What are the limitations of modified intention-to-treat (ITT) analysis?

From the Guidelines

Modified intention-to-treat (mITT) analysis has several important limitations, primarily that it compromises the randomization process by excluding certain participants after randomization, which can introduce selection bias and undermine the balance of known and unknown confounders between treatment groups. The limitations of mITT analysis include:

• Compromising the randomization process by excluding certain participants after randomization, which can introduce selection bias and undermine the balance of known and unknown confounders between treatment groups 1
• Lack of a standardized definition of mITT, resulting in inconsistent application across studies and making cross-study comparisons difficult 1
• Exclusion criteria in mITT analyses are often subjective and can be manipulated to produce more favorable results
• mITT may not reflect real-world effectiveness since it typically excludes non-adherent patients or those who don't receive the intervention as planned, yet these scenarios commonly occur in clinical practice
• Regulatory agencies like the FDA generally prefer strict ITT analysis for primary efficacy evaluations because it preserves randomization integrity 1

Key Considerations

When implementing or interpreting mITT analyses, researchers should:

• Clearly pre-specify exclusion criteria in the study protocol
• Report both ITT and mITT results for transparency
• Consider the potential impact of exclusions on the validity and generalizability of findings

Implications

The use of mITT analysis can lead to overestimation of treatment effects compared to strict ITT analysis, and may not accurately reflect real-world effectiveness 1. Therefore, it is essential to carefully consider the limitations of mITT analysis and to use it judiciously, with clear pre-specification of exclusion criteria and transparent reporting of results.

From the Research

Limitations of Modified Intention-to-Treat Analysis

• The modified intention-to-treat (ITT) analysis has several limitations, including the potential for bias when excluding participants who do not initiate treatment 2.
• The approach assumes that the intercurrent event (e.g., not initiating treatment) is not affected by the assigned treatment arm, which may not always be reasonable 2.
• The modified ITT analysis may not provide an unbiased estimate of the treatment effect if the assumptions are violated, such as in open-label trials where knowledge of the assigned treatment arm may affect the participant's choice to begin or not 2.
• The use of modified ITT analysis has been associated with post-randomization exclusions, industry sponsorship, and favourable findings, which may indicate potential biases 3, 4.
• The approach may not be suitable for all types of trials, such as those comparing an active intervention to usual care, where it may be difficult to identify which participants would have initiated the active intervention 2.

Methodological Limitations

• The modified ITT analysis requires dichotomizing adherence into Yes or No, which may not accurately reflect the complexity of real-world treatment scenarios 5.
• The approach may not account for intermediate adherence, which can lead to biased estimates of the treatment effect 5.
• The use of modified ITT analysis may lead to inconsistent reporting, with some trials using ambiguous terminology or not clearly describing the approach 3.

Implications for Trial Design and Analysis

• Trial designers and analysts should carefully consider the limitations of modified ITT analysis and choose the most appropriate approach based on the research question and trial design 5, 2.
• The use of modified ITT analysis should be transparently reported, with clear descriptions of the approach and any post-randomization exclusions 3, 4.
• Sensitivity analyses should be conducted to assess the robustness of the results to different analytical approaches and assumptions 5, 2.