Understanding the Policy Space for AgeTech: Implications for AI and Digital Health

A majority of older adults, those 65 and over, in the United States, desire to age in place (Binette & Farago, 2021), defined as the ability to live in one’s own home and community safely, independently, and comfortably, regardless of age, income, or ability level. However, a majority of older adults are unable to afford the personal care services needed to age at home (JCHS, 2023). Even if they can afford services, there is a growing gap in caregiver access. The caregiver support ratio for every person in the high-risk years of 80 plus is expected to decline to four to one by 2030 and to less than three to one in 2050 (Redfoot et al., 2013). There is a need for solutions to address these major barriers of unaffordability and caregiver shortfall.

Technologies and algorithms have shown promise in reducing caregiver burden (Huisman et al., 2022) and health care utilization costs (Gentili et al., 2022). Despite this promise, there is a limited number of technologies and decision-making tools that have been specifically developed for older adults, Alzheimer’s Disease and Alzheimer’s Disease Related Dementias (AD/ADRD) patients, caregivers, and their clinicians (Wang et al., 2019). In response to this need, the National Institute on Aging funded the creation of a network of centers through their AI and Technology Collaboratories for Aging program, including the Massachusetts AI and Technology Center (MassAITC) (Figure 1). The goal of these centers is to accelerate the R&D of AgeTech, that is, digital technologies built around the needs of older adults that include them in the design process.

Figure 1.

MassAITC affiliated institutions. MassAITC = Massachusetts AI and Technology Center for Connected Care in Aging & Alzheimer’s Disease.

Figure 2 illustrates some of the AgeTech solutions that the MassAITC pilot portfolio is supporting towards these goals. These projects leverage advances in smartphone sensing capabilities as well as wearable and passive sensors that can be leveraged in a home setting together with the latest advances in machine learning, including multimodal sensor analytics and large deep learning models such as Large Language Models (LLMs), and Vision models to address a broad range of challenges that older adults and their caregivers face in home settings.

Figure 2.

Focus areas, technologies, and analytics models that are representative of the broader MassAITC pilot portfolio. BP = blood pressure; EEG = electroencephalogram; fNIRS = functional near-infrared spectroscopy; HR = heart rate; ML = machine learning; spO2 = oxygen saturation.

Some examples of areas that are being explored by MassAITC include (a) low-burden technologies to enable older adults to remain independent and stay in their homes by preventing falls and more generally manage physical function such as by tracking sleep and mobility changes; (b) technology and algorithms that can be used in the home for older adults to detect early or monitor the risk of cognitive impairment; (c) providing better support to the caregiver, because caring for individuals with ADRD at home is challenging due to the need for continuous 24-hour supervision, and significant prevalence of behavioral and psychological symptoms of ADRD; and d) technologies to enable more informed chronic disease and co-morbidities management.

Within each of these areas, MassAITC is supporting multiple projects. For example, some of the pilot studies in the space of “Early detection & Management of Cognitive Decline” include (a) exploration of biomarkers for early detection of mild cognitive impairment (MCI) in home settings by leveraging a variety of signals including voice from smartphones, eye tracking and balance from Virtual Reality headsets, and walking cadence from smartwatches; (b) cognitive assistants that leverage the capabilities of LLMs to assist with memory issues of individuals with MCI; (c) new wearable platforms to image the brain and use of this information for both early detection and training to improve cognitive state; and (d) smartphone apps to measure “in the moment” cognitive state to improve safety, for example, by determining driving readiness.

Although the pilots funded by MassAITC are focused on early-stage technology development and validation on small cohorts of older adults, there are several policy challenges that need to be addressed as these technologies scale to larger longitudinal studies. This discussion’s focus is narrowed to the specific challenges that emerge in the interface between technology and the end-user, along with the potential solutions to address them.

Key Challenges to Address for Sustainable Adoption of AgeTech

This section asserts that when technology is used to interface with vulnerable populations, such as older adults, and particularly those with cognitive impairments, it is necessary to ensure that the devices, platforms, and associated algorithms are trustworthy, unbiased, interoperable, and adaptive to the user’s stage of functional and cognitive decline.

