Healthcare AI News 7/3/24

News

The American College of Radiology launches an AI quality assurance program for radiology facilities that covers governance, algorithm documentation, security and compliance adherence, documenting use cases, and monitoring algorithm performance.

Mayo Clinic receives a $20 million gift to fund AI projects as well as Mayo Clinic Platform. The donors are Nvidia SVP of Software Engineering Dwight Diercks and his wife Dian.

Business

PicnicHealth, which gives patients and life sciences companies access to medical records, announces an LLM that collects a patient’s medical records from all providers. The company says it product connects to 100% of US care sites, enabling patients to collect all their records and for life sciences companies to conduct observational studies. CEO Noga Leviner had no healthcare background when she co-founded the company in 2014.

Research

An MIT study finds that medical imaging AI models often use demographic shortcuts that cause them to issue biased predictions. The authors say that models may work when first trained (locally optimal), but may not maintain fairness in new environments (globally optimal). They caution that models can predict demographic information from images – such as race, sex, and age – and then use that and other data to make correlations that have no clinical basis.

Other

Anesthesiologist, informaticist, and AMA immediate past president Jesse Ehrenfeld, MD, MPH says that the AI in healthcare hype curve is peaking, but he notes that a recent survey of doctors found that 38% are using AI in their practices, almost all of it for back-end office tasks. He says that it was a mistake to design, develop, and deploy EHRs without enough physician involvement, and that same error could be repeated with AI development. He explains, “I see this with entrepreneurial companies, where there’s a physician who might be involved, but it’s an afterthought. They’re not really driving the development of the solution. That’s a problem.”

Rep. David Schweikert (R-AZ) introduces legislation that would require Medicare to pay for AI-powered remote devices. He previous introduced bills that would allow AI to prescribe as a practitioner and to amend the Social Security act to pay for telehealth consultations such as AI-monitored wearables. 

I read about this Google search error and replicated it: searching for “114/74 blood pressure” brings up an AI Overview that says that this is an elevated reading. I repeated the test on Google Gemini as well as ChatGPT and both correctly indicated that the BP is normal. Interestingly, the AI Overview was correct when searching for 113/73 and 115/74, incorrect for 114/71, but when I searched for 114/75, it correctly said that’s normal but added “along with a pulse of 89 beats per minute” for some reason.

Contacts

Mr. H, Lorre, Jenn, Dr. Jayne.
Get HIStalk updates.
Send news or rumors.
Contact us.

Healthcare AI News 6/26/24

News

The Coalition for Health AI publishes a draft framework for the responsible use of AI in healthcare and solicits public comment. CHAI also posted a draft document of Assurance Reporting Checklists for self-reporting and self-review for assuring that AI solutions meet criteria in five areas: (1) usefulness, usability, and efficacy; (2) fairness, equity, and bias management; (3) safety; (4) transparency and intelligibility; and (5) privacy and security.

In Japan, SoftBank Group founder Masayoshi Son will make a rare public appearance in a Thursday panel discussion on bringing AI to healthcare in his country. The investment group took a significant position in Tempus AI just before its recent IPO and has signed an agreement to set up similar services in Japan.

Business

Wolters Kluwer Health will add AI-powered adaptive learning to its nursing education products.

Oracle announces GA of Clinical Digital Assistant ambient documentation and voice command system for ambulatory clinics that use its EHR.

Solventum, formerly 3M Health Care, announces an AI-driven payment denials prediction solution.

Real-world data vendor Atropos Health expands its Evidence Network to 300 million patients with the addition of two new partners. The company’s products can generate publication-grade, AI-powered observational studies from clinical questions with a two-day turnaround.

Shares in AI-enabled precision medicine vendor Tempus AI have lost one-third of their value since the company’s June 14 IPO. The company’s $8 billion pre-IPO valuation in 2022 is at $4.5 billion. Shares rose Wednesday on news that FDA has cleared sale of its atrial fibrillation detection algorithm for 12-lead ECGs.

Research

NIH awards Cleveland Clinic and other organizations a $2.8 million grant to study the use of AI to analyze MRIs to determine how rectal tumors are responding to therapy. Researchers hope to develop radiomic signatures using radiology and pathology images to identify dying tumors that have responded to therapy, which would allow those patients to avoid unnecessary further surgeries and complications.

Other

Stanford Medicine develops Nuclei.io, which helps pathologists diagnose endometritis and metastatic colon cancer with higher speed and accuracy than pathologists alone. The AI system observes individual pathologist behavior – such as image clicking and enlarging —  to learn what they are looking for and which cells are important.

Contacts

Mr. H, Lorre, Jenn, Dr. Jayne.
Get HIStalk updates.
Send news or rumors.
Contact us.

Healthcare AI News 6/19/24

News

Republican lawmakers criticize the FDA’s partnership with the Coalition for Health AI for testing healthcare AI products, raising concerns that CHAI’s members include Microsoft and Google as well as large health systems that are incubating their own AI-related businesses.

The state of California seeks bids for an AI system to translate health and social services documents and websites as a first draft for human review.

The American Society of Nuclear Cardiology joins the AMA’s effort to prohibit the use of AI by insurance companies to deny coverage based on medical necessity. AMA’s policy would require prior authorization requests to be reviewed by a physician who is trained in the same specialty as the requestor.

Business

Cancer genetic testing vendor Color Health is working with OpenAI to use GPT-4o to identify recommended tests and to create detailed workup plans for human review. The company expects to generate personalized care plans for 200,000 patients through the second half of 2024. 

Cognizant launches its first set of healthcare solutions that use Google Cloud’s AI technology, including workflows for appeals resolution, contract management, marketing content, and health plan shopping.

Patient-centered communications technology vendor CipherHealth announces AI summarization capability for its rounding toolkit.

Sweden-based startup Tandem Health raises a $9.5 million seed funding round with investors that included OpenAI and Deepmind. The company, which was founded less than a year ago and has 50 providers using its system in Sweden, provides a clinician copilot that can transcribe visits and create medical notes. It is looking for partners to help it expand in Europe.

