The National Academy of Medicine publishes “Artificial Intelligence in Health Care: The Hope, the Hype, the Promise, the Peril.” Its major points:
- EHR and consumer data are widely available, but wider adoption of common data models and FHIR are needed to support AI projects.
- Inclusion and equity must be incorporated to prevent expanding existing health outcomes inequities as has occurred with other consumer-facing technologies.
- Transparency guidelines need to be developed to create trust.
- Near-term focus should be on supporting what clinicians already do rather than replacing them, such as by providing guidance to non-specialists, filtering low-acuity or normal cases, addressing inattention and fatigue, and automating business processes.
- AI training and education should be incorporated into continuing medical education.
- Health systems should implement AI solutions only if their IT governance process is mature and only if no low- or no-technology solutions already exist. National efforts will be required to support AI deployment in lower-resource environments to support healthcare equity.
- Regulatory challenges will remain for AI developers, but the FDA approach of considering the level of patient risk, the level of AI autonomy, and the level of static or dynamic AI behavior should be taken into account and post-marketing surveillance is needed to evaluate a given model’s ongoing learning.
Randy Bak, MD, JD added a comment — in response to my observation that doctors aren’t good at practicing evidence-based medicine – that is worth running here (with my edits):
Not all patients fit the target of a care guideline. They become care guidelines when most patients should be treated that way. You could call the result art rather than science, but some hardheadedness still applies. If you treat most patients as exceptions to the guideline, then you are out of bounds, just as you are if you treat all patients by the guideline. The key is understanding what makes an exception, and even then realizing that sometime you will be wrong.
Managers looking at how clinicians respond to guidelines need to look at actual practice, but they also need to apply the same kind of hardheadedness about measurement and its conclusions.
Small sample sizes don’t tell you a lot about a practice pattern. They say it takes about 30 samples to get a reasonable approximation of the normal curve of a phenomenon, so if you start judging physician practice based on 10-20 cases, you’re looking for trouble. Even when you get decent sample sizes, they remain just that– samples. Regression to the mean is a real phenomenon. Worse is that, especially in low sample-size settings, last year’s champion can be next year’s black sheep. Sampling must be repeated over time get to the “truth.”
Use case exists where real measurement can be applied, such as surgical procedures. Just about every practice has something that occurs frequently enough to allow reliable measurement. There is not infrequently a halo effect or inference that can be made from what is measurable to what is not, which can drive management of that clinician. Still, caution is due.
As in sports, individual measures may not tell the whole story. I am fascinated at how pro sports geeks have gone “moneyball” on metrics, trying to find measurements that tell them how to spend their team budget. Is there a way to get to “outcomes above replacement” or such things that tell you this clinician improves the care all around them?
HIStalk Announcements and Requests
Unrelated, other than seasonally: Rev, the transcription company I use for interviews, sent a holiday email that contains the perfect mix of humor, holiday cheer, and sly self-promotion.
Thanks for the cool holiday swag from Ellkay, which included several flavors of honey from the company’s rooftop beehives. I don’t usually get vendor marketing stuff other than at the HIMSS conference, but Ellkay’s is the best, and the honeybee connection is the most memorable, feel-good tie-in that I can think of.
Acquisitions, Funding, Business, and Stock
RCM vendor Streamline Health will sell its legacy enterprise content management business to Hyland.
- Baptist First Health (KY) will integrate ActX’s genomic decision support software with its Epic EHR.
Aspire Health co-founder Brad Smith will become the new head of the Center for Medicare & Medicaid Innovation.
Augmedix names Davin Lundquist, MD (CommonSpirit) as chief medical officer.
Announcements and Implementations
OhioHealth implements KitCheck’s Bluesight for Controlled Substances across 10 hospitals.
Rush University System for Health (IL) integrates Mytonomy’s Patient Experience Cloud care education software with Epic’s MyChart.
Online health and wellness company Hims & Hers will offer customers in Florida access to telemedicine for chronic conditions through Ochsner Health System (LA) beginning next year.
Employer-sponsored provider QuadMed implements Epic.
KLAS is apparently branching out from purely technology coverage given its new reports on worksite health services and value-based care consulting. I’m not all that interested in either, but the first report gives Cerner a B- and QuadMed – announced above as having implemented Epic – a D+, while the second puts Deloitte at the top as a transformational partner.
