Abstract
Background:
Gradual disengagement from daily physical activity (PA) could signal present or emerging mild cognitive impairment (MCI) or Alzheimer’s disease (AD).
Objective:
This study examined whether accelerometry-derived patterns of everyday movement differ by cognitive diagnosis in participants of the Baltimore Longitudinal Study of Aging (BLSA).
Methods:
Activity patterns, overall and by time-of-day, were cross-sectionally compared between participants with adjudicated normal cognition (n = 549) and MCI/AD diagnoses (n = 36; 5 participants [14%] living with AD) using covariate-adjusted regression models.
Results:
Compared to those with normal cognition, those with MCI/AD had 2.1% higher activity fragmentation (SE = 1.0%, p = 0.036) but similar mean total activity counts/day (p = 0.075) and minutes/day spent active (p = 0.174). Time-of-day analyses show MCI/AD participants had lower activity counts and minutes spent active during waking hours (6:00 am–5:59 pm; p < 0.01 for all). Also, they had lower activity fragmentation from 12:00–5:59 am (p < 0.001), but higher fragmentation from 12:00–5:59 pm (p = 0.026).
Conclusion:
Differences in the timing and patterns of physical activity throughout the day linked to MCI/AD diagnoses warrant further investigation into potential clinical utility.
INTRODUCTION
Alzheimer’s disease (AD) dementia is a neurodegenerative disorder associated with cognitive impairment and loss of independence in daily functioning [1]. Difficulty performing everyday activities and consequent reduced participation may precede diagnoses of mild cognitive impairment (MCI) and AD [2]. To this end, measuring detailed daily activity patterns sequentially over time may hold important potential to enhance detection of MCI and AD.
Physical activity has long been considered a positive health behavior conferring reduced risk of multiple chronic diseases and debilitating health outcomes. The advent and proliferation of research and commercial accelerometers that provide continuous monitoring of daily movement/activity offers the opportunity to examine patterns of activity as indicators of emerging disease pathology [3]. Initial evidence using a hip-worn accelerometry protocol that best captures ambulation [4] suggests that daily higher-intensity activity patterns degrade among persons with mild AD compared to normal controls. Placement of accelerometers in large scale research studies has begun to move from the hip to the wrist [5, 6], which might be more sensitive for capturing upper extremity movement [7, 8] that may better reflect upper body movement characteristics of activities common to daily living [9, 10]. Yet, wrist-worn daily physical activity patterns between normal cognition, MCI, and AD diagnoses have not been well characterized. Additionally, fragmentation or “breaking up” of daily physical activity (Fig. 1), as a proxy of compensation due to physiological decline [11] and a marker of adverse health [11–17], has not been investigated among persons with MCI or AD. Further, accelerometry offers the opportunity to examine these associations by time of day, an important exploration since those who experience greater cognitive declines have been shown to be more fatigable [18]—a state related to waning physical activity engagement later in the day [19].
Physical activity fragmentation concept. Each green segment represents 1 minute of activity. Each black segment represents 1 sedentary minute. The top row shows an individual performing a 5-minute activity (upright individual) continuously and then resting (sitting individual). The middle and bottom row shows increasing sedentary/sitting/rest intervals, breaking up the 5 minutes of activity into shorter, more fragmented bouts. More sitting/resting intervals needed for activity completion is thus conceptualized as physical activity fragmentation.
Participant flow. APOE, apolipoprotein; CES-D, Center for Epidemiologic Studies-Depression; BMI, body mass index.
Thus, this study aims to identify and characterize objectively measured physical activity patterns between older adults with and without MCI/AD using wrist-worn accelerometry over a 24-hour day. We hypothesized that older adults with MCI/AD would have lower overall physical activity and more frequent rest periods identified as activity fragmentation [11, 15] throughout the day, compared to their counterparts with normal cognition.
METHODS
Study design and participants
This cross-sectional study utilized data from the BLSA, an ongoing enrollment, population-based study of human aging that began in 1958. This study is conducted by the National Institute on Aging’s (NIA) Intramural Research Program. Enrollment and study design details of the BLSA are published elsewhere [20]. Briefly, recruitment criteria include being ≥20 years old and having no cognitive impairment, functional limitations, or major chronic diseases (except hypertension or cancer within the past 10 years). BLSA participants are followed for life once enrolled. Comprehensive health, cognitive and functional assessments are conducted during a 3-day stay at the NIA’s Clinical Research Unit located at Harbor Hospital in Baltimore, Maryland. Depending on the participant’s age, visits are regularly scheduled every 1–4 years. All assessments are administered following standardized protocols by trained and certified staff. Written informed consent is provided by each participant. The study protocol was approved by the Internal Review Board of the Intramural Research Program of the National Institutes of Health.
