AI in Geriatric Care: Promise, Proof, and the Pressure to Get It Right
Artificial intelligence can lighten the load for clinicians and caregivers - but only if it's built around patient needs and validated in the real world. That was the clear takeaway from researchers presenting at the Gerontological Society of America (GSA) 2025 Annual Scientific Meeting.
What worked: Identifying caregiver vs patient messages
In one study, researchers tested whether AI could tell if a patient portal message came from a person with dementia or a care partner. The model analyzed 1,973 messages and hit a high bar: an AUC of 0.92.
The hardest cases were messages that referenced more than one person, such as "my husband and I." That matters because misidentifying the sender can trigger the wrong follow-up or misrouting, according to Kelly T. Gleason, PhD, associate professor at Johns Hopkins School of Nursing. With clinicians under growing pressure to respond to portal messages, automation that routes and flags correctly could cut after-hours EHR time.
"AI use in healthcare is highly prevalent, but most tools are launched before being tested by clinicians, patients, or care partners," Gleason said. "Use of automation could help with shared access registration and caregiver identification - but there is still so much we do not know about AI."
What patients and caregivers want (and don't trust yet)
Gleason's team also interviewed people with dementia and their caregivers after inviting 650 participants; only 5% responded, underscoring how hard it is to reach older adults and care partners digitally. Respondents were open to AI-drafted messages if clinicians reviewed them and health systems were transparent about use.
Still, many doubted AI's ability to detect tone, urgency, or emotion. "It is important to make sure automation is done in a way that does not compromise patient trust in healthcare," Gleason said. Her team plans future trials to test whether these tools actually improve or degrade care quality.
The tension: End-user needs vs investor and developer priorities
In a separate qualitative study, Nancy L. Schoenborn, MD, associate professor of medicine at Johns Hopkins University, interviewed 49 stakeholders across patients, clinicians, insurers, investment firms, and tech companies. Each group weighed AI through a different lens.
- Older adults prioritized affordability and simple, accessible design.
- Clinicians emphasized seamless EHR workflow integration.
- Investors and developers focused on market size, revenue models, and clearing costly regulatory hurdles.
These competing pressures create "real tension between the priorities of end users and those of developers and investors," Schoenborn said. Many tools are still built around what the tech can do rather than what older adults need - "solutions in search of a problem."
Schoenborn also noted that AI tools are often funneled into lengthy approval routes intended for traditional medical devices at the FDA, and argued for a more fitting pathway for AI in health. For context, see the FDA's current approach to AI/ML software as a medical device here. More on the GSA can be found here.
Practical takeaways for teams building or buying AI in senior care
- Co-design with older adults and caregivers from day one. Validate with real users, not just synthetic datasets.
- Measure what matters: routing accuracy, time saved, response times, safety events, and patient-reported trust.
- Be explicit about roles: clearly detect and label whether a message is from a patient or caregiver; add guardrails for multi-person references.
- Keep humans in the loop. Allow clinicians to review AI-drafted messages and override routing decisions.
- Optimize for EHR fit. Reduce clicks, use standard vocabularies, and avoid creating new inbox burdens.
- Honor plain language and accessibility. Short prompts, large fonts, and multilingual support improve adoption.
- Price transparently. If patients or practices can't afford it, it won't scale.
- Plan for regulatory and auditability early. Document data sources, model updates, and performance drift.
Bottom line: AI can help reduce friction in care for older adults, but trust is earned with evidence, transparency, and simple design. The next phase is not more features - it's proof that these tools make care safer, faster, and easier for patients, caregivers, and clinicians.
If you're upskilling teams to implement responsible automation in healthcare workflows, explore focused training at Complete AI Training.
Your membership also unlocks:
