EPtalk by Dr. Jayne 3/5/26

A health system that went live on a system-wide EHR five years ago with promises to upend its best-of-breed strategy is sending notices that it is finally migrating its laboratory systems. The lab migration project has been underway for quite a while, although it seems like communication to end users is just beginning.

The first communication explained the what and why of the project. It also outlined the communication cadence, so that people can be on the lookout for more information as the go-live gets closer.

I suspect that the team is planning to over-communicate for two main reasons. First, no one likes surprises. Second, many physicians, especially those who are community-based rather than being employed by one of the health system’s entities, likely assume that the project has already been completed and aren’t anticipating changes.

We will see how the communications unfold as go-live approaches. I’m sure that the finance folks will be glad to stop paying maintenance to multiple vendors.

The hot topic around the virtual physician lounge this week was an article in Nature Medicine that looks at how ChatGPT Health performed at triaging medical emergencies. ChatGPT Health, which launched in January of this year, was designed specifically to handle consumer-driven, health-related queries.

The authors submitted a set of 60 clinical vignettes across multiple clinical domains and conditions to the chatbot and to a panel of three physicians. The physicians triaged them based on clinical guidelines and their own expertise.

The tool underperformed. It failed to correctly identify one-third of non-urgent cases and nearly half of emergency cases. It recognized stroke and anaphylaxis as emergencies, but failed to refer the user to the emergency department for the life-threatening conditions of diabetic ketoacidosis and impending respiratory failure.

Other scenarios tested biases, such as when family or friends minimize a patient’s symptoms.

The ability to appropriately generate crisis intervention messages was unpredictable. Interventions appeared more often when cases discussed suicide generally than when the discussion included a description of a specific method of self-harm.

The authors conclude, “Our findings reveal missed high-risk emergencies and inconsistent activation of crisis safeguards, raising safety concerns that warrant prospective validation before consumer-scale deployment of artificial intelligence triage systems.”

User access to ChatGPT Health is limited by waitlist. Parent company OpenAI says it will make it widely available when it has finished validating its safety and reliability.

I’m surprised by the tool’s poor performance. Triage protocols have been available for many years and are commonly used by nursing staff in primary care offices. I wonder if the model was trained using any of those references or if those weren’t included because of intellectual property concerns.

We’re partway through the spring conference season, with ViVE in the rearview mirror and HIMSS on the horizon. A fair amount of alcohol flows at health tech conferences and it’s not just during the after-hours parties. It seems like happy hour events on the show floor are an expectation rather than an exception. A timely piece in The Harvard Gazette examines the effects of binge drinking on the digestive system.

Authors of the study, which was published in November, found that a single episode of binge drinking, which they defined as four drinks in a two-hour period for women or five for men, can make it harder for the small intestine to keep bacteria from entering the bloodstream. The research was performed in mice with the alcohol administered by gavage, which is the research equivalent of a beer funnel. I’m going to have a hard time keeping that imagery out of my mind next week when I see people hitting the cocktail circuit at HIMSS.

A fair amount of literature shows that younger generations are consuming less alcohol than older groups such as Gen X and Baby Boomers. Given the amount of alcohol-fueled bad behavior that I’ve seen during my time in the industry, that’s probably a good thing.

Speaking of HIMSS, I’ve got my trusty sneakers packed so I can stroll the exhibit hall in comfort, but I’m also taking my dancing shoes for after-hours sparkle. I’ll be making at least one trip to the Bellagio to catch the fountains after dark, since they are my favorite of the excesses found on the Las Vegas Strip.

I will be stopping by the booths of our sponsors (anonymously, of course) and looking for the best booth décor and of course footwear. Stay sharp during those booth shifts and save your cell phones for scanning badges, sharing party invitations, and emailing me your cute shoe pictures.

From Jimmy the Greek: “Re: AI tools in remote meeting platforms. My organization allows us to use them to create transcripts and summaries. It’s been helpful, but I literally laughed out loud when this turned up in a recent recap of a section that the AI tool titled ‘Product Staffing Woes and Teen Sleep’”: 

Robert and Susan discussed Susan’s staffing challenges, with Susan noting she was down to 1 3/4 of her intended six developers. Robert offered to support, but explained he couldn’t help directly. Robert then shared his personal experience with his teenage son‘s morning routine difficulties, leading to a discussion about teenage sleep patterns and morning habits. The conversation concluded with Robert introducing the topic of the next big project.

Pre-meeting small talk is common, especially in organizations where meetings don’t start on time. Seeing it memorialized highlights how much time is actually spent talking about topics that aren’t moving the organization forward. If you’re seeing a lot of these types of items in your summaries, it might make sense to disable the feature that automatically starts recording and transcription, and instead, manually start the process when you are ready to begin the meeting.

What’s the most amusing thing you’ve seen in an AI-generated meeting transcript? Leave a comment or email me.

Email Dr. Jayne.