Darwin's Our Take 6.8.26: Feature: Health systems gain AI governance controls as collaborations proliferate

Feature: Health systems gain AI governance controls as collaborations proliferate

The Coalition for Health AI (CHAI) recently released a series of governance playbooks developed to provide healthcare organizations with baseline controls they can use to implement artificial intelligence safely and transparently.

The goal, CHAI said in a news release, is for organizations to be able to integrate the controls into existing processes. The playbooks offer examples, guidance for implementation, tools, and resources to assist organizations in applying the controls.

More than 150 health AI leaders participated in a series of workshops and workgroups to create the playbooks. The collaboration involved 100-plus academic medical centers, regional care facilities, community health centers, and other organizations.

“We’ve been hearing the increased need from our members for practical, consistent, and comprehensive guidance for responsible AI deployment. Our working groups set out to tackle this challenge, and I’m encouraged by how much collaboration from across the industry occurred to deliver today’s playbooks,” said Dr. Brian Anderson, CEO of CHAI.

“These efforts don’t just help to define responsible AI, they intend to make it usable for healthcare delivery organizations across the country — regardless of resource level and all with the goal of translating AI innovation and high-quality care for every patient,” Dr. Anderson said.  

The playbooks address what the workgroups identified as “critical elements of responsible AI use,” including organizational AI policy, structure, and resources; responsible AI lifecycle management; risk and impact assessments; responsible data management and use; third-party management; and education, training, and feedback.

“These resources bring much-needed structure to one of the most important challenges in healthcare AI: turning good intent into governed, measurable, and sustained practice,” said Taylor Rhodes, director of Mercy Health’s responsible AI program.

The playbooks are available here.  

In separate news, Mayo Clinic and Microsoft are developing a frontier AI model “purpose-built” for healthcare. It is designed “to synthesize diverse clinical data to support earlier diagnoses, more personalized treatment decisions, and better patient outcomes,” the organizations said in the announcement.

Mayo Clinic will own the model, which will be deployed initially within Mayo Clinic’s clinical environment to allow for continuous testing and refinement through real-world use.

Eventually, Microsoft will make the model available to other healthcare organizations through Azure Foundry APIs, giving them access to advanced healthcare AI capabilities.

For years, Mayo Clinic has been developing AI-powered models and tools across multiple medical disciplines in specialized research laboratories.

One of those models is an automated AI framework called the Radiomics-based Early Detection Model, or REDMOD. In late April, study results showed that REDMOD can help detect pancreatic cancer on CT scans up to three years before clinical diagnosis, identifying signs of the disease before tumors are visible.

In early May, Abridge published a blog post about the ambient documentation tool it codeveloped with nurses at Mayo Clinic. Instead of merely modifying an existing solution for use by nurses, Abridge said developing the tool involved listening as nurses described their workflows and the challenges they faced, and finding out what “meaningful support” looked like to them. Abridge calls the tool For Nurses, By Nurses.  

Meanwhile, Mass General Brigham reported in late April that an AI tool used to estimate biological age from facial photos might serve as a noninvasive biomarker for cancer prognosis.

The health system’s FaceAge tool can estimate a person’s biological age from just one photo. Using multiple photos, it measures face aging rate (FAR) — changes in a person’s biological age over time.

Researchers used the AI tool to analyze two photographs from each of nearly 2,300 patients with cancer taken at different points during their course of treatment. The team found that higher FAR was significantly associated with decreased survival probability.

“Our study suggests that measuring FaceAge over time may refine personalized treatment planning, improve patient counseling, and help guide the frequency and intensity of follow-up in oncology,” said Dr. Raymond Mak, a radiation oncologist at Mass General Brigham Cancer Institute and a faculty member in MGB’s Artificial Intelligence in Medicine program.

The University of Southern California announced a partnership in April with Chicago-based Tempus AI to advance precision medicine through the use of AI. Their collaboration entails integrating AI tools across the USC health system, including the USC Norris Comprehensive Cancer Center and Keck Medicine of USC.

For instance, Tempus’ advanced molecular diagnostics and genomic profiling will be integrated into Keck Medicine’s clinical workflows for clinical testing. Other efforts will focus on clinical trial matching, clinical care gap pathways, and research collaboration and codevelopment.

Cleveland Clinic and Aspira Women’s Health, a bioanalytics company based in Austin, Texas, are collaborating on an initiative to improve AI-powered diagnostics.

The initiative will involve research projects designed to discover and validate biomarker signatures. The organizations said in a news release they will also work on developing advanced analytical models to improve diagnostic accuracy, clinical utility, and patient care.

Through their joint efforts, the organizations envision expanding patient sample access, enhancing clinical validation capabilities, and supporting accelerated development and future commercialization of next-generation multiomic diagnostics across women’s health.

The agreement they entered into sets up a framework for platform expansion, including biomarker discovery, data generation, translational development, intellectual property collaboration, and clinical research initiatives.

