A Nomogram to Predict Postoperative Pulmonary Complications after Cardiothoracic Surgery

Abstract Objective Development of novel scoring system that would be predictive of postoperative pulmonary complications in critically ill patients after cardiac and major vascular surgery. Methods 17,433 postoperative patients after coronary artery bypass graft, valve or thoracic aorta repair surgery admitted to the cardiovascular intensive care units (ICU) at Cleveland Clinic Main Campus from 2009-2015. Primary outcome was the composite of postoperative pulmonary complication including pneumonia, prolonged postoperative mechanical ventilation (> 48 hours) or re-intubation occurring during the hospital stay. Elastic net logistic regression was used on the training subset to build a prediction model that included perioperative predictors. Five-fold cross validation was used to select an appropriate subset of the predictors. The predictive efficacy was assessed with calibration and discrimination statistics. Post hoc, out of 13,353 adult patients, we tested the clinical usefulness of our risk prediction model on 12,956 patients who had surgery from 2015-2019. Results Postoperative pulmonary complications were observed in 1669 (9.6%) patients. A prediction model that included baseline and demographic risk factors along with perioperative predictors had a C-statistic of 0.87 (95% CI: 0.86- 0.88), with a corrected Brier score of 0.06. Our prediction model maintains satisfactory discrimination (C-statistics of 0.87) and calibration (Brier score of 0.07) abilities when evaluated on an independent dataset of 12,843 recent adult patients who had cardiovascular surgery. Conclusions A novel prediction nomogram accurately predicted postoperative pulmonary complications after major cardiac and vascular surgery. Intensivists may utilize these predictors to allow for proactive and preventative interventions in this patient population.