This paper is in the following e-collection/theme issue:

Formative Evaluation of Digital Health Interventions (2226) Fitness Trackers and Smart Pedometers/Accelerometers (543) Artificial Intelligence (1216) Dementia and Cognitive Decline (429) Wearable Devices and Sensors (825) Machine Learning (1401)

Published on in Vol 8 (2024)

Preprints (earlier versions) of this paper are available at https://preprints.jmir.org/preprint/55575, first published .
Prediction of Mild Cognitive Impairment Status: Pilot Study of Machine Learning Models Based on Longitudinal Data From Fitness Trackers

Prediction of Mild Cognitive Impairment Status: Pilot Study of Machine Learning Models Based on Longitudinal Data From Fitness Trackers

Prediction of Mild Cognitive Impairment Status: Pilot Study of Machine Learning Models Based on Longitudinal Data From Fitness Trackers

Authors of this article:

Qidi Xu1 Author Orcid Image ;   Yejin Kim1 Author Orcid Image ;   Karen Chung2 Author Orcid Image ;   Paul Schulz2 Author Orcid Image ;   Assaf Gottlieb1 Author Orcid Image
Article Authors Cited by Tweetations Metrics

    Original Paper

    1McWilliams School of Biomedical Informatics, University of Texas Health Science Center at Houston, Houston, TX, United States

    2McGovern Medical School, University of Texas Health Science Center at Houston, Houston, TX, United States

    Corresponding Author:

    Assaf Gottlieb, PhD

    McWilliams School of Biomedical Informatics

    University of Texas Health Science Center at Houston

    7000 Fannin St

    Houston, TX, 77030

    United States

    Phone: 1 7135003698

    Email: assaf.gottlieb@uth.tmc.edu


    Background: Early signs of Alzheimer disease (AD) are difficult to detect, causing diagnoses to be significantly delayed to time points when brain damage has already occurred and current experimental treatments have little effect on slowing disease progression. Tracking cognitive decline at early stages is critical for patients to make lifestyle changes and consider new and experimental therapies. Frequently studied biomarkers are invasive and costly and are limited for predicting conversion from normal to mild cognitive impairment (MCI).

    Objective: This study aimed to use data collected from fitness trackers to predict MCI status.

    Methods: In this pilot study, fitness trackers were worn by 20 participants: 12 patients with MCI and 8 age-matched controls. We collected physical activity, heart rate, and sleep data from each participant for up to 1 month and further developed a machine learning model to predict MCI status.

    Results: Our machine learning model was able to perfectly separate between MCI and controls (area under the curve=1.0). The top predictive features from the model included peak, cardio, and fat burn heart rate zones; resting heart rate; average deep sleep time; and total light activity time.

    Conclusions: Our results suggest that a longitudinal digital biomarker differentiates between controls and patients with MCI in a very cost-effective and noninvasive way and hence may be very useful for identifying patients with very early AD who can benefit from clinical trials and new, disease-modifying therapies.

    JMIR Form Res 2024;8:e55575

    doi:10.2196/55575

    Keywords

    mild cognitive impairment; Fitbits; fitness trackers; sleep; physical activity 


    Alzheimer disease (AD) is the sixth leading cause of death in the United States and incurs a heavy economic burden of US $257 billion in direct costs [1]. The numbers are staggering—11.3% of Americans aged 65 years and older have Alzheimer dementia, and more than twice as many are anticipated to have Alzheimer dementia by 2050 [1]. Moreover, estimates suggest that 46.7 million Americans are already in a preclinical AD stage [2]. AD diagnoses are often significantly delayed to time points when brain damage has already occurred, owing to difficulty in detecting early signs of AD due to cost and effort, which makes it critical to identify efficient methods for early detection of AD signs. One pillar of AD programs, like the National Alzheimer's Project Act, focuses on early diagnosis of AD, as it allows patients to make lifestyle changes and consider new treatment options. Early diagnosis and identification of the trajectory of cognitive decline would also allow pharmaceutical companies to develop better therapeutics to delay or halt the progression to AD and support enrollment in experimental trials.

    A growing body of evidence indicates that cognitive, sensory, and motor changes may precede clinical manifestations of AD by several years [3]. Validated neuropsychological or cognitive tests are often less effective in the earliest stages of the disease and are impractical for frequent testing [4,5]. Frequently studied biomarkers, such as detection of beta-amyloid and tau proteins in cerebrospinal fluid [6,7] or using structural magnetic resonance imaging (MRI) [8] or positron emission tomography (PET) molecular imaging [9] promise earlier detection of disease, but these tests are infrequently performed in clinical settings since they are invasive or costly. Additionally, not all individuals with amyloid plaques will go on to develop dementia [10]. This calls for inexpensive and noninvasive methods that can provide clinicians with ways to augment these biomarkers and help triage patients eligible for these expensive and invasive tests.

    One biomarker that has been associated with early changes in patients with mild cognitive impairment (MCI) is disruption of sleep. Sleep disturbance is prevalent in mild cognitive decline and has been previously associated with cognitive decline in several meta-analyses when compared across different stages of cognitive status [11,12]. Previous studies suggest that poor sleep quality is associated with the presence of amyloid plaques in cognitively healthy individuals in AD-sensitive brain regions [13] and that even 1 night of sleep deprivation affects the levels of amyloid-β in the cerebrospinal fluid of healthy middle-aged men [14]. Other studies suggest that poor sleep may also be associated with increased levels of tau in the cerebrospinal fluid of cognitively normal people [15] and participants with the highest levels of tau had the lowest levels of nonrapid eye movement (NREM) across different cognitive statuses from unimpaired to maximal impairment [16]. Finally, high-frequency heart rate variability (HRV), especially during NREM sleep, may be an early biomarker for dementia detection when comparing older adults with MCI to patients with subjective cognitive impairment [17].

    A growing body of evidence also strongly suggests that physical exercise may attenuate cognitive impairment and reduce the risk of AD via a variety of mechanisms [18] and that some preventable lifestyle-related risk factors such as a sedentary lifestyle and lack of physical activity increase the risk of dementia and AD [19,20]. Proposed mechanisms that may be responsible for this effect include increased production of neurotrophic factors and neurogenesis, increased cerebral blood flow, reduced risk of cardiovascular diseases [21], increased cortical thickness, and enhanced cognitive function through improved cardiorespiratory fitness in MCI patients [22]. Although physical activity does not attenuate cognitive impairment, low levels are a risk factor for dementia [23].

