Logistic Risk Model for Prolonged Ventilation After Adult Cardiac Surgery

Background The aim of this study was to develop a multivariate risk prediction model for prolonged ventilation after adult cardiac surgery. Methods This is a retrospective analysis of prospectively collected data on 12,662 consecutive patients undergoing adult cardiac surgery between April 1997 and March 2005. Data were randomly split into a development dataset (n = 6,000) and a validation dataset (n = 6,662). A multivariate logistic regression analysis was undertaken using a forward stepwise technique to identify independent risk factors for prolonged ventilation (defined as ventilation greater than 48 hours). The area under the receiver operating characteristic (ROC) curve and the Hosmer-Lemeshow goodness-of-fit statistic were calculated to assess the performance and calibration of the model, respectively. Patients were split into low-, medium-, and high-risk groups based on their predicted probability of prolonged ventilation. Results Three hundred thirty-three patients had prolonged ventilation (5.5%). Independent variables, identified with prolonged ventilation, are shown with relevant coefficient values and p values as follows: (1) age 65 to 75 years, 0.7831, p p p p = 0.013; (5) current smoker, 0.5315, p = 0.001; (6) serum creatinine 125 to 175 μmol/L, 0.6371, p p p p p p = 0.004; (12) prior cardiac surgery, 0.8946, p p = 0.004; (14) emergency surgery, 0.7421, p = 0.005; (15) mitral valve surgery, 0.7715, p p p = 0.025; intercept, −4.7666. The ROC curve for the predicted probability of prolonged ventilation was 0.79, indicating a good discrimination power. The prediction equation was well-calibrated, predicting well at all levels of risk. A simplified additive scoring system was also developed. In the validation dataset, 5.1% of patients had prolonged ventilation compared with 5.4% expected. The ROC curve for the validation dataset was 0.75. Conclusions We developed a contemporaneous multivariate prediction model for prolonged ventilation after cardiac surgery. This tool can be used in day-to-day practice to calculate patient-specific risk by the logistic equation or a simple scoring system with an equivalent predicted risk.