M-17: Sample Size Planning in Bioequivalence Trials: A Systematic Review of Methodology
Poster Presenter
Junior Sinclair Awounvo
Student
University of Bremen Germany
Objectives
This review investigates the transparency of sample size considerations in bioequivalence (BE) trials: the statistical assumptions made on the true Test/Reference (T/R)-ratio, how these affected the choice of the other parameters and to compare the actual trial results with the assumptions.
Method
We searched PubMed for “crossover bioequivalence” and reviewed all Open Access reports on BE trials published between 01/13 and 07/18. We categorized them regarding the transparency of the sample size considerations and analyzed the planned and actual values of the power, T/R-ratio, CV and drop-out.
Results
We identified and reviewed a total of 126 articles reporting on crossover BE trials. The results of this review indicate that no information about the sample size considerations were provided in 54 (approx. 43%) of these articles; in 24 of them (approx. 19%) the information provided were insufficient and only 48 articles (approx. 38%) contained information on their sample size considerations which is sufficient to fully repeat the sample size determination.
Further, the planned T/R-ratio was missing in 15 of the 48 articles (approx. 31%) containing sufficient information and had to be derived from the other parameters. The planned coefficient of variation was missing in 2 articles (approx. 4%), whereas the remaining 31 articles (approx. 65%) provided transparent and complete information about the sample size planning.
We found that, in 75% of the articles with sufficient information, the authors adopted a conservative approach while planning the T/R-ratio, since they assumed that its true value deviates at least of 5% from 1.00. At the opposite, the other authors expected the true T/R-ratio to be exactly equal to 1.
Additionally, all the authors powered their study with at least 80%, with approx. 48% respectively approx. 44% of them actually powering it with 80% respectively at least 90%.
The systematic review was completed with a quality control of all results. Any disagreement between the reviewers was resolved with a consensus.
Conclusion
Sample size planning is a major task when performing BE trials as it has a direct impact on the success probability of the trial. Therefore, reports on BE trials should include a fully and transparently description of statistical parameters, such as the T/R ratio, the power, the drop-out rate and the CV as they play a key role in sample size planning, and thus in the outcome of the trial.
However, this review has shown that, more than 50% of the reports on BE trials written over the last 5 years, do not provide sufficient information on their sample size considerations, with almost half of them not providing any information at all.
This review also highlighted the problem of estimating the T/R ratio for the sample size planning and how this correlates with the power of BE trials. In order to confer the study satisfactory power, the vast majority of authors adopted a conservative approach by choosing a true T/R-ratio different from 1. Although a value of 1.00 is the most likely for the true T/R-ratio, it will lead to the smallest sample size and is therefore not a conservative choice. Any deviation from this value in the reality will result in a loss of power. Since specifications on the manufacture of drug products also address possible deviations from the optimal value of 1.00, it is advisable to choose a value such as 0.95. To a certain extent, estimating the T/R-ratio from pilot studies can be an alternative worth considering.
Nevertheless, all the studies were powered with at least 80% meaning that the authors who opted for an anti-conservative approach, compensated a potentially power loss by adjusting the other parameters.
The full Review will investigate in detail in which manner, the authors proceeded to overcome an eventual power loss. The full review will also assess the sample size planning by comparing the planned and actual parameters values as it is important to determine whether the assumed parameters range was met in the study or not.