Research

San Diego State University researchers develop an AI robot called Pepper that can detect emotions from live video analysis, allowing it to detect mental health issues and to provide insights to clinicians. The device, which was funded with a $5 million grant, is being reviewed by local hospitals.

Contacts

Mr. H, Lorre, Jenn, Dr. Jayne.
Get HIStalk updates.
Send news or rumors.
Contact us.

Healthcare AI News 6/12/24

News

Apple announces its long-awaited Apple Intelligence capability, which will be released in beta in IOS18 in September. It will work only on newer Apple hardware such as the IPhone 15 and IPads/Macs that are outfitted with the M1 chip. Its Siri voice assistant will use on-device AI to answer most questions, Apple-hosted LLMs for more complex questions, and OpenAI – with the user’s permission – to respond to more detailed questions. Planned features include a conversational Siri, word-based photo search, automatic photo editing, the ability to proofread and summarize documents, and the ability to summarize call audio.

The American College of Radiology asks members to weigh in on the request that was made to CMS by AI diagnostic software vendor Cleerly, which wants CMS to pay for its FDA-cleared CAD detection. The company says that using its $1,500 CT AI test reduced costs by avoiding invasive coronary angiography. Cleerly says that AI-enabled CCTA is better at diagnosing CAD than the traditional stress test.

Tampa General Hospital rolls out Nuance DAX Copilot to 500 of its affiliated physicians for ambient documentation.

UMass Memorial Medical Center uses Mednition’s Kate AI product for triaging ED patients.

University Hospitals deploys Aidoc’s AI platform across 13 hospitals and several outpatient locations. The company offers 17 FDA-cleared algorithms for triage, quantification, and care coordination.

Business

CCS launches PropheSee, an AI-powered predictive model that improves continuous glucose monitoring device adherence rates by up to 50%, saving an estimated $2,200 per patient per year.

Research

Researchers develop a publicly released trustworthiness benchmark for medical large vision language model that assesses trustworthiness, fairness, safety, privacy, and robustness.

Harvard-based pediatric physicians find that Google Translate and ChatGPT 4.0 perform expert-level translation of patient instructions to Spanish and Portuguese, but fall short of professional translations for Haitian Creole.

ChatGPT can generate expert-quality radiologic reports that are easier for patients to understand. The study authors note that it would be nearly impossible to expect radiologists to create patient-centered radiology reports at the required middle school reading level.

Researchers in China find that AI can accurately detect coronary artery disease using non-invasive facial temperature images. 

Contacts

Mr. H, Lorre, Jenn, Dr. Jayne.
Get HIStalk updates.
Send news or rumors.
Contact us.

Healthcare AI News 6/5/24

News

Researchers at George Mason University’s public health school create MeAgainMeds.com, a free AI-powered website that uses self-reported patient history and data that was extracted from 3.6 million patient records to help patients and clinicians choose the antidepressant that is most likely to be effective for them.

Kimberly Powell, healthcare VP/GM at Nvidia, says that she expects healthcare to be more affected by AI than any other industry, particularly in the area of drug discovery.

Business

Evolent acquires the software and development team of Machinify’s prior authorization and utilization management products. Machinify will continue its work with payer applications.

Innovaccer launches an AI-powered healthcare contact center that combines all available data to personalize the call center interaction of consumers.

Tampa General Hospital chooses Palantir Technologies to create an AI-powered care coordination system.

Spain-based AI medical triage chatbot vendor Mediktor acquires Sensely, which offers patients chat-based help with insurance.

Research

Researchers find that applying retrieval-augmented generation – where an AI system searches a specified database to craft its answers – improved the accuracy of ChatGPT-4 in answering cancer-related questions from 57% to 84%. However, it performed poorly when faced with conflicting advice statements from clinical trials, expert opinions, or committee recommendations.

Other

Children’s National Hospital hires Alda Mizaku, MS (Mercy) to the newly created role of VP/chief data and AI officer.

Montefiore Einstein Comprehensive Cancer Center doubles the colonoscopy completion rate among patients who cancelled or no-showed their colonoscopy appointments by using the AI tool MyEleanor from MyndYou to conduct voice call outreach. More than half of patients that the system called stayed on the line and half agreed to be transferred to a human navigator to reschedule.

Johns Hopkins palliative care doctor and cartoonist Nathan Gray, MD creates a fascinating “op-comic” for the LA Times that describes his experience with his hospital’s ambient documentation pilot project. He says that he had to learn to suppress the urge to type and had to remember to speak his findings out loud, but enjoyed “not having a screen stand as a dividing wall between doctor and patient.” He found that the program generated an accurate, coherent summary of the conversation even when it didn’t follow an obvious structure. His biggest complaint is that as a palliative care doctor, the AI ignored the hopes, dreams, and fears of the patient in seemingly mistaking that part of the conversation as chitchat.

Contacts

Mr. H, Lorre, Jenn, Dr. Jayne.
Get HIStalk updates.
Send news or rumors.
Contact us.

Healthcare AI News 5/29/24

News

Microsoft announces new features for Copilot that include expanded Teams capabilities, the ability to create custom copilots that act as agents, and extensions and connectors for developers.

OpenAI says in an unrelated blog post that it bas begun training the success to ChatGPT 4o, which it says will “bring us to the next level of capabilities on our path to AGI.”

XAI, which was founded in July 2023 by Elon Musk, will raise $6 billion in funding to bring its first products to market. The company’s first release was its Grok AI system.

Mayo Clinic will hold an AI summit on July 8-9 at its Rochester, MN campus and virtually, with a registration fee of $350 for either option.

Business

Google profiles 24 healthcare-related startups from Europe, the Middle East, and Africa that are part of the 2024 cohort of its health startup support program.

Videra Health, which offers an AI-driven mental health assessment platform, raises $5.6 million in seed funding. Its product uses patient self-assessments and AI analysis of video patient responses.

Atropos Health raises $33 million in a Series B funding round, which the company will use to expand the stake of its evidence-generating applications in value-based care.

Other

China’s Wuhan Union Hospital launches an AI pilot project in collaboration with Baidu Health that includes AI-powered patient triaging, appointment scheduling, and wait time management.

