Advancing Fair & Effective AI for Older Adults

Advancing Fair & Effective AI for Older Adults is sponsored by The SCAN Foundation (in partnership with Coalition for Health AI - CHAI). This initiative, in partnership with the Coalition for Health AI, focuses on laying the foundation for equitable, effective AI solutions for older adults, particularly those in underserved and marginalized communities.

Advancing AI for Older Adults The SCAN Foundation (TSF) and the Coalition for Health AI (CHAI) launched a collaborative planning grant to advance equitable AI solutions for older adults. This initiative, spanning 6 months from January – July 2025, was structured in three key phases: 1. A foundational literature review and gap analysis (Jan - Feb 2025) 2.

Individual stakeholder interviews with experts across the community (March – April 2025) 3.

A series of roundtables bringing experts together to discuss the 3 emerging core themes (May 2025) This project set out to lay a robust foundation for developing representative datasets and fostering best practices for applying AI equitably in healthcare for vulnerable populations, specifically older adults in underserved and marginalized communities.

By engaging healthcare practitioners and stakeholders, this grant was to identify gaps, clarify use cases, and establish pathways for deploying AI solutions that address health disparities effectively and ethically. Older adults exhibit significant interindividual variability in needs and multimorbidity, leading to heterogeneous data that results in inconsistent AI use, applicability, and performance.

Existing datasets fail to adequately represent older adults - particularly those over 85 - and from diverse socioeconomic backgrounds, increasing the risk of bias and inequitable AI applications. Current AI models targeting this population suffer from inconsistent testing, training, and validation methods, limiting real-world effectiveness and generalizability.

For example, AI applications in cancer diagnostics, frailty detection, and dementia risk prediction show promise but lack robust validation for older populations which limits their efficacy. The lack of standardized data means that there are variable definitions, such as how researchers define an older adult.

The wide variation in AI applications across medical specialties, use cases, and machine learning tasks further compounds these challenges, creating an AI landscape that fails to address the highest-priority needs of older adults. Adoption challenges, including usability concerns, digital access barriers, and limited AI literacy among older adults, must be addressed to enhance engagement and ensure meaningful impact.

The findings from Phase 2 & 3 reinforce a point that can be applied to communities outside older adults: that AI needs to be developed with a deep understanding of the population it’s being Certain actionable insights emerged for the design of more effective AI: - The full complexity of older adults’ lives demands linkage between clinical, behavioral, and social inputs, and a nuanced understanding of the settings people live in.

- We underutilize existing infrastructure for understanding social determinants of health (for example, there being little incentive and adoption of “z” billing codes 1) - Establishing consistent approaches to behavioral phenotyping and building a common data model are critical enablers - Governance models and community partnerships are also key to making data usable and trustworthy across settings Patient centricity should then be applied to deployment: - Real-world validation, both local and external, of models cannot be an afterthought but must be embedded in the design lifecycle and be actionable and responsive to diverse care settings - Deployment of AI shouldn’t assume high digital literacy, but be built into systems that are adaptive, useful, and non-invasive.

This grant crystallized a vision and roadmap for advancing AI in older adult health care. Armed with a deeper understanding of the challenges, potential strategies, and bringing together a network of stakeholders in this space, The SCAN Foundation and CHAI have laid an important foundation to catalyze action.

Ultimately, this initiative strives to ensure that AI serves the needs of older adults - improving health outcomes, supporting independent living, and reducing disparities - by building solutions that are inclusive, evidence-based, and anchored in the lived realities of our aging population. The healthcare landscape is undergoing a paradigm shift driven by the transformative power of artificial intelligence (AI).

Traditional (predictive) AI systems have already proven useful for analyzing medical data, diagnosis, treatment planning, and drug discovery. Nevertheless, the track record of new AI technologies in healthcare is one marked by a growing digital divide, with greater health disparities for underrepresented populations over time.

For example, facial > 1Z codes in medical billing are used to indicate factors influencing a patient's health status or need for healthcare services, rather than a disease or injury recognition algorithms have often demonstrated lower accuracy for people of color, with a 2019 NIST study finding that facial recognition technology disproportionately misidentified Asian Americans and African Americans at a rate 10 –100 times that of white American (Grother et al.

, 2019). In another example, the lack of older adult patients in clinical studies leads to unrepresentative datasets, which in turn leads to unfair and inequitable performance in models applied to older adult patient care (Mace et al. , 2022).

