Nvidia founder and CEO Jensen Huang tells attendees of the J.P. Morgan Healthcare Conference that he expects drug design to move nearly completely to AI systems. He also predicts that medical devices such as ultrasound and CT scanners will be “a device plus a whole bunch of AIs” and that will create incredible value and opportunities.
A UC Berkeley public health professor and psychiatrist warns that using AI to offset a shortage of mental health providers and to address growing problems such as loneliness raises concerns. Jodi Halpern, MD, PhD worries about the effect of companies marketing AI conversational bots as therapists and trusted companions to people who are depressed or vulnerable. She says that unlike therapists, app vendors are not regulated, companies may be motivated to use the “addictive engineering” of social media companies, and app users may rely on the app the point of withdrawing from human engagement.
A Georgia lawmaker introduces a bill that would make it illegal to use AI to discriminate against people or for making health insurance and public services decisions without human review.
Family-powered autism therapy AI platform provider Forta raises $55 million in a Series A funding round.
A Google-created AI system called AMIE that was optimized for diagnostic dialog outperforms primary care doctors in synchronous text chat with patient actors. The authors warn that while synchronous text chatting is not a common practice, their work is a milestone in developing conversational diagnostic AI to take patient histories, provide diagnostic support, and communicate with skill and empathy.
Investor Mark Cuban predicts that AI will transform healthcare, but will involve millions of models that are created by individual companies and providers instead of a handful of big-name models. He says that the biggest issue is intellectual property protection, and expects health systems to limit access to the models they develop.
Workers’ compensation insurer MEMIC will use AI technology from Clara Analytics to automate medical records transcription and data extraction for claims reviews.
The CEO of medical device maker Medtronic says that the company will transform its products with AI, giving recent examples of its solutions for intelligent endoscopy, digital surgery, and remote surgery assistance.
Mass General Brigham researchers show that large language models can be trained to automatically extract social determinants of health from clinician notes. The finely tuned model identified 94% of patients who have adverse SDoH versus the 2% that could be identified using diagnostic codes. The model was trained to identify sentences that refer to employment status, housing, transportation, relationships, the availability of social support, and parental status. Bias was less likely than with GPT-4. The model was much smaller than ChatGPT models and results were better even thought the model was trained on just a few hundred patient records.
Switzerland-based cardiologists and researchers use AI to improve the usefulness of the atrial fibrillation detection capabilities of consumer wearables that are made by AliveCor, Apple, Fitbit, and Samsung. The study authors say that the single-lead ECG functionality alone is not clinically useful because of high numbers of inconclusive tracings, but the model reduced the percentage of those from 16% to 1.2%. They caution that opportunities for further studies will be limited as those manufacturers enhance their algorithms in undisclosed ways and may restrict access to their raw data. The study used cardiac diagnostics AI products from PulseAI.
A Michigan Medicine team uses AI to review physician notes for signs of risky drinking, which identified three times the number of patients than would have been flagged by diagnosis codes alone.
ChatGPT failed to make an accurate diagnosis in journal-published pediatric case challenges 83% of the time. The authors note that systems such as Google’s Med-PaLM 2 that were trained on medical data would likely perform better.
Pharmacy professors find that ChatGPT performs well in answering questions that patients might ask a pharmacist who is working remotely. It also did a good job of answering general patient questions, offering dietary suggestions, encouraging patient adherence, and clarifying medical terms.
UCSF Health is developing AI tools for nurses, which include patient deterioration detection, patient placement, and matching patients with specialized nurse expertise. A representative of UCSF Health’s nursing union isn’t convinced, declaring that AI has created a “disaster-capitalism moment” in being used to improve efficiency with an end result of reducing access to skilled care.
Mr. H, Lorre, Jenn, Dr. Jayne.
Get HIStalk updates.
Send news or rumors.