Cambien F, Richard J L, Jacqueson A, Ducimétière P
Rev Epidemiol Sante Publique. 1981;29(3):281-8.
In a prevention trial, when the randomised units are groups of subjects, the first step in the analysis of the results is to check whether the groups differ, for the variables under study, ie, to test a possible group effect. If this effect is not significant, the results are analysed as if the subjects had been randomised. On the other hand, it this effect is significant, the comparison must be carried out between the groups and no more between the subjects. In the latter case, the loss of efficiency of the randomization of groups instead of the randomization of subjects can be computed. When a group effect is present and the number of subjects in each group differs, the analysis is considerably more complex. In the Paris cardiovascular prevention trial, 160 groups of young men, with variable numbers of subjects in each group have been randomised. The change in weight, blood cholesterol and cigarette consumption after two years of intervention are analysed in the present paper with the methodological principles mentioned above.
在一项预防试验中，当随机分组的单位是受试者群体时，分析结果的第一步是检查各群体在所研究的变量方面是否存在差异，即检验可能的群体效应。如果这种效应不显著，则按受试者被随机分组的情况来分析结果。另一方面，如果这种效应显著，则必须在群体之间进行比较，而不再在受试者之间进行比较。在后一种情况下，可以计算出群体随机分组而非受试者随机分组时随机化效率的损失。当存在群体效应且每组中的受试者数量不同时，分析会复杂得多。在巴黎心血管预防试验中，对160组年轻男性进行了随机分组，每组中的受试者数量各不相同。本文根据上述方法原则分析了干预两年后体重、血胆固醇和吸烟量的变化。
