Knowing the effect model of a therapy is useful in predicting the effect in non studied patients, i.e. in delineating the therapy target population. In some instances, the effect model provides evidence that the treatment is hazardous in low risk patients (class I antiarrhythmics, in post-MI patients, oral anticoagulant in patients with atrial fibrillation) However, there is no established approach to figure out the effect model but to explore the summarised results of a number of trials, as a by-product of a meta-analysis. We propose here an approach that can be used with the data from a single trial or a pooled individual records database. It consists in rearranging cases in N pseudo-trials, taking advantage of a block randomization (block-size = R), with N = n/kR (n = trial sample size ; kR: size of the pseudo-trials). Hence, both the randomization and the balance over time allocation are kept. Regression techniques or Macintosh's approach are used to explore Rt = f(Rc), where Rt is the outcome frequency in treated patients and Rc in the outcome frequency in control patients. Jackknife can be used to test the robustness of the findings. An example of such an approach is given with the EMIP-FR trial (n = 19725). EMIP-FR was a trial testing an antihoxydant intervention in acute MI patients. A strategy for choosing N is also examplified with these data.