Predictors of Failure to Rescue After Postoperative Respiratory Failure: A Retrospective Cohort Analysis of 13,047 Patients Using the ACS-NSQIP Dataset

Introduction: Death after suffering a postoperative complication (failure to rescue) is an area of concern, and its occurrence after postoperative respiratory failure (PRF) is not well defined. We seek to identify the predictors of failure to rescue in patients who develop PRF. Methods: A retrospective cohort study of adults undergoing noncardiac surgery who developed PRF (postoperative unplanned intubation and receipt of mechanical ventilation for longer than 48 h) was conducted using the American College of Surgeons National Surgical Quality Improvement Project database. Predictors of failure to rescue after PRF were identified using the Least Absolute Shrinkage and Selection Operator (LASSO)–penalized variable selection method, with the Bayesian information criterion, in the context of a multiple logistic regression model (with Firth's bias correction). Results: Of the 13,047 patients that formed our final evaluable study cohort, 3669 (28.1%) patients died within 30 days of surgery. We identified age, sex, American Society of Anesthesiologists physical status, presence of preoperative ascites, disseminated cancer, bleeding disorders, elevated preoperative creatinine, and low preoperative prealbumin levels as predictors of failure to rescue. The area under the curve for the final model was 0.6804, with a standard error of 0.0104 (95% CI area under the curve: 0.6600 to 0.7008). Conclusions: We observed that almost 30% of patients that develop respiratory failure after noncardiac surgery die within 30 days of surgery. The validated eight-variable perioperative predictive model provides a risk estimate for death after PRF and may be useful for the purposes of preoperative planning, prognostication, decision making and resource allocation in patients who develop this complication.