Fundamental Lack of Trust

Despite the recent increase in the use of at-home technologies, a sense of fear and mistrust remains a key barrier to scaling their adoption. Privacy concerns and the potential for data misuse are primary sources of this distrust (Vitak & Shilton, 2020). This is exacerbated by the fact that many family caregivers are turning to 24/7 monitoring technologies, such as wearables and at-home sensors, to ensure the wellness and safety of their older relatives. These devices, which can track a range of metrics over time, offer valuable insights into an individual’s overall quality of life. However, the use of such devices comes with inherent privacy risks and potential issues related to data sharing. For example, personal health information collected by these devices could be inadvertently shared in ways that might lead to denial of insurance coverage or increased premiums (Raber et al., 2019). This potential for data misuse has been identified as a key source of fear and mistrust among the older adult demographic regarding the use of these technologies (AgeTech Collaborative, 2024).

A significant contributor to this mistrust is the lack of transparency in privacy policies related to these technologies. Many users are unaware of what sensitive data is being collected, how it is being used, and with whom it is being shared. In fact, older adults express higher concerns that their data could be shared with people who do not have a right to it (Boise et al., 2013). Vague, ambiguous, or overly complex privacy policies exacerbate these concerns, leading to increased fear and reluctance to adopt these technologies.

Lack of Representation in Validation Datasets

Furthering this sense of distrust, at-home technologies and wearables often lack inclusion of older adults in the large-scale studies used to train the AI algorithms that underpin them. Older adults are disproportionately underrepresented in datasets collected in large scale wearable device studies, such as the Apple Heart Study, where less than 6% of the participants were above 65 (Perez et al., 2019). Ageism in trialing technology products on older adults is having a fundamentally negative effect on this population (Mace et al., 2022). This systematic bias in the training set leads to unfairness and inequity in the resulting performance of the model as borne out by a substantially decreased sensitivity for detection in subsequent validation studies specific to older adults (Ford et al., 2022). This issue of skewed datasets is further complicated in the case of MCI and ADRD, where devices and algorithms are rarely validated (Maresova et al., 2018) due to the cost and complexity of recruiting individuals with these conditions and running studies (Malzbender et al., 2020) that often involve older adult and caregiver dyads.

Additionally, within the limited older adult data sets from wearable device studies, there is a further bias in the geographical setting of the data collection (Pratap et al., 2020). This underrepresentation of individuals from rural or remote areas is a further source of inequity and limits the generalizability of the devices and may contribute to poorer outcomes in rural areas than metropolitan areas. Overall, this creates an environment where the consumer may not be aware that the performance of the device or the algorithm’s outputs may not work as communicated.

Lack of Interoperability and Data Standards Leading to Poor Care Delivery

Monitoring older adults in a home setting is also challenging due to the need to integrate data from a range of wearable, mobile, and internet of things (IoT) devices that need to be leveraged to obtain a complete picture of the health of an individual. These devices are often from many different vendors (health care companies, big tech, and startups), and often these devices are not necessarily designed specifically for health monitoring—for example, smart home sensors such as thermal cameras designed for occupancy detection can be repurposed towards mobility monitoring and fall detection. This creates a significant interoperability challenge because there is a lack of unified data output standards (Sedlakova et al., 2023) that broadly applies to at-home devices of different types that are useful for health monitoring.

The absence of consistent data standards leads to data fragmentation and data siloing within devices, which challenges the efficient development of care services that are tailored to the individual needs of the older adults. The absence of a standard data framework that integrates service platforms leads to additional burden for caregivers and cognitive strain for older adults, as they must navigate multiple log-in portals and research into which combination of devices for their specific needs are interoperable with their care provider’s electronic medical record system.

Limited Adaptability of Technologies and Consent Procedures

A key challenge to the sustainable utilization of technologies is the need for interfaces that account for the evolving capacity and functional ability of older adults (Genge et al., 2023). From the time of initial adoption of a given technology, the user will likely experience changing capabilities related to health conditions coupled with aging, such as frailty, musculoskeletal disorders, cognitive decline, and sensory impairments. Many of the current IoT and wearable devices are designed without consideration of these age-related changes and therefore present usability challenges.

Likewise, consent for data sharing from these devices should be cognizant of the age-related changes of cognition. Given the general timeframes for transitions from MCI to ADRD, it is likely that an older adult who had the cognitive capacity to provide informed consent when they started using a device may lose that capacity subsequently. By the time they are asked to re-consent, they may no longer fully comprehend the changes to the policy, making it difficult for them to provide continued informed consent. This in turn raises concerns about the validity of their ongoing consent.

Potential Solutions to Address Policy Challenges

Having described the challenges involved when deploying AgeTech in the home, the focus now shifts to potential strategies for mitigating these issues (Figure 3). It is crucial that researchers, developers, and policy makers work together to create a framework that will temper these challenges to facilitate a greater sense and ease of use that is necessary to ensure large-scale adoption of AgeTech.

