—How good are doctors for prediction in medicine? —We don't know.
—How good is #AI for prediction in medicine?
—We don't know.
All 15 papers I've summarized are in silico, retrospective, many w/ statistical methodologic issues, and we've yet to see a prospective validation study in a real world clinical environment
https://www.nature.com/articles/s41591-018-0300-7 … #OApic.twitter.com/4HbrJu4zoK
-
-
-
We do know, They/we are not good at that, substantiated by many studies. And that has set the stage for help from machines,
#AI. - 1 more reply
New conversation -
-
-
Rec.: "High-performance medicine: the convergence of human and artificial intelligence"––
@EricTopol 's January 2019 in@NatureMedicine A breathtaking overview of the wide ranging research into applications of#AI in#medicine –– and how/why it has fallen short of clinical use.pic.twitter.com/I1QlVyqc2x
- 1 more reply
New conversation -
-
-
Isn’t this kind of like asking, how good is Excel or R? ML/DL is really just (sophisticated) math. If the data is good, the algorithms will be good. How we use them is the more interesting question, IMO.
-
Yes, but no one ever suggested replacing doctors with Excel or R. Yet that's what is being suggested by some with
#AI… - 1 more reply
New conversation -
-
-
#AI will become valuable for prediction in medicine, ONLY when every citizen will have a LONGITUDINAL medical record. From conception to death. Think about the opportunities.....https://www.linkedin.com/pulse/health-care-re-imagined-robert-gergely-md/ … -
That would also require the person to live with no changes, no science, no new developments. No life changes, no economic changes, random events---and that hasn't been human experience for hundreds of years.
- 1 more reply
New conversation -
-
-
AI will only ever be as good as the data that we feed it. Medical data is neither granular enough or accurate for an AI that would surpass a physician’s overall clinical assessment. But in data such as imaging, AI already surpasses us...
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
Shamelessly plugging in our result: Heart Failure, n=1,150,000, AUC=0.82, Rasmy et al https://www.sciencedirect.com/science/article/pii/S1532046418301175?via%3Dihub …
-
Wow, only 3% more accurate than logistic regression? Clearly the data didn't contain sequential, non-linear, nor hierarchical patterns almost at all.
-
Admittedly, data is very noisy. Maybe a bit lack of sequential. But non-linear embedding helps: Logistic regression without embedding is 6% lower (76%), while LR on top embedding is 79%.
End of conversation
New conversation -
-
-
#AI could transform stock trading - one of the purest data sectors And there are many use cases in medicine, but most data geeks don’t realize just how much of healthcare is an art, how hard it is to get clean data or how most isn’t clinical and how complex the human body is -
New conversation -
-
-
I think we can significantly benefit from researchers sharing the
#AI algorithms they used, maybe after patenting, so that other investigators can validate/reproduce the results in prospective studies.Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
Agreed 1. Data is still limited; 2. External validation in minimum two or more data sets is needed; 3. RCT for efficacy/effectiveness is needed e.g. we are (NCT03610165) evaluating an prediction algorithm. 4. PPV and NPV should always be reported.
Thanks. 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.