DLyons May 7, 2015:

@Audrey That seems to fit quite well with my answer. Treatment A being significantly better than B is the primary criterion (critère primaire d'efficacité) and that being attained is the boundary that is being crossed. So, here, it's a significance boundary, not a stopping boundary.

Audrey Bernard-Petitjean (asker) May 7, 2015:

@DLyons In fact, there are 3 arms in the study: arm A with combined treatment A (= x + y), arm B with treatment B (treatment of reference) and arm C with treatment C (= x alone).
The primary endpoint (1) is to evaluate the efficacy of treatment A against B according to the PFS. Then, there are several secondary endpoints, among them are the following : (2) to evaluate the efficacy of treatment A against B according to the PFS in a sub-population of patients and (3) to evaluate the efficacy of treatment A against C according to the PFS in the same sub-population of patients.
So, the first statistical analysis will be performed to study the primary endpoint (1), then "if the boundary is crossed", the second statistical analysis will be performed to study the endpoint (2) and then "if the boundary is crossed" for these 2 first analyses, the third statistical analysis will be performed to study the endpoint (3)...
I hope that I am clear enough...

DLyons May 7, 2015:

@Audrey What is the second statistical analysis? Knowing that should help to decide what boundary has been crossed.

Is there, for example, a postprogression arm to the trial?

Audrey Bernard-Petitjean (asker) May 7, 2015:

Et concrètement, Adrien, pourrais-tu m'expliquer ce que cela signifie ? Je comprends "si le seuil est dépassé" comme étant "l'analyse statistique n'est pas significative" (on est au-delà du seuil).
Ce qui va à l'encontre de ce que j'ai pu lire jusqu'à présent concernant la notion de "boundary"... mais bon, je peux me tromper.
J'ai vu notamment que l'on parlait de "significance boundary" et de "stopping boundary"...

Platary (X) May 7, 2015:

Je pense que ce qui suit correspond à ta problématique. Il s'agirait donc bien du seuil de signification. Donc dans le genre "si le seuil est dépassé", mais largement adaptable :
http://www.spc.univ-lyon1.fr/polycop/principe du test statis...

Audrey Bernard-Petitjean (asker) May 7, 2015:

Bonjour Adrien, Et bien, je ne sais pas bien... avant cela, on définit l'analyse de la comparaison de la survie sans progression (PFS) des traitements A et B telle que "If PFS at 12 months was 30% with A and 10% with B (with 5.8 month median PFS) and assuming the true average PFS hazard ratio (HR) is 0.59, the trial will have >90% power to demonstrate statistical significance at the 5% level (using a 2-sided test) for the comparison of A versus B, with the smallest treatment difference that could be statistically significant being an average HR of 0.81."
Ensuite, pour décrire la 2e analyse statistique, j'ai "If the boundary is crossed..."

Platary (X) May 7, 2015:

Bonjour Audrey Est-ce que l'on fait ici référence au "seuil de signification" ?