Between July 2015 –December 2019, 758 participants had measured objective physical activity and cognitive diagnoses assessments (Fig. 2). Exclusions included < 3 valid days of accelerometer wear-time (n = 33), age < 50 years (n = 79), having non-AD dementia (n = 5), and missing covariate data (n = 56). The final analytic sample was 585 participants.
Diagnosis of mild cognitive impairment and Alzheimer’s disease
Clinical and neuropsychological data from BLSA participants were reviewed at a consensus case conference for participants who either: 1) scored ≥4 on the Blessed Information Memory Concentration test [21]; 2) scored ≥0.5 on the Clinical Dementia Rating test through subject or informant reporting [22]; 3) had concerns raised about their cognitive status during the study visit; or 4) died or withdrew from the study. MCI was determined using Petersen criteria [23]. MCI diagnosis was defined either when cognitive impairment was evident for a single domain (typically memory) or cognitive impairment in multiple domains occurred without significant functional loss in activities of daily living. Dementia diagnosis was based on the Diagnosis and Statistical Manual of Mental Disorders (3rd edition, revised) [24]. AD diagnosis was based on the National Institute of Neurological and Communication Disorders and Stroke-Alzheimer’s Disease and Related Disorders Association criteria [25]. Estimated age of MCI/AD onset was determined from the case conference.
Physical activity measures
Physical activity was measured with an Actigraph GT9X monitor (Actigraph, Pensacola, FL). This device contains a tri-axial accelerometer sensor and was set to collect movement in units of gravity (g) at a sampling rate of 80 Hz. During the last day of their BLSA clinic visit, participants were fitted with the device on the non-dominant wrist and instructed to wear the Actigraph monitor for 7 consecutive days, 24 hours per day. Participants returned the monitor to the Clinical Research Unit via express mail after the collection period. Data were downloaded and pre-processed into 1-minute epoch level vector magnitude activity counts (unitless quantities of movement) using ActiLife Software (version 6.13.4). Vector magnitude activity counts were derived as the root mean squared of the counts collected on three axes.
Based on Choi and colleagues’ algorithm, non-wear data were determined for each minute register-ing zero activity counts, using a ≥90 minute threshold of consecutively occurring zero-count minutes [26]. A valid wear day was defined as < 10% non-wear data [15]. Only participants with ≥3 valid days were included in this analysis [15].
Three accelerometer metrics were extracted. The first was mean total activity counts per day, calculated by summing the total amount of activity counts across all valid minutes and dividing by the number of valid days. The second was active minutes per day calculated by summing each minute that registered ≥1,853 counts [27] and dividing by the number of valid days. The third variable was activity fragmentation, operationalized through the active-to-sedentary transition probability (ASTP) [11]. To calculate ASTP, the number of active bouts (defined as contiguous minutes registering ≥1,853 counts) was divided by total active minutes, in other words, the reciprocal of the mean active bout length. A higher ASTP value (presented as a percentage) represents higher activity fragmentation (shorter active bouts normalized to total active time). For reference, an earlier publication showed 301 cognitively normal BLSA participants had an average fragmentation of 25% (SD = 5, ranging 18–47%), translating to a 25% probability of transitioning from an active state to a sedentary state [15]. Accelerometry data processing was performed in R (ver. 4.0.2) with the ARCTOOLS package [28].
Covariates
The relationship between cognitive diagnosis and physical activity patterns might vary by confounders that include sociodemographic (age, sex, race, and education) [29–31], body mass index [32, 33], gait speed [34, 35], medical conditions [36, 37], and apolipoprotein E (APOE) ɛ4 carrier status [38–40]. Self-reported age, sex, self-identified race, and years of education were collected via questionnaire by staff. Measured weight (kg) and height (m) were used to calculate body mass index (BMI; kg/m2). Usual gait speed (m/s) was measured over a 6-m course, with the faster of two trials used for analysis. Participants were asked if they were ever told by a doctor or other health professional that they had any of the following conditions: cardiovascular disease including angina, myocardial infarction, congestive heart failure, peripheral arterial disease, and vascular-related procedures; lung disease including chronic bronchitis, emphysema, chronic obstructive pulmonary disease, or asthma; hypertension or high blood pressure; high cholesterol or triglycerides; diabetes, glucose intolerance, or high blood sugar; cancer, malignant growth, or malignant tumor; arthritis or osteoarthritis; kidney disease, nephritis, or renal insufficiency; liver disease including cirrhosis and hepatitis; stroke and transient ischemic attack; Parkinson’s disease. Responses were summed to a count of morbidities. The 20-item Center for Epidemiologic Studies-Depression (CES-D) scale was used to measure depressive symptoms [41]. Scores ranged from 0–60 where a higher score represents higher depressive symptoms. APOE ɛ4 carrier status was determined as the presence of ≥1 ɛ4 allele versus 0.
