PLAN DE RESCATE Y REUBICACIÓN DE FAUNA:

The right choice of the analysis population is another crucial point in two-arm non-inferiority trials that is closely related to the assessment of assay sensitivity. However, this issue has not

1.3 METHODOLOGICALPROBLEMS 13

been discussed in the literature to the same extent as the previously mentioned issues. In gen- eral, there exist two types of analysis sets, the full analysis set (FAS) on the one hand and the per protocol set (PPS) on the other.

The FAS is defined according to the intention-to-treat (ITT) principle mentioned in the ICH E9 guideline, where the treatment effect is evaluated on the basis of the intention to treat a patient instead of the actual treatment given. The FAS is as close as possible to this ideal and is generated by the set of all randomised patients with only minimal and justified eliminations defined prior to the trial. For instance, common reasons for an exclusion from the FAS are a failure to take at least one dose of study medication or violations of major inclusion criteria. The use of the FAS preserves the value of randomisation and, moreover, provides results that are more likely to reflect reality.

The PPS is derived as the subset of patients in the FAS who sufficiently complied with the study protocol and is often characterised by the following criteria: 1. Certain minimal exposure to the treatment; 2. Available measurements of the primary variable(s); 3. No major protocol violations such as violations of entry criteria. The main advantage of an analysis based on the PPS is the ability to estimate the drug’s efficacy potential under optimal conditions. However, the exclusion of patients who do not adhere to the study protocol breaks the randomisation, so that a per protocol analysis can be biased considerably. Depending on the relationship be- tween adherence to the study protocol and treatment or outcome, this bias can be in both directions. Most often, however, a per protocol (PP) analysis leads to over-optimistic estimates of the treatment effects, as ‘problematic’ patients tend to be excluded from the PPS.

In superiority trials the FAS is commonly accepted as the primary analysis set of choice, as an ITT analysis provides a conservative analysis approach. Non-compliers will generally diminish the difference between the two treatment groups, resulting in a bias towards the null hypothesis of no treatment difference. Thus, it will usually be more difficult to demonstrate superiority with an ITT analysis than based on the PPS. In contrast, there is still no consensus on the role of the FAS in non-inferiority trials. In the ICH E9 guideline it is stated that “in an equivalence or non-inferiority trial the use of the full analysis set is generally not conservative and its role should be reconsidered very carefully”. This has often been mistakenly interpreted to mean that the PPS is a conservative choice and should be the primary analysis set in non-inferiority trials. However, analysing a non-inferiority study based on the PPS is not conservative per se, as e.g. major protocol deviations might be related to the treatment or outcome. For this reason the CPMP (2000) recommends that “in a non-inferiority trial, the full analysis set and the PP analysis set have equal importance and their use should lead to similar conclusions for a robust interpretation”. But this strategy also not necessarily guarantees valid conclusions, as it was shown by Sanchez and Chen (2006). Their simulation study revealed that analyses based on the FAS and the PPS in non-inferiority studies can be both conservative and anti-conservative, depending on the types of protocol violations and missingness. They proposed that a so-called

addresses the missing data as in an ITT analysis, would result in more reliable study results. Some other interesting issues regarding the choice of the analysis set in non-inferiority trials have been mentioned by Wiens and Zhao (2007). They think that the justifications to use the FAS in superiority trials also carry over to non-inferiority trials. For instance, they argue that an analysis based on the FAS preserves the value of randomisation and estimates the “real-world” effectiveness. Moreover, the use of different analysis sets for superiority and non-inferiority comparisons could lead to inconsistencies. The question arises whether an α-adjustment is necessary for a subsequent superiority test, based on the FAS, after non-inferiority has been demonstrated, based on the PPS. The adequate handling of missing data is also closely related to the right choice of the analysis population, as it was also mentioned by Sanchez and Chen (2006). This topic has become more and more important in the recent years, but publications regarding this matter almost exclusively deal with the superiority objective. Yoo (2010) con- ducted a comprehensive simulation study to investigate the impact of different types of miss- ingness on six different statistical analyses in a non-inferiority trial. It turned out that none of the six statistical methods uniformly outperformed the others in terms of controlling the type I error rate. Nevertheless, there is a need for further investigations on methods dealing with missing data in non-inferiority trials.