Sample size determination using composite binary endpoints
Author
Bofill, M.; Gómez Melis, Guadalupe
Type of activity
Presentation of work at congresses
Name of edition
3ra Reunión General de Biostatnet
Date of publication
2017
Presentation's date
2017-01-20
Book of congress proceedings
3ª Reunión Nacional de la Red Biostatnet: afrontando retos de investigación bioestadística con proyección internacional: Santiago de Compostela, España: enero 20-21, 2017
One of the key issues in clinical trial design is to calculate the suitable sample size to detect a given treatment effect in the main response or primary endpoint, for given significance level and power. In studies where the primary endpoint is assessed by binary endpoints, as success versus failure, the standard procedure of sample size calculation is based on the normal approximation to the binomial, often taking into account finite sample size correction. Several formulas can be used dependi...
One of the key issues in clinical trial design is to calculate the suitable sample size to detect a given treatment effect in the main response or primary endpoint, for given significance level and power. In studies where the primary endpoint is assessed by binary endpoints, as success versus failure, the standard procedure of sample size calculation is based on the normal approximation to the binomial, often taking into account finite sample size correction. Several formulas can be used depending on whether the odds ratio, the probability risk or the probability difference is defined for the effect. We restrict this presentation to the odds ratio formulation. Composite binary endpoints (CBE), defined as the union of two individual binary endpoints, are frequently used as the primary endpoint in a clinical trial. The derivation of the sample size for a CBE requires the specification of its odds ratio which is uniquely determined by the odds ratio and event rates of its marginal components and the degree of association between them in each treatment group. While the marginal parameters can be presumably anticipated, the association’s degree between marginal endpoints is usually unknown. The goal of this presentation is two–fold. First, we discuss the conditions under which the normality assumption is considered reliable given the parameter constellation of the CBE and which is the role of the level of association on these conditions. Second, aiming to provide a practical formulation in terms of the marginal parameters, we study different formulations for the sample size for a CBE taking into account the unknown association between components.