| | 1. likely not data mining - BECAUSE
2. they were pre-determined subgroups AND
3. p values, number of events and number of patients in the subgroup point to real results; SPECIFICALLY
*** 39% reduction in risk of death or first cardiovascular hospitalization in treated Class II heart failure patients (patients=689; events=216; p=.0003)
This next one may be accused of data mining, [not pre-defined subgroup] but the number of patients and p values are excellent. Results look real.
Note: these results took place over 500 days; this further supports real data.
*** 72% of treated patients with no prior history of heart attack experienced a statistically significant reduction in risk of death or cardiovascular hospitalization (patients=1747; events=560 p=.005)
>>> What they are doing is called data mining. They are looking through the various subgroups of a trial that failed to meet the primary endpoint. If you have a good knowledge of statistics, you will understand why this is a questionable technique, and why the FDA very seldom accepts this kind of analysis. If you look hard enough, you will usually be able to find some group in a trial that has 'good' results. The number of possible subgroups is huge (male/female, young/old, very ill/less ill, smokers/nonsmokers, the list is almost endless). The likelihood is very high that by random chance, there will be some subgroup that shows a 'positive' result. If you do not understand this, you don't have a good appreciation of what VAS is doing here. |
|
