The advent of networks of centers to encourage collaboration in randomized trials both in North America and internationally (for example, the British Association of Perinatal Medicine Perinatal Clinical Trials Group, the Canadian Perinatal Clinical Trials Network, and the Perinatal Trials Service of the National Perinatal Epidemiology Unit)¹ has been a major innovation in recent years. Whether these networks have increased the number and improved the quality of neonatal trials has not been formally evaluated but the signs are surely encouraging.

Data collected by these networks are often voluminous, and the temptation to analyze ancillary data collected during the conduct of trials is compelling. One of the best-known networks, the National Institute of Child Health and Human Development (NICHD) Neonatal Research Network, has already published two sequential papers based on patients entered from 1987 through 1988² and 1989 through 1990,³ and, in this issue of Pediatrics, data from the largest network, the Vermont Oxford Network, are reported.⁴ It is not, of course, the randomized comparisons that are typically analyzed in these secondary reports. With respect to these analyses, the fact that the centers were brought together to conduct randomized trials is irrelevant. These cross-sectional secondary analyses of large data sets are almost never hypothesis-driven and use data neither planned nor collected for purposes of the current analysis. Are these, then, simply large data-dredging exercises, or can useful observations be made from them?

Observational studies can provide interesting insights into iatrogenic and other institutional and environmental influences on neonatal mortality. For example, national data from England and Wales, and Norway both point to the increased perinatal mortality owed to being delivered in . . . [Full Text of this Article]