Read 3 CMS New Tech Reviews: (1) AI for Head CT (Pass), (2) AI for Pulm. Emb. (Proposed), (3) Pneumonia Pathogen Panel (Fail)

 As I noted in an earlier blog, on April 28, 2022, CMS released its initial assessment of 22 technologies proposed for FY2022 hospital inpatient new tech payments (here).

Two Technologies for Hospital Inpatient "AI"

Of note, one of the proposals today is for hospital inpatient use of artificial intelligence software in radiology.  This is the AIDOC BRIEFCASE FOR PE system by AIDOC.  

Last winter, CMS granted add on payments to a system for hospital inpatient use of artificial intelligence in stroke management.   This is the ContaCT software from VIZ.AI.

I've put both of these inpatient AI technologies in a zip file; link at bottom.   Both of these AI systems pre-read CT scans in queue and can highlight those that are 90% likely to be positive for first-in-line review.  This, in turn, should speed therapy for the stroke or for the pulmonary embolism.

CMS's Critique of Biofire Outcomes Data

In addition, and making a contrast, adjacent to the December 2020 "ContaCT decision" I noticed that CMS had published a negative analysis and negative decision for the Biofire Pneumonia Pathogen Panel.   

CMS criticized the Biofire panel has having inferred or speculated benefits rather than confirmed ones.   This is a frequent criticism of diagnostic tests, and for anyone planning diagnostics outcomes data or clinical utility arguments for CMS review, this example may be worth a read.  And interestingly CMS nixed extra payments for the Biofire pneumonia panel in 2H2020, at the height of the COVID epidemic, even knowing that a significant portion of the most complex cases likely had coinfections that would be hard to culture in patients on broad spectrum antibiotics.

Zip file of the three items, HERE.

See CMS announcement re a "AI CHALLENGE" contest, April 28, here.

Flurry of News: AI in Medicine and Creative Destruction

German economist Joseph Schumpeter coined the term Creative Destruction (schöpferische Zerstörung) to describe the growth and collapse of businesses and industries under capitalism.  (Today "creative destruction" is sometimes used enthusiastically to describe dynamic change and progress.  In his day, Schumpeter saw this less as upward growth and more as an eventual downward spiral, a "fall-of-the-west" or "implosion-of-capitalism" viewpoint.) 

Today with AI (and other new technologies in healthcare) we often construe "creative destruction" as an interesting idea, investment, bubble, hype, and then disappointment a la the Gartner Hype Cycle.   Is IBM Watson becoming an example?

First Example:  Derek Lowe's Assessment of IBM Watson in Drug Discovery

See an open access article by Derek Lowe at his Science blog, In the Pipeline.  He reports on April 18 that STAT reports that IBM has has canned its much-vaunted "Watson for Drug Discovery" now.   But here, Watson for Drug Discovery can probably stand in, almost with canned text, for many efforts to turn big data into useful health outcomes.   IBM wrote...

Watson for Drug Discovery reveals connections and relationships among genes, drugs, diseases and other entities by analyzing multiple sets of life sciences knowledge. Researchers can generate new hypotheses using the resulting dynamic visualizations and evidence-backed predictions. . .Pharmaceutical companies, biotech and academic institutions use Watson for Drug Discovery to assist with new drug target identification and drug repurposing. Connect your in-house data with public data for a rich set of life sciences knowledge. Shorten the drug discovery process and increase the likelihood of your scientific breakthroughs.

Clearly, this could be easily edited into "Harvard for Drug Discovery" or "Mayo for Health Outcomes" or "Optum for Public Health" -- to give some wholly hypothetical names to applications for AI where the existence of real projects wouldn't surprise you.

Lowe's blog has a lot of detail about IBM Watson for Drug Discovery, and if you don't subscribe to STAT, he also links to a recent open access article at IEEE Spectrum "How IBM Watson Overpromised and Underdelivered on AI Health Care," here.   (IEEE first chimed this note in 2015, here.)

"Overpromised and Underdelivered" is a truly impressive deep dive piece by IEEE senior editor Eliza Strickland, who also manages the May 2018 IEEE series, "Hacking the Human OS," which is a wealth of interesting open access articles on similar themes.   Find it here and track their running medtech blog here.

For addition IBM Watson bad news, see Forbes on a fiasco with MD Anderson, 2017, here.  See a 2017 negative STAT article on IBM Watson, here.  See a 2018 article on layoffs, here, and a 2018 article on scaling back the IBM Watson hospital services business here.  Over 5 years, IBM stock has slipped from $190 down to $140.  Microsoft is up from $45 to $120, and Apple up from around $100 to around $200.   During that time, Dow Jones is up from 16,000 to 26,000.  (GE has slipped from $25 to $10.  GE spunout its biopharma business to Danaher (here) but is working with ACR on AI in imaging...here...see next story.)

Second Example:  AI in Imaging

Creative destruction is a dynamic mix of upswings and downswings.   Here's an upswing.  At the same time, we read that AI medical imaging startup AIDOC has just raised a new $27M for medical imaging based on AI, here.

And institutions are buying in, including prestigious bodies like NIH and American College of Radiology.   See an article in MedTechDive here, that links to a press release from Radiological Society of North America, that leads to a new NIH/ACR position paper on the expected rapid and important growth of AI in imaging.  That last document is here, Langlotz et al., and it's $30.  (As mentioned earlier, GE is tying in with ACR re AI R&D, here.)

See also the recent trade press here, about a new 20 page FDA white paper and guidance document on AI in devices, here, which was released with a press release by Scott Gottlieb, here.

FDA has begun approving AI-driven devices, such as for retinopathy diagnostics (here).  Those approvals have been for locked-software devices; the new guidance moves FDA toward self-updating AI systems. 

Digital Pathology, Too

CAP Today ran a long cover story on digital pathology in February 2019 - by Karen Titus, here.   (Admittedly, focused more on digital storage and interpretation than machine learning or AI.)  But see a very interesting article that we don't need "digital pathology," but rather "intelligent digital pathology," by Acs and Rimm in JAMA Oncology in March 2018 (here).   See some additional autumn 2018 headlines in digital pathology at blog, here.

See a 2018 review article on AI and digital pathology by Tizhoosh and Pantanowitz, open access, here or here.  See a 2018 article in IEEE Spectrum, "The First Frontier for Medical AI is the Pathology Lab," here.  Similarly, see a trade press article in Healthcare-In-Europe, October 2018, here.

See a May 2019 article by consultancy DeciBio on Digital Pathology and Multiplex Spatial Tissue Analysis, here.

In another area of medical technology, Edwards Lifesciences inked a deal in December 2018 with Bay Labs for AI-assisted product development in cardiac devices (here).


If It's in NEJM, It's Probably True

For another institutional endorsement, see the brand new review on machine learning in medicine by Rajkomar et al. in New England Journal, April 4, 2019, here. 

For recent webinars from Rock Health about "How to Exit [financially] in Digital Health" and from Accenture/Medtronic on "How AI Can Change the Future of Healthcare" - see here and here, respectively.


If the UK National Health Service Is Doing It, We Hope the Economics Are Sound

For a "Topol Report" which is a 50-page roadmap to digital health plans for the NHS over the next decade - here.