Controlled trials in single subjects.

S. Day

DOI: https://doi.org/10.1136/bmj.303.6801.522-a

1991-08-31

British Medical Journal

Abstract:error is not a major problem with the data on alcohol consumption. Furthermore, as none of the potential confounders other than cigarette smoking had any consistent effect on the odds ratios these other factors are unlikely to be important confounders despite inevitable measurement bias. The hypertension variable included in the analyses related to drug treatment, not the actual blood pressure, and was included because people being treated for hypertension mav have changed their alcohol consumption after the hypertension was diagnosed. In this study controlling for antihypertensive treatment had no consistent effect on the odds ratios. There is a strong social class association with cigarette smoking in New Zealand,4 and this, not selection bias, explains most of the social class association with coronary heart disease in our study. Professor A G Shaper and colleagues suggest that changes in alcohol consumption may have occurred before the period we specifically examined. This is irrelevant because we categorised current non-drinkers on the basis of lifetime drinking behaviour. The reliability of case-control studies depends on the design and conduct of the specific study. The validation studies we conducted were more extensive than those in most previous reports. Furthermore, despite the potential susceptibility of case-control studies to many biases they have the advantage of assessing both exposure and confounders close to the time of the outcome of interest and therefore may provide more accurate information than cohort studies entailing assessment of exposure often years before a disease event occurs. The issue of"more radical" exclusion criteria for prevalent cardiovascular disease is also dealt with in our paper. We report additional analyses restricted to people without current self reported medical problems, which show the same results. It would be inappropriate to evaluate the data presented in tables II and III statistically as these are raw figures. The statement that a complete lack of a dose response is biologically implausible is not only a weak argument but also false. We suggest a biologically plausible mechanism for the relation observed, although the study did not specifically examine biological mechanisms. As most of the participants drinking less than four units a week drank more than once a week the combination of a small high density lipoprotein effect and a haemostatic effect could account for the protective association observed in this group. Finally, our findings are consistent with those of most cohort studies that have examined this issue.'" The extremely high proportion of former drinkers in the non-drinking group in the British regional heart study conducted by Shaper et al5 suggests that their findings cannot be generalised. The weight of evidence suggests that the migration theory proposed by Shaper et al is wrong.