1. As already noted, there were some problems in comparing cases and controls in that cases born before 1971 would be included only if diagnosed after 1970, and births in the 1980s would be included only if diagnosed at a relatively early age. We therefore carried out an analysis of only those babies born between 1971 and 1980, all of whom would have been followed for at least 10 years. We selected those who had had cancer diagnosed before their 10th birthday (89 cancers, 214 controls). Although neither year of birth (x2 = 38, df=4) nor hospital,(x2=
2. This study provides an example of the unexpected generation of a hypothesis as the result of a data trawl followed by the testing of that hypothesis on a different data set (none of the index cases in this study were included in the 1970
3. birth cohort (when it was almost certainly given entirely intramuscularly) had a similar odds ratio (2-6; 95% confidence interval 1-3;the possibility that the association may be causal.,1970
4. As Elwood has described, epidemiologists have identified several criteria which taken together would make it "more provident to act on the basis that the association is causal rather than to await further evidence."'4 The most important criteria, he notes, are biological plausibility as well as strength and consistency between studies. Many epidemiologists would describe an odds ratio of 2 as strong. The fact that two out of two studies have produced similar results provides consistency, though more studies would be even better
5. A recent British study identified cases in the British Isles during 1987-90.2' The authors showed that the risk varied from 4 4 per 100 000 in those given no vitamin K prophylaxis to 1-4 per 100 000 in those given oral vitamin K and 0 11 per 100 000 in those given intramuscular vitamin K. A similar study in Germany produced similar results,2,