These examples underline the need for increased representation in training datasets for vulnerable populations to ensure the implementation of trustworthy, responsible AI for all. Meanwhile, the global population of older adults is growing rapidly, with projections indicating that by 2050, more than 2.

1 billion people will be over the age of 60, with 80% living in low-and middle-income countries, according to the World Health Organization . For the first time in history, older adults now outnumber children under five. In the U.S., nearly a quarter of the population will be over 65 by 2060, the Census Bureau reports .

And aging populations face significant health challenges: the National Council on Aging says that nearly 93% of adults 65 and older have at least one chronic condition, while nearly 79% have two or more. And The U.S. Surgeon General has declared loneliness a public health crisis, equating its impact on mortality to smoking 15 cigarettes a day.

As AI promises to transform healthcare, ensuring its equitable application to older adults and other priority populations, such as rural communities and marginalized groups, is critical. However, the lack of standardized methods for AI development, testing, and validation limits the effectiveness and real-world applicability of these technologies.

Demand for older adult healthcare is surging, and so are the risks of algorithmic bias in how AI is being deployed. We cannot afford to leave this age group behind.

Phase 1: A foundational literature review and gap analysis (Jan - Feb 2025) To assess the current landscape of AI applications in healthcare for older adults and identify key research gaps, we started the project by conducting a structured literature review and dataset cataloging process.

A total of approximately 100 papers were reviewed, covering a range of AI applications across different medical domains and use cases relevant to aging populations. To further analyze the state of AI research for older adults, we extracted and cataloged datasets referenced in the reviewed papers. The datasets were systematically classified based on geographies, scale, age ranges and data sources.

This dataset catalog provided a first insight into the geographic distribution, demographic inclusivity, and structural composition of data used in AI research for older adults. For a more detailed methodology and bibliography, see Appendix 1. We found that the reason AI solutions do not perform effectively or fairly on older adults is because of three main factors: 1.

High variability within data leading to low generalizability of models: Many models lack local or external validation, limiting their effectiveness across different healthcare settings, especially given the wide variability in older adults’ health profiles. Put simply, just because you’re over 65, you won’t necessarily have s imilar characteristics to others in your age category.

Models deployed on older adults need to be trained, tested and validated on highly variable datasets. 2. Low representation in datasets leading to bias performance: Datasets often underrepresent individuals aged 85 and older, use overly broad or inconsistent age groupings, and fail to capture the diversity of aging experiences - particularly among those with complex medical and social needs.

This lack of representation and poor data stratification leads to biased AI outcomes and reduced clinician trust. 3. Poor training data leading to poor adoption: Care data is highly fragmented, with most information in unstructured formats that hinder interoperability and coordinated care.

There are positive retirement and longitudinal datasets being created at the state-level, but open-access datasets for aging populations remain scarce. Even when tools are developed, nearly 40% have shown no benefit over standard care in trials, and adoption among older adults is low due to usability barriers and limited digital access.

This Phase underscored several critical issues: highly heterogeneous health data in older populations, older adults (especially 85+ are significantly underrepresented in current healthcare datasets , fragmented data sources, inconsistent definitions (e.g. what counts as “older adult”), and widespread omission of social and behavioral factors in data collection.

These findings established a compelling need for targeted efforts to improve data quality, representation, and validation methods for AI in geriatric care. Phase 2: Individual stakeholder interviews with experts across the community (March – We went out to CHAI’s network to test these findings with leading experts in the field.

We held structured focus interviews with nearly 30 individuals representing 22 organizations, spanning health data companies, community health providers, and aging-focused associations. A full list of interviewees are listed in Appendix 2. Four key focus areas have emerged as a framework for the Phase 2 findings: 1.

Establishing standardized best practices and techniques for behavioral phenotyping There is currently no consistent method for defining older adults’ health and behavioral profiles in AI development. Data on aging populations is highly fragmented across clinical, home-based, and community systems, making it difficult to develop accurate or personalized AI tools.

Most models lack integration of key social and behavioral determinants of health. 2. Leveraging/creating a common data model for data standardization at scale Interviewees identified the urgent need for a shared data model to harmonize disparate datasets and facilitate cross-institutional learning.