Contacts

Mr. H, Lorre, Jenn, Dr. Jayne.
Get HIStalk updates.
Send news or rumors.
Contact us.

Healthcare AI News 5/22/24

News

Epic is using Microsoft Azure’s Phi-3 small language model to summarize patient histories faster and cheaper than other generative AI models. Small language models can be run offline when extensive reasoning isn’t required and fast responses are needed. Phi-3 can also analyze images.

A VA official says that AI and predictive analytics have been a “game changer” in identifying and supporting veterans who are at risk for suicide. The REACH VET system, which was launched in 2018 and is active at 28 sites, identifies 6,700 veterans each month who need additional support.

Microsoft introduces Copilot+, which are AI-optimized Windows 11 PCs that feature high performance and a Recall function that takes regularly scheduled, locally stored screenshots that can be searched across applications, websites, and documents.

Y Combinator-backed Bland AI releases an AI phone agent that can carry on human-like conversations with callers. Website visitors can enter their name and phone number to initiate a sample call with AI agent Blandy. The platform can also perform live call transfers and add live data into phone calls. Suggested users are inbound sales, customer support, and B2B data collection.

Business

Remote patient monitoring solution vendor Implicity earns FDA’s 510(k) clearance for its AI-powered risk assessment for heart failure. The vendor-agnostic SignalHF analyzes data from implanted devices such as defibrillators, pacemakers, and cardiac resynchronization therapy devices. The company notes that 75% of the alerts that are triggered for patients who eventually require hospitalization are issued at least 14 days in advance, allowing time to adjust medications or take other measures.

Precision medicine system vendor Tempus AI files for an IPO at an estimated deal size of $600 million. The company reported $562 million in revenue for the 12 months ending March 31, 2024.

Research

Researchers review the technical ways that healthcare AI could be improved to provide more generalizable information about patient subsets, especially for underrepresented groups. They provide breast cancer as an example, where AI models underrepresent male patients and therefore should be selectively deployed where they are known to perform well, but call out the ethical challenges in adapting models based on sociocultural factors. They conclude that AI models should be trained on datasets that exclude samples that don’t meet carefully considered exclusion criteria.

Other

Two students of The University of Texas Health Science Center at San Antonio will graduate this month with dual degrees, the MD as well as an MS in artificial intelligence, a program that was launched last year as the first of its kind. Click the graphic for an enlarged view of the program’s timeline.

Robert Pearl, MD, Stanford University School of Medicine professor and former CEO of The Permanente Medical Group, offers some pearls (no pun intended) from his new book, “ChatGPT, MD”:

• AI can help clinicians override confirmation bias, where they might favor data elements that fit their mindset and ignore others.
• The short-term use of AI will be to manage chronic conditions, such as to analyze data from wearables, and to guide patients with personalized health recommendations.
• AI’s ability to analyze EHR data will transform medicine by helping clinicians understand the optimal ways to manage diseases and perform procedures.
• The biggest challenges to reap AI benefits are (a) clinicians have to be willing to empower patients; and (b) the fee-for-service payment model must shift to value-based care because doctors won’t do something that reduces their income and they aren’t paid to keep people healthy.
• Pearl lists ChatGPT as a co-author of his book, which he says shortened its completion time from two years to six months. He fed all of his writing into ChatGPT so it could understand his voice and writing style, then had it suggest changes to his draft versions.

Educator Henry Buchwald, MD, PhD says that AI can help overcome “natural stupidity,” comparing it to his time in the Air Force where he flew his plane upside down because he trusted his instinct instead of the instrument panel. He offers this:

In contrast to AI, let us examine natural stupidity. Unfortunately, there is an abundance of that in our world, perhaps a preponderance. In medicine, we hope that every physician is intelligent, or at least competent. But that may not be the case. When I was still in active academic practice, conducting patient rounds, I asked a medical student for his thoughts on a patient’s differential diagnosis and how he would proceed to narrow the potential options. He whipped out his iPhone to consult an algorithm. I told him to put the instrument away and to speak from the knowledge base of his nearly four years of medical training to make his own analysis of variables. He proved that he had little knowledge of established facts and, even when prompted, could not produce a reasonable thought sequence.

After graduation, this student became someone’s doctor, treating afflictions, counseling fellow human beings. Fortunately, after a patient consultation, this doctor will be able to postpone diagnosis and therapy by waiting for laboratory and imaging results, allowing him time to go to his iPhone, consult the algorithms and then come to a conclusion for the patient. In essence, AI may be the patient’s ghost doctor. In some instances, medical AI might prevent the physician from crashing into an unappreciated mountain.

Contacts

Mr. H, Lorre, Jenn, Dr. Jayne.
Get HIStalk updates.
Send news or rumors.
Contact us.

Healthcare AI News 5/15/24

News

OpenAI releases GPT-4o (the “o” stands for “omni”):

• All users, including those who use the free plan, will have access to the new version.
• Users can carry on voice conversations with ChatGPT with faster responses and can interrupt it when it is talking.
• The system can perform real-time language translation.
• It can interpret photos, screen shots, and other documents in real time.
• It can return real-time information from the web.

Google announces enhancements related to Gemini AI:

• Addition of 1.5 Flash, which is optimized for speed and efficiency with a lighter weight than 1.5 Pro.
• Adding audio understanding to Gemini 1.5 Pro.
• Extending Project Astra, the next generation of AI assistants that can understand real-time video and conduct conversations about what it sees, which can be extended to phones or glasses.
• Adding Gemini-powered AI Overviews to results from Google Search.
• Enhancing Google Photos with Ask Photos, which can find photos in response to user queries.
• Enhancing Gmail with generation of summaries and replies, complex search and the ability to analyze attachments.

Google launches AlphaFold3, which offers a 50% improvement in analyzing and predicting molecules for biology-related research and drug discovery.

Business

CLEW Medical’s learning models for predicting patient deterioration receive FDA’s 510(k) clearance

An Amazon Web Services blog post describes how Japan’s Fujita Health University used Amazon Bedrock’s managed AI service to generate discharge summaries, which it says reduced time requirements by 90 minutes to about 1 minute per patient.

