Objectives: To improve the precision of currently available models for predicting length of stay of individual patients in the intensive care unit, to assist in directing patients into fast-track management after coronary artery bypass graft (CABG) surgery.
Setting: ICU in an Australian teaching hospital.
Design and participants: Prospectively collected data from 333 patients who underwent elective CABG surgery were analysed by univariate and multivariate regression, to develop models of increasing power through the addition of factors covering the operative and early ICU phases (1, 4 and 8 hours postoperatively) to traditional preoperative risk of patient care. The model that gave the best combination of precision and availability for clinical decision-making was then validated on a series of 117 patients who underwent CABG surgery. Overall competence of this model was assessed.
Results: Addition of intraoperative factors to the first (preoperative only) model (R2 = 0.07) doubled the precision of the analysis (R2 = 0.18). Addition of factors derived from the first 4 hours of ICU management increased precision fivefold (R2 = 0.38). This model was satisfactorily validated: regression of actual versus predicted ICU stay from the validation set gave a slope of 0.85 and y intercept of 2.60 hours. The 95% confidence levels of individual predictions obtained from the development set, for an estimated ICU stay of 12 hours, spanned 43 hours.
Conclusions: Although the optimal model greatly increases precision, it is still inadequate for scheduling fasttrack patients, where wrong predictions for individuals can cause major problems in resource allocation.