Map of life expectancy at birth from Global Education Project.

Tuesday, July 13, 2021

The bias of science, part the next

Before I get to the subject of this post I should ask, if you don't think government-funded research should be conducted by university-based scientists, and you don't think it should be done by government itself, who do you think should do it? Maybe award research grants at random? 




This is  generally applied, not basic research. Although at one time some corporations supported basic research enterprises -- notably Bell Labs -- that's largely a thing of the past. You can't patent basic knowledge so you can't really profit from it, and profit is all that matters nowadays. In most industries, that's okay as far as it goes. There aren't any incentives to fake your results because if it doesn't work, it doesn't work, and you can't sell it. I say as far as it goes because they aren't investing in a better planet, they're investing in higher profits, and we all know that saleable and profitable technologies don't necessarily end up being net positive social goods, to say the least. But that's another discussion.


One industry that does have an incentive fake it, however, is the pharmaceutical industry. The effects of pharmaceuticals are not obvious, go or no/go. The silicon chip either processes the data properly or it doesn't. The jet engine either makes the plane fly or it blows up. But medications aren't like that at all.


The way you get a drug approved by the FDA, meaning a patent, exclusive marketing rights, and maybe billions of dollars, is by convincing FDA regulators through randomized controlled trials (RCTs). The idea seems straightforward at first glance. You round up a bunch of people with Creeping Crud, assign half of them at random to get Thanatos, and half to get a placebo or an already approved competing drug, and after a specified interval you measure the average amount of crud on the intervention and control people. Less crud with Thanatos, it must work!


Sadly, no. Or not at all that easy./ For one thing, the average difference probably masks huge variability. Some people probably do worse than average on Thanatos, while some people on placebo clear up entirely. Another problem is that the people eligible for trial are not at all typical. They may have to meet requirements such as age range, no co-occurring conditions,  ability to speak English, not pregnant or likely to become pregnant . .. When we take it out into the real world we don't know what might happen. Also they are strictly monitored to ensure that they "adhere to the regimen" which also doesn't happen in the real world. And it isn't necessarily the case that less measurable crud means better quality of life. There may be adverse effects that go undetected because the follow-up is too short, or the investigators don't bother to look for them. 


Also, what you are assessing is statistical significance of the difference. What is the likelihood you would see a difference of this size, just by coincidence, if Thanatos has no real effect? There's usually an arbitrary threshold of 5%, and it's easy to play with the numbers if you need to push it down a little. And a statistically significant difference isn't necessarily a clinically significant difference -- it may be too small to matter to people. 


Also, too, in the past if you didn't like the result you could just hide it in a drawer and do another study and maybe another until you got one that came out with the precious p value. They used to do that all the time, although the FDA has now made it harder. I could write a textbook on all the ways to game RCTs  -- and there are plenty more I haven't mentioned -- but you can study up on it if you're interested.The bottom line is that pharmaceutical companies have no soul and they all cheat as much as they can get away with. So we're probably taking a lot of drugs that are not a good deal for us. 


But that's the easy part. Next time I'll get more philosophical.

No comments: