If the pharmaceutical industry were willing to fake efficacy results, surely they could do better than having over ½ their drugs fail Phase III efficacy trials. Especially since one thing everyone agrees on is that the best way to lower drug costs is to figure out which drugs are going to fail Phase III much, much earlier.

> real data about ... collected and exposed

Which is not a machine learning issue. Just collecting the information is a costly / time consuming issue which can't be outsourced to servers.

Right but the current innovation with regards to ml for medicine will force more data to be collected.

We have lots of ways to process and learn from data already. ML may make some things easier, but we're ready to learn more things right now with existing tools. Getting the data is the problem.

The amount of data you would require to do proper machine learning using current gradient descent techniques are orders of magnitude greater than that done in a drug study.

For drugs, we already do postmarketing surveillance. Whatever you think of pharmas, drugs are not a farce.

I'll just leave this here.

https://rationalwiki.org/wiki/Extraordinary_claims_require_e...

There's scores of evidence and studies about this. Don't feel like looking all of them up.

https://www.webmd.com/mental-health/news/20080227/antidepres...

And yet there's https://www.sciencedirect.com/science/article/pii/S014067361... from 2018 (a decade later) which concludes "All antidepressants were more efficacious than placebo in adults with major depressive disorder. Smaller differences between active drugs were found when placebo-controlled trials were included in the analysis, whereas there was more variability in efficacy and acceptability in head-to-head trials."

Along with commentaries like https://www.cambridge.org/core/journals/the-british-journal-... where the abstract is:

> Although antidepressant drugs are commonly effective, several meta-analyses of antidepressant drug trials undertaken decades after their introduction suggested that they were effectively acting as placebos. A recent meta-analysis concluded that they were effective. Both conclusions have been widely taken up by the media. This paper seeks to explain the disconnect.