7 - Monitoring (on) Your Mind

Summary

Advances in biomarker science are transforming the treatment of Alzheimer’s disease. First, clinicians can now identify Alzheimer’s disease neuropathology in situ, years, or even decades, before the onset of cognitive and functional impairment using biomarkers, particularly measures of amyloid, tau, and neurodegeneration. Second, there is growing evidence that other digital markers – for instance, changes in banking and driving patterns – can be used to identify older adults at increased risk of cognitive decline. As a result of these two advances, there is growing interest in monitoring older adults for incipient changes in cognition that might affect their wellbeing. This monitoring can be no-tech, such as when an older adult asks a friend or family member to let them know if they start repeating questions or retelling stories. Or it can be high-tech, such as installing a device in one’s car that tracks sudden accelerations and decelerations and the distance traveled from home. While there is great enthusiasm for this kind of monitoring among clinicians and researchers, and even, as our own research shows, among some older adults and their family members, other older adults deem monitoring to be an unwanted intrusion on their privacy. This chapter discusses the changing understanding of Alzheimer’s disease risk; presents data on the desire for monitoring, highlighting the potential for intrafamilial conflict; considers the ethical obligations of non-health businesses to collect or disclose information about Alzheimer’s disease risk; and, finally, identifies potential ethical and legal challenges – such as privacy, stigma, and discrimination – with possible solutions.

I Introduction

What first comes to mind when you hear the words “Alzheimer’s disease?”

For many, those words evoke images of an older adult who exhibits troubling changes in memory and thinking. Perhaps the older adult has gotten lost driving to church, although it’s a familiar route. Perhaps they have bounced a check, which is out of character for them. Perhaps they have repeatedly left the stove on while cooking, worrying their spouse or adult children. Perhaps they have trouble finding words or are confused by devices like iPhones and, as a result, have lost touch with longtime friends.

Alzheimer’s disease (AD) has traditionally been understood as a clinical diagnosis, requiring the presence of symptoms to be detected. The older adult we just envisioned might make an appointment with their physician. The physician would likely listen to the patient’s medical history – noting the characteristic onset and pattern of impairments – and determine that the patient has dementia, a loss of cognitive and functional abilities severe enough to interfere with daily life. Dementia can have numerous causes, and so the physician would also conduct a comprehensive physical and cognitive examination, perhaps ordering lab tests or brain imaging scans, as well as neuropsychological testing. After excluding other causes, the clinician would diagnose the patient with “probable” AD, a diagnosis that can only be confirmed postmortem via autopsy. This approach to diagnosis interweaves the patient’s experience of disabling cognitive and functional impairments (i.e., dementia) with the label of AD.¹

Yet, our ability to measure the neuropathology of AD is rapidly evolving, as is our understanding of the preclinical and prodromal stages of disease. Thus, it is now possible to identify individuals who are at risk for developing dementia caused by AD years or even decades before the onset of cognitive decline through clinical but also digital monitoring. A key premise of this article is that, in the future, the identification of at-risk individuals will continue to occur in clinical settings using traditional biomarker testing; but, the identification of at-risk individuals will also increasingly occur closer to – or even in – one’s home, potentially using digital biomarkers.

When you hear “Alzheimer’s disease,” in-home monitoring should come to mind.

Here, we argue that because AD affects the mind, the challenges associated with monitoring aimed at understanding the risk for disabling cognitive impairments are heightened as compared to the challenges of monitoring for physical ailments. In Section II, we discuss the biomarker transformation of AD, which is allowing us to see AD neuropathology in living persons and to identify individuals at increased risk for developing dementia caused by AD. In Section III, we outline empirical evidence regarding five different digital biomarkers; these digital biomarkers offer further insights into an individual’s risk for cognitive impairment and could soon be used for in-home monitoring. Finally, in Section IV, we identify six challenges that are particularly pronounced when monitoring for AD.

