3 ms·
Well, the hope I have (and OP probably as well) is that machine learning techniques properly applied to the raw data could uncover a lot of statistical errors i
by drcode 7y ago
Well, the hope I have (and OP probably as well) is that machine learning techniques properly applied to the raw data could uncover a lot of statistical errors in the historical literature, which could be both profitable and disruptive.
... Unfortunately, this may happen first in countries with poor human rights records (i.e. China) because there is likely a greater availability of patient data under those circumstances.
- lukeschlather 7y agoML doesn't fix bad statistics. ML is really just a sexier way of saying real-time statistics with lots of data.
- drcode 7y agoI would agree that ML is just statistics, but I also kind of think that a median approach to statistics in medicine are currently at an even poorer level of quality than the low bar set by typical ML, so there could be an improvement nonetheless.
- sidlls 7y agoML is not in a state to be of much use in this context. Source: me, I work as a data scientist to apply ML and other techniques, with claims, EHR, and a bunch of other data, specifically for the purpose of identification of conditions, effectiveness of treatments and non-medical intervention. The tech world is very much in a condition of being unjustifiably arrogant about its ability to solve problems in this space. The problems are very hard, and the answers aren't gonna be found by some egotistical software eng who tries to shift his toy kaggle crap into this space, or the kinds of startups that are built up around that kind of thinking.
- drcode 7y agoI kind of agree- A rigorous solution to this problem would probably involve running large numbers of (human?) cell cultures with deep analysis fed through ML systems. i.e. we first need to have a full understanding of cell chemistry/physiology and then scale it up to larger organs and the whole human organism.
- marcinzm 7y agoML will simply be used to find all sorts of new spurious correlations that will spawn a whole new flood of unnecessary medications and treatments. After all, statistics puts a whole lot more care into avoiding those sorts of errors than ML and, well, here we are anyway. Technology doesn't solve societal problems.
- rscho 7y agoYeah you see that's the problem. The raw data is not there, because we don't have the necessary information retrieval capability yet. Data is really sparse and of the worst quality. It is interesting that the hype is exactly the opposite: big bad hospitals keep essential info for themselves.