| | | | |  | | By Carmen Paun, Erin Schumaker, Daniel Payne and Ruth Reader | | | | | 
Former British prime minister Tony Blair. | Ben Stansall/Pool Photo via AP | With his party back in power in the United Kingdom for the first time in 14 years, former Labour Prime Minister Tony Blair has a plan for the country’s ailing public health care system: a digital health record for each citizen, which could then be used to train artificial intelligence tools. The digital record he envisions would pull together patient data from hospitals, primary care doctors, pharmacies and phones, the Tony Blair Institute for Global Change said in a report Monday. Patients could share their digital records on patient portals in exchange for more personalized and accurate health advice, the report argues, citing the United States’ 21st Century Cures Act. That law enables people to access their health records from any provider. The report also advocates creating a centralized store of digitized health records that could be used to power “AI doctors” that would interact with citizens through a chatbot, POLITICO’s Laurie Clarke reports. Why it matters: Blair and his institute see AI as a solution to problems at the U.K.’s National Health Service, which is plagued by long waiting lists. “Investment in the country’s digital and data infrastructure may seem like a second-order consideration at a time when elective waiting lists stand at more than 7 million and there are an estimated 250 people dying prematurely in [the emergency department] every week — but without it, long waits and care failures will continue,” the report argues. Blair sees AI as a silver bullet not only for the NHS, but also for other ailing public services, government inefficiency and a stagnant economy, Clarke reports. But his institute’s link to tech billionaire Larry Ellison, who has pledged a total of $375 million to Blair’s group and owns cloud computing company Oracle, which stands to benefit from an AI boom, has made some in the U.K. wary of Blair’s AI plans. “Rather than seeking to cut costs using wholly unproven and overhyped AI tools that seem designed to further open up the NHS to large corporate profiteers, patients want to see the investment in the health service that would mean they can see a [general practitioner] — a human that knows them and has time to listen when they need help,” said Diarmaid McDonald, director of Just Treatment, an organization that campaigns against NHS privatization.
| | | 
New York, N.Y. | Erin Schumaker/POLITICO | This is where we explore the ideas and innovators shaping health care. Even lawyers hate "legalese." So why are legal documents so hard to understand? MIT researchers’ conclusion: They often feature performative "center-embedding" — long definitions in the middle of sentences — signaling that the document is special and authoritative. Share any thoughts, news, tips and feedback with Carmen Paun at cpaun@politico.com, Daniel Payne at dpayne@politico.com, Ruth Reader at rreader@politico.com, or Erin Schumaker at eschumaker@politico.com. Send tips securely through SecureDrop, Signal, Telegram or WhatsApp.
| | |  A bot struggled to understand injury reports about scooter riders. | AP | When it comes to accurately deciphering doctors’ clinical notes, large language models aren’t up to the challenge, new research suggests. To come to that conclusion, Columbia University researchers analyzed emergency department records for 54,569 patient visits to 96 hospitals for injuries on e-bikes, hoverboards and powered scooters between 2019 and 2022. The researchers tasked the ChatGPT-4 bot from San Francisco-based OpenAI to assess whether riders were wearing protective helmets at the time of their injuries based on doctors’ notes related to the accidents. They then compared those results with their own assessments of randomly sampled notes in 400 records. The bottom line: ChatGPT-4 struggled to correctly determine whether riders were wearing helmets when they were injured, particularly if doctors used negative phrases, such as “w/o helmet” or “unhelmeted” in their notes. The model also had trouble replicating its work and tended to make things up, instead of duplicating correct readings. “When we used highly detailed prompts that included all of the text strings related to helmets, on some days ChatGPT-4 could extract accurate data from the clinical notes,” Andrew Rundle, senior study author and epidemiology professor at Columbia Mailman School of Public Health in New York, said in a statement. Being accurate on some days isn’t enough, according to Rundle. Extracting data from doctors’ notes accurately would help researchers and likely accelerate their work, but the time required to define and test text, coupled with ChatGPT-4’s repeated inability to replicate work, means the large language model isn’t ready for prime time — just yet. The study was published this month in the journal JAMA Network Open.
| | | | 
Was that puff today or a week ago? Researchers want to know. | AFP via Getty Images | Researchers at the National Institute of Standards and Technology are partnering with scientists at the University of Colorado Boulder to test whether two breath measurements indicate recent marijuana use — a first step in developing reliable breath tests for cannabis intoxication. They face a big challenge: People can test positive for marijuana use for days, weeks or even months. That makes developing any sort of cannabis impairment test difficult, our Mona Zhang reports. The researchers want to test whether taking two breath samples spaced apart can help determine whether a person recently used cannabis, explained NIST materials research engineer Kavita Jeerage. A minor change in the amount of THC, the compound in marijuana that effects the brain, between breath measurements would indicate that a person did not recently use cannabis, while a greater change would indicate more recent use. “Our hope is that by focusing on a change in THC over time, this type of approach could apply to people who are both frequent or infrequent users,” Jeerage told Mona. Why it matters: The research collaboration with CU Boulder is part of a larger effort at NIST to learn about breath THC measurement — for example, what happens when someone exhales more forcefully compared with a lighter exhale. While NIST is not developing breathalyzer technology itself, researchers hope that their work can help those who are developing those devices. The National Institute of Justice, a research agency under the Justice Department, is funding the work. “All of our research is trying to help support the industry as a whole, that their devices work as intended,” Jeerage said. What’s next? Researchers are recruiting about 45 study participants who are 25- to 50-years-old and have experience both smoking and vaping cannabis. It will take at least another year or year and a half to collect the data, Jeerage said. | | | | Follow us on Twitter | | | | Follow us | | | | |