II The Evolving Understanding of AD

The field of AD research is rapidly moving from a syndromal definition of AD (see, e.g., the diagnostic process described in Section I) to a biological one. This shift reflects a growing understanding of the mechanisms underlying the clinical expression of AD.²

Biomarkers can now be used to identify AD neuropathology in vivo. A biomarker is a “defined characteristic that is measured as an indicator of normal biological processes, pathogenic processes or responses to an exposure or intervention.”³ Individuals are understood to have a biomarker profile, which (as we’re using it here) describes the presence or absence in their brain of three AD biomarkers: Beta-amyloid, pathologic tau, or neurodegeneration. These biomarkers can be measured using various modalities, including positron emission tomography (PET), cerebrospinal fluid (CSF) sampling, or magnetic resonance imaging (MRI); moreover, that blood-based biomarker tests are now available.⁴

In addition to the biomarker profile, a second, independent source of information is the individual’s cognitive stage. An individual may be cognitively unimpaired – within the expected range of cognitive testing scores and functioning in daily life, have mild cognitive impairment (MCI) – a slight but noticeable decline in cognitive skills, or have dementia. The patient’s biomarker profile can then be used in combination with the patient’s cognitive stage to characterize the patient’s place – and likely progression – along the Alzheimer’s continuum. The continuum spans the preclinical (i.e., clinically asymptomatic with evidence of AD neuropathology) and clinical (i.e., symptomatic) phases of AD.⁵

Individuals in the preclinical stage have AD biomarkers but do not have clinically measurable cognitive impairment. They may be truly cognitively unimpaired, or they may have subjective cognitive decline – a self-experienced decline in cognitive capacity as compared to baseline.⁶ Those with preclinical AD are understood to be at an increased risk of short-term cognitive decline.⁷ An estimated 46.7 million Americans have preclinical AD (defined by amyloidosis, neurodegeneration, or both), though it’s important to emphasize that not all of them will progress to a dementia-level of impairment.⁸

At present, preclinical AD remains a research construct. It is not yet diagnosed clinically. Researchers hope, however, that intervening earlier rather than later in the course of the disease will allow them to delay or prevent the onset of cognitive and functional impairment. Therefore, they are conducting secondary prevention trials, which recruit individuals who are asymptomatic but biomarker-positive for AD – that is, who have preclinical AD – to test new drugs or novel interventions. It is reasonable to assume that if the preclinical AD construct is validated and if a disease-modifying therapy for AD is found, preclinical AD will move from the research to the clinical context.

In the future, people who receive a preclinical AD diagnosis will have insight into their risk of developing MCI or dementia years or even decades before the onset of impairments.⁹ Monitoring digital biomarkers in the home, the focus of Section III, will likely be complementary to clinical assessment. For instance, monitoring may be used to watch for incipient changes in cognition after a preclinical AD diagnosis. Or, conversely, data generated by in-home monitoring may suggest that it is time to see a clinician for an AD workup.

III Digital Biomarkers of AD

In parallel with our evolving understanding of beta-amyloid, pathologic tau, and neurodegeneration as “traditional” biomarkers of AD, there have been advances in our understanding of digital biomarkers for AD. Efforts to concretely and comprehensively define digital biomarkers are ongoing.¹⁰ For the purposes of this chapter, we use the following definition: “Objective, quantifiable, physiological, and behavioral data that are collected and measured by means of digital devices, such as embedded environmental sensors, portables, wearables, implantables, or digestibles.”¹¹

Digital biomarkers have the potential to flag uncharacteristic behaviors or minor mistakes that offer insights into an older adult’s risk of cognitive and functional decline or to indicate early cognitive decline. As noted above, the preclinical stage of AD is characterized by the presence of biomarkers in the absence of measurable cognitive impairment. Despite going undetected on standard cognitive tests, subtle cognitive changes may be present. There is, in fact, a growing body of evidence that subjective cognitive decline in individuals with an unimpaired performance on cognitive tests may represent the first symptomatic manifestation of AD.¹² These small changes from the individual’s baseline may have downstream effects on complex cognitive and functional behaviors. Digital biomarkers offer a means of capturing these effects.

Here, we discuss five digital biomarkers for AD, highlighting both promising opportunities for monitoring the minds of older adults and limitations in our current knowledge and monitoring abilities. Crucially, these opportunities primarily reside outside of routine clinical settings. These examples are not meant to be exhaustive, but rather have been selected to highlight a range of monitoring modalities involving diverse actors. Moreover, they reveal a variety of potential challenges, which are the focus of Section IV.

A Driving Patterns

Due to the complex processes involved in spatial navigation and vehicle operation, an assessment of driving patterns offers an avenue for detecting changes in thinking and behavior. Indeed, prior studies demonstrate that those with symptomatic AD drive shorter distances and visit fewer unique destinations.¹³ Research also suggests that detectable spatial navigation deficits may precede AD symptom development in cognitively normal individuals with AD biomarkers.¹⁴ A limitation of this work is that it was conducted using simulators, which only measure performance in very controlled settings and so are limited in their generalizability.¹⁵ Studies have, therefore, shifted to a naturalistic approach to data collection to characterize daily driving patterns. Researchers can passively collect data using global positioning system (GPS) devices installed in participant vehicles. The resulting information includes average trip distance, number of unique destinations, number of trips with a speed of six miles per hour or more below the posted limit (i.e., underspeed), and a variety of other measures to quantify driving performance.¹⁶ These studies have found differing behavior and driving patterns between cognitively unimpaired participants with and without AD biomarkers, including a greater decline in the number of days driving per month for those with AD biomarkers.¹⁷

These findings suggest that assessing driving patterns – as some insurers already do through standalone devices or apps¹⁸ – may help identify individuals at risk for cognitive decline due to AD.

B Banking and Finances

Instrumental activities of daily living (IADLs) are complex activities necessary for individuals to live independently, such as managing one’s finances. As AD progresses, IADLs become increasingly impaired. A 2021 study examined longitudinal credit report information for over 80,000 Medicare beneficiaries.¹⁹ The researchers found that those with an AD or related dementia diagnosis were more likely to have missed bill payments over the six years prior to their dementia diagnosis. They also found that individuals with a dementia diagnosis developed subprime credit scores two-and-a-half years before their diagnosis. In a prospective study of cognitively unimpaired older adults, researchers found that a low awareness of financial and other types of scams was associated with an increased risk for MCI and dementia, though the measure was too weak for prediction at the individual level.²⁰

More work is needed to characterize the timeframe of changes in financial management, but detecting changes such as missed payments, bounced checks, or altered purchasing behavior (e.g., repeated purchases) presents another opportunity to identify individuals with preclinical AD. Banking and financial institutions already use algorithms, behavioral analytics, and artificial intelligence (AI)-powered technology to identify unusual transactions or spending behaviors that may be suggestive of fraud.²¹ Similar techniques could be adapted to monitor older adults and notify them of behaviors indicative of dementia risk.

C Smart Appliances

Sensors can be deployed in the home to detect cognitive changes in older adults.²² In a task-based study, individuals with MCI have been shown to spend more time in the kitchen when performing a set of home-based activities.²³ While in the kitchen, participants with MCI open cabinets and drawers, as well as the refrigerator, more often than cognitively unimpaired participants.²⁴ Researchers are exploring whether it is possible to use similar techniques to differentiate between healthy controls and individuals with preclinical AD.²⁵ A challenge for such monitoring studies (and, by extension, for real-life uptake) is the need to deploy multiple sensors in the home. One study attempted to circumvent this issue by focusing on passive in-home power usage for large appliances; the team found, on average, lower daily and seasonal usage of appliances among people with cognitive impairment.²⁶

Smart appliances, like refrigerators and ovens, connect to the internet and can sync with smartphones or other devices. They are already in many homes and are another alternative to sensor-based systems for detecting early cognitive changes. Smart refrigerators could track the frequency with which they are opened and for how long. Similarly, smart ovens may track the time they are left on. Such usage information could then be shared with the consumer by the appliance itself, for example, via an app.

D Speech

Changes in speech have been used to characterize the progression of AD. Studies have often used active data collection in which individuals are recorded on a smartphone or similar device as they complete tasks associated with verbal fluency, picture description, and free speech. The voice recordings are then processed, sometimes using machine-learning techniques. Studies have found that short vocal tasks can be used to differentiate participants with MCI from those with dementia.²⁷ It remains an open question whether preclinical AD presents with detectable changes in speech. Yet, one study of speech changes found that cognitively unimpaired participants with AD biomarkers used fewer concrete nouns and content words during spontaneous speech.²⁸

The interest in modalities for passive speech data collection – for example, conversations over the phone, communication with digital assistants, and texting related information – is mounting.²⁹ Improvements in machine-learning to reduce the burden of speech analysis, coupled with broad access to devices with microphones, are increasing the potential of passive speech data collection. Automatic speech recognition used for digital assistants like Amazon Alexa and Apple Siri has made strides in accuracy. As technological advancements further streamline transcription and analysis, speech data may be used to characterize changes related to preclinical AD. Simply put, Alexa may soon diagnose progression along the Alzheimer’s continuum from preclinical AD to MCI to dementia.³⁰