An interesting Forbes article features digital health Investor Glen Tullman:

• He says that telehealth is dead, declaring that “all of the virtual care we’re doing now is going to move to chat.”
• His latest venture Transcarent – which has reached a $2.2 billion valuation for its chat-based primary care services —  is creating an AI chatbot that can answer a patient’s questions about their health plan, such as those that are related to deductibles and in-network doctors.
• Tullman says that Transcarent’s $100 million acquisition of 98point6 in 2023 gave it 1 million doctor-patient chat conversations that it is using for AI validation and testing.
• He says that Teladoc Health – onto which he unloaded Livongo for $18.5 billion in 2020 – saw its market cap shrink to around $2 billion because of poor leadership and the high cost of conducting video visits.

Sift Healthcare raises a $20 million Series B funding round to expand its AI tools that tie clinical information to financial outcomes to predict “adverse payment outcomes” and calculate patient responsibility upfront.

SmarterDx, which uses AI to audit and review medical claims, raises $50 million in a Series B funding round.

Other

A Wall Street Journal report on how professionals are using AI profiles a Mass General primary care doctor who says that ambient documentation is “incredibly accurate,” never requires major editing of the notes it creates, and allows her to interact with patients rather than typing.

America in a nutshell: a researcher finds that ChatGPT creates more compelling GoFundMe pitches for donations from strangers to cover catastrophic medical expenses. He notes, however, that sometimes AI creates stories and images that are too polished, which turns off potential donors. A research team used ChatGPT to rewrite the stories of 900 medical crowd-funding projects, where it added emotion and direct calls-to-action to move people to open their wallets.

Contacts

Mr. H, Lorre, Jenn, Dr. Jayne.
Get HIStalk updates.
Send news or rumors.
Contact us.

Healthcare AI News 5/8/24

News

The American Hospital Association asks Congress to employ flexibility in any decisions to regulate AI in healthcare, proposing use of a sliding scale that is based on risk and the level of human oversight that is involved. AHA says that a one-size-fits-all regulatory approach could stifle innovation, saying that a better model is that of the FDA’s guidelines for defining Software as a Medical Device.

Apple announces the 2024 IPad Pro, the company’s first product to feature the newly announced, AI-optimized M4 chip. Prices start at $999 before trade-in credit for the 11-inch, one-pound model with Ultra Retina XDR display, 256 GB, and 12 MP front and back cameras. The company is speculated to be preparing all of its hardware to run AI without further upgrades, and the Wall Street Journal reports that Apple is working on a chip to run AI software in data centers.

Business

Karius, whose early warning system diagnoses infectious disease by using AI to analyze a form of DNA from blood samples, raises $100 million in Series C funding. The company will use the proceeds to expand beyond its 400 US hospital customers of Karius Test, which can detect 1,000 pathogens from a single blood sample.

Opmed.ai, which uses AI to optimize surgery scheduling and resource allocation, raises $15 million in a Series A round.

Expion Health, which offers pharmacy and medical cost management solutions for payers, develops an AI-driven tool for responding to RFPs, which increases efficiency and reduces errors in its contracts in which it accepts financial risk.

OpenAI is working on a search feature for ChatGPT that would perform real-time web searches and return detailed information with citations, competing with Google’s search that includes paid ads and SEO-optimized, low-quality sites. Google generates $200 billion in annual search engine revenue, which represents about 80% of its total.

Research

Experts from drug maker Lilly say that AI will soon be able to design new drugs by using its own thinking instead of being trained on human-created datasets. They believe that even AI’s hallucinations will provide value by analyzing non-existent proteins within models.

Ohio State researchers train AI on de-identified medical claims data to predict the impact of available medical interventions on patient outcomes, which could supplement randomized clinical trials and explore new uses for existing drugs.

A UCSF study of emergency room visits finds that AI performed triage based on acuity as accurately as physicians. The authors note that since the commonly used Emergency Severity Index (ESI) is a rules-based triage scoring system that is straightforward for clinicians to calculate, AI might outperform them when making decisions that are based on both ESI score and patient details, such as answering difficult questions such as, “We have two patients and one available resuscitation bed – which one should get it?”

Other

A Mother Jones article says that AI chatbots not only haven’t eliminated health misinformation, they have made it worse. It says that GPT can’t tell the difference between scientific information and the information it crawls from questionable websites, refers users to sites that sell scientifically unproven health products, and is easily fooled by information that sounds authoritative but isn’t. A science communication professor says that AI, unlike humans, doesn’t have the capacity for critical thinking, applying skepticism, and seeking out facts that disprove inaccurate information. I replicated the authors question to Microsoft Copilot about alternative cancer treatment programs, which included a link to the website of for-profit, cash-only Cancer Center for Healing, whose physician owner sells cancer treatments that have not been FDA approved.

Contacts

Mr. H, Lorre, Jenn, Dr. Jayne.
Get HIStalk updates.
Send news or rumors.
Contact us.

Healthcare AI News 5/1/24

News

Google highlights the medical capabilities of its unreleased Med-Gemini AI models. The four models can provide text, image, and video outputs and are integrated with web search.

CLEW announces GA of its FDA-cleared, AI-powered system for early detection of patient deterioration. The company introduced its first AI prediction models for virtual ICUs in 2018 and has since expanded to cover all inpatient settings. CLEW’s CEO is industry long-timer Paul Roscoe, who has held leadership roles with Sentillion, The Advisory Board Company, Visionware, and Docent Health.

Several hundred nurses picket Kaiser Permanente San Francisco Medical Center to express their concerns that KP, which promotes itself as a leader in healthcare AI, will use the technology to improve its margins rather than patient care.

HHS OCR and CMS publish a rule that prohibits the use of patient care decision support tools –- their new name for what they previously called “clinical algorithms” – to discriminate based on race, color, national origin, sex, age, or disability. It requires Covered Entities to identify and review their use of any tools that take those factors into account to make sure they do not discriminate. It lists areas such as prior authorization, medical necessity determination, screening, risk prediction, diagnosis, prognosis, clinical decision-making, treatment planning, healthcare operations, and resource allocation in which race or ethnicity inputs might create discrimination.