A well-designed CDM can bridge clinical, behavioral, and community data sources and enable scalable phenotyping. However, data standardization alone isn’t enough - semantic alignment, quality checks, and data governance are needed to ensure true interoperability.

Linking CDM frameworks to longitudinal aging studies and real- world datasets (like OCHIN’s 13 -million-patient warehouse) could dramatically improve data inclusivity for rural, underserved, and oldest-old populations. 3. Model training, validation and local testing of “agetech” AI solutions Without meaningful evaluation in real-world settings, AI tools for older adults often fall short of their intended impact.

Participants emphasized embedding model validation directly into the local testing lifecycle, ensuring solutions are safe, equitable, and fit-for-purpose. A four-tiered framework (see Appendix 3) - ranging from local threshold adjustment to federated retraining -offers pathways for adapting AI tools to diverse older adult populations.

Trust-building partnerships with community groups are critical to ensure that local testing includes meaningful engagement with patients, caregivers, and community leaders. 4. Playbooking to ensure patient-centric design and deployment of AI solutions.

Current AI tools are often built around technical capabilities rather than the lived experiences of older adults. Interviewees noted a widespread lack of older-adult-specific design features, leading to poor adoption and low utility. Voice-based interfaces, passive monitoring, and ambient technologies were cited as promising alternatives to app-based tools.

A shared playbook for patient-centered design and deployment - drawing from lessons across industry and community contexts - could help developers avoid common pitfalls and build more inclusive, sustainable solutions for aging populations. Phase 2 confirmed the major gaps we’ve found - from data fragmentation and lack of standard practices, to challenges in validation, deployment, and user-centric design.

The emerging framework from focus interviews with key experts are a first step in solutionizing what we would be taking to future funders. Phase 3: Three-part roundtable series (May 2025) CHAI hosted three 60-minute virtual roundtables focused on key areas from the Phase 2 framework. Each session’s goal was to refine solution design for the upcoming RFP deliverable and to build a network of potential delivery partners.

We had around 20 experts attend each roundtable, themed according to the framework: 1. Data Structure & Standardization 3. Local Testing & Patient Engagement The notion site with full agenda, context and objectives created for attendees can be found here.

Key findings from Roundtable 1: Data Structure and Standardization The conversation centered on the importance of patient trust and purpose-driven approaches to social determinants of health (SDoH) data collection. Older adults are often willing to share sensitive social risk information when approached with clear intent and personalized care, in a trusted environment.

However, there are significant gaps in the utilization of structured coding systems like Z codes . While relevant SDoH data exists, it is rarely documented in structured formats due to limited incentives, provider burden, and concerns around legal implications or discriminatory use.

This challenge is compounded by the limitations of electronic health records (EHRs), where social and behavioral data are inconsistently captured - often embedded in unstructured narratives - leading to biased, fragmented datasets and limited interoperability.

To advance equity in healthcare AI, models must integrate multi-modal data sources - including clinical, behavioral, and social information - and be validated for bias, especially to ensure that the diverse experiences of aging populations are accurately represented and addressed.

Key findings from Roundtable 2: Model Validation – Local Testing and Performance Tuning When reimagining AI validation infrastructure for community-based aging care, participants emphasized that validation processes must be lightweight, automated, and scalable, avoiding reliance on high-resource environments.

The group explored the benefits of hybrid validation approaches, combining centralized validation across diverse sites with optional, workflow-specific local testing. A major challenge discussed was the unsustainability of manual ground truth generation 2, which often depends on labor-intensive chart reviews and clinician input.

Instead, participants proposed standardizing endpoint definitions 3 and developing automated pipelines to generate validation data more efficiently. Finally, the roundtable underscored the critical importance of incorporating unstructured, social, and conversational data - often the most effective way to detect hard-to-surface risks such as social isolation, transportation barriers, or abuse.

Participants agreed that for validation to be equitable and scalable, it must integrate non-traditional data sources, support low-lift deployment, and be co-developed with communities from the outset.

Key Findings from Roundtable 3: Patient Engagement, Co-Creation, and Local Testing The third roundtable focused on the critical role of patient and caregiver engagement in AI development, especially for older adults in underserved or hard-to-reach settings.

Participants emphasized that human-centered co-design must be foundational, with tools like entrepreneur-in-residence programs and design workshops allowing developers to build trust, observe real workflows, and co-create solutions alongside older adults and staff that doesn’t disrupt too much.