E Device Use

The ways people use their smartphones – including the amount of time spent on certain apps, login attempts, patterns of use, and disruptions in social interactions – may reveal signs of cognitive decline.³¹ Studies examining patterns of smartphone use in older adults with and without cognitive impairment suggest that app usage is related to cognitive health.³² There is much interest in leveraging device use as a potential marker of cognitive decline. The Intuition Study (NCT05058950), a collaboration between Biogen and Apple Inc., began in September 2021 with the aim of using multimodal passive sensor data from iPhone and Apple Watch usage to differentiate normal cognition from MCI; a secondary aim is to develop a function for predicting between individuals who will and will not develop MCI. With 23,000 participants, this observational longitudinal study will be the largest study to date collecting passive device use data.

Devices, like smartphones, could soon flag usage patterns that are suggestive of an increased risk of decline. Further, specific apps may be developed to detect concerning behavior changes by accessing meta-data from other apps and devices; this may streamline access to information and improve consumer friendliness.

IV Challenges Ahead

Here, we identify six ethical and legal challenges that will accompany the monitoring of digital biomarkers for AD. These are not exclusive, and many issues associated generally with measuring digital biomarkers will apply here as well. Moreover, the challenges outlined herein are not unique to digital biomarkers for AD. Rather, we would argue that they are heightened in this context because AD is a disease not just of the brain but the mind.

A Consent to Collect and Consent to Disclose

Although we hypothesize that preclinical AD will not be diagnosed clinically (i.e., using traditional biomarkers) until there is a disease-modifying therapy that renders the diagnosis medically actionable, in-home monitoring of digital biomarkers is not subject to this constraint. In fact, potential means of collecting and analyzing digital biomarkers for AD are already in our homes.

Yet, it is unlikely that individuals are aware that the GPS devices in their cars, the smart appliances in their kitchens, and the online banking apps on their phones can provide insights into their risk of cognitive and functional decline. Plugging in the GPS device, using the smart oven, or paying bills online, therefore, does not imply consent to having one’s brain health measured. Nor can consent be presumed. Many individuals do not want to know about their risk of developing dementia caused by AD because there is little to be done about it.³³ Others eschew learning their dementia risk to avoid existential dread.³⁴ This all suggests that, if digital biomarkers for AD are to be collected, there must be explicit consent.

Even if individuals agree to having their digital biomarkers for AD monitored, they may ultimately choose against learning what is revealed therein. Some individuals who undergo testing to learn whether they are at risk for dementia caused by AD – whether due to genes or to biomarkers – subsequently decline to learn the results.³⁵

This contrasts with – drawing an analogy to emergency medicine – our ability to presume consent for an Apple watch to monitor for and alert us to a possibly fatal arrythmia. But even there, where there is greater reason to presume consent, the evidence suggests we ought to eschew a “more is more” approach to disclosure. Apple watch monitoring for arrythmia can unduly worry people who receive a notification and subsequently follow-up with doctors, undergoing invasive and expensive tests only for the results to come back normal.³⁶ When the rate of false positives is unknown – or remains high – and when there are risks and burdens associated with disclosure, caution must accompany implementation.

B Communicating Digital Biomarker Information

To date, traditional AD biomarker information has been disclosed to cognitively unimpaired adults in highly controlled environments, mostly through research studies and with specialist clinical expertise.³⁷ Substantial work has gone into developing methods for disclosure, and the recommended steps include preparing people to learn about their biomarker information, maintaining sensitivity in returning the results, and following-up to ensure people feel supported after learning the results.³⁸ Digital biomarkers present an opportunity for individuals to learn that they are exhibiting subtle signs of cognitive decline or are at risk for dementia in the future from an app or from their banker or insurance agent – and without the option to speak directly and quickly about the results with a medical professional.

Although the disclosure of AD biomarkers has generally been found to be safe in pre-screened populations,³⁹ care should be taken when disclosing digital biomarker information more broadly. Here, the field may learn from discussions of direct-to-consumer genetic or biochemical testing.⁴⁰

Another concern is that the monitoring of digital biomarker data could lead to the inadvertent disclosure of dementia risk. Imagine, for instance, that your changing device usage is flagged and then used to generate targeted advertisements for supplements to boost brain health or for memory games. You could learn you are at risk simply by scrolling through your social media feed. And, in that case, any pretense of thoughtful disclosure is dropped.

