Caveat #1, the following data is from formulas and lots of assumptions
, not real studies results.
Some probability numbers from Statistica Microsoft software calculator of the difference between parcentages, the use of the Chi Square test for non-parametric data (the data from Xoma studies is non-paremetric, not all of it but the one we are interested).
This assumes that they are using a two tail test (more discrimination than a single tail test).
No POWER was specified, but most studies use 80% or 90% power.
If the DSMB will review the data on 130 patients, and I will not assume the posibilty of zero death in the treatment group, since this is a very improbable outcome:
N1 is non BPI number of patients, N2 is the BPI number of patients
% is the percentage of deceased patients.
P value of less than 0.05 is statistically significant.
Also, assuming that the distribution of patients is even in both groups, this is not the case but by pure ramdomization it tends to be approximately that way. Then:
Total of patients = 130
A:
N1 65-----30%
N2 65-----15% p value is 0.0426 (less than the commonly use 0.05 but not overwheelmingly so)
B:
N1 65-----30%
N2 65-----10% p value 0.0051
C:
N1 65-----30%
N2 65------5% p value 0.0003
If this does not happen then they will review at 150 patients?
D:
N1 75-----30%
N2 75-----10% p value 0.026
E:
N1 75-----30%
N2 75-----15% p value 0.0294
If this does not happen then at 200 patients?
F:
N1 100----20% here I went for the best the non BPI could do
N2 100----10% and double the death from the PII and still one gets statitistical significance: p value 0.0491
If one look at sample A, the difference is 50% and even with the close to 0.05 p value it could be stop here by the DSMB. If one remembers the PII the BPI group was 4% deaths (0ne in 26). Sample A is very conservative, it assumes a relatively low death rate for non treated patients (30%) and almost 4 times higher failure for BPI compared to the PII study ( 15/4 = 3.75 )
Judicious wait and see must prevail and irrational exuberance must be contained.
Caveat #2: the numbers are good, but not necessarily appropriate, I am trying to contact a friend with a good knowledge of Statistics and a better software program to confirm this numbers and/or my interpretation of them.
Not a fly to the catcher. |
