Methods for Handling Non-Compliance in Randomized Trials

Published in BMC Medical Research Methodology, the study assesses how four different summary measures of treatment effect perform when standard assumptions of trial adherence are violated. The work addresses a methodological challenge in clinical research, where non-compliance in randomised controlled trials may distort the estimate of treatment effect in intention-to-treat and per-protocol analyses.

The study adapts two established causal inference techniques, Inverse Probability of Censoring Weighting and the Rank Preserving Structural Failure Time model, to handle one-sided non-compliance. In the context of time-to-event outcomes, such as survival data, these methods are proposed to address treatment switches or deviations across trial arms. The authors specifically modify these models to account for one-sided non-compliance affecting the experimental arm.

For the Inverse Probability of Censoring Weighting approach, the methodology involves censoring non-compliant participants in the experimental arm at the specific time they deviate from the protocol. For the Rank Preserving Structural Failure Time model, the causal model was modified to reconstruct the counterfactual survival time for non-compliers in the experimental arm.

The researchers quantified the treatment effects using four distinct summary measures: the hazard ratio, the difference in restricted mean survival time, the ratio of restricted mean survival time, and the milestone survival difference.

To validate these adapted methods, the team conducted a rigorous simulation study alongside a real-world data analysis based on a randomized trial for nasopharyngeal cancer. The simulation results indicated that models related to the Rank Preserving Structural Failure Time approach generally produced smaller bias in estimating the treatment effect. In contrast, the Inverse Probability of Censoring Weighting method generally performed well under high non-compliance for all summary measures.

However, the study also highlighted trade-offs in statistical power. Under idealized proportional hazard settings, the estimates based on restricted mean survival time demonstrated lower power compared to the hazard ratio estimate.

When applied to the real-world nasopharyngeal cancer data, the different methods yielded varying estimates of the treatment effect depending on the summary measure used. The Rank Preserving Structural Failure Time models produced larger estimates for the difference and ratio of restricted mean survival time. On the other hand, the Inverse Probability of Censoring Weighting method generated larger estimates when the treatment effect was quantified using the hazard ratio and milestone survival difference.

The authors conclude that handling treatment non-compliance in randomized controlled trials requires careful consideration of both the statistical model and the summary measure.

Subject of Research: Methodological assessment of statistical methods for handling non-compliance in randomized controlled trials

Article Title: Implications of non-compliance in the experimental arm on treatment effect estimation in randomised controlled trials: a methodological assessment of four summary measures

Article References: Gu, T., Zhu, J., Li, J., Soon, Y. Y., Wee, J., & Tai, B.-C. (2026). Implications of non-compliance in the experimental arm on treatment effect estimation in randomised controlled trials: a methodological assessment of four summary measures. BMC Medical Research Methodology. https://doi.org/10.1186/s12874-026-03059-w

Image Credits: AI Generated

DOI: 10.1186/s12874-026-03059-w

Keywords: randomized controlled trials, non-compliance, treatment effect estimation, inverse probability of censoring weighting, rank preserving structural failure time, hazard ratio, restricted mean survival time, clinical methodology, nasopharyngeal cancer, statistical simulation, Implications, experimental