A predictive index for length of stay in the intensive care unit following cardiac surgery

Abstract

Objective: To develop a predictive index for length of stay in the intensive care unit (ICU) following cardiac surgery.

Design: Univariate and multivariate logistic regression analysis of a cohort of 1404 patients divided into a derivation set of 713 patients and a validation set of 691 patients.

Setting: St. Michael's Hospital, Toronto, a tertiary care cardiovascular centre.

Patients: A consecutive sample of all patients undergoing cardiac surgery between Jan. 1 and Dec. 31, 1990 (derivation set), and Jan. 1 and Dec. 31, 1991 (validation set).

Main outcome measure: A long ICU stay (more than 2 days). Other outcomes analysed were ICU stays over 4, 7 and 10 days, and death.

Results: In the derivation set increasing age, female sex, left ventricular function, type of surgery, and urgency of surgery were found to be independent risk factors for a long ICU stay in a multivariate logistic regression analysis. A predictive index was created by assigning risk scores based on the odds ratios of the significant variables in the logistic regression analysis. The predictive index was found to predict lengths of ICU stay greater than 2, 4, 7 and 10 days, and patient death in the validation set.

Conclusions: Length of ICU stay and death following cardiac surgery can be predicted with a multivariate predictive index. The index has potential application as a means of stratifying cardiac surgical risk as well as in optimizing ICU resource planning when resources are limited.