Remote Digital Technologies for the Early Detection and Monitoring of Cognitive Decline in Patients With Type 2 Diabetes: Insights From Studies of Neurodegenerative Diseases

Computer-, Tablet-, and Browser-Based Assessments

Smartphone-Based Assessments

Smartphone-based assessments of cognition, motor functioning, speech, and language are under evaluation for ADRD. Smartphones are attractive for conducting widely accessible remote cognitive evaluations given their high prevalence globally. ⁵³ Assessments deployed through smartphones have been found to be reliable^40,41,53 and valid, as supported by evidence of strong correspondence between smartphone cognitive tests and gold-standard cognitive tests,^{40,61 -65} everyday functioning,^{62,66 -68} brain imaging, and fluid biomarkers of AD pathology.^{40,69 -72}

Motor deficits are a common manifestation of a range of neurodegenerative, neurological, and other medical conditions including T2D.^{73 -77} Applications have been developed to utilize the smartphone’s touch screen, accelerometer, magnetometer, and gyroscope to quantify gait, balance, dexterity, and coordination. Smartphone-based assessments of finger tapping, gait, and balance can detect treatment response, predict disease severity, and are correlated with standard clinical motor assessments.^78,79

Language impairment is a central feature of many neurodegenerative disorders, and subtle speech changes may occur early in the disease process either due to changes in speech/language networks or secondary to other cognitive deficits. Several applications leverage smartphones’ audio recording capabilities to capture speech during conversation and completion of various tasks.^36,80 Automated processing pipelines extract linguistic (e.g., syntactic complexity, semantic content, and lexical diversity) and acoustic (e.g., pause length, pitch) features from audio recordings.^{80 -85} Digital speech analyses are sensitive tools to detect early signs of cognitive impairment and predict transition from normal cognition to MCI.^{80,81,86 -88} One study comparing healthy older adults with and without T2D suggested temporal speech parameters (e.g., utterance length and total pause duration), but not standard neuropsychological measures, were sensitive to early cognitive changes in those with T2D. ⁸⁹ Replication and extension are needed to understand the utility of speech recordings for early detection of cognitive decline in T2D.

A few platforms collect several data streams including self-administered cognitive tests, motor, and speech collection using a single smartphone app such as the NIH-funded ALLFTD Mobile App, built on the Linkt app in partnership with Datacubed Health.^36,90

To overcome the noise inherent to testing people in an unsupervised environment, innovative testing paradigms with repeated measurements are employed in smartphone testing studies. Some studies utilize randomized independent parallel test versions for each memory task to reduce confounding practice effects,^41,53 whereas others administer the same test multiple times and capitalize on diminished practice effects as an early indicator of pathological brain changes. ⁵³ The “burst” approach is a paradigm for deploying repeated assessments delivered at varying times per day using a pseudorandomized ecological momentary assessment method and is promising for extracting highly reliable data from unsupervised evaluation.^{43,70,71,91 -93}

Limitations of Active Tools

Despite many benefits, active remote monitoring is not without limitations and unique considerations. A common criticism of unsupervised self-administration is the difficulty discerning a participant’s level of effort and engagement, the presence of distractions, and whether assistance completing measures was received. ⁴⁵ Distractions are reported to occur in a small (~7.4%) yet nonnegligible proportion of participants completing unsupervised remote cognitive testing. ⁴⁹ For patients with cognitive impairments or low digital familiarity, it may be difficult or impossible to engage them in testing without a proctor. Additional ethical concerns must be considered if attempting to integrate caregivers as proctors for participants with more substantial cognitive deficits—current recommendations are to only engage caregivers in setting up digital assessments but not proctoring any tasks. ⁹⁴

The lack of widespread access to technology and internet connectivity is a significant barrier, particularly in older adults, low-income households, and ethnic diverse ethnic groups. However, studies show a steadily increasing rise in technology adoption, smartphone ownership, and accessibility among these populations, especially in recent years.^{95 -103} Investigators and clinicians must also decide whether to use a “managed” or “bring-your-own” device (BYOD). A managed device (i.e., a dedicated device for the cognitive assessment rather than the examinee’s own device) may reduce variance associated with hardware and software, but requires additional resources for device purchase and maintenance. ³⁶ Differing levels of familiarity with the chosen managed device may also affect performance. ³⁶ Alternatively, BYOD may reach a wider and more diverse audience at a lower cost, although concerns regarding the effects of hardware and software differences remain. ¹⁰⁴ Finally, privacy concerns must be considered while using digital data capture as cybersecurity is paramount.^35,105,106 Despite its limitations, active testing paradigms are reliable and valid methods for conducting ADRD research and are relevant for observational and interventional diabetes studies.

In-Home Sensor Monitoring

A common approach to passive data collection involves using sensors to monitor movement inside and outside of the home. In-home passive infrared (PIR) motion sensors record indoor mobility frequency and variability (room-to-room transitions/day) and walking speed.^{107 -116} Fewer outings and decreased indoor movement were found in people with incident cognitive impairment. ¹¹⁷ Graded reduction in gait velocity, collected unobtrusively from in-home PIR sensors, is both associated with and predictive of cognitive and physical decline.^{107,110,118 -122} PIR sensors can also detect the nonamnestic subtype of MCI in older adults with high precision. ¹²³

Other in-home technologies include bioelectrical impedance analysis of a digital scale to track weight fluctuations, a sleep mat placed under a patient’s mattress to measure heart and respiratory rate variability, sleep duration and sleep stage measurements, and a 7-day electronic pillbox recording medication adherence.^115,116 A study comparing healthy older adults with slight differences on cognitive testing, observed large differences in medication adherence—thereby suggesting real-world monitoring may be more sensitive than traditional gold-standard paper-and-pencil tasks.^116,124,125 Audio recordings can also be collected passively (e.g., during phone calls or interviews, in-home microphone use).^{126 -128} We anticipate the array of behaviors captured by in-home sensors will continue to expand as sensor technology develops.