Business

AI drug developer Xaira Therapeutics launches with $1 billion in capital. The company’s founding CEO is former Stanford University President Marc Tessier-Lavigne, PhD, who left that post in August 2023 when several of the research papers he published while working for Genentech were found to have used manipulated data.

Startup Dionysus Digital Health develops an AI-analyzed genetic test that it says can detect post-partum depression before symptoms appear. DoD has provided a $6 million grant for further study. The company hopes to earn FDA approval for the $250 test and have it covered by health insurance.

Quest Diagnostics will acquire PathAI’s Memphis-based digitized laboratory – the former Poplar Health, which it acquired in July 2021 — and convert it to Quest’s AI and digital R&D and solutions center to support its AmeriPath and Dermpath Diagnostics specialty pathology businesses. PathAI will continue to offer clinical trials services from its biopharma lab, which is a separate business.

Drug maker Moderna launches its own instance of ChatGPT that has been used by 80% of its employees. The company is using advanced analytics, image generation, and a library of 750 self-developed GPTs that help drive automation and productivity.  One of the GPTs evaluates optimal vaccine doses based on standard criteria and principles, supporting its conclusions with rationale, references, and charts that illustrate key findings, all of which are then reviewed by humans. The company identified its top 100 AI users to serve as champions and created an active internal AI forum. Moderna says that 40% of its active users have created GPTs and each user averages 120 ChatGPT Enterprise conversations each week.

Research

Researchers find that using AI to analyze the high-risk ECGs of hospitalized patients was associated with a 31% drop in all-cause mortality, the first randomized clinical trial to show that AI saves lives.

A study finds that ChatGPT-4 performed poorly in evaluating the data of chest pain patients to determine whether they should be admitted. Researchers noted particular concern that the model would give different risk estimates and test ordering recommendations for the same set of patient data that it reviewed more than once.

University of Texas Southwestern Medical Center researchers find that ChatGPT 3.5 can extract pathological classifications from lung cancer pathology reports with 89% accuracy, with most misclassifications being caused by lack of specialized pathology terminology and its faulty interpretation of TNM staging rules (tumor, number of affected nearby lymph nodes, and the presence of metastasis). The test was repeated using ChatGPT 4, which improved the model’s performance by 5% but at a higher cost.

Other

A Colorado attorney is suspended from the bar for one year and is fired by his law firm employer after admitting that he used ChatGPT to create legal documents. He says that he was trying to save his client money by using AI to perform case load research. ChatGPT generated what he says were “seemingly impeccable case law and case citations” that he didn’t double check, and a judge noticed that some of the cases cited were either incorrect or non-existent.

Contacts

Mr. H, Lorre, Jenn, Dr. Jayne.
Get HIStalk updates.
Send news or rumors.
Contact us.

Healthcare AI News 4/17/24

News

Kontakt.io raises $47.5 million in a Series C funding round led by Goldman Sachs. Launched in 2013, the multi-vertical company offers patient flow analytics and optimization software and hardware that leverages AI and RTLS technologies.

Business

MemorialCare (CA) selects Abridge’s generative AI software for clinical documentation.

Research

Researchers from University Hospitals Cleveland Medical Center, University of Southern California, and Johns Hopkins University use machine learning to develop a risk assessment model for bedsores that increases prediction accuracy to 74%, a 20% increase over current methods.

UMass Chan Medical School and Mitre launch the Health AI Assurance Laboratory, which will work to ensure the safety and efficacy of AI in healthcare through the evaluation of healthcare AI tools.

A Gartner survey of 600 enterprise software engineers finds that 75% say they’ll be using AI code assistants by 2028. Sixty-three percent of organizations are already using the technology in some way.

Wolters Kluwer Health publishes findings from a new survey focused on provider perceptions of AI. A few snippets:

• Eighty percent believe generative AI will improve patient interactions.
• Forty percent say they’re ready to start using generative AI in those interactions.
• Over 50% believe generative AI will save them time when it comes to summarizing patient data from the EHR, or looking up medical literature.
• Despite their enthusiasm for the technology, at least 33% say their organizations don’t have guidelines on how to use it.
• Providers are more enthusiastic about the technology than patients, with the majority of surveyed consumers in a previous study reporting they’d be concerned about using generative AI in a diagnosis.

AI-drafted physician messaging may not reduce response time to patient messages, but it does lessen cognitive burden, according to research out of UC San Diego Health. Lead researcher Ming Tai-Seale, PhD, explains: “Our physicians receive about 200 messages a week. AI could help break ‘writer’s block’ by providing physicians an empathy-infused draft upon which to craft thoughtful responses to patients.”

Other

OSF HealthCare (IL) will pilot personalized customer engagement AI assistants from Brand Engagement Network at several facilities as a part of its continuing education simulation training for its Advanced Practice Provider primary care fellowship participants.

Contacts

Mr. H, Lorre, Jenn, Dr. Jayne.
Get HIStalk updates.
Send news or rumors.
Contact us.

Healthcare AI News 4/10/24

News

Google opens Gemini 1.5 Pro to developer preview via API, which includes native speech understanding and a file handling API. The system was tested with a 400-page transcript from the Apollo 11 mission, about which it could answer questions and details from conversations and images using up to 1 million tokens.

Meta says it will release its Llama 3 open source LLM within one month, with double the parameter size of Version 2 and about the same as Open AI’s GPT-4.

Business

Bayer and Google Cloud will collaborate on developing AI solutions for radiologists.

Northwestern Medicine and Dell’s AI Innovation Lab create an LLM solution for drafting x-ray reports. They are also working on a predictive model for the EHR.

All of India’s major hospital chains are using AI, with Apollo Hospitals testing it for patient monitoring, cardiovascular disease prediction, symptom checking, radiology workflow, and in-room automation. Apollo says that some of the technologies have earned US FDA clearance.