C Conflicting Desires for Monitoring

Studies suggest that some cognitively unimpaired older adults share their AD biomarker results with others because they would like to be monitored for – and alerted to – changes in cognition and function that might negatively affect their wellbeing.⁴¹ Often, these individuals feel it is ethically important to share this information so as to prepare family members who might, in the future, need to provide dementia care or serve as a surrogate decision maker.⁴² Other older adults, however, perceive monitoring as intrusive and unwelcomed.⁴³

In an interview study of the family members of cognitively unimpaired older adults with AD biomarkers, some family members described watching the older adult more closely for symptoms of MCI or dementia after learning the biomarker results.⁴⁴ This may reflect family members’ evolving understanding of themselves as pre-caregivers – individuals at increased risk for informal dementia caregiving.⁴⁵ Technology can allow family members to remotely monitor an older adult’s location, movements, and activities, in order to detect functional decline and changes in cognition, as well as to intervene if needed. Despite these putative advantages, monitoring may be a source of friction if older adults and their families do not agree on its appropriateness or on who should have access to the resulting information.

V Stigma and Discrimination

Dementia caused by AD is highly stigmatized.⁴⁶ Research with cognitively unimpaired individuals who have the AD biomarker beta-amyloid suggests that many worry that this information would be stigmatizing if disclosed to others.⁴⁷ Unfortunately, this concern is likely justified; a survey experiment with a nationally representative sample of American adults found that, even in the absence of cognitive symptoms, a positive AD biomarker result evokes stronger stigmatizing reactions among members of the general public than a negative result.⁴⁸

Discrimination occurs when stigmatization is enacted via concrete behaviors. Cognitively unimpaired individuals who have beta-amyloid anticipate discrimination across a variety of contexts – from everyday social interactions to employment, housing, and insurance.⁴⁹ It is not yet known whether – and if so to what extent – digital biomarkers will lead to stigma and discrimination. However, we must be aware of this possibility, as well as the scant legal protection against discrimination on the basis of biomarkers.⁵⁰

VI Information Privacy

Digital biomarker information is health information. But it is health information in the hands of bankers and insurance agents or technology companies – individuals and entities that are not health care providers and are therefore not subject to the privacy laws that govern health care data. The Health Insurance Portability and Privacy Act (HIPAA) focuses on data from medical records; it does not, for instance, cover data generated by smartphone apps.⁵¹ The need for privacy is intensified by the potential for stigma and discrimination, discussed above.

Further, older adults with MCI and dementia are vulnerable – for example, to financial scammers. It is important to ensure that data generated by monitoring is not abused – by those who collect it or by those who subsequently access it – to identify potential targets for abuse and exploitation. Abuse and exploitation may occur at the hands of an unscrupulous app developer but also, or perhaps more likely, at the hands of an unscrupulous family member or friend.

VII Disparities in Health and Technology

The older population is becoming significantly more racially and ethnically diverse.⁵² Black and Hispanic older adults are at higher risk than White older adults for developing AD, and they encounter disproportionate barriers to accessing health care generally, and dementia care specifically.⁵³ Health disparities are increasingly understood to reflect a broad, complex, and interrelated array of factors, including racism.⁵⁴ There are well-reported concerns about racism in AI.⁵⁵ Those are no less salient here and may be more salient, given disparities in care.

Further, monitoring may be cost prohibitive, impacted by the digital divide, or reliant on an individual’s geographic location. For instance, older adults, especially adults from minoritized communities, may not have smart devices. According to a Pew report using data collected in 2021, only 61 percent of those aged 65 and older owned a smartphone and 44 percent owned a tablet.⁵⁶ As many of the digital biomarkers described in Section III require a smart device, uptake of monitoring methods may be unevenly distributed and exacerbate, rather than alleviate, disparities in AD care.

VIII Conclusion

The older adult we envisioned at the beginning of this chapter will not be the only face of AD much longer. We may soon come to think, too, of adults with preclinical AD. These individuals may learn about their heightened risk of cognitive and functional impairment from a clinician. Or they may learn about it because the GPS device plugged into their car has detected slight alterations in their driving patterns, because their smart refrigerator has alerted them to the fridge door staying open a bit longer, or because their phone has noted slight changes in speech.

AD is undergoing a biomarker transformation, of which digital biomarkers are a part. AD is a deeply feared condition, as it robs people of their ability to self-determine. Care must therefore be taken to address the multitude of challenges that arise when monitoring our minds.