Statistical analysis
Participant characteristics were descriptively examined by cognitive diagnoses. ANOVA and Fisher’s exact test were used to compare continuous and categorical covariates by cognitive diagnoses, respectively. Means and standard deviations were calculated for all accelerometer outcomes by cognitive diagnoses. Due to limited sample size, those with MCI and AD were collapsed into one category and physical activity was compared to those with normal cognition. Linear regression models were used to estimate the differences of each accelerometer outcome between those with MCI/AD versus normal cognition. Linear regression models were first adjusted for age, sex, race, and education, and then fully adjusted for BMI, APOE ɛ4 status, depressive symptoms, usual gait speed and number of morbid conditions. Interactions of cognitive diagnosis with age, sex, and usual gait speed (respectively) were also explored. Linear mixed effects models were used to estimate cognitive diagnoses differences across four sequential 6-h time periods (12:00 am–5:59 am; 6:00 am–11:59 am; 12:00 pm–5:59 pm; 6:00 pm–11:59 pm). The 6-h time period variable was treated as a nominal indicator variable. Physical activity metrics in each time interval were treated as repeated measures. The interaction between cognitive diagnoses by time interval was tested to whether the change in physical activity metric between two consecutive intervals differed between MCI/AD versus normal adults. Random intercepts and random slopes by time intervals were included in the model. Unstructured covariances were used for random effects. Mixed effects models were adjusted for age, sex, race, education, BMI, APOE ɛ4 status, depressive symptoms, usual gait speed, and number of morbid conditions. Two-tailed hypothesis testing with an alpha level = 0.05 was used to determine statistical significance. All statistical analyses were performed using Stata IC 15.1 (Stata Corporation, College Station, TX).
RESULTS
The mean age was 75 (SD = 11) years, 56% were women, and 68% were White among the total 585 participants (Table 1). Participants averaged 17.8 (2.5) years of education, 27% were APOE ɛ4 carriers, and had an average usual gait speed of 1.1 (0.2) m/s. Those with MCI/AD (n = 36) tended to be older and have slower gait speed compared to those with normal cognition (n = 549). The 5 participants with AD were slightly older (mean age of 89.3 years), had slower gait speed (mean 0.5 m/s), and had higher prevalence of cardiovascular disease history (40%) compared to those with MCI. Estimated AD onset for 3 of the 5 participants was within the same year of accelerometry collection, and 1 year before accelerometer collection for the other 2 participants.
Baseline participant characteristics for normal cognition and MCI/AD
MCI, mild cognitive impairment; AD, Alzheimer’s disease; BMI, body mass index; APOE, apolipoprotein; CES-D, Center for Epidemiologic Studies-Depression, a 20-item scale where higher values indicate more depressive symptoms.
Among the 585 participants, mean total activity counts/day were 2,105,145 (SD = 595,120), mean active minutes/day were 418 (107), and mean activity fragmentation was 25% (6%) (Table 2). Descriptively, those with MCI/AD had fewer total activity counts/day, fewer active minutes/day, and greater activity fragmentation. Participants with AD appeared to have slightly fewer activity counts/day, fewer active minutes/day, and great activity fragmentation compared to participants with MCI (Supplementary Table 1).
Physical activity characteristics for normal cognition and MCI/AD
MCI, mild cognitive impairment; AD, Alzheimer’s disease.