    Although previous studies have shown that sleep and physical activity are individually associated with AD [11,12,17,20,24-26], their measurements were either based on collection via self-reported questionnaires, which tend to overestimate objective measurements [27-30], or collected over very short periods of time (typically 2-6 days) via an actigraph [15,16,25], limiting the ability to assess temporal variability and track gradual changes in these traits. Additionally, actigraphs typically do not measure heart rate, and acceleration features show only modest performance in differentiating between sleep stages [31,32]. Wearable technologies present a unique opportunity to capture these early changes preceding clinical decline while addressing limitations of the aforementioned biomarkers and circumvent inaccuracy in self-reported traits, but studies have thus far relied mostly on costly medical devices that were used only for short time periods. A review discussing the potential of digital markers for early detection of AD indicated that accurate algorithms require longitudinal observational studies with confirmed MCI and healthy controls [3].

    Using cost-effective, widely available fitness trackers (Fitbit Charge 4), we captured sleep, physical activity, and heart rate. Passive collection is well-suited for older adults, especially those who are already diagnosed with MCI, as it avoids relying on subjective recollection and does not interfere with their day-to-day activities. Studies have determined that Fitbits provide similar estimates of sleep outside the laboratory to a research-grade actigraph [33], and another study showed that consumer-grade multimodal sensor streams can detect MCI status [34].

    Here, we report a pilot study that demonstrates that sleep, physical activity, and heart rate recorded from a fitness tracker can distinguish between MCI and age-matched controls with a high accuracy. Furthermore, we show that 11 days of recording is sufficient to reach high accuracy. These methods may be used for early detection of decline in cognitive function, which can prompt clinicians to conduct further testing.


    Recruitment

    This study included participants who visited the University of Texas Health Science Center at Houston (UTHealth) Neurosciences Neurocognitive Disorders Center (NDC) at Houston, TX between January 2022 and September 2023. Per the approved recruitment protocol, patients and their spouses visiting the NDC who matched the inclusion criteria (see the Cohort Definition section) were approached by a study coordinator to participate in a fitness tracking study.

    Interested participants were consented by the study coordinator in a private room and were given a written consent form and time to ask any questions. The study was discussed, including the purpose, procedures, associated risks and benefits, time commitments, and costs. Participants were also told that participation was voluntary and they could withdraw at any time. If the participant agreed to be in the study, the participant signed the informed consent form. A signed and dated copy was given to the participant. The same procedures were followed for consent of the study partner. Patient privacy was protected by assigning unique study IDs to each participant, and the mapping was available only to the study coordinator.

    After participants consented, the recruiting study coordinator set up their fitness tracker (Fitbit Charge 4) and installed the corresponding free Fitbit app on the participant’s smartphone. Participants were also provided with instructions on how to wear and charge the fitness tracker and were instructed to keep their Bluetooth on their phones so the Fitbit app could synchronize with the fitness tracker. Each participant wore their fitness tracker between 2 consecutive visits to the clinic, returning it during their next visit.

    Ethical Considerations

    The study was approved by the UTHealth institutional review board (HSC-MS-21-0719).

    Cohort Definition

    Cohort

    The recruited cohort included 12 cases, including 11 patients with amnestic MCI and 1 diagnosed with early AD. Additionally, 8 age-matched healthy controls were recruited. All the recruited controls were spouses of the recruited MCI, enabling minimization of variability in lifestyle affecting physical activity and sleep times. Having the spouse wear a fitness tracker concomitantly also helped verify that they were charged on time.

    Inclusion Criteria

    The eligibility criteria included cases and controls aged 60 years to 80 years.

    MCI Diagnosis

    Cases were diagnosed by a neurologist who runs the NDC clinic with 35 years of experience diagnosing MCI. The diagnostic process starts with surveying the patient’s previous medical history (eg, comorbidities), patient background, family history, social history, sleep patterns, mood issues, and other neurologic changes, such as gait, from the patient and their family. The clinical team asks directed questions to better identify areas of change in cognition, mood, behavior, and activities of daily living. Screening tests are administered, including the Montreal Cognitive Assessment [35] and Mini Mental State Examination [36] to identify the level of cognitive impairment. Depending on these results, other tests are ordered to clarify the diagnosis and rule out alternative diagnoses. The initial tests most often include neuropsychological testing [37], MRI of the brain [38], and blood work. They can, however, also include sleep studies, electroencephalograms (EEGs), and lumbar punctures. The neurologist meets with patients and families and reviews the results. If appropriate, other testing is then ordered, including fluorodeoxyglucose PET [39], amyloid PET [40], or spinal taps. The neurologist or staff then meet the patient to explain the test conclusions. When testing is consistent with MCI due to AD pathology, patients were considered for inclusion in this study.

    Exclusion Criteria

    Participants were excluded if they were hospitalized within 1 month prior to screening, had suspected or known drug or alcohol abuse (ie, more than approximately 60 g of alcohol [approximately 1 liter of beer or 0.5 liter of wine] per day), had acute intercurrent infections, or were unable or unwilling to comply with the protocol and follow-up requirements. Patients taking sleep-inducing medication such as benzodiazepines were also excluded.

    A list of comorbidities and prescribed medications used by the patients with MCI is available in Tables S1 and S2 in Multimedia Appendix 1. We did not collect this information for the controls since the controls were not patients of the NDC.

    Data Collection

    Data, including daily summaries of sleep, activity, and heart rate, were collected through the Fitbit application programming interface using proprietary scripts coded in Python 3.9.

    For activities, we adopted the All of Us definition for a valid day as a participant wearing the Fitbit for at least 10 hours per day and reporting at least 100 steps per day [41]. Days not considered valid based on this threshold were excluded, including the days of the clinic visits when the participants were recruited or returned their fitness trackers.

    Fitbit algorithms differentiate between the main sleep epoch and naps with regard to capturing sleep stages. For the main sleep epoch, Fitbit measures periods of light, deep, rapid eye movement (REM), and wake stages. For naps, Fitbit detects awake, asleep, and restless stages. Due to differences in captured features, the main sleep epoch and naps were separated for the purpose of modeling.