Research

Researchers document the steps that were involved in Duke University Health System’s design, development, and maintenance of the AI-powered SepsisWatch system. They hope to illustrate the lessons that were learned in developing algorithms, involving stakeholders, and setting up an organizational structure.

Researchers find that an algorithm can predict hospital-acquired pressure injury with 74% accuracy, extending EHR-calculated Braden Scales into an early warning system that could save a 500-bed hospital up to 90,000 labor hours and $18 million in expense each year.

Other

Penn’s medical school names Marylyn Ritchie, PhD, former director of biomedical informatics, its first vice dean of AI and computing. University of Pennsylvania Health System also promotes former radiology chief Mitchell Schnall, MD, PhD to SVP of data and technology solutions, with a focus on AI.

International health leaders list four action areas that should be addressed to fully realize AI’s healthcare potential:

• Identify high-priority data elements that are needed for AI applications and ensure that availability of those elements is reliable.
• Test AI tools to ensure that they are safe and effective within specific patient populations and are free of bias.
• Standardize business processes so that data can be shared between institutions, such as connecting encounter data to outcomes, and explore privacy-preserving data sharing approaches.
• Advocate for paying for value (quality, safety, health, and costs) to align the financial incentives for using AI.

Contacts

Mr. H, Lorre, Jenn, Dr. Jayne.
Get HIStalk updates.
Send news or rumors.
Contact us.

Healthcare AI News 4/3/24

News

Plastic surgeon Robert Pearl, MD, who was executive director and CEO of The Permanente Medical Group through 2017, predicts that ChatGPT’s eventual ability to remember and use information across multiple chat sessions — its “context window” — will be significant in healthcare. He says that while its current capacity won’t hold an average medical record, generative AI systems are expected to become 30 times more powerful in the next few years, which along with GPTs, will allow records from multiple providers to be consolidated to provide patients with diagnosis and treatment suggestions. He also expects the extended memory to improve patient monitoring with its capability to look back into historical clinical status and to issue real-time alerts when established best practices are not followed.

The World Health Organization launches Sarah, an AI-powered health information avatar for consumers. Sarah replaces a previous WHO non-AI tool called Florence that provided public health information. Sarah is available online.

Epic is developing an AI validation suite to help health systems evaluate and monitor the performance of AI models, both third party and self developed.

OpenAI offers access to ChatGPT 3.5 without requiring registration, although functionality is limited and content safeguards are expanded over the paid version.

A new KLAS report looks at early use of healthcare AI, concluding that:

• ClosedLoop’s predictive modeling leads the market in performance, especially with ACO and health plan users.
• Customers of second-place Epic are expanding beyond clinical use cases into operational and financial areas, with good employee adoption because of integration with EHR workflows. Some customers complained about Epic’s pricing and its subpar ability to incorporate outside data.
• Oracle Health customers were split. Satisfaction was good with the company’s cloud plans and its use of clinical data, but some customers complain that support quality and nickel-and-diming has gotten worse since Oracle acquired Cerner.
• Health Catalyst customers say the product is robust, but half of them believe that the product fails to deliver on its potential because of insufficient training and support.

Business

Amazon ends its Just Walk Out checkout-free experience at its Amazon Fresh grocery stores, which will be replaced by in-cart scanners that customers use themselves. A report by The Information says that while the Just Walk Out system seemed automated and AI powered, 1,000 offshore workers served as low-paid remote cashiers in monitoring shoppers via video camera and manually ringing up items. Social media wags labeled the failed experiment as API – A Person in India. Amazon rolled the system out to hospitals in January 2024 for cashier-less employee purchases, with St. Joseph’s / Candler as its first customer.

Healthcare AI company John Snow Labs releases a no-code solution that allows healthcare domain experts to train, tune, test, and share models without help from data scientists. The company says that creating small, task-specific models can be used to train other models, are cheaper to run, and can operate behind an organization’s firewall. They can also include human-in-the-loop oversight.

FDA grants marketing clearance to the sepsis detection tool of Prenosis, which applies AI to biomarker and EHR clinical data to assess sepsis risk.

Research

Penn researchers find that AI analysis of Facebook postings is three times less likely to accurately predict depression among black American users as compared to white. The model uses prior findings that white people who have higher levels of depression use more first-person singular pronouns — along with words that suggest isolation, self-criticism, worthlessness, anxious-outsider, and despair — but the model performed poorly when analyzing posts from people who are black. The authors note that their study illustrates the need for caution when applying AI in mental health settings.

Other

HFMA notes the “battle of the bots” as hospitals that have been outgunned by payers that used technology to deny or delay payment are implementing AI to beef up revenue cycle management. Examples of provider use of AI:

• Using robotic process automation to ensure that claims are clean and to manage work queues (Luminis Health).
• Reducing RCM headcount by 30 FTEs by using AI bots (Mayo Clinic).
• Reducing authorization-related denials by using bots to notify payers when patients are admitted (Care New England).
• Increasing administrative efficiency by using Copilot for Microsoft 365 and using RPA to save labor in authorization, registration, credentialing, and billing workflows (Corewell Health).

Contacts

Mr. H, Lorre, Jenn, Dr. Jayne.
Get HIStalk updates.
Send news or rumors.
Contact us.

Healthcare AI News 3/27/24

News

OpenAI files a trademark application for a voice engine and digital voice assistants, possibly signaling that the ChatGPT developer plans to complete with Apple’s Siri and Amazon’s Alexa voice assistants.

An official with the Congressional Budget Office says that AI could reduce healthcare costs by identifying patients who could benefit from early treatment, but it could also increase costs by spurring the development of beneficial but expensive technologies or identifying more patients who need treatment.

Business

Amazon will invest another $2.75 billion in generative AI startup Anthropic, which it had already backed with $1.25 billion, valuing the maker of the Claude chatbot at $18 billion.

Healthcare business process technology vendor Sagility acquires BirchAI, which offers AI-powered call center technology.

Tennr, which use AI to extract and route data from faxes, raises $18 million in Series A funding.

