Birth-weight-adjusted infant mortality in evaluations of perinatal care: towards a useful summary measure.

Since birth-weight-specific infant mortality rates are widely considered to be measures of the effects of perinatal medical care, birth-weight-adjusted and birth-weight-specific infant mortality rates have often been used in comparisons across hospitals and geographic areas. Wilcox and Russell have provided a model which leads to the conclusion that birth weight adjustment is biased against populations with heavier birth weights and that birth-weight-specific infant mortality rates can yield misleading results. Nevertheless, evaluators and health planners still need a summary measure of the infant mortality rate in which some sort of birth weight adjustment is used to generate an (appropriately) weighted mean of birth-weight-specific relative risks. We used the 1983-85 national linked files of live births and infant deaths to investigate two new methods that extend the Wilcox-Russell approach, 'mean adjustment' and 'Z-adjustment', to compare birth-weight-specific infant mortality rates among the white populations of the states. Colorado was used as the reference population in logistic regression analyses. Statistical interactions between state and birth weight on the conventional kilogram scale, the mean-adjusted scale, and the Z scale were examined. There were substantial interactions with birth weight on all three scales. In addition, when Colorado was used as the reference population, mean adjustment shifted the odds ratios downward to implausibly low values, especially at lower weights. The Z-adjustment method incorporating the Wilcox-Russell approach appears to be a useful alternative to birth weight adjustment. However, because birth-weight-specific mortality rates do not differ uniformly across all birth weight groups, multiple summary measures are needed.