Preoperative prediction of intensive care unit stay following cardiac surgery

Abstract

Objective: Following cardiac surgery, a great variety in intensive care unit (ICU) stay is observed, making it often difficult to adequately predict ICU stay preoperatively. Therefore, a study was conducted to investigate, which preoperative variables are independent risk factors for a prolonged ICU stay and whether a patient's risk of experiencing an extended ICU stay can be estimated from these predictors.

Methods: The records of 1566 consecutive adult patients who underwent cardiac surgery at our institution were analysed retrospectively over a 2-year period. Procedures included in the analyses were coronary artery bypass grafting, valve replacement or repair, ascending and aortic arch surgery, ventricular rupture and aneurysm repair, septal myectomy and cardiac tumour surgery. For this patient group, ICU stay was registered and 57 preoperative variables were collected for analysis. Descriptives and log-rank tests were calculated and Kaplan-Meier curves drawn for all variables. Significant predictors in the univariate analyses were included in a Cox proportional hazards model. The definitive model was validated on an independent sample of 395 consecutive adult patients who underwent cardiac surgery at our institution over an additional 6-month period. In this patient group, the accuracy and discriminative abilities of the model were evaluated.

Results: Twelve independent preoperative predictors of prolonged ICU stay were identified: age at surgery>75 years, female gender, dyspnoea status>New York Heart Association class II (NYHA II), unstable symptoms, impaired kidney function (estimated glomerular filtration rate (eGFR)<60 ml min(-1)), extracardiac arterial disease, presence of arrhythmias, mitral insufficiency>colour flow mapping (CFM) grade II, inotropic support, intra-aortic balloon pumping (IABP), non-elective procedures and aortic surgery. The individual effect of every predictor on ICU stay was quantified and inserted into a mathematical algorithm (called the Morbidity Defining Cardiosurgical (MDC) index), making it possible to calculate a patient's risk of having an extended ICU stay. The model showed very good calibration and very good to excellent discriminative ability in predicting ICU stay >2, >5 and >7 days (C-statistic of 0.78; 0.82 and 0.85, respectively).

Conclusions: Twelve independent preoperative risk factors for a prolonged ICU stay following cardiac surgery were identified and constructed into a proportional hazards model. Using this risk model, one can predict whether a patient will have a prolonged ICU stay or not.