george w, I managed to find how to use the statistical software from your link:
soton.ac.uk
Download is free and relatively easy to use (it took me just couple of weeks, but due to different terminology). One could even download a manual (I did not, maybe next time I will.
Numbers vary a little bit with the ones in post # 5214, at the time, I
did not know the Power of the test (Statistica from Microsoft either does not provide it or I did not manage to elucidate it, in any case I think the Power use in that post is 90%).
NEW NUMBERS (CAVEAT, this is total speculation, some educated guess involved, actual events could be as predicted by george w or RobertS).
Test is to estimate the size of a sample for two proportions (percentages).
Ratio of group 1/group 2 = 1 (for all the examples it will be one, I will assume an exact number of placebo/bpi patients.
Power 80%, this is usually 80% or 90% in most studies, when time is critical and study is expensive and mortality is an issue and recruitment is slow 80% is appropriate and justifiable.
This test is akin to Chi square for non-parametric nominal variables, (not ordinal)like survival, and p value less than 0.05 is expected.
Alpha of 0.05 (kind of the reversal of p value 1 in 20 anyway) to decrease the chances for a type I error. Confidence level 95% (this could probably be reduced and alpha increase resulting in less number of subjects needed, but this 95% is common use so I keep to be conservative in estimates).
Total Number of subjects: 384
68% in the low Glasgow score: 262
32% in the high Glasgow score: 122
Total recruitment = 384
Each group is stratified for placebo vs bpi.
Low Glasgow score placebo N1 131, mortality 18%
Low Glasgow score bpi N2 131, mortality 6%
p value = or less than 0.05
subtotal events = 24 + 8 = 32
High Glasgow score placebo N1 61, mortality 30%
High Glasgow score bpi N2 61, mortality 12%
p value = or less than 0.05
subtotal events = 18 + 7 = 25
Total mortality : 57
It is highly improbable to get better numbers, worse numbers are more probable.
Even O mortality in the bpi group still will leave us with
24 + 18 = 42 total events. This is a highly improbable outcome.
Let the recruitment continue.
For the followers of the 10% mortality in the placebo group:
Placebo N1 382, mortality 10%
Bpi N2 382, mortality 5%
Total number of patients to recruit = 764
p less than 0.05, power 80%.
total mortality: 38 + 19 = 57 (again?).
All disclaimers apply, check with your personal Statistician,
I could not assure of the accuracy and/or applicability of this numbers, no claims of validity, no relationships with any corporate sponsor, only product is free from a government agency.
Disclosure: the author holds Xoma shares and they are less than 10% of his total holdings. The Author is extremely bias for positive results, despite his long position and his long face.
For entertainment purposes only. |
