AI-based models accurately predict, show promise for expansion in headache

Key takeaways:

• AI currently may be utilized best as a predictor of headache and migraine attack.
• The future of AI-based predictive tools likely exists in rapid diagnosis, treatment and health care access expansion.

SAN DIEGO — AI-based models are currently best utilized as a predictor of headache, but show promise for expansion into the health care continuum, according to a speaker at the American Headache Society Annual Scientific Meeting.

“There’s a high demand among headache specialists or neurologists, so having such AI tools can really help get patients diagnosed earlier and get treatment earlier,” Chia-Chun Chiang, MD, an assistant professor of neurology at the Mayo Clinic in Rochester, Minnesota, said during her presentation.

Particularly for non-specialists, she said, AI can boost accuracy in headache diagnosis. Citing a study in which more than 4,000 persons aged 15 years and older were examined for headache along with 17 variables presented in a questionnaire, the AI model’s accuracy for diagnosis was 83.20% while accuracy without the model was 46%.

For headache specialists, building precision models can arise from information as simple as social media posts, Chiang said. One natural language processing framework, crafted from users on Twitter and Reddit about their headache timing, prevalence, status, duration and pain level, yielded two different models that resulted in a high degree of precision in headache diagnosis.

In addition, larger language models (LLM), based upon review of a high volume of clinical notes from electronic health records, can be utilized. Given a few simple questions such as “What is the headache frequency per month?”, a vocabulary can be built from which generative answers may be found from the records themselves.

Regarding research, Chiang noted that AI-based models can simplify chart review efforts that typically are the most labor intensive, citing one LLM-based generative model that extracted headache frequency data from clinical notes that yielded 92% accuracy.

As headache complaints comprised the largest single division of inbox volume in the neurology department at the Mayo Clinic, Chiang added that utilizing AI led to quicker message retrieval. That, in turn, led to a quicker refining of the portal itself and development of more detailed categories of headache treatment that may result in easier development of care plans for patients.

In the future, AI will likely be utilized to streamline communication and access among a broad range of health care professionals, to remove traditional barriers to specialty care and precision, individualized medicine.

“Hopefully, we could develop an AI-based diagnostic and treatment-decision support tools to help in partnering with primary care or other specialties, to expand access to headache specialty care,” Chiang said.