Fitting logistic, probit and extreme value models for the proportion of neonates dying from hyaline membrane disease

Abstract

In this thesis, generalized linear models (GLMs) are going to be considered in which the outcome variables (or categorical response variables) are measured on a binary scale. Binary response models are used for example in toxicology to describe the effect of a toxic chemical dosage on whether a subject dies. The data was obtained from the Department of Biomedical Sciences and compares the number of deaths of premature neonates in relation to length of gestation (completed weeks) and describes the use of surfactant treatment in reducing the overall neonatal mortality resulting from HMD (Hyaline Membrane Disease). It is our objective to find a suitable relationship between the probability that a neonate survives against the length of gestation for different years (1992 - 1994 and 1995 - 1997), whilst taking into consideration the effect of surfactant on the probability of neonates surviving during the years 1995 - 1997. From the data, one will be able to conclude that the longer the period of gestation, (i.e. the longer a baby stays in his mother's womb before being born), the greater is the chance for him to survive. Moreover, if the baby is born premature, (with a short gestation period), he is treated with surfactant (the treatment) which improves his chance of surviving. This is tested by running the chi-squared test for independence to determine whether there is any association between gestation period and state for the years 1992 - 1997 and then, by running another test, we investigate if surfactant played a major role in decreasing the number of dead neonates.