LAUSR

This issue of Clinical Trials presents noteworthy insights into protocol adherence and hypothesis confirmation in non-inferiority randomized trials. Noninferiority trials usually compare two interventions with similar expected efficacy, but possible differences in convenience or cost. A typical example in oncology is the comparison between an oral and an intravenous agent from the same class. In these situations, it is inappropriate to declare non-inferiority on the basis of a non-significant P-value arising from a superiority comparison; rather, non-inferiority has to be demonstrated formally by showing that the experimental treatment is not worse than control by more than the prespecified non-inferiority margin in the comparison involving the primary endpoint. Therefore, the choice of the non-inferiority margin is probably the most critical aspect in the design and interpretation of noninferiority trials. However, other aspects of noninferiority trials may play a differential role in their conduct and interpretation, when compared with superiority trials. One such aspect is the rate of patient adherence to the protocol. In superiority trials, lack of exposure to treatment or outcome assessment is expected to bias an intention-to-treat (ITT) analysis toward the null hypothesis, which is considered as appropriately conservative. Conversely, bias in that same direction would be seen as anti-conservative in non-inferiority trials. Bamat et al. compare the rates of protocol adherence and hypothesis confirmation between superiority and non-inferiority trials published in three highimpact medical journals over a period of 10 years. Among nearly 3000 parallel-group, two-arm trials with an active control and more than 50 patients randomized (10% of which non-inferiority trials), they selected for analysis 120 trials of each type (superiority and non-inferiority). This sample size gave the study enough power to test the hypothesis that protocoladherence rates would differ by 5% in absolute terms between the two trial types. Contrary to their hypothesis, adherence rates did not differ significantly between superiority and non-inferiority trials (medians of 91.5% and 89.8%, respectively), a result that was corroborated by multivariable analysis using a sensibly small number of covariates. Notwithstanding the plausibility of their initial hypothesis, these findings are reassuring. However, they are not surprising if one considers the potential role played by patients in adherence rates. In terms of the interventions being compared, the randomized arms of a non-inferiority trial are likely to be viewed by patients as a priori comparable; conversely, in superiority trials, at least if they are open-label (usually the case in oncology and arguably the majority in the current work), patients may have more incentive to drop out because of a differential perception of efficacy and safety between experimental and control arms. Hence, the finding of similar dropout rates in both types of trials is rather good news, even if the median overall rate of non-adherence around 10% may be considered somewhat high. Another finding from the study by Bamat et al. is that non-inferiority trials have a significantly higher success rate (i.e. rejection of the null hypothesis). Reassuringly, such a higher rate was not associated with the adherence rate. Once again, the higher success rate in non-inferiority than in superiority trials is not surprising (incidentally, they reproduce our own findings some years ago in a more limited setting). While all of the potential explanations raised by the authors may have played a role in these findings, it may well be that non-inferiority trials are only undertaken if they have a very reasonable chance of success, in contrast to superiority trials that are often conducted based on evidence from undersized or poorly controlled phase 2 trials, or even in some cases immediately after phase 1 trials. The stakes and financial rewards are generally far higher if a superiority trial of an experimental agent succeeds, as compared with a non-inferiority trial. There is still some controversy about whether the ITT or the ‘‘per-protocol’’ (PP) population should be