    Assessing the Sex Imbalance Effect on Model Performance

    The sex distribution in the MCI and control groups was imbalanced, with 4 (4/12, 33%) women in the MCI group and 7 (7/8, 88%) women in the control group. To isolate the effect of sex on predictions, we modeled the MCI prediction as the following regression model:

    MCI status=a·sex+b

    After obtaining the coefficient a and residual b, a separate regression model was built on the residuals b as the dependent variable. This 2-step approach discerned whether other features, apart from sex, could effectively capture the unexplained variability in the residuals. The logistic regression model of sex yielded an area under the curve (AUC) value of 0.58, and a regression model based on the residuals b obtained an R2 of 0.96, which meant that sex plays a minor role in differentiating between cases and controls. This led us to exclude sex from future analysis to reduce bias from recruitment imbalance.

    Aggregation Model and Daily Model

    We tested 2 sets of prediction schemes (Figure S1 in Multimedia Appendix 1). The first scheme (“aggregation model”) aggregated the daily records for each participant to a single record. Summary statistics, including minimal, maximal, mean, sum, standard deviation, and skewness were applied to the daily data. The final set of features included age and 329 engineered summary statistics features from the fitness trackers.

    The second scheme (“daily model”) classified individual days per participant. The classifier predicted MCI status per day of each participant, followed by a majority vote on the percent of days predicted to be MCI per participant.

    We included age in the aggregation model, although there was no statistically significant difference between MCI and controls. To avoid introducing artificial bias into the daily model resulting from the difference in the number of days measured per participant (ie, the number of days a specific participant age was replicated depends on the number of days the fitness tracker was worn), we excluded age from the daily model, but this exclusion did not alter the performance.

    Feature Selection

    In order to deal with the large number of variables, we built a prediction model to select important features. These selected features were subsequently fed into a final prediction model.

    We tested several machine learning algorithms to select the features based on the best performing algorithm, including Gaussian naïve Bayes, logistic regression, support vector machine (SVM), random forest, XGBoost, and CatBoost. In each aggregation and daily data set, we used all features to train classifiers using 5-fold cross-validation.

    For each machine learning algorithm, we calculated Shapley Additive Explanations (SHAP) values to illuminate the impact of each feature on the model's prediction. We conducted independent feature selection on each training fold (alpha=0.02) and tested them on their respective testing folds. Finally, we identified features as important if they appeared in at least 2 of the 5 folds.

    Final Models and Evaluation

    The final aggregation and daily models were trained using only the features selected in the feature selection stage. We built prediction models with naïve Bayes, logistic regression, SVM, random forest, XGBoost, and CatBoost.

    In order to gain robustness, we conducted both 5 repeats of 5-fold cross-validation, each time with different data splits, and a leave-one-out cross-validation technique. The mean AUC values across the 5 repeats served as the standardized metric to identify the best model. When dealing with daily models, we used the majority vote on the number of days a participant had days classified as MCI, classifying a participant as MCI if the proportion of days classified as MCI out of their total number of measured days exceeded 50%. The proportion of MCI-classified days per participant also served for computing the AUC value for each testing fold. As before, we computed the mean AUC values to select the most effective model.

    Statistical Analysis

    Age differences between the MCI patients and controls were computed using the Wilcoxon ranked sum test, and sex differences were calculated using the Fisher exact test.


    Participant Characteristics

    We collected fitness tracker data from 12 patients with MCI or early AD and 8 age-matched controls between 2 clinic visits (as described in the Methods section; Figure 1).

    ‎
    Figure 1. Illustration of data collection over a period of ~1 month, followed by a machine learning classifier to detect mild cognitive impairment (MCI).

    After removing partial activity days (for more information, see the Methods section), a total of 530 tracked heart rate days (MCI: mean 25.6, SD 10.1 days; control: mean 27.9, SD 8.5 days), 510 activity days (MCI: mean 24.4, SD 10 days; control: mean 27.1, SD 8.2 days), and 499 sleep days or nights (MCI: mean 23.6, SD 8.8 days; control: mean 27, SD 9 days or nights; Figure S2 in Multimedia Appendix 1) remained. One MCI patient returned their fitness tracker early (Figure S2 in Multimedia Appendix 1).

    The average age of the patients in the MCI group was higher by 1.3 years than the control group’s age (Table 1), but the difference was not statistically significant (P=.6). Additionally, 55% (11/20) of all the participants were women. However, significantly more women were recruited to the control group than to the MCI group (P=.03; Table 1).

    Table 1. Demographic characteristics of the recruited patients to wear the fitness tracker (all participants were White).
    GroupParticipants, nWomen, nAge (years), mean (SD)
    Mild cognitive impairment12471.8 (7.3)
    Controls8770.5 (5.7)

    Predicting MCI Status

    We tested models for predicting MCI status that considered 2 types of processing the data: a model that aggregated the daily data per participant and a model that classified individual daily records (see the Methods section). We also tested the effect of using different combinations of data modalities from the 3 groups of measures—sleep, physical activity, and heart rate.

    In the aggregation model, we chose to use CatBoost as the preliminary model because it outperformed other algorithms on our aggregation data, while SVM was chosen in the daily model due to its superior performance in daily data.

    Regarding the aggregation model, different machine learning algorithms obtained the top accuracy for some data modality combinations of activity, heart rate, and sleep data (Tables S3 and S4 in Multimedia Appendix 1). However, the CatBoost algorithm was the most consistent algorithm at achieving high accuracy across combinations of modalities and across different cross-validation schemes.

    The classification performance of different combinations of data modalities is displayed in Table 2. Adding data modalities increases the accuracy of the algorithm, and the combination of physical activity, heart rate, and main sleep achieved perfect separation (AUC=1.0; Table 2; Figure 2; Figure S3 in Multimedia Appendix 1). The combination of physical activity, heart rate, and naps also achieved good separation between MCI and controls (AUC=0.94; Table 2); however, it included only 15 participants (15/20, 75%: 11 MCI and 4 controls) that had taken any nap during the time they wore the fitness tracker.

    Table 2. Performance of classifiers differentiating between patients with mild cognitive impairment and controls using different combinations of data modalities captured by fitness trackers for an aggregated data model.
    DataImportant features selected, nAUCa of the repeated 5-fold CVb, meanAUC of the LOOCVc
    Activities150.760.58
    Activities + heart rate100.820.82
    Main sleep150.810.77
    Napsd60.910.86
    Activities + main sleep150.870.84
    Activities + napsd110.970.93
    Activities + heart rate + main sleep151.01.0
    Activities + heart rate + napsd130.940.88

    aAUC: area under the curve.

    bCV: cross-validation.

    cLOOCV: leave-one-out cross-validation.

    dClassification of 15 participants (11 MCI and 4 controls) who had recorded naps within the time window.