In linear regression models adjusting for age, sex, and race, those with MCI/AD had 209,000 fewer total activity counts/day than those with normal cognition (Table 3, Model 1). This difference was not statistically significant after full covariate adjustment (Table 3, Model 2). Between MCI/AD and normal cognitive diagnoses, there were no statistically significant differences in active minutes/day. However, persons with MCI/AD had 2.7% higher activity fragmentation than those with normal cognition (Table 3, Model 1)—which remained statistically significant after full covariate adjustment. In fully adjusted sensitivity analyses focused only on persons with MCI, the magnitude of the association with fragmentation remained similar but statistical significance was lost (p = 0.067) (Supplementary Table 2). Separate interactions with age, sex, and usual gait speed were not significant (p > 0.08).
Linear regression models with continuous PA outcomes comparing MCI/AD (n = 36) and normal cognition (n = 549)
aModel 1: adjusted for age, sex, race, education years. bModel 2: Model 1 + BMI, APOE ɛ4 status, depressive symptoms, usual gait speed, and number of morbid conditions. Bold indicates p < 0.05; MCI, mild cognitive impairment; AD, Alzheimer’s disease; BMI, body mass index; APOE, apolipoprotein.
Differences in diurnal (24-h) patterns of activity were assessed between MCI/AD and normal cognitive diagnoses. Descriptively, the diurnal patterns of those with MCI/AD appeared to be diminished by total amount (Fig. 3), with slight delays in awakening (6:00–7:00 am) and extended activity later in the night (11:00 pm–12:00 am) (Fig. 3) in MCI/AD versus normal cognition.
Diurnal patterns of physical activity by diagnosis status. MCI, mild cognitive impairment; AD, Alzheimer’s disease.
Using linear mixed effects models, time-of-day differences in each physical activity metric were tested by cognitive diagnosis (Table 4). During the overnight hours (12:00 am–5:59 am), mean total activity counts and active minutes did not differ between MCI/AD and normal cognition statuses. However, those with MCI/AD had 9.6% lower (95% Confidence Interval [CI]: –13.9, –5.2) activity fragmentation compared to those with normal cognition. During the morning hours (6:00 am–11:59 am), persons with MCI/AD had 116,000 fewer (95% CI: –178,000, –55,000) mean activity counts and 16 fewer (95% CI: –27, –4) active minutes, respectively, than those with normal cognition. Activity fragmentation was similar between diagnoses. During the afternoon (12:00 pm–5:59 pm), those with MCI/AD had 127,000 fewer (95% CI: –195,000, –59,000) mean activity counts, 21 fewer (95% CI: –34, –8) active minutes, and 3.4% higher (95% CI: 0.4, 6.4) activity fragmentation respectively, than those with normal cognition. During the evening hours (6:00–11:59 pm), there were no differences in any of the three accelerometer metrics.
Estimated physical activity pattern differences by normal cognition (n = 549) and MCI/AD (n = 36) and by 6-hour time of day intervals using linear mixed effects modeling
Models adjusted for age, sex, race, education years, BMI, APOE ɛ4 status, depressive symptoms, usual gait speed, and number of morbid conditions. MCI, mild cognitive impairment; AD, Alzheimer’s disease; BMI, body mass index; APOE, apolipoprotein. *p < 0.05; **p < 0.01; ***p < 0.001.
This study compared physical activity patterns between older adults with and without an adjudicated MCI/AD diagnosis in their free-living, community settings. After adjusting for a multitude of sociodemographic, behavioral, and medical factors, participants with an MCI/AD diagnosis had similar overall amounts of physical activity engagement but accumulated their daily physical activity in a more fragmented manner compared to those with normal cognition. In time-of-day analyses, there were little physical activity patterning differences in the morning between those with MCI/AD diagnosis and those with normal cognition. Yet, those with MCI/AD had earlier afternoon declines in physical activity and more movement during late evening hours. Collectively, these results suggest small but potentially important differences in patterns of daily physical activity that may be informative when assessing progression of cognitive change.
This study did not find overall differences in physical activity between MCI/AD and normal cognition diagnoses. Yet, the magnitude of the differences support published evidence that shows lower accelerometer-measured physical activity associated with higher MCI/AD risk [4, 42]. Review studies suggest interventions that increase physical activity provide benefit towards preventing cognitive decline that leads to MCI/AD, though the strength of the evidence ranges widely [43–45]. These studies suggest that physical activity is a modifiable protective factor against declines into MCI/AD. However, the possibility that declines in cognition partially drive physical activity declines, should be considered. Observed deficits in instrumental activities of daily living have been observed in persons with MCI [46] that are progressively worse in those with AD [47], suggesting a related gradual disengagement or inability to perform more complex daily activities, even while maintaining the ability to perform basic activities of daily living. This is further supported by the burgeoning evidence suggesting poor adherence to physical activity interventions may be due to altered brain structure and function [48]—that reduces cognitive function and, in turn, the ability to plan, perform and adhere to complex physical activity behaviors [49].
Disengagement of complex movement in everyday activity might partly be captured within time-of-day differences measured by wrist accelerometry. Those with MCI/AD tended to have lower daily physical activity from morning to late evening but higher levels of physical activity later in the evening and overnight hours. This redistribution of normal to abnormal diurnal patterning supports published literature that suggest the degradation of circadian rest/activity rhythms might be symptoms of or contributors to MCI and AD [50–52]. Potential reasons for this redistribution of diurnal patterning include increased fatigue [53], daytime sleepiness that results in napping [54, 55], and insomnia [56]. Together, these results highlight the utility of time-of-day activity differences (as well as the limitations of undifferentiated 24-h monitoring) to potentially detect early signs of MCI/AD. Still, the bidirectionality of this relationship warrants further exploration as some studies observed abnormalities in rest-activity cycles as potential risk factors for MCI/AD [57, 58].
A novel finding of this study was the association between physical activity fragmentation and MCI/AD. Physical activity fragmentation is conceptualized as a behavioral compensatory mechanism to preserve energy [59] and maintain inde-pendent living in the face of functional decline and mobility loss that occurs with aging [60]. Our results suggest that activity fragmentation might be a more effective sign of MCI/AD progression than total physical activity. One reason is that activity fragmentation may capture subtle losses of motor function that occur with AD progression [61]. It is also possible that during waking periods, more fragmented activity manifests with behavioral signs of subtle cognitive deficits and other neuropsychiatric symptoms that occur with MCI/AD progression [62, 63]—factors that might contribute to the eventual replacement of routine activity with wandering, rest, and sedentariness. The finding that MCI/AD participants had lower activity fragmentation during typical sleep times (i.e., when active, more likely to remain active) supports this study’s diurnal pattern findings showing the flattening of overall diurnal curves with parallel increases in activity during normal sleep times. More longitudinal research is needed to disentangle the reasons fueling differential activity fragmentation patterns occurring with MCI/AD onset and progression.
Our findings should be interpreted within the context of the study’s limitations. The first limitation is that the sample consisted of a limited number of MCI and AD participants who were collapsed into one group to increase statistical power. The second limitation is that a higher proportion of participants with either MCI or AD with insufficient valid accelerometry collection appear to be excluded from the analysis, possibly attenuating the observed associations towards the null. The third limitation is that accelerometry does not capture context of movement behavior (e.g., standing, sitting), leading to a possible underestimation of the observed associations. The fourth limitation is that the cross-sectional study design does not account for temporality, allowing for reverse causation. The fifth limitation is that BLSA participants tend to have higher physical functioning and levels of education compared to the general older adult population, limiting generalizability. The sixth limitation is that defined intervals across the 24 h used in this study do not capture individual-level waking, napping, and sleeping times. The seventh limitation is that the study sample consists mainly of persons with MCI so results do not clearly translate to persons with AD. The strengths of the study include having a sample of participants with well-characterized MCI and adjudicated AD diagnoses, detailed physical activity patterns collected from wrist accelerometry using a 24/7 wear protocol, statistical adjustments for a large number of covariates, and reduced residual confounding due to the recruitment of older adults without initial functional decline and major disease.
Time-of-day and fragmented daily physical activity patterns were observed among older adults with MCI/AD diagnosis compared to counterparts with normal cognition. Future directions include exploring the temporal relationship these daily physical activity differences, in an effort to identify possible trajectories towards MCI/AD onset earlier in the pathological process to increase preventative and therapeutic efforts against AD. These findings also warrant further investigation into physical activity patterning among different subtypes of MCI and dementia.
Footnotes
ACKNOWLEDGMENTS
We thank all the staff and participants in the Baltimore Longitudinal Study of Aging for their important contributions. Also, we acknowledge artist Becca Reeves Tillett for the activity fragmentation concept diagram.
This research was supported by the Intramural Research Program of the National Institute on Aging of the National Institutes of Health. Data used in the analyses were obtained from the Baltimore Longitudinal Study of Aging, a study performed by the National Institute on Aging Intramural Research Program. JS and AW are supported by R01AG061786 and U01AG057545. AS is supported by U01AG057545. RD is supported by T32AG027668.
