Artificial intelligence (AI) could transform care for older adults and people living with dementia, but few tools have successfully moved from research into routine clinical practice.
A new national initiative at Duke aims to close this gap.
The Duke Artificial Intelligence and Technology Collaboratory for Aging: Governance and Evaluation (Duke AITC-AIGE) is one of five centers in the National Institute on Aging’s Artificial Intelligence and Technology Collaboratories program selected to receive funding to create a national resource for advancing trustworthy AI in aging and Alzheimer’s disease and related dementias.
Building shared infrastructure
Duke will receive nearly $25 million over five years to build shared infrastructure, fund research projects nationwide, harmonize aging-related databases, and expand expertise to help promising AI innovations move toward real-world use.
Duke AITC-AIGE is the only center led from the Southeast, reflecting Duke’s strengths in geriatrics, dementia research, translational science and responsible AI oversight, said Juliessa Pavon, MD, associate professor in the Division of Geriatrics, who is one of three Duke principal investigators.
Pavon leads the Administrative Core, and Rishi Kamaleswaran, PhD, associate professor of surgery and anesthesiology, leads the Clinical Implementation & Data Science Core, and Sheng Luo, PhD, professor of biostatistics and bioinformatics, leads the Research Innovation Core.
Making AI real-world ready
“This work is about making AI real-world ready, building the infrastructure and the collaborations to be able to move AI in a very trustworthy way into practice,” Pavon said. “We do not have a shortage of AI algorithms. We have a shortage of AI solutions that have been rigorously evaluated, safely implemented, continuously monitored — and shown that it really improves real-world outcomes for older adults. Our goal is to build the pathway that moves promising AI from an idea to a trusted, useful tool in practice.”
For people living with dementia, trustworthy AI could help detect early cognitive decline by analyzing speech, behavior, and wearable device data. Other applications could identify risks such as falls or wandering, support medication optimization and deprescribing, and help caregivers monitor systems and access personalized guidance.
“We’re already seeing promising research in these areas,” Pavon said. “Our center will help investigators rigorously evaluate these technologies in real-world settings and determine which approaches truly improve outcomes and quality of life for older adults and people living with dementia.”
The center will receive about $5 million per year, with about $4 million annually dedicated to research projects nationwide. At least 80% of those research funds will support Alzheimer’s disease and related dementias, with the remainder supporting broader aging research, Luo said.
Housed within Duke Clinical and Translational Science Institute, the center will provide pilot funding, data science infrastructure, implementation support and responsible AI safeguards. It will also develop resources that investigators and organizations can adopt and adapt, including evaluation methods, model cards, safety-monitoring approaches and implementation tools.
AI tool transparency
The model cards, for example, will provide standardized information about an AI tool, much like a food package nutrition label. The cards will provide a clear, standardized summary of what an AI tool is designed to do, the data it was trained on, how well it performed during testing, where it works well, and where it may be less reliable.
Model cards can also describe potential biases, limitations, and appropriate uses so that clinicians, health systems, researchers, and patients can make informed decisions about whether a tool is suitable for a particular situation. The goal is to make AI more transparent, understandable, and trustworthy.
“Just as clinicians would not prescribe a medication without understanding its risks and benefits, we believe users of AI should have clear information about how a tool works and when it should or should not be used,” Pavon added.
The Research Innovation Core will develop statistical and AI methods to identify performance limitations of AI tools, understand their causes, and improve the reliability and generalizability of AI systems.
Through NeuroAI, a secure analytics and evaluation platform, researchers will be able to bring together different types of health data such as medical records, brain images, laboratory results, and genetic information to study aging and dementia and develop and evaluate AI tools.
From academia to patients
Kamaleswaran said the effort will extend beyond major academic centers. “We want to democratize the AI expertise built at large academic institutions, extending it to community researchers and even patients who may have an idea for a tool but lack the resources to build it,” he said
Through a national pilot funding program, investigators across the country will be able to access expertise and support in areas such as AI evaluation, data science, implementation, ethics, governance, and stakeholder engagement without needing to be part of a large academic AI center.
By providing a secure, shared platform with standardized data and ready-to-use methods, our AITC resources can reduce the time, cost, and technical expertise needed to develop and test AI tools, allowing community researchers, smaller institutions, patients, and caregivers to contribute ideas and participate in innovation, not just large academic centers,” Pavon said.
Improving lives
Ultimately, the overall goal for AITC-AIGE is better lives, through technologies that help older adults maintain independence, support caregivers, improve care delivery, and enhance quality of life..
“We’re about improving lives, not just improving these algorithms,” she said. “Five years from now, success will mean that researchers and health systems have a clearer, reproducible pathway for turning AI innovation into safe, and meaningful benefit for older adults.”
Research reported in this publication was supported by the National Institute on Aging of the National Institutes of Health under Award Number P30AG103094. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.