Early testing in simulated environments, followed by gradual deployment in real-world clinical or community settings, was suggested as a best practice for responsible and inclusive implementation. > 2Manual ground truth generation involves human experts labeling data to create a reference set for training and evaluating machine learning models.

> 3Standardizing endpoint definitions is crucial for ensuring comparability and interpretability of results across different studies. This involves establishing clear, objective criteria for measuring outcomes, particularly for events like death, myocardial infarction, stroke, and revascularization A key theme was the need to involve caregivers and reflect community diversity from the outset.

Caregivers bring invaluable lived experience and must be compensated for their time and insight. Moreover, developers should engage local networks - such as borough-specific organizations, immigrant groups, and community-based service providers - before initiating product development. This approach ensures that solutions are contextually relevant and equitable.

Participants also addressed persistent barriers to patient access, particularly outside of institutional settings. Older adults who are socially isolated or without caregivers often fall through the cracks. Most AI pilots still rely heavily on university or health system data, missing broader real-world representation.

Targeting programs like PACE and working with faith-based organizations were recommended as effective strategies to reach more diverse populations. Ultimately, equity demands thoughtful design, compensation for engagement, and standards that ensure inclusion of underserved groups in both development and deployment.

The three roundtables reinforced the need for trust-based, purpose-driven data collection, lightweight and scalable validation infrastructure, and meaningful patient and caregiver engagement from the outset. AI models must be built on more inclusive, multi-modal datasets and validated for bias using real-world, non-clinical inputs such as social and conversational data.

Validation approaches should be automated, standardized, and locally adaptable so everyone can participate, regardless of resources. AI solution co-design with patients and caregivers, flexible deployment across community-based settings, and intentional outreach to underserved groups are essential to ensure relevance, usability, and impact.

Appendix 1: Literature Review Methodology To assess the current landscape of artificial intelligence (AI) applications in healthcare for older adults and identify key research gaps, we conducted a structured literature review and dataset cataloging process. Literature Search Strategy We refined the search criteria with The SCAN Foundation to ensure relevance to older adults and healthcare AI applications.

A systematic search was conducted on PubMed , using the following search terms: • "artificial intelligence older adults healthcare" • "artificial intelligence older adults healthcare" [Title/Abstract] • "artificial intelligence older adults"[Title/Abstract] A filter was applied to limit results to studies published within the past year , ensuring that the findings reflect the most recent advancements in the field.

A full list of the identified research papers is provided in the bibliography. Review and Analysis of AI Applications From the curated paper set, we documented: • Successful applications of AI in healthcare for older adults, focusing on areas such as diagnostics, treatment optimization, risk prediction, and decision support.

• Gaps in research, including limitations in model validation, dataset diversity, generalizability across populations, and integration into clinical workflows. A total of approximately 100 papers were reviewed, covering a range of AI applications across different medical domains and use cases relevant to aging populations.

Dataset Cataloging and Classification To further analyze the state of AI research for older adults, we extracted and cataloged datasets referenced in the reviewed papers.

The datasets were systematically classified based on the following attributes: • Age range of the study population • Data source , including clinical records, wearable sensors, genomics, and survey-based datasets This dataset catalog provides insight into the geographic distribution, demographic inclusivity, and structural composition of data used in AI research for older adults.

The full list of datasets is available in a publicly accessible spreadsheet: Dataset Inventory . While this methodology captures a literature review of recent AI research for older adults, limitations include: • Potential publication bias , as studies reporting successful AI applications may be more likely to be published than those with negative or inconclusive findings.

• Dataset accessibility issues, as not all referenced datasets are publicly available for independent validation. • Non-systematic nature - as decided with The SCAN Foundation, we did not do a fully scientific systematic review, so there may be gaps in knowledge. We were instructed to read an indicative set of papers to lead us to a set of hypotheses that we could test and validate in upcoming phases.

1. AI is demonstrating promise in key medical specialties, though challenges remain in applicability and real-world validation for older adults: a. Oncology : AI shows promise in cancer diagnostics, treatment response prediction, and therapeutic dose optimization, but faces challenges in adapting to older adults' needs and analyzing complex biological datasets.

b. Orthopedics and Physical Therapy : Wearables and machine learning models have demonstrated potential for fall risk prediction and sarcopenia prevention, yet require validation in real-world settings for better data on effectiveness. c.

