Trump previews next Medicare drug price negotiations

A version of this text first appeared in CNBC’s Healthy Returns newsletter, which brings the newest health-care news straight to your inbox. Subscribe here to receive future editions.

The Trump administration is already gearing up for one more round of Medicare drug price negotiations, but it is going to look just a little different this time around. 

The U.S. Centers for Medicare and Medicaid Services on Monday issued recent draft guidance for that third cycle, because the second round of negotiations is underway. The method was established under the Biden administration’s signature Inflation Reduction Act as a option to rein in high health care costs for older Americans. 

CMS plans to announce an inventory of 15 drugs eligible for the third round of price talks by February 2026, which is able to then kick off months of backwards and forwards between the federal government and manufacturers in the event that they comply with participate. The brand new negotiated prices for those products will go into effect in 2028.

But listed below are the largest changes this time around: 

• Medicare Part B drugs – For the primary time, the list would come with drugs payable under Medicare Part B – which covers medicines administered in a health care provider’s office or hospital – along with pharmaceuticals covered under Medicare Part D. Previous rounds only targeted Part D medications. 
• Renegotiation process – CMS may decide to renegotiate the costs for certain drugs that already had prices set for the primary and second cycles of talks, including those with recent approved uses or changes in “monopoly status.” The agency will announce any medicines chosen for the primary cycle of renegotiation, with revised prices for those products taking effect in 2028. 
• Transparency – CMS is aiming to spice up transparency around the method, in search of public feedback on topics reminiscent of how the agency determines an initial price offer for a drug. 

“This draft guidance is critical to making a transparent, competitive, and fair prescription drug market that puts American patients first,” Medicare Director Chris Klomp said in a release.

But Wall Street analysts are focused on one other a part of the guidance that might cause issues for Merck, Bristol Myers Squibb and another pharmaceutical firms. 

The guidance document suggests that the Trump administration could end a workaround that those firms are using to tug out revenue from top-earning cancer drugs, reminiscent of Merck’s Keytruda and Bristol Myers Squibb’s Opdivo.

The plan had been to shift patients to newer injectable – or subcutaneous – versions of their cancer drugs and keep charging Medicare higher prices for them, even after their original intravenous versions are subject to recent negotiated prices under this system. Drugmakers have been banking on those subcutaneous versions as a option to dampen the revenue they might lose from Medicare drug price negotiations, together with upcoming patent expirations for the unique types of their drugs. For instance, key patents for Keytruda start expiring in 2028. 

Under the present rules, complex drugs referred to as biologics are eligible for the negotiation process after 13 years, however the clock restarts for a new edition of the drug – like a subcutaneous form – that adds an extra lively ingredient.

Subcutaneous versions of medicine like Opdivo are combination products that include an extra ingredient, allowing them to be injected quickly as an alternative of being slowly infused like the unique intravenous form.

But on Monday, CMS said it’s “soliciting comments” on the way it “might consider” grouping these combination drugs with their original versions — if the added ingredient doesn’t affect how the drug treats the underlying disease. In other words, the agency is considering whether to count two versions of a drug as a single product in certain cases. 

That appears to be “somewhat targeted” at products reminiscent of subcutaneous Keytruda and Opdivo, JPMorgan analysts said in a note on Monday. They said the guidance results in “at the least the potential for inclusion” of those drugs in future negotiations. 

Still, no changes are final yet, so it could be too soon to predict the impact on drugmakers like Merck and Bristol Myers Squibb. 

Be happy to send any suggestions, suggestions, story ideas and data to Annika at annikakim.constantino@nbcuni.com.

It’s common for CEOs who transformed their firms to step back into leadership when things veer off target. 

This week, UnitedHealth Group Chairman Stephen Hemsley took a page from Bob Iger’s playbook at Disney, and took back the CEO position at the corporate following the abrupt departure of Andrew Witty.

Be happy to send any suggestions, suggestions, story ideas and data to Bertha at bertha.coombs@nbcuni.com.

“If developed and deployed effectively, large language models have the potential to expand access to health information, support clinicians in delivering high-quality care, and help people advocate for his or her health and that of their communities,” OpenAI said in a blog post. “To get there, we’d like to make sure models are useful and secure.”

The corporate said HealthBench was developed alongside 262 doctors from 60 countries. It’s based on 5,000 conversations that simulate interactions between individual users or clinicians and AI models. The discussions are split into seven different themes, including global health, emergency situations and handling uncertainty. 

When a model responds to a prompt, each response is graded against a set of “physician-written rubric criteria specific to that conversation,” OpenAI said. HealthBench accommodates 48,562 unique rubric criteria.

OpenAI included one example where a user said they found their 70-year-old neighbor unresponsive on the ground. The AI model in that instance told the user to take motion immediately, and included eight steps they might follow. HealthBench gave this answer a 77% based on its rubric criteria. 

OpenAI said HealthBench responses were evaluated against responses written by doctors to know how the model in comparison with their clinical judgement. The corporate found that HealthBench “closely aligns” with physicians’ grading. 

OpenAI said it used HealthBench to guage several existing models, including its own o3, GPT-4.1, o1, GPT-4o and GPT-3.5 Turbo models, xAI’s Grok 3, Google’s Gemini 2.5 Pro, Anthropic’s Claude 3.7 Sonnet and Meta’s Llama 4 Maverick. 

The corporate found that o3 outperformed other models, and it said its models have improved by 28% on HealthBench. 

OpenAI said the total evaluation suite and underlying data for HealthBench is offered in its GitHub repository. 

“We hope this supports shared progress toward using AI systems to enhance human health,” the corporate said.

Read the total blog post here.

Be happy to send any suggestions, suggestions, story ideas and data to Ashley at ashley.capoot@nbcuni.com.