3 ms·

> but for a study with a small number of sites where the patients are treated with the drug, there's enough inter-site variability that may obscure any effect,

> but for a study with a small number of sites where the patients are treated with the drug, there's enough inter-site variability that may obscure any effect, or create a false effect. The simple solution is to spread it over enough sites so that the law of large numbers quantifies that variability into the dataset. For example if we run trials over 10 hospitals in 20 different states, 200 sites in the dataset would almost certainly capture inter-site variance.

akiselev 7y ago
That only normalizes natural variances in your patients, not artificial biases introduced by the healthcare system. Pump and other hardware measuring out dosages might come from different manufacturers or be calibrated and serviced by different technicians. Maybe the study includes one too many hospitals in a region dominated by one manufacturer whose hardware tends to error more in one direction than other manufacturers. The generics used by pharmacies and hospitals might be slightly different because of local relationships. Exercise, ability to have visitation in quarantine, and other amenities in hospitals will have differing impacts on confounding factors like exercise and morale. And on and on. There are a bunch of "contract research organizations" all over the world, quite a few of them making billions of dollars a year, who specialize only in running clinical trials because these things are so complicated. The documentation for clinical trials are often thousands of pages long describing tiny details like device calibration across hospitals trying to eliminate as many risks as possible. These organizations tend to have long standing relationships with hospitals and host many different trials in each one with a bunch more subcontractors who do patient selection.

3 ms·

> but for a study with a small number of sites where the patients are treated with the drug, there's enough inter-site variability that may obscure any effect,

by dcolkitt 7y ago

akiselev 7y ago
That only normalizes natural variances in your patients, not artificial biases introduced by the healthcare system. Pump and other hardware measuring out dosages might come from different manufacturers or be calibrated and serviced by different technicians. Maybe the study includes one too many hospitals in a region dominated by one manufacturer whose hardware tends to error more in one direction than other manufacturers. The generics used by pharmacies and hospitals might be slightly different because of local relationships. Exercise, ability to have visitation in quarantine, and other amenities in hospitals will have differing impacts on confounding factors like exercise and morale. And on and on. There are a bunch of "contract research organizations" all over the world, quite a few of them making billions of dollars a year, who specialize only in running clinical trials because these things are so complicated. The documentation for clinical trials are often thousands of pages long describing tiny details like device calibration across hospitals trying to eliminate as many risks as possible. These organizations tend to have long standing relationships with hospitals and host many different trials in each one with a bunch more subcontractors who do patient selection.