Frailty : Emerging tools like the Electronic Frailty Index (eFI) and AI-enabled frailty prediction for heart failure patients can enhance care, but integration with clinical workflows remains a challenge, limiting data on real-world impact. d.

Neurology and Psychology : AI-based models for cognitive impairment, depression, and dementia prediction have shown high accuracy, though gaps remain in addressing diverse subgroups and real-world applications. 2. Insights for effective data curation, AI development, and AI use in aging populations: a.

High quality multimodal data: AI models that integrate diverse data sources (e.g., pathology slides, radiologic images, biomarkers, EHRs, wearables, genetic data) tend to have higher performance. b. Real-world variability: Some of the most successful AI-driven interventions (e.g. oncologic treatment response models, and frailty screening tools) have been validated in clinical or real-world settings.

c. Importance of transparency and explainability: The effectiveness of AI in aging populations depends on clinician and patient trust —especially in fields like cancer treatment planning, where models like Watson for Oncology and ChatGPT have fallen short due to lack of explainability and adaptability. d.

Precision AI shows promise: AI-driven dose optimization, tailored sarcopenia interventions, and personalized dementia risk models highlight the need for individualized AI-driven decision-making. e. Wearables can reduce hospitalizations but need more user-centered design: Wearable-based fall prediction and real-time health monitoring have demonstrated success in reducing hospital admissions and adverse events.

However, multiple studies cite the discomfort and suspicion of wearables by the population. f. AI-augmented decision-making support : The most successful applications (e.g., cancer diagnostics, frailty screening, cognitive decline detection, polypharmacy management) enhance clinical decision-making by providing predictive insights based on large datasets, rather than attempting to fully automate decision processes.

g. Gaps in validation/generalizability of models need to be addressed: AI models for fracture risk, disability prediction, and social robotics for cognitive health show great promise but need broader validation across diverse populations. 3.

Key over-arching challenges: a. Generalizability & external validation i. Many AI models lack external validation which limits their transferability across healthcare settings.

ii. The variability in older adults’ health needs complicates AI standardization and integration into clinical workflows. b.

Data representation & bias i. Older adults, especially those aged 85+, are underrepresented in AI training datasets. ii.

The majority of datasets do not include age-disaggregated data. Some even use binary distinctions of ‘young’ and ‘old’. iii.

Many datasets treat individuals 65+ as a single category, or have defined ‘older adults’ inconsistently (e.g. 40+, 50+, 65+) disregarding critical age -specific differences. iv. Training data lack diversity required to ensure AI models perform consistently in minority populations and in individuals more representative of the aging majority (e.g. those with complex social and medical needs).

v. Effective balancing techniques are inconsistently applied to data. vi.

Disproportionate exclusion of aging populations from clinical trials leads to limited clinical and real-world data on treatment effectiveness — especially in those with multiple co-morbidities, social support limitations, and limitations in activities of dai ly living. This also impacts clinicians’ trust in using or applying an AI solution in these populations, and similarly limits the adoptability/use of direct to patient tools. c.

Data interoperability & accessibility i. Fragmented care data and inconsistent patient tracking reduce AI’s effectiveness in care coordination. ii.

The dominance of unstructured text data (up to 80% of health records) means data is not standardized, highly variable, and largely uninteroperable, which limits AI’s applicability. iii. There is a scarcity of open-access health datasets available for research and development.

d. Limited clinical utility & adoption i. A review of 65 randomized controlled trials found that nearly 40% of AI tools for aging populations provided no added clinical benefit over standard care.

ii. Acceptability of AI technologies is also low among older adults due to usability challenges and limited access to digital infrastructure. • Disease prevalence vs research: Many conditions that disproportionately affect older adults receive little AI-focused research.

• Increasing data diversity: Large-scale longitudinal studies, such as UK Biobank and All of Us, integrate clinical, genomic, and environmental data to advance precision medicine. Studies are increasingly merging clinical, genetic, and environmental data to create a holistic view of health. • Age group definitions vary , mixing summary statistics (e.g., median, mean) with categorical age ranges (e.g., 40+, 50+, 65+).

• Need for Global Inclusion: While studies in South Africa (HAALSI), Brazil (ELSI), and India (LASI) expand representation, research on aging populations in low- and middle-income countries remains scarce.

Mass-scale longitudinal datasets (year by number of participants) This proposal outlines a clear pathway from standards development to data development, providing a structured approach that we can practically advance in collaboration with NIA, NIH, and other key stakeholders.

Our next step is for CHAI, in its role as convener, to bring together a diverse set of stakeholders —including 5-10 researchers, 5-10 model developers, 5-10 health systems serving older adults, and 5-10 older adult patients. This convening will ensure that our standardization efforts are directly aligned with the real-world needs of those impacted.

We will focus on high-impact conditions such as diabetes, cardiovascular disease, and dementia to maximize the relevance and utility of these standards in improving health outcomes for older adults.

• ASFNR Multicenter Dataset • Comprehensive Gerontology Survey • MAP (Memory and Aging Project) • Health and Retirement Study (HRS) • English Longitudinal Study of Ageing (ELSA) • Survey of Health, Ageing and Retirement in Europe (SHARE) • CHARLS (China Health and Retirement Longitudinal Study) • ADNI (Alzheimer's Disease Neuroimaging Initiative) • Korean Longitudinal Study of Ageing • The Irish Longitudinal Study on Ageing • Mexican Health and Aging Study • Indonesia Family Life Survey • Japanese Study on Aging and Retirement • Study on Global Ageing and Adult Health • Global Aging Data Repository • Costa Rican Longevity and Health Ageing Study • HAALSI (Health and Aging in Africa: A Longitudinal Study of an INDEPTH Community in South Africa) • NICOLA (Northern Ireland Cohort for the Longitudinal Study of Ageing) • ELSI (Brazilian Longitudinal Study of Aging) • HART (Health, Aging, and Retirement in Thailand) • LASI (Longitudinal Aging Study in India) • SPS (Chilean Social Protection Survey) • MARS (Malaysia Ageing and Retirement Survey) • OpenSafely Literature Bibliography • Akyon SH, Akyon FC, Yılmaz TE.

Artificial intelligence -supported web application design and development for reducing polypharmacy side effects and supporting rational drug use in geriatric patients. Front Med. 2023;10:1029198.

doi: 10. 3389/fmed. 2023.

1029198 • Alty J, Goldberg LR, Roccati E, et al. Development of a smartphone screening test for preclinical Alzheimer’s disease and validation across the dementia continuum. BMC neurology.

2024;24(1). doi:https://doi. org/10.

1186/s12883-024-03609-z • Andersson G. The latest developments with internet-based psychological treatments for depression. Expert Review of Neurotherapeutics.

2024;24(2):171-176. doi:https://doi. org/10.

1080/14737175. 2024. 2309237 • Angermueller, C.

; Lee, H. J. ; Reik, W.

; Stegle, O. DeepCpG: Accurate Prediction of Single-Cell DNA Methylation States Using Deep Learning. Genome Biol.

2017, 18, 67. • Bai C, Mardini M. Advances of artificial intelligence in predicting frailty using real-world data: A scoping review.

Ageing Research Reviews. Published online October 2024:102529. doi:https://doi.

org/10. 1016/j. arr.

2024. 102529 • Brand YE, Kluge F, Palmerini L, et al. Self-supervised learning of wrist-worn daily living accelerometer data improves the automated detection of gait in older adults.

Scientific Reports. 2024;14(1). doi:https://doi.

org/10. 1038/s41598-024-71491-3 • Caravagna, G. ; Giarratano, Y.

; Ramazzotti, D. ; Tomlinson, I. ; Graham, T.

A. ; Sanguinetti, G. ; Sottoriva, A.

Detecting Repeated Cancer Evolution from Multi-Region Tumor Sequencing Data. Nat. Methods 2018, 15, 707 –714.

• CDC. Older Adults. Chronic Disease Indicators.

Published June 12, 2024. https://www. cdc.

gov/cdi/indicator-definitions/older-adults. html • Choudhury A, Renjilian E, Asan O. Use of machine learning in geriatric clinical care for chronic diseases: a systematic literature review.

JAMIA Open 2020; 3: 459 –71. • Chu CH, Nyrup R, Leslie K et al. Digital ageism: challenges and opportunities in artificial intelligence for older adults.

Gerontologist 2022; 62: 947 –55. • Clegg A, Bates C, Young J, et al. Development and validation of an electronic frailty index using routine primary care electronic health record data.

Age Ageing. 2016;45(3):353-360. • Comparison of Claims Based Frailty Indices in US Veterans 65 and Older for Pre-Diction of Long-Term Institutionalization and Mortality.

• D. Su, X. Zhang, K.

He, Y. Chen. Use of machine learning approach to predict depression in the elderly in China: a longitudinal study.

• Daunt R, Curtin D, O’Mahony D. Optimizing drug therapy for older adults: shifting away from problematic polypharmacy. Expert Opinion on Pharmacotherapy.

2024;25(9):1199-1208. doi:https://doi. org/10.

1080/14656566. 2024. 2374048 • Elliott J, Stolee P, Boscart V, Giangregorio L, Heckman G.

Coordinating care for older adults in primary care settings: understanding the current context. BMC Fam Pract 2018; 19: 137. https://doi.

org/10. 1186/s12875-018-0821-7 . • Ferreira RN, Ribeiro NF, Santos CP.

Fall risk assessment using wearable sensors: a narrative review. Sensors (Basel) 2022; 22: 984. https://doi.

org/10. 3390/s22030984 . • Gallo RJ, Shieh L, Smith M, et al.

Effectiveness of an Artificial Intelligence –Enabled Intervention for Detecting Clinical Deterioration. JAMA internal medicine. Published online March 25, 2024.

doi:https://doi. org/10. 1001/jamainternmed.

2024. 0084 • Georgescu AL, Cummins N, Molimpakis E, Giacomazzi E, Marczyk JR, Goria S. Screening for Depression and Anxiety Using a Non-Verbal Working Memory Task in a Sample of Elderly Brazilians: A Preliminary Analysis of the Transferability of AI Models (Preprint).

JMIR Formative Research. Published online February 1, 2024. doi:https://doi.

org/10. 2196/55856 • Hall A, Money A, Eost-Telling C. et al.

. Older people’s access to digitalised services. NIHR ARC GM: University of Manchester, 2022.

• He S, Wei M, Meng D, Wang Z, Yang G, Wang Z. Self-Determined Sequence Exercise Program for Elderly with Sarcopenia: A Randomized Controlled Trial with Clinical Assistance from Explainable Artificial Intelligence. Archives of Gerontology and Geriatrics.

Published online December 21, 2023:105317. doi:https://doi. org/10.

1016/j. archger. 2023.

105317 • HIT Consultant. Why Unstructured Data Holds the Key to Intelligent Healthcare Systems. 2015.

https://hitconsultant. net/2015/03/31/tapping-unstructured-data- healthcares-biggest-hurdle-realized/ . (24 April 2023, date last accessed).

• Hong N, Hong N, Hong N, Hong N. The potential role for artificial intelligence in fracture risk prediction. The Lancet Diabetes & endocrinology.

Published online June 1, 2024. doi:https://doi. org/10.

1016/s2213-8587(24)00153-0 • Hu M, Shu X, Yu G, Wu X, Valimaki M, Feng H (2021) A Risk Prediction Model Based on Machine Learning for Cognitive Impairment Among Chinese Community-Dwelling Elderly People With Normal Cognition: Development and Validation Study. J Med Internet Res 23:e20298. https://doi.

org/10. 2196/20298Return to ref 9 in article • Huang H, Wang J, Dunk MM, et al. Association of Cardiovascular Health With Brain Age Estimated Using Machine Learning Methods in Middle-Aged and Older Adults.

Neurology. 2024;103(2). doi:https://doi.

org/10. 1212/wnl. 0000000000209530 • Javeed A, Dallora AL, Berglund JS, Ali A, Ali L, Anderberg P.

Machine learning for dementia prediction: a systematic review and future research directions. J Med Syst 2023; 47: 17. https://doi.

org/10. 1007/s10916-023-01906-7 . • Joon Sung Park, Bernstein MS, Brewer R, Kamar E, Meredith Ringel Morris.

Understanding the Representation and Representativeness of Age in AI Data Sets. arXiv (Cornell University). Published online July 21, 2021.

doi:https://doi. org/10. 1145/3461702.

3462590 • Kahkoska AR, Freeman NLB, Jones EP et al. Individualized interventions and precision health: lessons learned from a systematic review and implications for analytics-driven geriatric research. J Am Geriatr Soc 2023; 71: 383