Historically, the overall risk of infant mortality in the United States has declined. However, the rate of decline has slowed and even reversed in 2002 the most recent year reported. Further, racial and ethnic disparities have increased. Consequently, infant mortality remains a serious public health problem. Moreover, the cause of the racial and ethnic differentials has remained elusive. It is our view that effective theoretical tools, e.g. the proximate determinants model of infant mortality, are available, but that the statistical tools to fully operationalize these models are not. Ideally, these tools should accommodate direct and indirect (through the proximate determinants) effects, allow for non-linear effects, appropriately parameterize the proximate determinants (the most important of which are birth weight and gestational age), and account for """"""""hidden"""""""" heterogeneity in the birth cohort. The objective of this proposal is to continue development of Covariate Density Defined mixtures of logistic regression (and other GLMs) a statistical methodology that can operationalize the proximate determinants model. The objective of our previous work was to validate the statistical model, examine a number of issues concerning the structure of the proximate determinants component, and demonstrate the utility of the model for studying exogenous covariates. We will extend these results here to test specific assumptions concerning the """"""""causal"""""""" roll of birthweight with respect to infant mortality, i.e. how the """"""""structure"""""""" of the proximate determinants interacts with exogenous covariates and infant mortality. The """"""""causal"""""""" roll of birth weight has been questioned, an issue of extreme importance for designing interventions. The method we have developed is capable of statistically testing these assumptions. We will examine a number of potential covariates in the context of the proximate determinants model on a sample of ethnically diverse birth cohorts. We will develop and make available programs for conducting Covariate Density Defined mixtures of GLMs.
Showing the most recent 10 out of 12 publications
