4 ms·
My first reaction to these things is that there should be a large database of diseases and symptoms along with some statistics about how strong the correlation
by phkahler 9y ago
My first reaction to these things is that there should be a large database of diseases and symptoms along with some statistics about how strong the correlation is between the disease and symptoms. I don't mean "database" like pubmed. I mean something more machine readable with consistent coding so a symptom might be "elevated IgM" in english. The trick is coding something like that with consistent names for the tested item and consistent definitions of "elevated". Perhaps the definition of elevated would depend on the disease?
One should be able to then filter the disease database based on symptoms. Doctors seem to do this in their head, but they are not comprehensive stores of information. I suppose the problem is in the encoding of symptoms, it's not that simple.
- pjc50 9y agoThis was one of the early areas of AI research, which seems to have gone nowhere: https://en.wikipedia.org/wiki/Mycin https://en.wikipedia.org/wiki/Mycin
- jdmichal 9y agoThe only appropriate definition of elevated is higher than normal for that patient. Unfortunately, doctors rarely have a baseline to work from, so they use statistics instead. A relative of mine had a PSA test that read statistically normal. Luckily he looked at the actual number and saw that, while it was within statistical normal, it was in fact higher than any other prior reading for him. Further testing came up positive for prostate cancer. If he hasn't been paying attention, that would have been another year of cancer until the next check.
- bluesign 9y agoThe main problem here is human body works in mysterious ways. We are by definition reacting different, and unfortunately statistics are not helping on this case. Basically, we have: - too many illnesses (I guess 10% of top common health problems, covering 90% of all problems) - for each illness, we have N number of symptoms (which unfortunately all or only a subset of this can occur) - then we have fake symptoms (statistically out of threshold values but we are living fine with them) this is generating a big number of combinations. With this much combinations, it is requiring extensive amount of tests. As everything health is also optimized for the greater good. Which requires optimum cost management. So basically medicine today, with statistical approach, is happy to skip 1%, if it can fix 99% with 90% less cost.
- T2_t2 9y ago> - too many illnesses (I guess 10% of top common health problems, covering 90% of all problems) I think this is probably the easy thing to fix when it comes to searching, as you could pretty much leave out the common diagnoses as they'd likely have been tried - Imma call this the "Dr House, is it Lupus?" filter. From there, you could have "try this" filter, and see what the results are. I wonder what the overlap is between tests and symptoms, but surely with 10 tests and 10 symptoms, most illnesses could be mostly identified?
- eismcc 9y agoIt’s hard to convince institutions to release data locked up in EHRs (eg EPIC) due to the labor involved (EPIC is not well structured for this), risk of accidentally leaking PHI, and general motivation (getting hold of useful data is the bread and butter for career researchers).