The London-based non-profit Institute for Public Policy Research makes a worst-case prediction that AI could eliminate 8 million jobs with no gain in gross domestic product. The authors predict that a Phase 2 of AI implementation in which AI is allowed to execute tasks could affect financial workers, shop owners, and IT managers, while advancing to Phase 3 – where processes are built around AI and people accept interacting with avatars – could transform the work of teachers, doctors, and hospitality workers. They recommends creating policies that protect tasks where human involvement is valued, consider imposing wealth taxes or social security assistance, and raising taxes companies that replace people with AI.

Research

NIH provides funding for Treatment.ai to develop a culturally sensitive AI approach for collecting family medical history. The company says that approaches for collecting medical histories for African-American families are biased because they do not adequately capture information about blended families, non-traditional relationships, and the lack of patient information about the health of relatives.

Researchers find that AI can successfully design new drugs to overcome bacterial resistance.

ChatGPT generates discharge documents for orthopedics case that are comparable in quality to those that are created by junior orthopedic surgeons and orthopedics residents, but 10 times faster. Interestingly, while reviewers found hallucinations in four documents that ChatGPT created, that was fewer than the six that were found in the physician-generated notes.

A West Virginia University School of Pharmacy study will look at using AI to collect patient medication information from clinical and billing systems, including clinician notes, to help with medication reconciliation and to determine the risk of readmission.

Other

The National Bureau of Economics Research publishes a grant-supported book titled “The Economics of Artificial Intelligence: Health Care Challenges” that analyzes barriers – incentives, management, data availability, and regulation – that will impact the efficiency and cost disruptions that are possible in the 20% of the US economy that involves healthcare.

Contacts

Mr. H, Lorre, Jenn, Dr. Jayne.
Get HIStalk updates.
Send news or rumors.
Contact us.

Healthcare AI News 3/20/24

News

Google lists its recent accomplishments in applying generative AI to healthcare:

• The introduction of MedLM for Chest X-ray to classify images, which is being tested by users.
• Tuning a version of its Gemini model the medical domain to determine its capabilities for advanced reasoning and using high context volumes, which it is testing for analysis of images and genomics information.
• Working with its Fitbit business to develop a personal health LLM that uses data from the Fitbit wearables and app and from Pixel devices.
• Working with a healthcare organization to test the capability of its Articulate Medical Intelligence Explorer chatbot for conducting text-based consultations.

Nvidia Healthcare launches 25 healthcare microservices that allow developers to integrate AI into new and existing applications.

CommonSpirit Health launches an internal AI assistant to help its employees generate written content.

Zephyr AI, which offers explainable AI-powered precision medicine solutions, raises $111 million in a Series A funding round.

Toronto-based Pocket Health raises $33 million in Series B funding. The nine-year-old company offers a patient-centric image exchange platform that explains medical terms, detects follow-up recommendations, and suggests questions that the patient should ask their doctor. Patients pay $49 per year for the service. 

Business

Hippocratic AI raises $53 million in a Series A funding round that values the company at $500 million. It also announced that it is performing safety testing of a healthcare staffing marketplace where AI software agents can be “hired” to perform non-diagnostic, patient-facing tasks. The agents, use of which the company says cost less than $9 per hour compared to high nurse salaries, are evaluated in detail and scored by nurses and physicians for bedside manner, the ability to provide patient education, lack of bias, and safety. The company has said that AI will reduce the incremental cost of human-powered access and intervention activities to nearly zero as the only scalable way to address healthcare worker shortages.

Assort Health raises $3.5 million in funding and launches an AI solution for healthcare call centers. The company uses AI and natural language processing to understand a caller’s intent, then uses available data – including that from the EHR – to resolve their inquiries. Common tasks are call routing, patient registration, appointment scheduling, managing appointment cancellations and confirmations, and answering frequently asked questions.

Saudi Arabia will create a $40 billion investment fund for AI, making it the world’s largest AI investor as it moves to diversify its economy.

Nvidia announces an AI-powered humanoid robot system that can understand natural language and learn to move by observing humans.

Microsoft hires the co-founders and most of the employees of Inflection, which had raised $1.3 billion – with Microsoft as the lead investor — to create the “first emotionally intelligent AI” chat tool called Pi that it launched just 10 months ago. Mustafa Suleyman, who was CEO of Inflection and also a co-founder of DeepMind, was announced as CEO of Microsoft’s consumer AI business. The future of Pi was not addressed.

Research

Harvard Medical School and MIT researchers determine that individual radiologists react differently to AI-powered assistance, with some of them showing worse performance and accuracy when assisted by AI. They were surprised to find that radiologist experience didn’t correlated to AI-enabled changes and that AI didn’t help low-performing radiologists improve. The authors recommend further study of how humans interact with AI and that the technology be personalized based on clinician expertise, experience, and decision-making style.

Researchers train AI systems to think carefully before responding to user requests, which improved the reasoning ability of those systems. Users of ChatGPT have reported that the system provides better answers to prompts if they include instructions to think carefully or that offer an imaginary reward for a better answer.

Other

A JAMA op-ed piece questions whether clinician double-checking should be the primary protection against healthcare-related AI harming patients, noting that “humans are terrible at vigilance.” The authors suggest five approaches based on their use in other industries:

1. Let AI assess its own level of certainty and present its results that are color-coded based on its confidence, including a warning if the specific patient is not representative of the population that was used to train the model.
2. Detect clinicians who exhibit automation bias by nearly always accepting AI’s recommendations or suggested text.
3. Use any AI time savings to address clinician burnout instead of raising productivity expectations.
4. Program the AI system to deliver “deliberate shocks,” similar to TSA airport screening systems that randomly add an image of a firearm to keep operators vigilant.
5. Instead of using AI on the front end, program it to work like a spell checker to analyze clinician conclusions and highlight areas that seem to be at odds with its analysis.

A patient who was diagnosed with Type 1 diabetes uses ChatGPT to determine that he instead has a rare genetic condition that is treated by daily medication instead of careful diet monitoring and an insulin pump. His endocrinologist says that MODY (maturity-onset diabetes of the young) is often misdiagnosed because of the high prevalence of Type 1 and Type 2 diabetes.

