Variability of birth-weight distributions by sex and ethnicity: analysis using mixture models

Birth weight is the most important proximate determinant of the level of infant mortality. However, the association between birth weight and infant mortality is not constant among populations. For example, the mortality of African American infants is lower at low birth weight but higher at high birth weight compared with European American infants. One possible explanation is that birth cohorts are heterogeneous even after controlling for birth weight, ethnicity, sex, and multiple births. The analyses presented here use Gaussian mixture models to explore the interpopulation variation in the shape of the birth-weight distribution for evidence of intrapopulation heterogeneity. The results suggest that a two-component mixture model provides an excellent description of human birth-weight distributions. Further statistical analyses of sex and ethnic differences indicate (1) that the birth-weight distributions and heterogeneity within the distribution vary between the sexes and among ethnic groups and (2) that one specific component is more closely associated with the overall level of infant mortality. The results support the hypothesis that birth cohorts can consist of two or more subpopulations at differential risk of mortality. Differences in the subpopulation composition of birth cohorts (i.e., differences in the level of heterogeneity among the various ethnic groups) might partially explain the interethnic variation in birth-weight-specific mortality. Further development of these mixture models should provide important additional information concerning the biological, environmental, and social determinants of birth weight and infant mortality.