Hertz-Picciotto, Irva (1997). A New Insight into Pregnancy Loss. Epidemiology, 8(6)
In studies of reproductive outcomes, the presence of multiple pregnancies for the same woman introduces nonindependence among observations, a violation of most standard statistical modeling techniques. The work of Watier et al, reported in this issue, builds upon the vast literature addressing nonindependent observations that has emerged over the last decade and represents a significant methodologic advance for the epidemiology of spontaneous abortion. Watier et al propose a mixed model approach. The method extends beyond the examples given, since within-woman correlations in pregnancy outcomes arise whenever the data draw from an extended period of time. Where reproductive histories are collected retrospectively (a typical design in pregnancy outcomes studies), or prospectively for more than a year or so, multiple pregnancies per woman are almost certain to occur. These statistical techniques eliminate the need for perfectly independent observations and thereby render unnecessary the wasteful practice of selecting one pregnancy per woman for analysis.