Prediction of in-hospital mortality following pulmonary resections: improving on current risk models.

OBJECTIVES

Using a large, prospectively collected and independently validated thoracic database, we created a risk-prediction tool for in-hospital mortality with the aim of improving on the accuracy of Thoracoscore.

METHODS

A prospectively collected and independently validated database containing lung resections was utilized, N = 2574. Logistic regression analysis with bootstrapping, and by the use of a random training and test set was utilized. Comparisons against the Thoracoscore, ESOS.01 and the Society of Thoracic Surgeons (STS) models were performed.

RESULTS

A logistic model identified age [odds ratio (OR) 1.1, 95% confidence interval (CI) 1.0-1.2, P = 0.0002], sex (OR 0.34, 95% CI 0.14-0.83, P = 0.02), predicted postoperative FEV1 (OR 0.96, 95% CI 0.94-0.99, P = 0.002), emphysema (OR 3.2, 95% CI 1.0-9.9, P = 0.04), excess alcohol consumption (OR 1.0, 95% CI 1.0-1.0, P = 0.04), pre-existing renal disease (OR 4.3, 95% CI 1.1-17.1, P = 0.04), predicted in-hospital mortality with an receiver operating curve (ROC) of 0.81 and a Hosmer-Lemeshow test of 0.9. Bootstrap analysis confirmed the above risk factors (ROC 0.82 and Hosmer-Lemeshow 0.2). Comparisons between Thoracoscore, ESOS.01 and the STS risk models demonstrated that none was very accurate, as all had low ROC values of 0.69, 0.70 and 0.61, respectively. The STS risk model does not apply to our population (ROC 0.61, Hosmer-Lemeshow, P = 0.004), and the ESOS.01 has poor predictive power (Hosmer-Lemeshow, P < 0.0001).

CONCLUSIONS

Logistic regression based on age, sex, predicted postoperative FEV1, alcohol consumption and pre-existing renal disease predicts in-hospital mortality with improved accuracy compared with the use of Thoracoscore, ESOS.01 and the STS risk model.