Contacts

Mr. H, Lorre, Jenn, Dr. Jayne.
Get HIStalk updates.
Send news or rumors.
Contact us.

Healthcare AI News 3/13/24

News

Sixteen large health systems and Microsoft form the Trustworthy and Responsible AI Network (TRAIN), which will share best practices, register the use of AI in clinical settings, provide AI outcomes measurements, and facilitate creation of a federated national AI outcomes registry. The group will work with OCHIN and TruBridge to make sure that all organizations, regardless of resources, can benefit from TRAIN’s work. The recently formed Coalition for Health AI (CHAI) is a partner and says that TRAIN will operationalize its principles in health systems.

Google launches Google Cloud Vertex Search for Healthcare, a set of search-and-answer tools that integrate with its MedLM, Healthcare Data Engine and Cloud Healthcare FHIR APIs. It offers medically-tuned search of EHR data for clinician presentation, configurable cloud APIs, the ability to answer clinician questions while understanding medical terminology, and providing links to source data points to provide transparency.

The government of China is working with private companies to develop a medically focused AI chatbot based on Meta’s Llama 2.0. It is testing an AI assistant model for neurosurgeons at seven hospitals in Beijing. The CARES Copilot 1.0 was trained on medical literature and can process images, audio, text, MRI, ultrasounds, and CT scans.

Microsoft previews healthcare-specific safeguards for Azure AI Health Bot services that integrate with Copilot Studio to allow healthcare organizations to develop their own copilots.

Business

Symplr announces SymplrAI, which will offer healthcare efficiency tools via Amazon Web Services AI/ML.

Peter Thiel-backed Cognition AI unveils Devin, an AI-powered software engineer that can generate, debug, and deploy code for websites and videos from a single command. It has its own command line, code editor, and browser.

Amazon Web Services lists the generative AI solutions that its healthcare customers have created:

• Digital pathology (Philips).
• Phone-based, voice-enabled progress note generation (Pieces Technologies).
• Hospital operations efficiency (Symplr).
• Enterprise imaging (Konica Minolta).

Biotech startup Insilico Medicine describes how it used its AI platform to create a drug candidate for idiopathic pulmonary fibrosis that is undergoing Phase II clinical trials. The company says the process took 2.5 years versus the six years that would have been needed without AI. The company used biology AI to discover the drug, then generated the molecule using chemistry AI.

Research

In England, University of Oxford researchers warn clinicians that using ChatGPT and Bard to create care plans could compromise patient privacy or mislead users with incorrect information.

In Israel, Sheba Medical Center researchers develop an AI platform to help mental health professionals diagnose and manage conditions efficiently, concluding that its diagnostic accuracy is comparable to that of experience psychiatrists.

A study finds that large language models can translate discharge summaries into readable, patient-friendly language, which helps providers comply with requirements to give patients immediate access to their discharge notes. However, the authors found examples of key information being omitted, most often history of present illness and procedures, which they attribute to prompt engineering that optimized readability in the form of shorter documents.

Other

CommonSpirit Health EVP/CIO Daniel Barchi, MS says that healthcare’s adoption of AI should follow the Maslow’s Hierarchy of Needs in first being applied to technical needs, then moved up the triangle to improve healthcare efficiency and safety. (click graphic to enlarge)

Contacts

Mr. H, Lorre, Jenn, Dr. Jayne.
Get HIStalk updates.
Send news or rumors.
Contact us.

Healthcare AI News 3/6/24

News

UPMC physicians develop a AI-powered smartphone app that can diagnosis acute otitis media by analyzing video from a phone-connected otoscope.

Pediatric clinicians at Mass General Brigham create a series of 45 instructional smartphone videos for clinicians, then use ChatGPT to create Spanish language versions that they distributed to nurses in Guatemala and Colombia.

The Coalition for Health AI names co-founder Brian Anderson, MD as CEO and co-founder John Halamka, MD, MS as board chair. The company added federal leaders to its board and will work with the government to develop quality and safety standards.

The American Medical Association’s house of delegates defines AI as “augmented intelligence” rather than “artificial intelligence” to focus on AI’s role of enhancing human intelligence rather than replacing it.

Business

UMass Memorial Health will use Google Cloud to build patient care and research tools.

Health and human services software vendor Healthy Together acquires Kinsa Health, which offers AI-powered epidemiological prediction models using consumer smart thermometers as well as healthcare demand forecasting. Kinsa had shut down several months ago after 12 years, with founder and CEO Inder Singh expressing hope that he could find a new home for the company’s work.

Machinify launches an AI-powered healthcare claims system that includes apps for prior authorization approval, claims auditing, and claims review and correction. Founder and CO Prasanna Ganesan, PhD was a co-founder of digital video store Vudu, which was acquired by Walmart in 2010 and then in 2020 by Fandango, which now brands it as Fandango At Home.

Research

The federal government awards Philadelphia-based non-profit Every Cure a $48 million grant to use AI to find new uses for existing drugs. Penn medical school professor and immunologist David Fajgenbaum, MD, MBA, MSc co-founded the organization after saving his own life by finding a “repurposed” drug, a process he described in his book.

Other

Attorneys list potential areas of liability exposure for using healthcare-related AI, urging providers to avoid Practice Fusion type kickback situations by vetting the product’s explainability and the governance and privacy practices of its vendor:

• Payers using AI to manage prior authorization while denying legitimate claims or improperly influencing the process.
• Using AI to analyze medical images that could raise questions about how the system were trained, whether their vendors are paid for the volume and value of the resulting referrals, and kickback implications of AI systems that are tied to specific treatments.
• Possible clinical studies fraud in which AI could be tweaked to overstate drug efficacy.

Open Notes launches OpenNotes Lab, which will advocate and study the use of AI to enhance trust and communication between patients and their care teams. The company is interesting in partnering with vendors, health systems, patient advocacy groups, professional associations, and regulatory agencies.

Contacts

Mr. H, Lorre, Jenn, Dr. Jayne.
Get HIStalk updates.
Send news or rumors.
Contact us.