    ‎
    Figure 2. Receiver operating characteristic curve for 5-fold cross-validation of the CatBoost classifier; different curves correspond to different combinations of data modalities captured by fitness trackers. AUC: area under the curve.

    As a corollary question, we were interested in knowing the minimal number of days that a fitness tracker needed to be worn in order to obtain sufficient performance from the models. Using the same models described before, we observed that, after only 11 days of wearing it, the classification performance surpassed an AUC of 0.90 (Figure 3).

    ‎
    Figure 3. The machine learning model performance (area under the curve [AUC]) as a function of the maximal number of days recorded from fitness trackers for each participant in (A) 5-fold cross-validation and (B) leave one out cross-validation.

    Regarding the daily model, we evaluated the accuracy of classifiers when classifying each day and taking the majority vote per participant (see the Methods section). The daily models performed worse than the aggregation models (Table S5 in Multimedia Appendix 1). Specifically for naps, many participants who had any naps had fewer than 3 days on which these naps occurred, contributing to much lower performance in daily combination compared with the aggregation data sets.

    Features Predictive of MCI Status

    The selected features predictive of MCI status differed between different data modalities.

    Aggregated Model

    The aggregated model included 15 selected features related to sleep (deep sleep minutes), heart rate (resting heart rate and cardio, fat burn, and peak heart rate zones), and physical activity (light active minutes and distance; Figure 4; Table S6 in Multimedia Appendix 1). When considering naps instead of main sleep (including only 15 participants who had naps; see the Methods section), we found 13 top features, including more features from physical activity and fewer features from heart rate (Table S6 in Multimedia Appendix 1). Notably, most of the top features from heart rate and physical activity data types were different when we replaced main sleep with naps, but lightly active physical activity minutes and distance and heart rate zone cardio time were common to the set of top features from the 2 data modalities.

    ‎
    Figure 4. Top features captured from fitness trackers and used in the model to differentiate between mild cognitive impairment cases and healthy controls using aggregated daily data. SHAP: Shapley Additive Explanations.
    Daily Model

    The top performing daily data features included 13 top-selected features from sleep (eg, average of light sleep and REM minutes), heart rate (resting heart rate and several heart rate zones), and physical activity (calories expended and lightly active minutes; Table S7 in Multimedia Appendix 1). In contrast to the aggregated model, the models using main sleep and naps shared multiple features, including physical activity light and very active activities; calories out; and several heart rate zones, including resting heart rate and cardio, fat burn, and out-of-range zones.


    Principal Findings

    In this pilot study, we recruited and tracked 20 MCI and healthy control participants using fitness trackers for a typical time of 2 weeks to 4 weeks in order to test the ability of the data captured by the fitness tracker to predict MCI status. We tested the use of features from physical activity, sleep, and heart rate measurements and trained a machine learning classifier to predict MCI status, obtaining perfect separation in predicting MCI status (AUC=1.0) when all the data types were combined. We compared 2 models with different engineered features: a model that classifies MCI status based on aggregated daily data for each participant and a model that classified individual days and determined MCI status based on a majority of days classified as MCI or control. We found that the model aggregating all days into a set of variables per participant performed better than when classifying each day in a daily model.

    The top features identified in the aggregated model agree with those in previously published studies. The majority of the top features were related to heart rate zones, including peak, cardio, and fat burn zones. A systematic review of HRV and cognitive function suggested that HRV is considered a promising early biomarker of cognitive impairment in populations without dementia or stroke [42]. Furthermore, HRV was suggested as a marker of self-regulatory processes in neurodegenerative conditions [43]. The association between higher HRV and better cognitive performance was also confirmed in a multi‐ethnic cohort of aging adults [44] and across groups of different fitness levels [45,46]. In addition to heart rate zones, another top heart feature is variability of resting heart rate. A higher resting heart rate has been associated with an increased risk of cognitive decline [47-49].

    The top physical activity features were light physical activity (distance and time). Several studies have established the relationship between light physical activity and risk for dementia. Increased physical activity level, including a low amount of light-intensity physical activity, has been associated with a reduced risk of dementia [50], and increasing light physical activity helps to maintain cognitive function among community-dwelling older adults [51]. Additionally, light activity as low as 10 minutes per day measured through accelerometers was previously shown to be associated with better cognitive function, while light physical activity performed ≥3 hours per day could be a protective factor for maintaining cognitive function in older adults [52,53].

    Finally, the top feature associated with sleep was the average length of deep sleep. A comparison of Fitbit’s deep sleep stage with EEG signals found that it aligns with the N3 stage of the NREM sleep stage [54]. There is interest around the role of N3 sleep in dementia given its pivotal role in memory consolidation, but evidence about this association is still limited [55]. One study that focused only on patients with obstructive sleep apnea found that a shortened N3 stage of sleep led to declarative memory deficit [56].

    Although all but 1 MCI participant had occasional naps, only one-half of the controls had naps. Focusing only on the participants who had naps, we obtained good performance, suggesting that information about naps could be important for identifying MCI status, where the length of the nap is one of the selected predictors. Previous reports on the associations between naps and cognitive function were inconclusive. No association was found between napping and cognitive decline in a meta-analysis of observational studies conducted with healthy individuals [57], and there was only a modest causal association between habitual daytime napping and larger total brain volume [58]. However, afternoon napping was related to better cognitive function in the aging Chinese population [59]. This has been related to higher levels of inflammatory markers that may induce cognitive impairment [60] and may disrupt circadian rhythms at the time of day of best coordination and fastest reaction time [61].

    A major concern we had with this older adult population was whether they would continuously wear the fitness tracker and keep it charged. Fortunately, all participants except 1 who returned it early wore and charged their fitness tracker for the full term. When alerts were received that their battery was low, a research coordinator contacted the participant or their caregiver to remind them to recharge it.

