The risk of low birth weight. Alternative models of neonatal mortality

The authors examine the use of the infant's weight at birth to estimate its risk of mortality by the 28th day of life. The performance of several commonly employed statistical models is compared for the population of single births to resident South Carolina mothers in 1975-1980. A log-linear function, fitting the natural logarithm of the probability of neonatal mortality to birth weight, performs far better in this analysis than a quadratic or a logistic model or a model using the logarithm of the probability of survival. The parsimonious log-linear model also appears preferable to more complex models with additional parameters. The authors use the model in an analysis of data in two-year periods to demonstrate its value as an indicator of underlying changes in neonatal prospects and its easily interpreted parameters. The results highlight the importance of changes in neonatal mortality affecting low birth weight infants, which produce a noticeable shift in the range in which low birth weight is associated with the risk of mortality. These have proven more important in accounting for the decline in mortality rates than have changes affecting only neonates with more typical birth weights in South Carolina in 1975-1980.