What else you need to know

More hospitals are offering provider-sponsored health plans (PSHPs), indicating “a strategic evolution toward the payvider model,” according to a JAMA research letter. Traditionally, large insurers have led payer-provider integration in an effort to control the cost of care and steer plan members to clinics they own. However, the ongoing shift to value-based care is driving more hospitals to integrate PSHPs into their financial risk strategies.

An analysis of data from American Hospital Association Surveys conducted from 2018 through 2023 showed that, nationwide, hospital PSHP ownership increased significantly during that period, from 692 to 834 (or an increase from 18.3% to 27.2% of the hospitals surveyed). Regionally, the West had the most growth in PSHP ownership, while the South maintained lower PSHP prevalence and experienced only modest growth.

In 2023, nonprofit and nongovernmental hospitals, as well as those affiliated with medical schools, had a higher percentage of PSHP ownership as compared with their counterparts.

The researchers noted that newer PSHPs frequently take an “asset-light” approach by partnering with national insurers in joint ventures rather than purchasing additional clinical assets. They also pointed out that PSHPs are different from “legacy closed-loop systems, such as Kaiser Permanente,” as plan members are not restricted to certain hospitals.

Further, the researchers said that while PSHPs may mitigate the inefficiencies associated with fragmented care, they were unable to determine whether those gains led to lower costs or improved quality.    

Pfizer and Innovent Biologics entered into a deal worth up to $10.5 billion. The licensing and collaboration agreement they signed covers the research and development of 12 early-stage and de novo cancer drugs — specifically, next-generation antibody-drug conjugates with differentiated payloads and multi-specific antibodies. Eight of the programs originated with Innovent, a biopharmaceutical company based in China, and the other four are “Pfizer-proposed” discovery programs.

Innovent will be responsible for developing the drugs through Phase I; Pfizer will then lead global development. Additional details on how the companies will share development costs, licensing, and commercialization of the drugs are available in the press release. Innovent will receive $650 million up front and potentially another $9.85 billion in milestone payments. The transaction, which is subject to regulatory approvals, is expected to close in the third quarter.

Servier is acquiring Edgewise Therapeutics’ muscular dystrophy business, which includes an investigational drug called sevasemten, in an agreement worth up to $2.65 billion. A first-in-class orally administered therapy, sevasemten inhibits fast skeletal myosin, a protein that causes fast contractions of muscle fibers. The drug is designed to preserve and protect unstable muscle from contraction-induced injury in individuals living with Becker or Duchenne muscular dystrophy.

The agreement calls for Servier to pay Boulder, Colo.-based Edgewise $1.55 billion up front and as much as $1.1 billion in milestone payments. Servier’s governing foundation and Edgewise’s board of directors have approved the acquisition. If the deal receives regulatory approval and satisfies other customary closing conditions, it is expected to close in the third quarter.

Eli Lilly may begin withholding 340B pricing this week from hospitals that fail to submit claims data for in-house pharmacy dispensing. In a letter to the Health Resources and Services Administration dated June 1, Lilly said it had informed an initial group of 340B covered entities that they had five days to start submitting the claims data or risk losing access to the program’s discounted drug pricing, Becker’s Hospital Review reported.

Lilly said in the letter it would tell wholesalers the noncompliant entities are not eligible for 340B pricing until they submit the data and noted that the company intends to “follow the same course for additional covered entities in the weeks ahead.” According to Becker’s, Lilly said it had given covered entities months to implement steps to comply with the requirement, which was announced in January.

Rick Pollack, CEO of American Hospital Association, said in a statement that Lilly’s “unlawful” claims data policy would force hospitals to “divert resources away from care and toward onerous and expensive administrative burdens.” According to Pollack, Lilly did not respond to AHA’s recent offer to work together on a common solution. He called on the Department of Health and Human Services to keep Lilly from proceeding with the “illegal and harmful policy.”

WVU Health System and Independence Health System signed a definitive agreement, taking the next step toward WVU Health’s acquisition of Greensburg, Pa.-based Independence Health. The combined health system will serve patients in Maryland, Ohio, Pennsylvania, and West Virginia. As part of the agreement, WVU Health will invest $800 million in Independence Health’s five hospitals. The funds will be used for a new electronic medical record system, facility upgrades and expansions, and enhancements to improve access to care, according to the press release.

The health systems anticipate completing the acquisition this fall. Although WVU Health obtained certain regulatory approvals during the due diligence phase, additional approvals will be required. Based in Morgantown, W.Va., WVU Health has 25 hospitals and is the state’s largest employer.

What we’re reading

The Remarkable Proliferation of Cancer Immunotherapies. Substack: Eric Topol MD, 5.24.26

Breaking Up Health Care Organizations — The Right Fix Right Now? JAMA, 5.28.26 (subscription required)

The Interaction Between The Manufacturer Discount Program And Medicare Drug Price Negotiation. Health Affairs, 5.29.26

‍