AI could help strengthen home hospital programs by supporting clinical decision-making and improving diagnostic tools, Northeastern researchers say.
At the height of the COVID-19 pandemic, hospitals faced overwhelming demand and limited space. To meet the need, many turned to a model that had existed for decades but had never gone mainstream: hospital-at-home care.
The concept is simple — deliver the same level of medical treatment patients would receive in a hospital, but inside their homes.
Massachusetts General Brigham was an early adopter, launching its home-hospital program in 2016 and expanding it significantly in the years that followed. In 2020, Medicare and Medicaid began reimbursing at-home hospital care at rates similar to traditional inpatient care.
That shift helped validate the model, says professor Gene Tunik, director of AI + Health Sciences at Northeastern University’s Institute for Experiential AI.
“Home hospitals had a renaissance period in this country because doctors were able to keep patients in their homes, not requiring the brick and mortar space, but still provide a [strong] level of care,” he says. “A lot of people prefer to be in their own home — you eat your own food, sleep in your own bed, and are with your own family.”
Today, home-hospital care remains popular, but it still presents challenges. Researchers like Tunik believe artificial intelligence could help by improving decision-making and diagnostic tools.
Next week, Northeastern will host a two-part workshop at East Village on the Boston campus to help health care providers do just that. The workshop is a precursor to ICD’s annual Hospital@Home Leadership Summit.
Unlike a visiting nurse or an occasional house call, home-hospital care involves hospital-grade equipment and around-the-clock attention from physicians and nurses. It’s designed for patients who are sick enough to need admission but not intensive care, Tunik explains.
AI tools could support this model by helping clinicians assess which patients are right for home care. At Northeastern, researchers are experimenting with generative AI models like ChatGPT, Claude, Gemini and Perplexity to simulate clinical decision-making based on patient medical histories.
Another promising application of AI is in ultrasound imaging, a critical tool for evaluating heart and lung function and identifying conditions like blood clots — key factors in both hospital and home-based care.
But using ultrasound equipment can be difficult and requires specialized training, says Alycia Markowski, a professor in Northeastern’s Bouvé College of Health Sciences.
“They are finding that a couple of techniques work and AI has brought it to a whole new level,” she says. “You just need to understand how to turn the machine on, how to hold the probe, and the area and position that you need the patient — the machine actually guides you through finding the correct image.”
Markowski says clinicians with minimal experience can begin making assessments with the help of AI-powered machines. “We’re going to let participants find their own assessment with very little knowledge and see how easy it is to use AI to support your clinical diagnosis with ultrasound imaging,” she says.
Still, both she and Tunik stress that AI is not a replacement for trained providers. Human expertise remains essential to interpreting results and making sound judgments.
“That’s why you have a clinician,” Markowski says. “AI is generating information, but we still need to train our medical providers to have their own clinical reasoning thought process, so when the answer comes and it doesn’t fit the picture, they need to go back. We still need to train clinical providers to think.”