[A model to predict death after lung cancer resection: applicability to individual cases]

To evaluate the reliability of a logistic regression model to predict individual risk of death related to lung cancer resection.

A study of 515 consecutive patients undergoing anatomical pulmonary resection (lobectomy or pulmonectomy) for lung cancer between January 1994 and December 2001. Dependent variable: death in or out of hospital within 30 days of surgery. Continuous independent variables: age, body mass index, and percent of predicted postoperative FEV1. Binary independent variables: ischemic heart disease, diabetes mellitus, preoperative arrhythmia, induction chemotherapy, type of resection (lobectomy or pneumonectomy), chest wall resection, tumor extension (localized or extended tumor) and perioperative blood transfusion. All data were gathered prospectively. A univariate analysis was performed using contingency tables for binary variables and analysis of variance for continuous ones; stepwise logistic regression analysis was then performed and the likelihood of death for each individual was calculated. A receiver operating characteristic (ROC) curve was constructed with the data, using surgical death as the state variable.

The following variables were found to be independently related to death in the univariate analysis: age (p < 0.001, odds ratio 1.11); tumor extension (p = 0.002; OR 3.47) and perioperative transfusion (p = 0.004; OR 3.87). The area under the ROC curve was 0.77, attributable to high specificity given that none of the complications could have been predicted.

Although some variables are related to surgical death, the described model is not able to give a prediction. Therefore, the model is of little use for application in making decisions about individual cases.