Patients turn to AI to interpret lab tests, and mixed results: Shot

When Judith Miller performed a routine blood test in July, she was alerted by a phone call the same day that the lab results were posted online. So the next day, when her doctor messaged her that her entire test was OK, Miller wrote back to ask about rising carbon dioxide and what is called the “low anion gap” mentioned in the report.

The 76-year-old Milwaukee resident was waiting for a reply, but patients were doing more and more than they could get to the medical team, saying Miller couldn't reach the medical team. She put the test results into Claude and asked her AI assistant to evaluate the data.

“Claude helped me to clearly understand the anomaly,” Miller said. The generative AI model did not report anything surprising, so she was not worried while awaiting a response from a doctor, she said.

Patients are receiving unprecedented access to medical records, as federal law requires that electronic health information be released immediately, such as doctor visits and notes about test results.

Also, many patients use LLMs such as Openai's ChatGpt, Anthropic's Claude, and Google's Gemini to interpret records. But there is a certain risk to that help. Doctors and patient advocates warn that AI chatbots can generate the wrong answers and that sensitive medical information may not remain personal.

But does AI know what it's talking about?

However, most adults pay attention to AI and health. 56% of people using or interacting with AI are not convinced that the information provided by AI chatbots is accurate, according to a 2024 KFF poll. (KFF is a health information nonprofit organization that includes KFF Health News.)

That instinct is born through research.

“LLMs are in theory very strong and can give great advice, but they can give really bad advice depending on how they are encouraged,” said Adam Rodman, an internist at Beth Israel Deacones Medical Center in Massachusetts and chairman of the Generator AI steering group at Harvard Medical School.

Justin Honce, a neuroradiologist at Uchealth in Colorado, said that it can be extremely difficult for patients who are not medically trained to know if an AI chatbot makes a mistake.

“In the end, LLMS needs to be taken care of overall. With the latest models, these concerns continue to be less and less, but they are not completely resolved,” Honce said.

Rodman has seen a surge in AI use among patients over the past six months. In one case, the patient took a screenshot of the results of the hospital lab on Mychart and then uploaded it to ChatGpt to prepare questions before appointment. Rodman welcomed patients to demonstrate how to use AI and said their research would create opportunities for discussion.

According to a recent University of Michigan poll, one in seven adults, over 50 adults use AI to receive health information, while one in four adults under the age of 30 does so, according to a KFF poll.

Using the Internet to advocate for better care for yourself is nothing new. Patients have traditionally used websites such as WebMD, PubMed, and Google to search for the latest research and have sought advice from other patients on social media platforms such as Facebook and Reddit. However, AI Chatbots' ability to generate individual recommendations and second opinions in seconds is novel.

Things you need to know: Beware of “hagaku” and privacy issues

Liz Salmi, director of Communications and Patient Initiatives at OpenNotes, the Beth Israeli Butler's academic lab defending healthcare transparency, was wondering what AI, particularly for patients, is a great interpretation.

In a proof-of-concept study published this year, Salmi and colleagues analyzed the accuracy of ChatGpt, Claude, and Gemini's responses to patient questions regarding clinical notes. All three AI models worked well, but how patients framed the questions was important, Salmi said. For example, instructing an AI chatbot to take on a clinician persona and asking one question at a time increased the accuracy of their answers.

Privacy is a concern, Salmi said, so it's important to remove personal information such as your name and Social Security number from the prompt. The data went directly to the tech companies that developed the AI ​​model and added that they were not aware of it being compliant with federal privacy laws or taking into account patient safety. Openai CEO Sam Altman warned in a podcast last month about putting personal information into ChatGpt.

“Many people who are not used to using large-scale language models may not know about hallucinations,” Salmi said. For example, Openai's Whisper, an AI-assisted transcription tool used in hospitals, has introduced fictional treatments into transcription, according to an Associated Press report.

Using generated AI requires a new type of digital health literacy, including asking questions in a specific way, verifying answers with other AI models, discussing with healthcare teams, and protecting privacy online.

Doctors should be careful about AI too

It's not just the patients who use AI to explain their test results. Stanford Health Care has launched an AI assistant to help physicians draft and send interpretations of clinical tests and LAB results to patients.

Colorado researchers studied the accuracy of the ChatGPT-generated summary of 30 radiation reports and the satisfaction of four patients. Of the 118 valid responses from patients, 108 indicated that the CHATGPT summary clarified details regarding the original report.

However, ChatGpt may be overemphasized or unemphasized, indicating that a small but significant number of responses have become more confusing after reading the summary, says Honce, who participated in the preprint study.

Meanwhile, after four weeks and several follow-up messages from Miller on Mychart, Miller's Doctor ordered her blood test and repeated additional tests that Miller proposed. The results have returned to normal. Miller was relieved and said she was better informed due to her AI enquiries.

“It's a very important tool in that respect,” Miller said. “It helps me organize my questions, do my research and level the arena.”

KFF Health News It is a newsroom nationwide that creates deep journalism on health issues and is one of the core operating programs at the country. KFF .