Predictors and statistical models in congenital diaphragmatic hernia

Despite advances in intensive care, congenital diaphragmatic hernia (CDH) maintains a risk of death >35%. Mortality predictors have been claimed, using logistic regression. When the outcome of interest is rare (i.e. <20%), the odds ratio, measured by logistic regression, is approximately equal to the relative risk, calculated by log-binomial model. However, for common events, the odds ratio misleads the exposition risk. The aims are to identify independent predictors of mortality in high-risk CDH, using the log-binomial model and disclose if the exposition risks could differ applying in comparison the logistic regression. Details of 113 consecutive high-risk CDH neonates, baseline demographics and disease features were collected retrospectively. Log-binomial model and logistic regression were applied and compared. Overall mortality rate was 41.6% (47/113). The log-binomial model identified preoperative pneumothorax (pnx), birth weight < or = 2,500 g and liver herniation as independent mortality predictors; female gender, an advanced gestational age and a PaO(2) > or = 90 mmHg as protective factors. Using logistic regression, liver herniation and birth weight did not maintain the significance. The exposition risks for pnx, female gender and gestational age were overestimated. The odds ratio measured by logistic regression overestimated the exposition risks. Since the mortality rate is confirmed to be >20% and the exposition risks, measured by logistic regression, are misleading, the log-binomial model should be consider in CDH binary outcome studies. According to the mortality predictors identified, making every effort to plan delivery at term and lowering the risk of pnx could improve the outcome.