68 million adults in the USA have two or more chronic diseases. To meet their demands for future healthcare, care capacity will have to increase at least tenfold, estimates Munjal Shah, co-founder and CEO of Hippocratic AI, based in Palo Alto, California. At present, he says, “we only have the resources to provide chronic care support to the top 20% of the sickest of the sick. Put simply, we don’t have enough nurses.”

AI can analyze hundreds of mammogram images to predict metastatic cancer. Credit: Pulsar Imagens / Alamy Stock Photo

With this in mind, Shah’s company (which has received US$50 million in venture capital support) is using LLMs to help nurses do their work. Such administrative support is “a safer use case for LLMs than diagnostic use cases,” Shah explains, and he believes that AI tools “can be scaled to truly solve the healthcare staffing crisis.”

The aim is to create virtual nurses for chronic care that use an automated ‘voice’ to speak to, and listen to, patients. Although the virtual nurses will not be involved in clinical diagnosis, says Shah, they could remind patients to take their medicine, follow through with care plans, schedule follow-up appointments, review medication issues and help patients navigate care-access issues.

An LLM-backed chronic-care nurse can pass the NCLEX licensing exam for nurses and the NAPLEX licensing exam for pharmacists, says Shah. An LLM nurse can “speak every language, and remember every conversation with each patient,” he says.

To train its LLM nurse, Hippocratic AI built its own model using text from care plans, regulations and other medical manuals. The company then trained the model on how to speak like a chronic-care nurse using conversations between registered chronic-care nurses and patient actors.

To safety-check its LLM nurse, Hippocratic AI has set up a safety council, and it has recruited more than a thousand nurses to evaluate the model in a double-blind randomized trial. The product, says Shah, will be released one sub-condition or procedure at a time — for example, for patients with chronic kidney disease, then for patients with arthritis, and so on.