    Limitations

    Our study has 3 major limitations. The primary limitation of our pilot study is our small sample size and the limited racial diversity in our population. We attempted to address potential biases between the patients with MCI and controls by including couples, who were both age-matched but also had similar daily schedules. Although our selected features agree with established risk factors for cognitive impairment, the sample size may lead to 2 limitations, including the “curse of dimensionality,” where the number of features exceeds the number of samples, and co-linearity between different data types. For example, when a specific heart zone is elevated on a specific day, other zones will correspondingly be lower on the same day. We attempted to address the “curse of dimensionality” by having an independent step of feature selection that selected features that appear in more than 1 fold. A result of the sample size is limited generalizability. In order to generalize our MCI prediction models, larger and more diverse populations are needed to establish their use in the general population.

    A second limitation is possible selection bias. Several potential participants declined to participate because they already had a fitness tracker or a smartwatch. It is possible that the population who already uses this technology are different in relevant physical activity traits than the population we recruited.

    Finally, our clinical study may have interfering comorbidities and medications that may affect sleep; for example, 3 patients with MCI had sleep apnea and were taking medications that may affect sleep such as donepezil, rivastigmine, sertraline, escitalopram, melatonin, duloxetine, and gabapentin. Some of these medications, such as antidepressants (sertraline, escitalopram, and duloxetine) and gabapentin, have been reported to affect deep sleep, which is our prominent sleep feature [62-65], while the rest primarily influence only the REM stage [66-73].

    Conclusions

    Our pilot study demonstrates the use of fitness trackers for determining MCI status. This study is intriguing in that it suggests that it may be possible to use an inexpensive, convenient digital biomarker applied serially to achieve the critical goal of detecting MCI due to AD pathology as early as possible because that is when new disease-modifying therapies have the greatest efficacy. Clearly, more study is necessary to ascertain the sensitivity and specificity of this digital biomarker. Another important goal of this study is to continue to follow these and other participants to learn more about ongoing digital biomarker measurements as they relate to future cognitive decline.

    Acknowledgments

    We would like to thank Susan Cooper for helping with institutional review board approval of the study and to Alexa Bavero and Christine Farrell for helping initiate the recruitment process.

    The study was supported by the Alzheimer’s Association award (AARG-NTF-21-847409). YK is also supported by National Institutes of Health (NIH) awards R01AG082721 and R01AG066749, and QX is supported by NIH award R01AG082721.

    Data Availability

    Deidentified data and code are available at [74].

    Conflicts of Interest

    None declared.

    Multimedia Appendix 1

    Supplementary Tables and Figures.

