ABSTRACT
As stated above, in an ideal world the total population of health authority births would have been used as a denominator for a whole range of variables. However, having exhausted the potential of the data which is available for the total population, a control group provides an alternative sample population which can be used to generate further data for comparative purposes. Although the NPEU made this recommendation, no further guidance was given as to which variables to control for in the matching process between cases and controls. Surprisingly too, given the tremendous reliance on case-control trials in epidemiological research, there was very little guidance on the question of matching in the literature. In general, there appear to be an assumption that in perinatal research one controlled for factors such as age of mother, parity, social class and ethnicity. However, once these variables are matched between cases and controls they are then eliminated from any subsequent statistical analysis and this may mask important local idiosyncrasies. For example, I wanted to test out the assumption that the mothers whose babies died (the cases) were different from those whose babies survived (the controls) for a range of social, demographic and economic variables. If I had matched the social class, ethnicity, marital status and age of the control mothers with those of the case mothers, then I would have had no way of knowing whether there was any difference between the cases and the controls in these important respects. Indeed, because a vital aspect of any research lies in the nature of the questions that are selected for asking, if certain questions are eliminated by virtue of assuming a priori that the answers are self-evident then an important element of bias is introduced into the research infrastructure.