Monitoring Driving and GPS

Driving is a complex functional activity requiring the integration of multiple cognitive domains. Impaired driving possesses high-stake safety implications, making it a clinically relevant target for passive monitoring. GPS data can be combined with software to monitor driving behaviors (i.e., rapid acceleration and braking). Everyday driving behavior, such as driving shorter distances and fewer visits to unique destinations, is associated with and predictive of preclinical AD.^129,130 Even in nondriving patients, GPS may be a method to study changes in travel, social activities, and other aspects of movement that might reflect underlying health issues.

Wearable Devices

Wearable devices (e.g., wristbands and smartwatches) enable continuous monitoring of physical activity, movement via GPS, and a vast range of physiological functions including sleep, gait, heart rate, electrocardiogram, blood oxygen, and continuous glucose monitoring (CGM). Combinations of mobility metrics (e.g., steps, turning, and transition from sit-to-stand) obtained from wearable devices are related to specific health outcomes such as MCI, Parkinson’s disease, and AD-dementia.^131,132 Heart rate variability is related to mental stress and cardiac functioning over time,^{133 -135} which may be a feasible tool to assess and monitor cognitive and brain health-related physiological changes.^136,137 A study found adherence to wearing a Fitbit device (89% of study days) was a predictor of memory functioning among older adults, ¹³⁸ while wearable-quantified physical activity was more predictive of brain health and cardiometabolic indices than self-report. ¹³⁹ High adherence in daily wear, incremental validity over self-report measures, and relatively low costs make wearable devices a viable tool for both research and clinical monitoring of T2D patients.

Computer Usage Monitoring

Interacting with personal electronic devices requires a complex orchestration of cognitive skills. Software installed to monitor computer and smartphone usage generates densely sampled data to identify metrics sensitive to early cognitive changes. For example, participants with MCI demonstrate lower overall computer usage and greater day-to-day usage time variability relative to cognitively normal controls. ¹⁴⁰ Compared with cognitively intact older adults, older adults with MCI generate fewer, more variable, and less efficient mouse movements, which are significantly associated with gold-standard cognitive test scores. ¹⁴¹ In another study, participants with MCI spent less time on the computer and on various computer programs (e.g., email, word processing applications, search functions) compared with healthy controls. ¹⁴² In Huntington’s disease, mutation carriers are more inconsistent than controls in keyboard strokes and typing cadence—both of which are correlated with disease severity. ¹⁴³

Smartphone Usage Monitoring

Given that over 83% of the world owns smartphones globally, ¹⁴⁴ and these devices are often carried throughout the day, smartphones offer a unique inroad to daily activity and functioning. Integrated sensors and usage meta-data (e.g., steps, user commute routine) are collected automatically in smartphones and are often accessible for research purposes.^145,146 Examples of extractable data types include typing patterns, call frequency, social media app interactions, and environmental decibel levels. Open-source software packages and digital phenotyping platforms (e.g., Beiwe app) ¹⁴⁷ are available for investigators to extract data from participants’ interaction with their smartphones (e.g., finger tapping, keystroke metadata).^{148 -150} The Linkt/ALLFTD App also captures several aspects of passive data including step count, screen time, social media app usage, GPS, and battery life.^{36,151 -153} Extracting smartphone metadata is a promising and scalable passive monitoring approach, although more research is required to understand how best to utilize this rich data source and identify best metrics for reliable early detection and monitoring of cognitive and functional changes.

Passive Data Input Integration

Rather than relying on a single sensor type or software package for passive data collection, time-stamped integration across multiple passive data streams may enable the most comprehensive and sensitive view of daily functioning and cognitive status. ¹⁵⁴ The Collaborative Aging Research using Technology (CART) Initiative ¹²⁴ platform integrates data captured from various sensors and software discussed above and weekly questionnaires completed by participants to contextualize data (e.g., presence of visitors, medication changes). Integrative platforms easily incorporate with CGMs and smart insulin pen smartphone app data to better understand cognitive fluctuations in relation to blood glucose level fluctuations as a mechanism to determine predictors of cognitive decline.

Limitations of Passive Tools

Despite its benefits, passive remote monitoring methods have several limitations. For example, wearables may be linked to perceived social stigma associated with diabetes as visible devices could reveal health status to others^{155 -157}; new data, however, suggests increasing willingness to wear devices and positive experiences with wearable use among adults and older adults.^{136,158 -161} Digital sensors can also be expensive and in-home installation must be consistently executed to obtain accurate naturalistic data. Like active technologies, cybersecurity and ethical implications of private home monitoring must be considered. The black box nature of commercial software and firmware algorithms can change, sometimes unbeknownst to the user, during a study. As new versions of devices are released, consistency with prior versions cannot be assumed and time-consuming harmonization studies are often required. Finally, challenges may arise when contextualizing collected data as sensors and digital devices must accurately discern sensors activated by the patient and not by other people, animals, or external forces.