Using Generalized Linear Models in Analysis of Risk Factors to Infant Death Considering from Birth Weight of Infant

Abstract

The analysis of risk factors to infant death considering from birth weight of infant has been performed by classifying the birth weight into 3 categories: risky group, monitored group, and normal group. The risk factors obtained from literature reviews consist of mother age, smoking during pregnancy, premature labor, historical hypertension, and uterine irritability from the real data set of 189 babies (Hosmer and Lemeshow, 1989). The models under consideration are cumulative logit model, logit model for binary response, loglinear model, and Poisson regression model compared with generalized Poisson regression model. The results reveal that every model provides significance at 0.05 for goodness-of-fit test. The three risk factors affecting the response of cumulative logit model and the logit model are the same: premature labor, historical hypertension and uterine irritability. From loglinear model, history of hypertension and presence of uterine irritability are obtained. From Poisson regression model, the age of mothers is obtained. Hence almost all risk factors under studies, except for the smoking during pregnancy, are statistically affecting the birth-weight risk of death significantly at 0.05 with a highly predicted probability of a corresponding model.