    PDF File (Adobe PDF File), 493 KB
    1. Wiley J. 2021 Alzheimer's disease facts and figures. Alzheimers Dement. Mar 2021;17(3):327-406. [CrossRef] [Medline]
    2. Brookmeyer R, Abdalla N, Kawas CH, Corrada MM. Forecasting the prevalence of preclinical and clinical Alzheimer's disease in the United States. Alzheimers Dement. Feb 06, 2018;14(2):121-129. [FREE Full text] [CrossRef] [Medline]
    3. Kourtis LC, Regele OB, Wright JM, Jones GB. Digital biomarkers for Alzheimer's disease: the mobile/ wearable devices opportunity. NPJ Digit Med. Feb 21, 2019;2(1):1. [FREE Full text] [CrossRef] [Medline]
    4. Reul S, Lohmann H, Wiendl H, Duning T, Johnen A. Can cognitive assessment really discriminate early stages of Alzheimer's and behavioural variant frontotemporal dementia at initial clinical presentation? Alzheimers Res Ther. Aug 09, 2017;9(1):61. [FREE Full text] [CrossRef] [Medline]
    5. Jutten RJ, Sikkes SA, Amariglio RE, Buckley RF, Properzi MJ, Marshall GA, et al. Identifying sensitive measures of cognitive decline at different clinical stages of Alzheimer’s disease. J Int Neuropsychol Soc. Oct 13, 2020;27(5):426-438. [CrossRef]
    6. Hampel H, Blennow K. CSF tau and β-amyloid as biomarkers for mild cognitive impairment. Dialogues in Clinical Neuroscience. Apr 01, 2022;6(4):379-390. [CrossRef]
    7. Lee JC, Kim SJ, Hong S, Kim Y. Diagnosis of Alzheimer's disease utilizing amyloid and tau as fluid biomarkers. Exp Mol Med. May 09, 2019;51(5):1-10. [FREE Full text] [CrossRef] [Medline]
    8. Leocadi M, Canu E, Calderaro D, Corbetta D, Filippi M, Agosta F. An update on magnetic resonance imaging markers in AD. Ther Adv Neurol Disord. Sep 04, 2020;13:1756286420947986. [FREE Full text] [CrossRef] [Medline]
    9. Schilling LP, Zimmer ER, Shin M, Leuzy A, Pascoal TA, Benedet AL, et al. Imaging Alzheimer's disease pathophysiology with PET. Dement. neuropsychol. Jun 2016;10(2):79-90. [CrossRef]
    10. Zolochevska O, Bjorklund N, Woltjer R, Wiktorowicz JE, Taglialatela G. Postsynaptic proteome of non-demented individuals with Alzheimer’s disease neuropathology. JAD. Aug 21, 2018;65(2):659-682. [CrossRef]
    11. da Silva RAPC. Sleep disturbances and mild cognitive impairment: A review. Sleep Sci. Jan 2015;8(1):36-41. [FREE Full text] [CrossRef] [Medline]
    12. Bubu OM, Brannick M, Mortimer J, Umasabor-Bubu O, Sebastião YV, Wen Y, et al. Sleep, cognitive impairment, and Alzheimer's disease: a systematic review and meta-analysis. Sleep. Jan 01, 2017;40(1):1. [CrossRef] [Medline]
    13. Sprecher KE, Bendlin BB, Racine AM, Okonkwo OC, Christian BT, Koscik RL, et al. Amyloid burden is associated with self-reported sleep in nondemented late middle-aged adults. Neurobiol Aging. Sep 2015;36(9):2568-2576. [FREE Full text] [CrossRef] [Medline]
    14. Ooms S, Overeem S, Besse K, Rikkert MO, Verbeek M, Claassen JAHR. Effect of 1 night of total sleep deprivation on cerebrospinal fluid β-amyloid 42 in healthy middle-aged men: a randomized clinical trial. JAMA Neurol. Aug 01, 2014;71(8):971-977. [CrossRef] [Medline]
    15. Holth JK, Fritschi SK, Wang C, Pedersen NP, Cirrito JR, Mahan TE, et al. The sleep-wake cycle regulates brain interstitial fluid tau in mice and CSF tau in humans. Science. Feb 22, 2019;363(6429):880-884. [FREE Full text] [CrossRef] [Medline]
    16. Lucey BP, McCullough A, Landsness EC, Toedebusch CD, McLeland JS, Zaza AM, et al. Reduced non-rapid eye movement sleep is associated with tau pathology in early Alzheimer's disease. Sci Transl Med. Jan 09, 2019;11(474):1. [FREE Full text] [CrossRef] [Medline]
    17. Kong SDX, Hoyos CM, Phillips CL, McKinnon AC, Lin P, Duffy SL, et al. Altered heart rate variability during sleep in mild cognitive impairment. Sleep. Apr 09, 2021;44(4):1. [CrossRef] [Medline]
    18. Ahlskog JE, Geda YE, Graff-Radford NR, Petersen RC. Physical exercise as a preventive or disease-modifying treatment of dementia and brain aging. Mayo Clin Proc. Sep 2011;86(9):876-884. [FREE Full text] [CrossRef] [Medline]
    19. Sink KM, Espeland MA, Castro CM, Church T, Cohen R, Dodson JA, et al. LIFE Study Investigators. Effect of a 24-month physical activity intervention vs health education on cognitive outcomes in sedentary older adults: the LIFE randomized trial. JAMA. Aug 25, 2015;314(8):781-790. [FREE Full text] [CrossRef] [Medline]
    20. Prakash RS, Voss MW, Erickson KI, Kramer AF. Physical activity and cognitive vitality. Annu Rev Psychol. Jan 03, 2015;66(1):769-797. [CrossRef] [Medline]
    21. Krell-Roesch J, Pink A, Roberts RO, Stokin GB, Mielke MM, Spangehl KA, et al. Timing of physical activity, apolipoprotein E ε4 genotype, and risk of incident mild cognitive impairment. J Am Geriatr Soc. Dec 2016;64(12):2479-2486. [FREE Full text] [CrossRef] [Medline]
    22. Reiter K, Nielson KA, Smith TJ, Weiss LR, Alfini AJ, Smith JC. Improved cardiorespiratory fitness is associated with increased cortical thickness in mild cognitive impairment. J Int Neuropsychol Soc. Nov 19, 2015;21(10):757-767. [CrossRef]
    23. López-Ortiz S, Lista S, Valenzuela PL, Pinto-Fraga J, Carmona R, Caraci F, et al. Effects of physical activity and exercise interventions on Alzheimer's disease: an umbrella review of existing meta-analyses. J Neurol. Feb 07, 2023;270(2):711-725. [CrossRef] [Medline]
    24. Brachem C, Winkler A, Tebrügge S, Weimar C, Erbel R, Jöckel KH, et al. Associations between self-reported sleep characteristics and incident mild cognitive impairment: The Heinz Nixdorf Recall Cohort Study. Sci Rep. Apr 16, 2020;10(1):6542. [FREE Full text] [CrossRef] [Medline]
    25. Lim ASP, Kowgier M, Yu L, Buchman AS, Bennett DA. Sleep fragmentation and the risk of incident Alzheimer's disease and cognitive decline in older persons. Sleep. Jul 01, 2013;36(7):1027-1032. [FREE Full text] [CrossRef] [Medline]
    26. Ebrahimi K, Majdi A, Baghaiee B, Hosseini S, Sadigh-Eteghad S. Physical activity and beta-amyloid pathology in Alzheimer's disease: A sound mind in a sound body. EXCLI J. 2017;16:959-972. [FREE Full text] [CrossRef] [Medline]
    27. Jackson CL, Patel SR, Jackson WB, Lutsey PL, Redline S. Agreement between self-reported and objectively measured sleep duration among white, black, Hispanic, and Chinese adults in the United States: multi-ethnic study of atherosclerosis. Sleep. Jun 01, 2018;41(6):1. [FREE Full text] [CrossRef] [Medline]
