Estimating odds ratios through the connected comparative experiments

In Japan rarely repeated are the comparative clinical trials with the same pair of the test and control drugs. The simple approach fur a complete iaudomized block design regarding the trials as blocks cannot therefore be applied It; strengthening the evidence uf the difference between the two drugs. In tins paper a method is discussed lo KTOVCI minimal lull hum the ruliliei ted trials in which those two drugs are respectively compared with the common third drug. The method consists of testing homogeneity of trials, forming a combined est illicit oi uf the n odds latius if the homogeneity is yeiified aiid giving the asymptotic variance of the combined estimator.In particular an approach of the multiple comparisons is taken so as to give a homogeneous subset of trials when the overall homogeneity is not satisfied. Although the paper is motivated from the comparative clinical trials, the resulting method can be applied to general incomplete block experiments if the outcomes are binoinial variables and the nil-interaction hetn-ern thc trtatnlent and block is suspicious. On the other hand in the appplications to clinical trials the other. non mathenlatical conditions for a meta-analysis such as the coincidence of tllc protocol should be satisfied as discussed by Hedges and Olkin(1983).