Prediction of chronic neonatal lung disease in very low birthweight neonates using clinical and radiological variables.

Abstract

There are good theoretical reasons for earlier intervention in neonates likely to develop chronic neonatal lung disease (CNLD). Very low birthweight (VLBW) neonates who receive artificial ventilation are at high risk of CNLD. A test was therefore developed to predict CNLD based on clinical and radiological information readily available at 7 days of age in VLBW neonates. Logistic regression analysis was used to identify those factors significantly and independently associated with CNLD. For each neonate it was possible to insert the value of the independent factors into the equation, providing a probability value between 0 and 1. By selecting different cut off values between 0 and 1, and knowing which neonates had developed CNLD, it was possible to assess the use of varying probability values as a predictive test for CNLD. The variation in these two parameters was graphically represented by a receiver operator characteristic (ROC) curve. The area under the ROC curve was used to represent the discriminatory capacity of the test over its full range of values. The maximum area under an ROC curve is unity. The area under the ROC curve was similar in a model with and without radiographic information (0.926 and 0.913 respectively) and was 0.937 in neonates from another hospital.