Assessing the influence of treatment nonadherence on noninferiority trials using the tipping point approach

In noninferiority (NI) trials, an ongoing methodological challenge is how to handle in the analysis the subjects who are nonadherent to their assigned treatment. Some investigators perform the intent-to-treat (ITT) as the primary analysis and the per-protocol (PP) analysis as sensitivity analysis, whereas others do the reverse since ITT results may be anticonservative in the NI setting. But even when there is agreement between the ITT and PP approaches, NI of the experimental therapy to the comparator is not guaranteed. We propose that a tipping point method be used to further assess the impact of nonadherence on the results of a NI trial. In this approach, data from the nonadherers obtained after treatment discontinuation is not used, and their outcomes under the counterfactual situation of complete adherence are considered missing. The tipping point analysis indicates how sensitive the NI trial results are to the values of these missing counterfactual outcomes. The advantages of this approach are that a model or mechanism for the missing outcomes does not have to be assumed, and all subjects who were randomized are included in the analysis. We consider both binary and continuous outcomes and propose extensions to accommodate different types of nonadherence. The methods are illustrated with examples from two NI trials, one to evaluate different doses of radiation therapy to treat painful bone metastases and the other to compare treatments for reducing depression in adolescents.