HomeScienceInterpreting the results of noninferiority trials—a review

Interpreting the results of noninferiority trials—a review


A treatment is either noninferior or ‘not noninferior’. Curiously, when clinical noninferiority is established, the new ‘noninferior’ treatment can be inferior to the comparator treatment in the conventional sense of being statistically significantly worse (Fig. 1). This confusing language occurs because a noninferiority comparison allows a noninferiority margin, and if this is made large enough, a treatment which is significantly less efficacious than the standard treatment can still meet the noninferiority criterion. Thus, a full evaluation should also consider if the new treatment is statistically inferior to the conventional treatment. When this is the case, it is particularly important to ask, ‘What is the evidence that the new treatment is more efficacious than no treatment?’ Failure to address these points is a serious omission and can lead to an unbalanced evaluation of the value of the new treatment [2].

Such an outcome arose in the TARGIT-A trial of intraoperative radiotherapy (IORT) for localised breast cancer [14,15,16]. Here, a 2.5% absolute difference in 5-year local recurrence was the pre-specified noninferiority margin. This noninferiority margin was met [13], but IORT led to more than twice the number of local recurrences when compared with conventional external beam radiotherapy (EBRT), and this difference is conventionally statistically significant (i.e., the null hypothesis of no difference in local recurrence rate would be rejected) for the predefined 5-year follow-up (24 vs 11 cases; 5 y Kaplan–Meier 2.33% vs 1.02%, difference 1.21% (95% CI (0.33–2.09%); binomial RR = 2.22, P = 0.024). Further, using all available follow-up (median 8.6 years), the absolute difference between treatments was substantially larger (60 vs 24 cases, 5.26% vs 2.07%, difference 3.19%) and this difference in proportions was highly significant (P = 0.00004) [[13], Table 3]. Thus, IORT was noninferior but conventionally inferior to EBRT in terms of 5-year local recurrence. However, in a secondary analysis, the authors report that when all-cause mortality is included for a composite endpoint of local recurrence-free survival, no significant differences were seen between treatment arms ([15], Fig. 3).

IORT could either be superior or non-superior to no treatment. Historically, local recurrence rates were high, and the addition of EBRT reduced 10-year local recurrence by almost two-thirds, from roughly 30–10%, in node-negative disease treated by breast-conserving surgery [17]. In their 2014 overview, Houssami et al. [18] report median local recurrence of 5.3% with a median of 6.6 years follow-up with whole breast radiotherapy. In more recent times, even lower rates have been achieved: 5-year local recurrence in TARGIT-A was just 0.95% in the conventional treatment arm.

This example makes clear that to focus only on absolute differences without a discussion of relative differences can be misleading. Thus, while a 2.5% absolute inferiority margin might have been reasonable when local recurrence was around 6% in patients receiving radiotherapy (i.e., 42% worse), as was the case when the trial was planned, it may not be reasonable when only 1% of conventionally treated patients had local recurrence at 5 years. Here, a rescaling to preserve the relative difference would roughly correspond to a 0.5% absolute noninferiority margin.


Source link



Please enter your comment!
Please enter your name here

Most Popular

Recent Comments

%d bloggers like this: