Non-Inferiority Margin Setting from Indirect Comparison

Introduction: Non-inferiority trials (NI) test the efficacy of an experimental treatment in comparison to an active-controlled treatment and indirectly with the historical placebo to demonstrate that the new treatment is no worse than the active comparator. Setting the NI margin depends on the assumptions of constancy, assay sensitivity and the absence of placebo creep and bio-creep. Research Question: This PhD research will investigate the changes in the efficacy of the placebo and active control over time. It will show how this could affect the setting of the NI margin and the conclusion of non-inferiority. The context is where there is a wish to make a retrospective indirect comparison of the experimental treatment with historical placebo. Methods: An overview of Cochrane reviews of placebo-controlled trials was conducted to measure the correlations between the placebo, active treatment and the treatment difference with the year of publication. From the constructed dataset from the Cochrane reviews, a weighted regression model was built to investigate factors affecting the estimate of the future trial from a meta-analysis of historical trials, followed by proposing a method for the use of meta-regression to adjust for time while setting the NI margin. Results: The correlations between the placebo, active treatment and treatment difference and the year of publication varied from strong negative to strong positive correlations. The median correlation for the treatment difference = - 0.1. The estimate of any future trial could be predicted from a meta-analysis of historical trials with a coefficient of 0.92 and range from 0.75 to 1.047 of the historical trials. Moreover, increasing the year of prediction and increasing the year difference in the meta-analysis will reduce the predicted estimate by 0.015 and 0.005 respectively. Pairwise meta-regression and network meta-regression can be used to assess the constancy, to set the adjusted non-inferiority margin and to analyse the non-inferiority trial when the constancy assumption does not hold. Conclusion: In NI trials, the constancy assumption needs to be assessed not assumed. Adjusting for the time will reduce the chance of the conclusion of non-inferiority of an inferior test treatment regardless of the constancy assumption.