I doubt that there's literally no info in that blood. Maybe one needs to rethink all the tests. Instead of testing one compound, make a model that predicts things using all the available info and a big, sophisticated ML model. BTW, would I look good in a black turtleneck?
-
-
Replying to @RokoMijicUK @halvorz
You need a biobank or hospital partnership to make that happen, but maybe!
1 reply 0 retweets 1 like -
But then you run into the same problem as uBiome. If you’re not measuring the levels of a specific molecule, but your own composite score, then you have to prove to the satisfaction of doctors and insurers that this score can be used to make actual medical decisions.
2 replies 0 retweets 0 likes -
Replying to @s_r_constantin @halvorz
I don't have a good intuition for how hard the regulatory hurdle is. But it's obvious to me that a composite score will outperform [M] for any single molecule M.
1 reply 0 retweets 0 likes -
-
Replying to @halvorz @s_r_constantin
For whatever you're testing for. I don't actually know what the outcomes of interest are, but presumably there are 1000's of them based on different diseases and conditions. Some of these probably leave a signature in even a small blood sample.
1 reply 0 retweets 0 likes -
Replying to @RokoMijicUK @s_r_constantin
what you're testing for is typically the level of some molecule of interest you seem to be saying that a composite score will better predict the levels of your molecule of interest, than measuring the level of the molecule
1 reply 0 retweets 0 likes -
Replying to @halvorz @s_r_constantin
No I mean a composite score will better predict a patient state like "does this person have disease X", and it might be *so much better* that the noisier sample from a fingerprick is good enough, so the patient doesn't have to give blood from a vein.
2 replies 0 retweets 0 likes -
Replying to @RokoMijicUK @halvorz
Well then you'll need a big patient sample of sick and healthy patients to prove it. (Big, because blood tests usually aren't *that* predictive of disease before the disease is obvious by other means. A biomarker with an OR of 2 is good.)
2 replies 0 retweets 2 likes -
Replying to @s_r_constantin @halvorz
Another way to put this is that the existing diagnosis process is probably very inefficient. Sending the concentration of specific chemicals to doctors who then manually make decisions? Doing it like modern ML would be much better. You "just" need the dataset.
1 reply 0 retweets 1 like
I agree in principle. Maybe with a clean slate medical system something like that would be done!
-
-
Replying to @s_r_constantin @halvorz
Well yes, the existing systems inevitably exert a drag on such things.
0 replies 0 retweets 0 likesThanks. Twitter will use this to make your timeline better. UndoUndo
-
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.