Government and Politics
CMS temporarily shuts down the Blue Button 2.0 system after a developer notifies the agency of a bug that may have exposed Medicare beneficiary data. CMS will restore service after it finishes a quality and validation review.
Privacy and Security
LifeLabs, Canada’s largest laboratory testing company, notifies patients of an October ransomware attack that compromised a server used for online appointment bookings. The company, which admits that it paid the hacker’s demanded ransom, says 15 million customers were affected and the lab results of 85,000 of them were exposed.
Standards developer NCPDP and Experian Health announce that they have assigned a Universal Patient Identifier to all 328 million Americans. Experian Health creates the UPI when a a provider, pharmacy, or lab sends it patient demographic information, then sends back specific identity information. The assigned UPI itself is not disclosed to the patient or provider to prevent its misuse.
Hospitals report that they are being inundated with requests to sell patient information to technology companies, many of them well-funded Silicon Valley startups that need to train their newly developed AI systems. Jefferson Health says companies that get a firm “no” from its executives then try to twist the arms of individual doctors and researchers. Jefferson Health’s cancer center director Karen Knudsen, MBA, PhD drily observes, “We often find, once we look deeper into the pitch, that it starts as a joint development project and ends up somehow with us being both the product and the customer that pays for the product.”
The CEO, a director, and four researchers of H. Lee Moffitt Cancer Center & Research Institute (FL) resign after its compliance department finds conflict of interest violations in their ties with research organizations in China. The cancer center’s founder says the group was found to be “secretly accepting money from China.”
Massachusetts General Hospital scientists say they can predict dementia by scanning their EHR data for a list of cognitive-related terms using natural language processing.
MIT researchers say their Gates Foundation-funded, patch-based vaccine delivery system would not only eliminate the need for syringes, it wouldn’t require an EHR for documentation either since the patch leaves a skin pattern that can be detected by smartphone.
- PatientPing makes Vynca’s advance care planning data available to providers within its real-time care collaboration network.
- Meditech releases a new video, “Meditech Expanse Delivers the Power of Mobility to Androscoggin Valley Hospital.”
- Pivot Point Consulting names Kyle McAllister (Nordic) director of strategic implementation.
- Greenway Health makes the Carequality Interoperability Framework available to its Prime Suite and Intergy EHR customers via the CommonWell Network.
- CHIME interviews The HCI Group’s VP of Operations and Strategy, Chris Belmont.
- PatientSafe Solutions is recognized in Gartner’s “Market Guide for Clinical Communication and Collaboration.”
- The New Pittsburgh Courier honors ConnectiveRx Director of Pharmacy Operations Natalie Tyler with a Women of Excellence Award.
- LaTonya O’Neal (Change Healthcare) joins The Chartis Group as principal.
- Vyne Medical and its Trace interaction capture solution are featured in KLAS’s “2019 Revenue Cycle Unicorns Report.”
- Cigna expands its relationship with MDLive to include virtual visits for behavioral healthcare.
- Three tips for maintaining an effective Master Training Environment (Nordic)
- How to Better Understand the Benefits of Health Data Standards (EClinicalWorks)
- 4 C’s of Millenial Healthcare (Optimum Healthcare IT)
- Clinical Communication Etiquette: When Should You Text, Call, Broadcast or Video Chat? (Mobile Heartbeat)
- Ensocare Product Showcase Creates Wow Factor for Team Members (Ensocare)
- How do you measure your return on analytics? (EPSi)
- What is a healthcare clearing house in medical billing? (Greenway Health)
- Using Data Analytics to Ensure Case Mix Index Accuracy (Hayes Management Consulting)
- Troubleshooting Slow Logons when User Profiles are Root Cause (Goliath Technologies)
- Healthfinch Resources: The Best of 2019 (Healthfinch)
- Has UCLA’s New PRIDE Study Perfected Shared Decision Making for Diabetes Patients? (Healthwise)
- Social Determinants of Health Data Analytics Top Disruptor in 2020 (Imat Solutions)
- Three tips to Jumpstart your Distressed IGA Deployment (Imprivata)
- How Providence St. Joseph Health is Digitizing the Consumer Healthcare Journey (Kyruus)