were statistically significant predictors. The area under the receiver operating characteristic curve (AUROC) of the model was 0.581 (0.558-0.601) [median (95% confidence interval (CI))] with fivefold cross-validation. For the model for postoperative respiratory complications, surgery duration (P = 0.001), history of hypertension (P = 0.028), surgery involving musculoskeletal system (P < 0.001), surgery involving integumental system (P = 0.034), surgery categorized to miscellaneous therapeutic procedures (P = 0.028), combined general and regional anesthesia (P = 0.033), ASA physical status 3 (P < 0.001), and ASA physical status >= 4 (P < 0.001) were statistically significant predictors, and AUROC of the model was 0.726 (0.712-0.737). Conclusions Multivariate logistic regression models including preoperative, and intraoperative characteristics as the predictors performed poorly to predict postoperative respiratory depression, and moderately for postoperative respiratory complications. Neither model is accurate enough to be subject to clinical use.