Meningo trial has Three Arms. Two treatment arms and one placebo
(check Tharos post 6523 for trial protocol)
Treatment arms stratification is based on Glasgow scale.
"Blindness" of the trial prevents the company from knowing which Arm does a deceased patient belongs to.
The total mortality is known by Xoma and they are not releasing it, and appropriately so.
Lets say 375 patients, each group is about 125:
Placebo - 125 patients, mortality 24% (conservative) deceased 31
Bpi 1 - (low Glasgow scale, lower mortality) 125 patients,
mortality 8% (double the PII trial 4% ) deceased 10
Bpi 2 - (high Glasgow scale, higher mortality) 125 patients,
mortality 12% (triple the PII trial 4%) deceased 15
Total mortality: 31 + 10 + 15 = 56
Is the difference statistically significant with the N:125 ? (N is number of patients on each arm which will provide the appropriate
P value for a study).
Bpi-1 vs placebo = 66% decrease in mortality
Bpi-2 vs placebo = 50% decrease in mortality
Bpi-1 and Bpi2 vs placebo = 58% decrease in mortality
These are highly CLINICALLY significant results. For this condition even a 15% reduction in mortality is highly Clinically significant.
Statistically significant? I will say yes (but I have to confirm with calculations). (some examples on post 5214, but these were calculated for a two arm trial). For a 15 % reduction in mortality let the trial go on and on.
The Epi6 (I downloaded from the link george w. provide in one of his references, anyone wants to try check post 9155) is a little more complex than the Microsoft Statistica I used before and I am still trying to figure things out.
The Bpi-meningo trial is not only blinded, but it is also RANDOMIZED.
Randomization is another technique to prevent bias.
Randomization works by pure chance and these days is determined from a computer program (before people will use a randomization table).
Patients are randomized from the start, but also centers are randomized.
It is not one patient placebo, one bpi, one placebo one bpi.....
It is not one center placebo, one center bpi, one....
It is not even one center/patient bpi, one center/patient placebo...
This is avoided because clinicians wants to know the results as early as possible and they could make their own personal notations and figured out how things are going and the randomization should be able to avoid this from happening.
The drug is named Neuprex in all cases, clinicians will make notations if the vials are different.
One example, during the Surfactant trials for neonatal Respiratory distress syndrome it was very difficult to blind anyone, the drug look like foamy water with soap, and the placebo was AIR, since the most common placebo is Saline water and it is not advisable to fill the lungs of a baby with saline water (it will kill the baby) then AIR was chosen as the placebo. But Foamy Surfactant is very different to Air, the randomization and blindness was in peril. The solution was the clinician that administer the drug will do it behind a barrier, and he/she will not be involved in management, only in giving the medication. Still there was a problem, clinicians could see foamy stuff coming out of everywhere even after the administration, and secretions of the mouth and throat of babies were very different, then the nurses were also change, then the respiratory technicians were noticing huge changes in ventilator machines that they have to account for, and so on... they got tired of all the details and agree that it was impossible to do full blindness. Still, clinicians,nurses and respiratory technicians were change after the dose, this made the trials expensive and a logistical nightmare.
The pharmacists role: they manage the cards that are assigned to each vial and distribute it. Some trials are "double blind" (patient does not know, neither the clinician)or "triple blind" when not even the pharmacists will know what are they providing but the code vials that belong to a code card that belong to a patient.
Randomization, is total chance. It could be even possible to have at one moment of the trial (by chance)90% placebo patients vs 10% treatment patients (mortality will be reach fast, but results will not be exactly useful) or the other way around 90% treatment vs 10% placebo (total mortality will be more difficult to reach if the product works), eventually over time the distribution will result in an approximately 50% placebo and 50% treatment patients.
So as one tries to calculate with even numbers of distribution of the samples at one time there is not but the belief in chance that the numbers are even at that moment, real numbers could not be like expected (by chance).
Check post 8469 for PARS case when the company DID provide the total mortality for pr purposes. They were lucky that it was a PII, and the tanking of the price after actual results were out are not determining the survival of the company. PARS product decreased mortality 26%, but not statistically significant for that N.
Are the DSMB computers emotional?
|
