‘A series of small steps:’ Artificial intelligence has arrived for caregivers

University at Buffalo researchers Dr. Andrew Baumgartner and Haowen Hsu are working on initiatives that include an instruction tool to condense and digest the information from discharge summaries into clear instructions and an AI chatbot to help older patients manage multiple chronic illnesses. (Photo by Joed Viera/Buffalo News)

Whether welcome or abhorred, artificial intelligence has entered modern healthcare.

Charts that write themselves and tools that can interpret a disorder from an X-ray are two of several ways AI has been used to accelerate treatment. Medical facts can be provided in milliseconds − typing a question on Google produces an AI-generated response right below the search bar.

But the patient or their caregivers still have to understand that information and know what to do with it. 

According to a Pew Research Center study, approximately half of Americans struggle with judging the accuracy of healthcare information, and over 75% have difficulty trusting information from contradicting resources. 

What if AI could help? 

Two University at Buffalo researchers are working to realize that vision for older adults, while three founders of a tech start-up are marketing a different take on the idea to employers.

Easier move 

A patient in her 80s, with two to three chronic conditions, had visited Dr. Andrew Baumgartner after she was discharged from the hospital. 

Her hospital discharge papers — follow-up instructions for outpatient care — took up 17 pages.

“As we’re going through it together, 90% of it was legalese, medical jargon, things that weren’t relevant to what she needed to know to be successful at home and be healthy at home,” Baumgartner said.

AI could make that transition of care from hospital to home easier.

Baumgartner, a family physician and researcher under UB’s “Team Alice,” is working on the development of an instruction tool that would condense and digest the information from discharge summaries into clear instructions. 

Haowen Hsu, another researcher on “Team Alice,” is piloting a similar initiative: an AI chatbot that would help older patients manage multiple chronic illnesses. It would tailor the treatment plan outlined in the discharge summary to the patient’s life.

“We have been focused on how we can provide personalized health intervention that is relevant to the patient’s priorities, their own life experiences. But unfortunately, we haven’t reached that point,” Hsu, who has a doctorate in pharmacy, said. “AI has the ability to make the intervention more personalized.”

Older adults are comfortable using AI and often already use platforms such as ChatGPT, Baumgartner said. When compared with a cohort of UB students in his study, they were more accepting of AI in their healthcare.

“There’s an ageist stereotype that older adults have an aversion to technology and don’t know how to use it well,” Baumgartner said. “When the reality is they just use it in a different way.” 

While ChatGPT tries its best to be personal, the answers it provides are still too generalized, according to Baumgartner. The websites are trained to cater to the general population, and as younger people use them substantially more than their older peers, a bias can be present against older adults.

“We want to make sure that when our tools are being trained and developed, they’re trained on information from older adults so that way that bias isn’t there as much,” Baumgartner said.

That feedback comes from focus groups of older adults, which the two researchers incorporate at each step. User interfaces − or even the full blueprints − are still early on in progress for both projects. So far, features for turning text into speech and providing instructional videos are on the table. 

Hsu describes the focus groups as “co-design sessions,” where the base model is brought in and the users provide the feedback.

“We demonstrate our developed user interface and provide an opportunity for them to provide the active input on what we try to develop,” Hsu said. 

Proactive party

Several caregivers were calling Allwel, a home care agency, when they faced a crisis. The person they cared for had fallen, received a diagnosis or was in an emergency and they didn’t know what to do.

Not knowing the specifics of the care plan, the agency was unable to help.  

Co-founders Priya Pinto and Chris Brown-Hall and CEO Jennifer Redding decided to create CareNGen, an AI platform, to do what Allwel could not.

“We’ve always wanted to make a change, but it’s so regulated that it’s tough to do that. The best way to make a change in home care is through hiring extraordinary people and making sure that they’re caring for people the best they can,” Redding said. “But this is something that can touch hundreds of thousands of people.” 

CareNGen, established in August 2024, would talk directly with caregivers caring for older adults on an individualized treatment plan and respond to questions on care with interventions tailored to the plan. Companies would register for the platform, paying $2 per employee. Anyone who participated would be anonymous to their employer.

The platform would help mitigate unseen productivity losses that come with employees taking time off or using work time to handle family emergencies, according to Pinto.  

“When they don’t use those avenues, that cost to the employer is hidden,” Pinto said. “The employer has to hire somebody else, or burden another employee with extra work to be able to do the job of the person who isn’t there.”

Brown-Hall’s mother had taken care of her brother in his last moments. She had tried to be her brother’s care manager, pharmacist, nutritionist and many more roles that prevented her from being his sister.

“All of these things took up every moment of her day, and then in the end, she wasn’t able to just sit by his bedside and hold his hand as he was dying,” Brown-Hall said. “My mission for CareNGen is that it alleviates a lot of those things so that we can just be a daughter, a son, a spouse and know that we’ve done everything we can to take care of everything else.”

The program can’t function without AI, according to Redding. Initial CareNGen prototypes prior to AI could only give a single care plan per condition and restarted from “ground zero,” providing a caregiver the same steps every time. 

“AI can be the proactive party,” Redding said. “The next time you log on, it can say, ‘Welcome back. Is there anything new with Mom?’”

CareNGen is in demo, with one pilot study completed and another beginning this month. The first study was without AI, but was still “beneficial,” Pinto said.

Redding and Pinto are marketing it to investors to fund further development, but Pinto says the three are not driven by the money.

“What a nice legacy it would be − a natural development out of the home care agency − being able to leave the world with a cogent, careful system that anyone can use to care for their elderly parents,” Pinto said.

Small, slow steps

AI does not replace the human touch, Pinto said.

CareNGen does not give clinical advice. The three founders are coding the program to respond with, “You should consult a doctor,” to questions about symptoms and medication.

It also cannot give financial or legal advice, and the three are in partnership with organizations such as the National Association of Senior and Specialty Move Managers and Aging Life Care Association, to have an “expert panel” feature on the platform to connect caregivers with professionals.

“We’ve been pretty careful about that,” Pinto said. “We’ve read articles recently about wrong advice given by ChatGPT, or something like that, and that going wrong.”

Nor would AI disrupt established protocols and workflow. Baumgartner and Hsu’s AI discharge tools need clinical oversight. AI would fit into the doctor’s routine, rather than the other way around.

The two researchers are conducting a safety analysis on their tools.

“We’re not ceding decision making to the AI tool at the end of the day,” Baumgartner said. “And we’re not outsourcing those decisions to an AI tool that, no matter how well you design it, is going to have some limitations with it.”

Hsu stressed the need for an interdisciplinary team. All information for AI models should be pulled from the target population, with their own needs, cultures and perspectives.“Healthcare providers, computers and scientists − we all need to collaborate with each other to tackle this kind of question,” Hsu said.

Before implementation, bringing visibility may be the first hurdle. Getting visibility for CareNGen is difficult, Pinto says.

“Visibility is where we have been like dogs with a bone,” Pinto said. “We won’t let that bone go until it happens, but it’s a series of small steps.”

The implementation of AI has to be slow, so it is implemented in a way that serves patients best, Baumgartner said.

“If you don’t implement them in the right way, you’re going to create more problems than it’s worth,” Baumgartner said.

Mylien Lai is a Buffalo News intern, funded by the New York & Michigan Solutions Journalism Collaborative. She is reporting on the experiences of caregivers and the loved ones they support in Western New York.