In a nutshell, no amount of lobbying will stop equivalent service for 100x cheaper. Tech does not permeate healthcare for 2 reasons: 1. mostly inapplicable. Everyone is focusing on ML model performance, but really information retrieval in healthcare is dismal and prevents the use of such new tech altogether except in very niche cases. 2. no integration in the workplace. Tech people and docs don't understand each other at all, so docs ask for impossible things, and tech people deliver perfectly functional, totally inapplicable tech.
If there are algos out there (and there are) that can accurately provide a strong heuristic on melanoma from a photo and this is being blocked by the state - that seems like an obvious instance of regulatory apparatus stopping an equivalent 100x+ cheaper service.
I've discussed with a number of people who work directly on DL for imaging at a major hospital system in Boston. They say that (outside of the doctors they work directly with) fear over competition and losing out on the pricier billings are one of the largest barriers to getting their (very accurate) tech deployed more widely.
Yes, so as usual it's so superior but it's never used. And the people building it say it's stellar, promise! Here's a clue: instead of building a tool, and try teaching people already practicing how and why they should use it, maybe we could actually go see what practitioners are doing and try to integrate into that without requiring 30 additional mouse clicks and the use of a new soft that nobody understands ?
You can find the studies on the recent melanoma classifiers. There are tons showing in various settings that they pretty clearly outperform physicians.
If 'additional mouse clicks' is a major barrier to physicians using a tool that leads to far better diagnosis outcomes of a fast-progressing and deadly disease, I'm not sure why that is an argument for why things should continue to be as physician-gated as they are.
I will happily perform the 30 extra clicks myself if it is my potential melanoma. But if I were to offer it as a self-serve app ($2 for melanoma diagnoses too cheap to meter), I would be thrown in jail.
Yes, we have so many models that completely outclass docs. It's really strange, they're not more widely available don't you think ? Providing so much value, one would think there would be a black market for those, at this point. Or maybe, just maybe, the setting necessary to make things work in large-scale realistic practice is more difficult than what the paper authors would have you believe? No, they would never do that...
I'm being very specific about melanoma because this is one of the cases with very compelling evidence. You can broaden the discussion if you want - but that is not what I'm discussing. Here's a study of this technology with 67k real-life practitioners showing obvious increases in accuracy. [0]
Apologies if the link to that article is one mouse click too many for you.
I never denied the performance. Now make it usable to the average doc. That's where we disagree. You believe it's usable, but you've never seen clinicians handle computer stuff.
exactly… this is tech that could be used by literally anyone if it were legal but I am supposed to believe it is just too difficult to bring into practice
I doubt that measurements outside a standardized environment would grant satisfactory performance. But, perhaps. I don't know.
You're just too sour, man. I'm not saying it won't work, not even saying with certainty it doesn't work now. I'm not refuting protectionism plays a role either. What I'm saying is just that clinical integration of new tech, especially involving computers, is much more difficult than you seem to believe. And that the primary reason for that is not the greed of docs, which in my experience holds far less political influence than you think. I'm all for new tech, so chill out a bit.
Last I heard, they were very sensitive to things like imaging equipment, so they could diagnose well if imaging was done by the same gear that provided the training data. It worked fine in the hospital that developed it, but unable to deploy widely. If that issue has been fixed, I look forward to an online service running from a less regulated region. It would be a money printer, even if the US blocked it.
Me and my team made a piece of successful software for patients and clinicians. It is really difficult, mostly for the reasons you state, but it is possible. It's used in about half the NHS, and I personally know three people who've used our app to successfully manage their pregnancy complication, which is great.
My company/team is an unusual exception. The UK has a much bigger problem with creating tech than the US does. That's why almost all NHS trusts in the UK use US-created EHR systems: Cerner and Epic, mostly.
The UK is bad at creating a pro-business/pro-investment environment, so we have to buy in stuff from elsewhere, even though it's not well-suited to our needs. Or best case we find US-based investment for our companies.
