Alzheimer's turns out to be intractable so far, and this has had a depressing effect on drug trials. Nothing has worked, so far. Well, may be something does work.
Today's reports about Lilly's failure, not the first in recent time, reveals that Lilly is trying the sample size bigger ploy. The snippets are from the NYT.
Lilly said, however, that when the results of the two trials were combined, creating a larger sample size, there was a statistically significant slowing of the decline in cognition.Being somewhat addicted to cop shows since adolescence, one of the standard lines from same: "A DA can get a ham sandwich indicted, if he wants to." In the world of quants, the analogue goes, "A quant can find a significant difference between ham sandwiches with a large enough sample size." In drug development, clinical trials can be time consuming and expensive, so individual trials tend to be powered to the smallest sample size indicated by previous data, and the assumed magnitude of difference between the drug and either placebo or some standard of care. What Lilly tried was to pool data from separate trials.
Pooled data is acceptable to math stats, but the requirements are pretty strict, principally with regard to variance within and between trials. Here's an historical criticism.
Lilly also went the way of post-hoc sub-group analysis, another not universally accepted gambit.
What this means for the drug's future is still unclear. The effect of a drug on a subgroup of patients in a trial is typically not sufficient grounds for a drug to be approved without further clinical trials involving just that subgroup.And finally, the coup de grace,
Also left unclear Friday was how big the effect on cognition was -- whether it would be meaningful for patients or merely meet some statistical test.
A biostat can make the slightest difference look colossal, given a large enough sample size. When does a ham sandwich look like a BLT? Measure enough of them, and you'll be able to prove it.
No comments:
Post a Comment