Evidence-based sizing of non-inferiority trials using decision models.

BACKGROUND

There are significant challenges to the successful conduct of non-inferiority trials because they require large numbers to demonstrate that an alternative intervention is "not too much worse" than the standard. In this paper, we present a novel strategy for designing non-inferiority trials using an approach for determining the appropriate non-inferiority margin (δ), which explicitly balances the benefits of interventions in the two arms of the study (e.g. lower recurrence rate or better survival) with the burden of interventions (e.g. toxicity, pain), and early and late-term morbidity.

METHODS

We use a decision analytic approach to simulate a trial using a fixed value for the trial outcome of interest (e.g. cancer incidence or recurrence) under the standard intervention (p) and systematically varying the incidence of the outcome in the alternative intervention (p). The non-inferiority margin, p - p = δ, is reached when the lower event rate of the standard therapy counterbalances the higher event rate but improved morbidity burden of the alternative. We consider the appropriate non-inferiority margin as the tipping point at which the quality-adjusted life-years saved in the two arms are equal.

RESULTS

Using the European Polyp Surveillance non-inferiority trial as an example, our decision analytic approach suggests an appropriate non-inferiority margin, defined here as the difference between the two study arms in the 10-year risk of being diagnosed with colorectal cancer, of 0.42% rather than the 0.50% used to design the trial. The size of the non-inferiority margin was smaller for higher assumed burden of colonoscopies.

CONCLUSIONS

The example demonstrates that applying our proposed method appears feasible in real-world settings and offers the benefits of more explicit and rigorous quantification of the various considerations relevant for determining a non-inferiority margin and associated trial sample size.