I am trying to design a noninferiority trial for a binomial outcome and have been disappointed to find that this appears to not be well supported by Proc GSdesign.
In trying to find a way around this, I opted to provide the sample size for a fixed-sample study in the samplesize model statement.
I am getting results that seem a bit higher than those provided by PASS, but the boundaries otherwise match.
To try to understand this better, I am looking for the formulas for the information Ix.
Questions are:
1. Can someone provide the formulas?
2. Has anyone successfully used Proc GSdesign for an NI trial? If yes, can you share the code.
Many thanks for any help that can be provided.
The SAS code for a non-binding futility margin and an efficacy bdry is pasted below
proc seqdesign errspend
pss(cref=0 0.4 0.7 1)
stopprob(cref=0 0.4 0.7 1)
plots=(asn power errspend) /*asn=average sample number*/
;
OneSidedErrorSpending: design nstages=10
method(alpha)=errfuncobf
method(beta)=errfuncobf
INFO=CUM(0.4 0.7 1.0)
alt=UPPER
stop=BOTH (BETABOUNDARY=NONBINDING)
alpha=0.025
beta=0.20
;
/* SAMPLE SIZE: Non-inferiority for two proportions. The sample size was derived for a fixed-sample noninferiority design FM statistic. See page 7045 of manual */
samplesize model=INPUTNOBS(N=270 SAMPLE=TWO);
run;
