Direct effects, indirect effects, and mediated effects models' predictions of low birthweight
The problem was whether a typical prenatal assessment instrument used by a variety of nurses can gather information useful in identifying the at-risk client by placing variables into models to discover retrospectively the predictors of low birthweight (LBW) outcomes.
The procedure examined records of prenatal clients with LBW outcomes within a 5-year (1987-1991) period. Variables predictive of LBW outcomes were placed into models using path analysis. In addition, the final model was cross-validated with another set of prenatal client records who delivered LBW infants in 1992 and 1993.
The POPRAS risk assessment instrument was used to gather data. Demographic data were analyzed using descriptive statistics including Pearson product-moment correlation, chi-square test of homogeneity, and t-tests. Path analysis was utilized to evaluate relationships among LBW determinants obtained from the POPRAS and estimate the amount of variance explained by the models. Finally, an F test was used to compare the differences between models for significance at the.05 or better level.
The demographic characteristics of age, income, prior obstetric history, marital status, and race were not found to be significant with this group of clients. The variables, as reported by the client, of inadequate transportation, inadequate financial resources, no perceived support, conflict with partner, feeling alone, smoking, and inadequate food were not directly correlated with the lowest birthweight outcomes although these variables correlated with each other. Feeling depressed was the only variable that had a direct correlation with LBW outcomes in the first model. This model was more predictive of the variable feeling depressed than LBW. The direct effects model (FD with LBW) and the mediated effects model (IT, IFR, NPS, CWP, FA, FD with LBW) were not significantly different,