    28. Herbolsheimer F, Riepe MW, Peter R. Cognitive function and the agreement between self-reported and accelerometer-accessed physical activity. BMC Geriatr. Feb 21, 2018;18(1):56. [FREE Full text] [CrossRef] [Medline]
    29. Sabia S, Fayosse A, Dumurgier J, van Hees VT, Paquet C, Sommerlad A, et al. Association of sleep duration in middle and old age with incidence of dementia. Nat Commun. Apr 20, 2021;12(1):2289. [FREE Full text] [CrossRef] [Medline]
    30. Benedict C, Byberg L, Cedernaes J, Hogenkamp PS, Giedratis V, Kilander L, et al. Self-reported sleep disturbance is associated with Alzheimer's disease risk in men. Alzheimers Dement. Sep 27, 2015;11(9):1090-1097. [FREE Full text] [CrossRef] [Medline]
    31. Kim H, Kim D, Oh J. Automation of classification of sleep stages and estimation of sleep efficiency using actigraphy. Front Public Health. Jan 9, 2022;10:1092222. [FREE Full text] [CrossRef] [Medline]
    32. Dietz-Terjung S, Martin AR, Finnsson E, Ágústsson JS, Helgason S, Helgadóttir H, et al. Proof of principle study: diagnostic accuracy of a novel algorithm for the estimation of sleep stages and disease severity in patients with sleep-disordered breathing based on actigraphy and respiratory inductance plethysmography. Sleep Breath. Dec 16, 2021;25(4):1945-1952. [FREE Full text] [CrossRef] [Medline]
    33. Hamill K, Jumabhoy R, Kahawage P, de Zambotti M, Walters EM, Drummond SPA. Validity, potential clinical utility and comparison of a consumer activity tracker and a research-grade activity tracker in insomnia disorder II: Outside the laboratory. J Sleep Res. Feb 03, 2020;29(1):e12944. [CrossRef] [Medline]
    34. Chen R, Jankovic F, Marinsek N, Foschini L, Kourtis L, Signorini A, et al. Developing measures of cognitive impairment in the real world from consumer-grade multimodal sensor streams. KDD '19: Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining. 2019:2145-2155. [CrossRef]
    35. Smith T, Gildeh N, Holmes C. The Montreal Cognitive Assessment: validity and utility in a memory clinic setting. Can J Psychiatry. May 24, 2007;52(5):329-332. [CrossRef] [Medline]
    36. Cockrell JR, Folstein MF. Mini-mental state examination. In: Copeland RM, Abou-Saleh MT, Blazer DG, editors. Principles and Practice of Geriatric Psychiatry, Second Edition. Hoboken, NJ. John Wiley & Sons Ltd; 2002:140-141.
    37. Jacova C, Kertesz A, Blair M, Fisk JD, Feldman HH. Neuropsychological testing and assessment for dementia. Alzheimers Dement. Oct 2007;3(4):299-317. [CrossRef] [Medline]
    38. Lombardi G, Crescioli G, Cavedo E, Lucenteforte E, Casazza G, Bellatorre A, et al. Structural magnetic resonance imaging for the early diagnosis of dementia due to Alzheimer's disease in people with mild cognitive impairment. Cochrane Database Syst Rev. Mar 02, 2020;3(3):CD009628. [FREE Full text] [CrossRef] [Medline]
    39. Tripathi S, Arroyo-Gallego T, Giancardo L. Keystroke-dynamics for Parkinson's disease signs detection in an at-home uncontrolled population: a new benchmark and method. IEEE Trans Biomed Eng. Jan 2023;70(1):182-192. [FREE Full text] [CrossRef] [Medline]
    40. van Maurik IS, van der Kall LM, de Wilde A, Bouwman FH, Scheltens P, van Berckel BN, et al. Added value of amyloid PET in individualized risk predictions for MCI patients. Alzheimers Dement (Amst). Dec 29, 2019;11(1):529-537. [FREE Full text] [CrossRef] [Medline]
    41. Master H, Annis J, Huang S, Beckman JA, Ratsimbazafy F, Marginean K, et al. Association of step counts over time with the risk of chronic disease in the All of Us Research Program. Nat Med. Nov 10, 2022;28(11):2301-2308. [FREE Full text] [CrossRef] [Medline]
    42. Forte G, Favieri F, Casagrande M. Heart rate variability and cognitive function: a systematic review. Front Neurosci. Jul 9, 2019;13:710. [FREE Full text] [CrossRef] [Medline]
    43. Liu KY, Elliott T, Knowles M, Howard R. Heart rate variability in relation to cognition and behavior in neurodegenerative diseases: A systematic review and meta-analysis. Ageing Res Rev. Jan 2022;73:101539. [FREE Full text] [CrossRef] [Medline]
    44. Schaich CL, Malaver D, Chen H, Shaltout HA, Zeki Al Hazzouri A, Herrington DM, et al. Association of heart rate variability with cognitive performance: the multi‐ethnic study of atherosclerosis. JAHA. Apr 09, 2020;9(7):1. [CrossRef]
    45. Luft CDB, Takase E, Darby D. Heart rate variability and cognitive function: effects of physical effort. Biol Psychol. Oct 2009;82(2):164-168. [CrossRef] [Medline]
    46. Luque-Casado A, Zabala M, Morales E, Mateo-March M, Sanabria D. Cognitive performance and heart rate variability: the influence of fitness level. PLoS One. Feb 20, 2013;8(2):e56935. [FREE Full text] [CrossRef] [Medline]
    47. Wang S, Fashanu OE, Zhao D, Guallar E, Gottesman RF, Schneider AL, et al. Relation of elevated resting heart rate in mid-life to cognitive decline over 20 years (from the Atherosclerosis Risk in Communities [ARIC] study). Am J Cardiol. Jan 15, 2019;123(2):334-340. [FREE Full text] [CrossRef] [Medline]
    48. Kim H, Jung YH, Han HJ. Resting heart rate and cognitive decline: a meta-analysis of prospective cohort studies. J Clin Neurol. Nov 2022;18(6):619-627. [FREE Full text] [CrossRef] [Medline]
    49. Imahori Y, Vetrano DL, Xia X, Grande G, Ljungman P, Fratiglioni L, et al. Association of resting heart rate with cognitive decline and dementia in older adults: A population-based cohort study. Alzheimers Dement. Oct 03, 2022;18(10):1779-1787. [CrossRef] [Medline]
    50. Yoon M, Yang P, Jin M, Yu HT, Kim T, Jang E, et al. Association of physical activity level with risk of dementia in a nationwide cohort in Korea. JAMA Netw Open. Dec 01, 2021;4(12):e2138526. [FREE Full text] [CrossRef] [Medline]
    51. Suzuki K, Niimura H, Kida H, Eguchi Y, Kitashima C, Takayama M, et al. Increasing light physical activity helps to maintain cognitive function among the community-dwelling oldest old population: a cross-sectional study using actigraph from the Arakawa 85+ study. Geriatr Gerontol Int. Aug 22, 2020;20(8):773-778. [CrossRef] [Medline]
    52. Hsiao C, Wen C, Yen H, Hsueh M, Liao Y. Association between accelerometer-measured light-intensity physical activity and cognitive function in older adults. J Nutr Health Aging. Mar 2022;26(3):230-235. [FREE Full text] [CrossRef] [Medline]
