y*****w
发帖数: 1350 | 1 Crossover design.
Subjects were randomized to Treatment A or Treatment B.
Outcome variable: a lab measure (continuous).
2 weeks 2 weeks
Treatment A -----------> lab test ----> Treatment B -----------> lab test
2 weeks 2 weeks
Treatment B -----------> lab test ----> Treatment A -----------> lab test
Since it's a study design, we don't have any pilot data at this time.
Literature search showed that for Treatment A, mean +/- standard deviation
was 41 +/- 21; for Treatment B, standard deviation was 13.
A difference of 4.2 is considered clinically significant.
Statistical significance level is 0.05. Two-sided.
Power is 80%.
I tried to calculate the sample size using nQuery Advisor with the above
data, where I would need to input the standard deviation of difference. I
have no idea how to efficiently estimate such a value based on the current
historical data.
Thanks! | a****g
发帖数: 8131 | 2 this is under the paired t-test one, right? | y*****w
发帖数: 1350 | 3 This is under ANOVA to test for equality (i.e. difference of means). I just
figured out there is nothing wrong with the small sample size, because this
is a crossover design. The difficult thing for me again is to efficiently
estimate a standard deviation of difference.
【在 a****g 的大作中提到】
: this is under the paired t-test one, right?
| a****g
发帖数: 8131 | 4 well, for a crossover design with 2 arms and 2 treatments each repeat once,
this is a paired t-test
if you analyze under anova, how are you going to consider the paired
property? Certainly you don't want to include the patients as another factor
enlighten me if I am wrong. Thanks
just
this
【在 y*****w 的大作中提到】
: This is under ANOVA to test for equality (i.e. difference of means). I just
: figured out there is nothing wrong with the small sample size, because this
: is a crossover design. The difficult thing for me again is to efficiently
: estimate a standard deviation of difference.
| y*****w
发帖数: 1350 | 5 (1) This is a "crossover ANOVA" which involves the feature of paired t-test.
It is exactly in the process of estimating "standard deviation of
differences" which involves the STDDEV of difference in AB, the STDDEV of
difference in BA, and the between-subject correlation between AB and BA.
(2) On the other hand, when directly applying paired t-test, we would need
to input the STDDEV of A, the STDDEV of B, and the correlation of within-
subject pair of measurements. See the link at
http://support.sas.com/documentation/cdl/en/statug/63347/HTML/d
Now, from historical data, we know mean and STDDEV of both A and B. In
scenario (1), we have no idea about the values of those parameters used for
estimating " standard deviation of differences". In scenario (2), however,
we know STDDEV of A and STDDEV of B, but we still don't know the within-
subject correlation. In both scenarios, we have to "guess" some values.
Some researchers have created a mathematical formula for calculating sample
size under 2x2 crossover design, which is exactly the mathematical basis of
scenario (1). Check out the formula at <http://hansheng.gsm.pku.edu.cn/pdf/2007/mean.pdf>. That's why I prefer to apply results of scenario (1).
,
factor
【在 a****g 的大作中提到】
: well, for a crossover design with 2 arms and 2 treatments each repeat once,
: this is a paired t-test
: if you analyze under anova, how are you going to consider the paired
: property? Certainly you don't want to include the patients as another factor
: enlighten me if I am wrong. Thanks
:
: just
: this
|
|
