Background: Classifications such as low birth weight, premature, and small for gestational age. i.e. compromised births, have been criticized because they depend upon arbitrary standards that may not be appropriate for all populations.
Aim: This study applies multivariate Gaussian mixture models with covariates to characterize birth weight by gestational age distributions.
Subjects and methods: The data consist of Asian, African, Hispanic and European American births in New York State in 1988. The analysis employs maximum likelihood methods.
Results: Birth cohorts appear heterogeneous and composed of at least two sub-populations. One sub-population accounts for the majority of births, has a higher mean birth weight and gestational age but small variances. The other sub-population has a lower mean birthweight and gestational age but very large variances. As a result of the large variances this sub-population accounts for compromised births. The model also suggests that a number of compromised births occur within the normal birth weight and gestational age range. Among normal births, birth weight increases and gestational age declines with maternal age. The effects on compromised births vary among populations.
Conclusions: Multivariate Gaussian mixture models provide a method of identifying compromised births that is not dependent upon arbitrary standards.