Thread by @EricTopol

There have only been 5 randomized clinical trials of #AI in medicine to date. Here's the summary:
4 in gastroenterology (2 @LancetGastroHep, 2 @Gut_BMJ)
1 in ophthalmology (@EClinicalMed)
All were conducted in China
(None in radiology, pathology, dermatology or other specialties)

@LancetGastroHep @Gut_BMJ @EClinicalMed What constitutes suitable evidence for clinical implementation (and surveillance after) for #AI in #healthcare?
Here's @d_spiegel's take:
hdsr.mitpress.mit.edu/pub/56lnenzj @pubpub @harvard_data

@LancetGastroHep @Gut_BMJ @EClinicalMed While there are now hundreds of in silico, retrospective dataset reports, the number of prospective (non-randomized) trials in a real clinical environment testing #AI performance is limited. I only know of 11. Let me know if I'm missing any.

@LancetGastroHep @Gut_BMJ @EClinicalMed ✓ that. There are 6 RCTs, one more just published
Updated via @DrOmerAhmad @GIE_Journal
(the pattern of specialty and place doesn't require #AI to recognize)

@LancetGastroHep @Gut_BMJ @EClinicalMed @DrOmerAhmad @GIE_Journal Even adding the published prospective trials (Table in thread), that's < 20 that reach "clinical grade" (assessed in a real clinical environment).

There are >50 #AI algorithms cleared/approved by @US_FDA, mostly from retrospective datasets; they are proprietary, unpublished.

@LancetGastroHep @Gut_BMJ @EClinicalMed @DrOmerAhmad @GIE_Journal When the results of randomized trials in #gastroenterology for machine vision #AI engender the "guardian angel" metaphor sciencedirect.com/science/articl… @GIE_Journal @KrishRagunath Sherman Picardo.
[a futuristic goal for supporting all clinicians and patients]

@LancetGastroHep @Gut_BMJ @EClinicalMed Why is #gastroenterology (unexpectedly) leading the charge of #AI randomized clinical trials?
Commentary by @tberzin @TheLancet
thelancet.com/journals/lance…
and now > 13 companies competing in this field