From the Guidelines
The use of a modified intention-to-treat (mITT) analysis can lead to an overestimation of efficacy in clinical trials. Unlike standard intention-to-treat analysis, which includes all randomized participants regardless of protocol deviations, mITT typically excludes certain participants after randomization, such as those who never received treatment or failed to meet specific post-randomization criteria. This selective exclusion can introduce bias by removing participants who might have had poorer outcomes, thereby artificially inflating the apparent effectiveness of the intervention. The magnitude of this overestimation varies depending on which participants are excluded and why. For example, excluding non-compliant patients may create a scenario where only those most likely to benefit from treatment remain in the analysis. This is particularly problematic when evaluating medications where adherence affects outcomes, as noted in a study on diabetes prevention, which suggested that reported effects may be overestimated due to missing data and non-adherence 1.
To minimize bias, researchers should clearly pre-specify mITT criteria in study protocols, report results using multiple analytical approaches (including standard ITT), and transparently discuss the potential impact of exclusions on efficacy estimates. Clinicians interpreting studies using mITT should carefully consider which patients were excluded before applying findings to their broader patient population. The importance of considering the analysis population and handling missing data appropriately is highlighted in recommendations for improving clinical trial design and analysis, including those for hospital-acquired pneumonia and ventilator-associated pneumonia trials 1. Furthermore, the choice of analysis population, such as modified intent-to-treat, should preserve the protections of randomization from bias, especially when the number of exclusions is relatively small, as discussed in the context of noninferiority trials 1.
Key considerations for researchers and clinicians include:
- Clearly defining and justifying the use of mITT analysis
- Reporting results from multiple analytical approaches to provide a comprehensive view of the intervention's efficacy
- Transparently discussing the potential for bias introduced by exclusions in mITT analysis
- Considering the implications of missing data and adherence on the estimated efficacy of an intervention, as suggested by studies such as the one on emerge guideline on medication adherence 1.
By acknowledging these considerations and taking steps to minimize bias, the use of mITT analysis can be optimized to provide more accurate estimates of efficacy, ultimately informing better clinical decisions and improving patient outcomes.
From the Research
Analysis of Modified Intention-to-Treat (mITT) Approach
- The use of a modified intention-to-treat (mITT) analysis may lead to an overestimation of efficacy in certain cases, as it excludes participants who do not initiate treatment 2.
- However, studies have shown that mITT analysis can be an unbiased estimator for the principal stratum estimand under certain assumptions, such as when the intercurrent event is not affected by the assigned treatment arm 3.
- A systematic review and meta-analysis found that mITT analysis over-estimated efficacy by up to 8-12% compared to intention-to-treat (ITT) analysis in chimeric antigen receptor T-cell therapy trials 2.
- In contrast, another study found that mITT analysis did not introduce bias compared to ITT analysis in rheumatoid arthritis trials investigating biological or targeted interventions 4.
Key Considerations for mITT Analysis
- The assumptions necessary for mITT analysis to be unbiased, such as the intercurrent event not being affected by the assigned treatment arm, must be carefully evaluated 3.
- The decision to use mITT analysis should be based on the specific research question and study design, and should be clearly justified and reported 5, 6.
- The potential for bias and overestimation of efficacy should be considered when interpreting the results of mITT analysis, particularly in cases where participants are excluded from the analysis 2.