In: Statistics and Probability
Margin of Acceptance for Non-inferiority Trial
Noninferiority trials are conducted when the underlying hypothesis is that the investigational treatment may not provide additional efficacy but may benefit patients or society in terms of quality of life and cost. For example, less toxicity should improve a patient’s quality of life and reduce the cost of supportive care; easier administration should be more convenient to patients, thereby improving their quality of life and reducing health service costs for administration; and a less expensive treatment would benefit society by reducing health-care costs. Demonstrating noninferiority shows that the new treatment has an acceptable level of efficacy to be adopted into clinical practice as an alternative to the current standard therapy. Use of such trial designs to study antineoplastic agents has increased in recent years; 17 noninferiority studies in patients with lung cancer were identified in a 2012 review.As drugs with better outcomes are produced, it will become more difficult to develop new drugs that are superior to existing drugs, so the objective of clinical trials is likely to change.
Key examples of noninferiority trials in lung cancer include the following: (a) a randomized phase III trial comparing pemetrexed with docetaxel for patients with advanced NSCLC previously treated with chemotherapy, which resulted in Food and Drug Administration approval for pemetrexed in this setting; (b) IRESSA NSCLC Trial Evaluating Response and Survival against Taxotere (INTEREST), a study in the same setting, but investigating gefitinib rather than pemetrexed; and (c) a randomized phase trial comparing cisplatin and pemetrexed with cisplatin and gemcitabine as first-line treatment for advanced NSCLC.
Special guidelines have been developed to address the unique challenges of designing noninferiority trials, as well as analyzing and interpreting data, and reporting the results.Because the crux of this type of trial is that the new treatment is noninferior to the standard treatment, defining what that means is a crucial aspect of the design and interpretation. The trial is meant to demonstrate beyond reasonable doubt that the new treatment is not worse than the currently accepted standard of care by a small, prespecified margin, called the noninferiority margin. There are various approaches to selecting this margin, such as the conventional method and the effect-retention method. The conventional method, also known as the fixed-margin method, is a subjective approach. A level of inferiority is selected that is considered to be of no clinical relevance or outweighed by other benefits of the experimental treatment. With the effect-retention method, also known as the percent-retention or putative-placebo method, a noninferiority margin is selected that ensures a substantial fraction (typically 50%) of the benefit demonstrated for the standard treatment over placebo is still retained by the new treatment. In a review of lung cancer trials, hazard ratios for the noninferiority margin ranged from 1.18 to 1.37.61 Ultimately, the value chosen has to be sufficiently small that the clinical community accepts the new treatment as the preferred option, given the other benefits demonstrated.
In terms of hypothesis testing, if a trial is set up to test the null hypothesis that there is no difference between treatments and the results are nonsignificant, then one can only conclude that the trial does not provide evidence of a difference between treatments. Although it is tempting to conclude that the treatments are equivalent, especially if the observed hazard ratio is close to 1, this conclusion is invalid. Therefore the hypotheses that are tested in a noninferiority trial are opposite to the hypotheses specified when testing for superiority. In a noninferiority trial, the null hypothesis is that the new treatment is inferior to the standard control treatment. The alternative research hypothesis is that the experimental treatment is not inferior to the standard control arm by the prespecified margin. The aim of a noninferiority trial is to collect data that provide sufficient evidence to reject the null hypothesis in favor of the research hypothesis; that is, if θ is the unknown true hazard ratio in the population, as specified previously, and k is the prespecified noninferiority margin, then the analysis will compare the null hypothesis that θ > k with the alternative hypothesis that θ < k. These hypotheses are one-sided, which is the statistical approach usually taken; however, two-sided hypotheses can be specified in a similar way to test for equivalence rather than noninferiority. The significance level and power for the trial are selected to minimize the chance of erroneously concluding that an inferior treatment is noninferior.
The final methodologic consideration in noninferiority trials is the choice of the population included for analysis. In traditional comparative designs, the criterion standard is the intent-to-treat population, which includes all patients within their randomly assigned treatment allocation, despite what treatment they may have actually received. The intent-to-treat population is a conservative analysis strategy for assessing potential treatment benefit. Sole use of this approach in a noninferiority trial may dilute a clinically important difference between treatments, leading to an incorrect conclusion about noninferiority. Therefore the recommendation is to perform an additional analysis of the per-protocol population, which includes only the patients who have received a prespecified minimum level of treatment. Guidelines state that a robust interpretation of noninferiority can be achieved only if both analyses lead to similar conclusions and that any differences between the intent-to-treat and per-protocol analyses need careful and close examination.