A linear model suitable for assessing graft patency in controlled clinical trials

Abstract In clinical trials carried out to assess the efficacy of different drugs in reducing the frequency of occlusion after coronary artery bypass, the ratio of the number of patients with at least one occluded anastomosis to the number of patients catheterized up to a given day is a widely adopted statistic. In the early evaluation (at 1 or 2 months after surgery), this is affected by the distribution of timing of angiography and it tends to underestimate the cumulative probability of occlusion because patients whose anastomoses are all patent at angiography and occlude between angiography and the day at which the ratio is estimated do not contribute their events to the numerator of the ratio. One may sensibly assume that this underestimate does not affect the evaluation of the efficacy of treatments tested “within” the trial. In contrast, since the distributions of the timing of angiography vary substantially from trial to trial, it can make comparisons “between” trials unclear and possibly biased. The aim of this article is to suggest an alternative approach of statistical analysis in terms of logistic regression. By modeling the dichotomous response given by each patient in function of time at angiography, type of treatment, and other possible covariates, asymptotically unbiased estimates of the cumulative probability of occlusion are attained. Furthermore, the pertinent hazard function can be estimated. The main features of the model are discussed and the results obtained by fitting early data collected in the Studio Indobufen Nel Bypass Aortocoronarico (SINBA) are given.