Dear Allstatters,
I was asked to check a sample size calculation
of a 2*2 factorial design (randomized trial).
The outcome is continuous, assumed to be normally distributed.
The patients are to be randomized to one of the 4 groups (equal numbers).
The following means are expected (the standard deviation
is assumed to be 14 in each group):
MEANS:
| Ordinary B |
-----------+------------------------+-
Placebo | 5 12 |
A | 30 44 |
-----------+------------------------+-
B is a adjunctive therapy. The researchers
want to show that its effect is larger when
given together with A (7 versus 14; that is
interaction effect is 7).
They estimated a sample size for two parallel
groups as 64+64=128 with a clinically relevant difference of 7,
SD=14, two-sided alpha=0.05, power=0.80 (It's OK). Then they claimed, that
this overall number of 128 is satisfactory to
detect the interaction effect - that is for each of the
4 groups is is enough to have 32 patients (32+32+32+32=128).
However, according to my simulations, actually approx. 128+128+128+128=514
patient is needed to detect this interaction with power=0.80.
Do you have a formula to calculate this sample size?
I appreciate any comment.
Thank you,
Regards,
Akos Pap
Bayer Biometry
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|