Predicting prolonged intensive care unit length-of-stay following coronary artery bypass surgery

Introduction: Intensive Care Unit (ICU) length-of-stay (LOS) following coronary artery bypass grafting (CABG) is influenced by patient condition and operating room events. Prolonged ICU LOS is associated with increased morbidity and costs. We developed a model to predict prolonged ICU LOS (>120 hours) based on the patient's condition at ICU admission. Methods: 2787 patients underwent primary or re-operative CABG between 1/1/93 and 3/31/94, either isolated or combined with other cardiac surgery. Characteristics of the 266 patients with ICU LOS >120 hours were compared to those with shorter ICU LOS. Factors significant at p<0.05 were analyzed by multiple logistic regression (MLR). Discrimination of the model was evaluated by C-statistic and calibration by Hosmer-Lemeshow goodness of fit. Results: 96 of 135 factors examined have a univariate correlation with ICU LOS >120 hours. Multiple logistic regression identifies 11 independent predictors, including use of intra-aortic balloon pump, emergency procedure, peripheral vascular disease, operation for mitral insufficiency, carotid stenosis or prior repair, elevated central venous pressure, advanced age, prolonged time on cardiopulmonary bypass, low preoperative serum albumin, and elevated alveolar-arterial oxygen gradient and decreased arterial bicarbonate level at ICU admission. The C-statistic (area under the ROC curve; a measure of discrimination) for this model is 0.824, and the model calibrates well (p=0.99). Discussion: Variables predicting ICU LOS >120 hours following coronary artery bypass include factors unique to the cardiac surgical patient (use of intra-aortic balloon pump, cardiopulmonary bypass time, mitral valve surgery) plus more general markers of physiologic reserve. Risk also increases with advancing age and evidence of generalized atherosclerosis. Identification at ICU admission of a patient subset at high risk for prolonged ICU stay may allow interventions to be directed at this group. Prediction of ICU LOS based on admission factors also has applications for quality improvement, bed triage, cost control, and ICU staffing.