    53. Spartano NL, Demissie S, Himali JJ, Dukes KA, Murabito JM, Vasan RS, et al. Accelerometer-determined physical activity and cognitive function in middle-aged and older adults from two generations of the Framingham Heart Study. Alzheimers Dement (N Y). Jan 2019;5(1):618-626. [FREE Full text] [CrossRef] [Medline]
    54. Kawasaki Y, Kasai T, Sakurama Y, Sekiguchi A, Kitamura E, Midorikawa I, et al. Evaluation of sleep parameters and sleep staging (slow wave sleep) in athletes by Fitbit Alta HR, a consumer sleep tracking device. NSS. Apr 2022;Volume 14:819-827. [CrossRef]
    55. Pase MP, Harrison S, Misialek JR, Kline CE, Cavuoto M, Baril A, et al. Sleep architecture, obstructive sleep apnea, and cognitive function in adults. JAMA Netw Open. Jul 03, 2023;6(7):e2325152. [FREE Full text] [CrossRef] [Medline]
    56. Djonlagic IE, Guo M, Igue M, Kishore D, Stickgold R, Malhotra A. Continuous positive airway pressure restores declarative memory deficit in obstructive sleep apnea. Am J Respir Crit Care Med. May 01, 2021;203(9):1188-1190. [CrossRef]
    57. Álvarez-Bueno C, Mesas AE, Reina-Gutierrez S, Saz-Lara A, Jimenez-Lopez E, Martinez-Vizcaino V. Napping and cognitive decline: a systematic review and meta-analysis of observational studies. BMC Geriatr. Sep 15, 2022;22(1):756. [FREE Full text] [CrossRef] [Medline]
    58. Paz V, Dashti HS, Garfield V. Is there an association between daytime napping, cognitive function, and brain volume? A Mendelian randomization study in the UK Biobank. Sleep Health. Oct 2023;9(5):786-793. [FREE Full text] [CrossRef] [Medline]
    59. Cai H, Su N, Li W, Li X, Xiao S, Sun L. Relationship between afternoon napping and cognitive function in the ageing Chinese population. Gen Psychiatr. Jan 25, 2021;34(1):e100361. [FREE Full text] [CrossRef] [Medline]
    60. Basta M, Koutentaki E, Vgontzas A, Zaganas I, Vogiatzi E, Gouna G, et al. Objective daytime napping is associated with disease severity and inflammation in patients with mild to moderate dementia. JAD. Apr 07, 2020;74(3):803-815. [CrossRef]
    61. Keage HAD, Banks S, Yang KL, Morgan K, Brayne C, Matthews FE. What sleep characteristics predict cognitive decline in the elderly? Sleep Med. Aug 2012;13(7):886-892. [CrossRef] [Medline]
    62. Wichniak A, Wierzbicka A, Walęcka M, Jernajczyk W. Effects of antidepressants on sleep. Curr Psychiatry Rep. Aug 09, 2017;19(9):63. [FREE Full text] [CrossRef] [Medline]
    63. Kluge M, Schüssler P, Steiger A. Duloxetine increases stage 3 sleep and suppresses rapid eye movement (REM) sleep in patients with major depression. Eur Neuropsychopharmacol. Jul 2007;17(8):527-531. [CrossRef] [Medline]
    64. Lo H, Yang C, Lo H, Lee C, Ting H, Tzang BS. Treatment effects of gabapentin for primary insomnia. Clin Neuropharmacol. 2010;33(2):84-90. [CrossRef] [Medline]
    65. Foldvary-Schaefer N, De Leon Sanchez I, Karafa M, Mascha E, Dinner D, Morris H. Gabapentin increases slow-wave sleep in normal adults. Epilepsia. Dec 11, 2002;43(12):1493-1497. [FREE Full text] [CrossRef] [Medline]
    66. Di Giacopo R, Fasano A, Quaranta D, Della Marca G, Bove F, Bentivoglio AR. Rivastigmine as alternative treatment for refractory REM behavior disorder in Parkinson's disease. Mov Disord. Apr 30, 2012;27(4):559-561. [CrossRef] [Medline]
    67. Moraes W, Poyares D, Guilleminault C, Ramos L, Bertolucci P, Tufik S. The effect of donepezil on sleep and REM sleep EEG in patients with Alzheimer disease: a double-blind placebo-controlled study. Sleep. Feb 2006;29(2):199-205. [CrossRef] [Medline]
    68. Hsieh C, Tseng P, Chen T, Lin P, Chen Y, Bo-Lin Ho, et al. The association of changes of sleep architecture related to donepezil: A systematic review and meta-analysis. J Formos Med Assoc. Aug 2022;121(8):1466-1477. [FREE Full text] [CrossRef] [Medline]
    69. Jindal RD, Friedman ES, Berman SR, Fasiczka AL, Howland RH, Thase ME. Effects of sertraline on sleep architecture in patients with depression. J Clin Psychopharmacol. Dec 2003;23(6):540-548. [CrossRef] [Medline]
    70. Kátai Z, Adori C, Kitka T, Vas S, Kalmár L, Kostyalik D, et al. Acute escitalopram treatment inhibits REM sleep rebound and activation of MCH-expressing neurons in the lateral hypothalamus after long term selective REM sleep deprivation. Psychopharmacology (Berl). Aug 21, 2013;228(3):439-449. [CrossRef] [Medline]
    71. Kostyalik D, Vas S, Kátai Z, Kitka T, Gyertyán I, Bagdy G, et al. Chronic escitalopram treatment attenuated the accelerated rapid eye movement sleep transitions after selective rapid eye movement sleep deprivation: a model-based analysis using Markov chains. BMC Neurosci. Nov 19, 2014;15(1):120. [FREE Full text] [CrossRef] [Medline]
    72. Kunz D, Mahlberg R, Müller C, Tilmann A, Bes F. Melatonin in patients with reduced REM sleep duration: two randomized controlled trials. J Clin Endocrinol Metab. Jan 2004;89(1):128-134. [CrossRef] [Medline]
    73. Schredl M, Weber B, Braus D, Gattaz WF, Berger M, Riemann D, et al. The effect of rivastigmine on sleep in elderly healthy subjects. Exp Gerontol. Mar 2000;35(2):243-249. [CrossRef] [Medline]
    74. Xu Q. QidiXu96 / Fitbit_MCI_Study. GitHub. URL: https://github.com/QidiXu96/Fitbit_MCI_Study [accessed 2024-06-29]

    ‎
    AD: Alzheimer disease
    AUC: area under the curve
    EEG: electroencephalogram
    HRV: heart rate variability
    MCI: mild cognitive impairment
    MRI: magnetic resonance imaging
    NDC: Neurosciences Neurocognitive Disorders Center
    NREM: nonrapid eye movement
    PET: positron emission tomography
    REM: rapid eye movement
    SHAP: Shapley Additive Explanations
    SVM: Support Vector Machines
    UTHealth: University of Texas Health Science Center at Houston

    Edited by A Mavragani; submitted 17.12.23; peer-reviewed by C Gordon, S Duffy; comments to author 25.01.24; revised version received 15.02.24; accepted 08.06.24; published 18.07.24.

    Copyright

    ©Qidi Xu, Yejin Kim, Karen Chung, Paul Schulz, Assaf Gottlieb. Originally published in JMIR Formative Research (https://formative.jmir.org), 18.07.2024.

    This is an open-access article distributed under the terms of the Creative Commons Attribution License (https://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work, first published in JMIR Formative Research, is properly cited. The complete bibliographic information, a link to the original publication on https://formative.jmir.org, as well as this copyright and license information must be included.