The lay abstract featured today (for Assessing efficacy in non-inferiority trials with non-adherence to interventions: Are intention-to-treat and per-protocol analyses fit for purpose? by Matthew Dodd, James Carpenter, Jennifer A. Thompson, Elizabeth Williamson, Katherine Fielding and  Diana Elbourne) is from Statistics in Medicine with the full Open Access article now available to read here.

Dodd M, Carpenter J, Thompson JA, Williamson E, Fielding K, Elbourne D. Assessing efficacy in non-inferiority trials with non-adherence to interventions: Are intention-to-treat and per-protocol analyses fit for purpose?. Statistics in Medicine. 2024; 1–18. doi: 10.1002/sim.10067

Abstract

In non-inferiority trials, participants are typically assigned at random to receive either a new intervention or a control intervention that is already known to be effective. The aim is to show that the new intervention is not worse (i.e. similar or better) than the control intervention with regards to the outcome of interest. This is because it is offers some other advantage, such as being cheaper, producing fewer side effects, or offering greater convenience. In these trials, some participants may not receive their assigned interventions as planned which is known as ‘non-adherence’ to the interventions. For example, in the context of a drug regimen for treating tuberculosis (TB), some participants may miss several doses of medication or discontinue the regimen completely.

Intention-to-treat (ITT) analyses, which include all participants within their randomly assigned groups, and per-protocol (PP) analyses, which typically exclude participants deemed to be non-adherent to the interventions, are performed frequently in non-inferiority trials. However, there is growing evidence that these approaches can potentially result in bias in the presence of non-adherence to interventions. For instance, non-adherence may result in trial arms appearing more similar than they should, potentially increasing the risk of incorrectly concluding that the new intervention is not worse than the control intervention.  

Computer simulations were used to compare the performance of different statistical approaches for handling non-adherence to interventions in non-inferiority trials. The simulations replicated a real non-inferiority trial comparing a standard 6-month TB treatment to a shortened 4-month TB treatment.¹ The statistical approaches used included ITT and PP analyses, multiple imputation of outcomes, inverse-probability-of-treatment weighting (IPTW), and a doubly-robust estimator.

The results showed that ITT and PP analyses can lead to an increased risk of incorrectly concluding that the new intervention is not worse than the control intervention or, alternatively, an increased risk of missing an effective intervention, depending on the observed patterns of non-adherence. In contrast, the multiple imputation, IPTW, and doubly-robust approaches were able to correct for the impacts of non-adherence under most scenarios.

In conclusion, the study showed that when non-adherence patterns differ between arms in non-inferiority trials, ITT and PP analyses can potentially lead to the acceptance of inferior interventions into clinical practice or effective interventions being missed. We recommend that IPTW or the doubly-robust estimator are used to supplement ITT and PP analyses to guard against this problem.

References:

1. Gillespie SH, Crook AM, McHugh TD, et al. Four-Month Moxifloxacin-Based Regimens for Drug-Sensitive Tuberculosis. New England Journal of Medicine. 2014; 371(17):1577-1587.