>>Return to Tell Me About Statistics!
In medical articles, we often see the results of trials called “non-inferiority trials.” This is because, in recent years, it has become more difficult to conduct comparative studies with placebo or no treatment groups due to ethical issues, and as a result, trials are often conducted using the standard treatment at the time as a control group. Even if a new treatment does not clearly provide superior efficacy, it may have clinical benefits, such as reduced side effects or significantly fewer drug doses.
In the previous issue (vol. 9), we discussed equivalence trials. In superiority trials, p-values and significance levels are used to determine efficacy, equivalence trials use confidence intervals and equivalence margins, and non-inferiority trials use confidence intervals and non-inferiority margins.
In this issue, we will explain non-inferiority trials.
Distinction between superiority, equivalence, and non-inferiority trials
・Superiority Trials: Trials that demonstrate that the intervention group is more effective than the control group.
・Equivalence Trials: Trials demonstrating that the efficacy of the intervention group is equivalent to that of the control group.
・Non-inferiority Trials: Trials demonstrating that the intervention group is not inferior to the control group in efficacy.
What is a non-inferiority trial?
In a non-inferiority study, for example, if a new drug is compared to a conventional drug, it is assumed that the new drug has more clinical benefits than the conventional drug, such as fewer side effects. This type of study is used in research where, even if efficacy superiority to a conventional drug cannot be demonstrated, it is sufficient to prove that the drug is not inferior to the conventional drug.
If equivalence is indicated, both the upper and lower limits of the confidence interval must fall within the width of the equivalence margin (Figure 1).
[Figure1] Confidence interval and equivalence margin
However, whereas the margin of equivalence is two-sided, non-inferiority trials focus only on one side of the confidence interval (upper or lower limit) (Figure 2). This is because non-inferiority trials focus only on whether the new drug is non-inferior to the conventional drug, and the superiority of the new drug over the conventional drug may or may not hold.
[Figure2] Confidence interval and non-inferiority margin
As is customary, to tighten the criteria, non-inferiority trials often take the 97.5% confidence interval instead of the usual 95% confidence interval.
Cautions when Interpreting Non-Inferiority Trial Results
There are two factors to be aware of when interpreting non-inferiority trials: analytical sensitivity and setting the non-inferiority margin.
Analytic sensitivity refers to “the ability to determine which treatments are ineffective and which are effective.” Analytical sensitivity is reduced in poor-quality trials, and the difference between the intervention and treatment groups will be small.
For example, if both the new and conventional drug groups had very poor adherence and took very few medications, the difference between the groups would be very small. In such a case, a superiority trial would not be able to show a difference, and it would be difficult to show that there was a significant difference. In contrast, in a non-inferiority trial, a smaller difference would instead favor a result showing non-inferiority.
This means, even if the new drug is inferior to the conventional drug by more than a meaningful margin, the poor quality of the trial may prevent a firm determination of the difference, resulting in a false non-inferiority result.
Next, what happens if the range (value) of the non-inferiority margin is large?
The non-inferiority margin works as a handicap in statistical analysis. The larger the handicap, the easier it is to prove non-inferiority, and the smaller the handicap, the harder it is to prove non-inferiority.
The non-inferiority margin is often set based on past studies and other factors and must be set prior to the trial, just as the margin is set for equivalence studies.
Thus, the results of superiority trials can be interpreted objectively, but the interpretation of non-inferiority trials involves some assumptions and subjective factors, such as the assumed existence of analytical sensitivity and how the non-inferiority margin width is set.
This is the biggest difference between a superiority trial and a non-inferiority trial. When reading an article, it is important to not just look at the results of the non-inferiority trial but also to read whether the non-inferiority margin was set objectively, before the trial.
Points to keep in mind when interpreting non-inferiority test results
In a non-inferiority trial, if the non-inferiority margin Δ = 0 is set, it corresponds to the criteria for a superiority trial (Figure 3).
[Figure3] Confidence intervals and verification of non-inferiority margin and superiority
So, can superiority be verified in a non-inferiority trial?
Not only can non-inferiority be proven beyond -Δ in a non-inferiority trial, but it can also exceed +Δ. In the field of oncology, for example, there are an increasing number of trials that are non-inferiority trials, but once non-inferiority is proven, the superiority is subsequently verified.
However, since the purpose of a clinical trial is to test a hypothesis that has been previously examined, it is usually unacceptable to claim superiority from a trial that tests a non-inferiority hypothesis.
In such cases, it is still considered necessary to replicate the superiority by other trials in order to clinically accept the superiority.
>>Return to Tell Me About Statistics!
Comments are closed