Figure 3.

Mapping of the potential solutions to the key challenges for sustainable adoption of at-home technologies.

AgeTech Assurance Labs to Certify at-Home Devices and Algorithms

One promising model for standardized evaluation of health technologies and algorithms is the Assurance Labs framework being developed by the Coalition for Health AI (CHAI). CHAI has developed a blueprint for establishing a network of certified assurance for the practical implementation of trustworthy AI (Shah et al., 2024). AgeTech researchers, developers, and funders can leverage this model to utilize assurance laboratories for wearables and sensor-based technologies to rigorously test and validate these devices and associated algorithms specifically for older adults to ensure the high standards of reliability, safety, and inclusivity are met. Evaluation facilities within the laboratories would simulate the continuum of home and care settings to validate device performance across a representative spectrum of older adults, covering the breadth of cognitive and physical health stages. Key features and functions of the AgeTech Assurance lab would include:

1. Setting and enforcing standards for interoperability and data handling to ensure devices can easily integrate across technological environments, including electronic health records

2. Engaging with stakeholders to inform research and development and to ensure the highest levels of usability and accessibility are obtained

3. Publishing transparency reports and ongoing performance evaluations over the lifecycle of the technologies to ensure the technologies maintain effectiveness as they adapt to the evolving landscape of aging

4. Creating a certification process to endorse products that adhere to the laboratory’s standards that would serve as an indicator of a trustworthy device

5. Curating existing disaggregated datasets, with an emphasis on including individuals over 65, for improved testing and evaluation of these technologies

6. Incentivizing the trialing of older adult participants for new datasets

AgeTech Fact Labels for Increased Transparency

Communication of the AgeTech device performance, specifications, and data-sharing practices needs to be transparent to allow for informed decisions in purchasing and to solidify trust by older adult users. In parallel with assurance laboratories, the development of AgeTech facts labels is encouraged as an expansion of the model cards that are a key pillar of trustworthy AI (Sendak et al., 2020). The labels should, at a minimum, contain the following features: intended use, operational guidelines, data it collects, expected benefits, and potential risks (such as those due to biases). Additionally, these labels would include a digest of the data privacy and sharing protocols, informing users about who has access to the data, how it is used, and what the safeguards are to protect sensitive information. The format of these labels should be informed by codesign principles to ensure usability, accessibility, and adaptability needs are met. For example, in the context of older adults with MCI, these labels should be intuitive and usable throughout the stages of cognitive decline.

Ethical Hubs for Trustworthy AgeTech

The creation of ethics hubs is proposed to provide critical oversight for the assurance laboratories, AgeTech fact labels, and adaptive consent procedures, supporting the highest level of transparency and trustworthiness. This central body would serve the following functions:

1. Reducing biases, enhancing fairness, and promoting equity by overseeing the inclusion of diverse demographics, including geographical, in validation data sets

2. Enforcing transparency in how devices collect data and how algorithms process it to make predictions through the incorporation of explainable AI principles

3. Conducting audits to monitor compliance with required ethical standards as a condition of technologies maintaining the AgeTech Assurance Lab Certification

Consent Policies that Adapt with Cognitive Status

Privacy and data sharing policies should work towards consent systems that evolve with the user’s cognitive status. Continued advancements in digital cognitive assessments and AI-enhanced passive monitoring systems that can predict changes in cognitive status could be leveraged by an adaptive consent system to determine how frequently to reaffirm consent and, when appropriate, transition to proxy consent from a caregiver.

To ensure that the caregiver honors an individual’s wishes before their cognitive status degrades, the initial consent in an adaptive user agreement can walk the user through a series of hypothetical outcomes spanning 5–10 years and, through a series of questions, generate a personalized proxy consent blueprint akin to advance health care directives. This would better ensure that the caregiver or family member has a base understanding of the user’s wishes and would reduce chances of unintentional coercion or undue influence.

Conclusion

In closing, the widespread implementation of AgeTech in homes represents a promising frontier for enhancing the quality of life and independence for older adults, especially as they face cognitive and physical health challenges. However, the deployment of such technologies must navigate significant policy challenges, including ensuring privacy, equity, and usability while addressing the unique needs of this demographic. It is imperative that stakeholders from across sectors collaborate to establish a robust framework that integrates assurance laboratories, ethical oversight, and adaptive consent policies. Prioritizing the trustworthiness and accessibility of AgeTech will help foster an environment that empowers older adults to age independently, comfortably, and affordably.