Abstract
Objectives:
To examine the association between self-reported physical exercise and cognitive function.
Design:
Secondary data analysis of a prospective cohort study.
Participants:
Community-dwelling older adults with normal cognitive function at baseline.
Measurements:
Data on self-reported physical exercise, immediate and delayed recall, clock drawing test, and orientation to time and current events were collected at baseline and annual follow-up visits. Generalized estimation equation method was used to determine the association between physical exercise and decline in cognitive function.
Results:
A total of 4,227 participants were included in the analysis, 58% women, 78% White, mean (SD) age 75.7 (7.1) years. The risk of cognitive decline was lower among those who reported moderate-to-high intensity exercise (odds ratio [OR] confidence interval [CI] = 0.45 [0.20, 0.69]; p < .001) and low-intensity exercise (OR [CI] = 0.62 [0.44, 0.83]).
Conclusion:
These results reaffirm the positive impact of physical exercise in maintaining cognitive function among older adults.
Previous studies have reported a high prevalence of major neurocognitive disorder among older adults (Matthews et al., 2019) and with the projected increase in this segment of the population (Bureau, 2017), this poses a significant public health burden. Both risk factors and protective factors for cognitive impairment in old age have been previously described. For example, prevalence of cognitive impairment has been found to be higher among older adults who report non-White race (Matthews et al., 2019; Weuve et al., 2018) and in individuals in low socioeconomic strata (Matthews et al., 2019). Previous studies have also suggested possible protective factors that include healthy dietary habits (Gardener & Rainey-Smith, 2018) and intellectual, social, and physical activities (Anaturk et al., 2018; de Souto Barreto et al., 2016; Gomez-Pinilla & Hillman, 2013). Of these protective factors, the association between physical activity and cognitive function is the most studied and some of these studies are mentioned below.
In both animal and human studies, physical exercise (PE) has been shown to increase the levels of neurotransmitters such as acetylcholine in the brain, and these neurotransmitters mediate the release of brain-derived neurotrophic factor (BDNF) which has been shown to improve brain plasticity and blood flow (Liu & Nusslock, 2018; Mandolesi et al., 2018). Increase in neurons and glial cell proliferation rate in the hippocampus and neocortex (Steiner et al., 2004), as well as increase in brain volume (Spartano, Davis-Plourde, et al., 2019) has also been reported in association with PE. Furthermore, PE has been shown to have a positive impact on the optimal management chronic medical conditions which are associated with cognitive function, such as hypertension and diabetes mellitus (Leritz et al., 2011; Pattyn et al., 2013).
The relationship between physical activity and cognitive function has been examined in clinical and epidemiological studies as well. However, results of these studies have not been consistent. For example, studies that used direct method to measure physical activity such as accelerometer have reported a positive association between moderate-to-vigorous physical activity (MVPA) and higher cognitive function (Spartano, Davis-Plourde, et al., 2019; Zhu et al., 2017), as well as lack of association between MVPA and cognitive function (Vasquez et al., 2017). PE in the form of Tai Chi that lasted 12 weeks has been shown to have a positive impact on cognition (Zhou et al., 2019); however, a systematic review of clinical trials that included PE such as Tai Chi, aerobic exercise, and resistance training and lasted for 6 months or more did not show significant association between PE and cognitive function (Brasure et al., 2018). Another systematic review has suggested that PE done during a person’s adult life may have beneficial effect on cognitive function in old age (Engeroff et al., 2018). In contrast, a controlled trial in which 1,635 older adults at eight study centers in the United States were randomized to a supervised PE program (intervention group) or a health education program (control group) did not find significant difference in cognitive function between the two groups at the end of a 24-month follow-up period (Sink et al., 2015). However, this trial may have some shortcomings that could have contributed to these results. In the design of the trial, the control group was scheduled to have educational activity at regular intervals which included group discussion of different health-related topics and interaction of study participants with each other. These educational and social activities could potentially have a positive effect on cognition as mentioned above. Furthermore, it is not clear whether the participants in the control group engaged in PE on their own outside of the study or whether they remained sedentary. These factors could explain at least in part the absence of significant difference in cognition status between the intervention and control groups.
Similar conflicting results were reported from epidemiological studies as well. For example, a cross-sectional study have shown a positive association between self-reported physical activity and cognitive function (Landi et al., 2007). Self-reported physical activity at baseline has also been shown to be associated with higher cognitive function assessed during follow-up (Landi et al., 2007; Middleton et al., 2008). Conversely, one large epidemiological study in which over 10,000 participants were followed up for over 20 years, with both physical activity status and cognition status assessed periodically during follow-up, did not find statistically significant association between PE and dementia during follow-up (Sabia et al., 2017). In this study, dementia diagnosis was obtained from medical records and it is possible that this may have underestimated the true extent of dementia in this study. However, the authors report that physical activity of participants who were diagnosed with dementia started to decline up to 9 years before the diagnosis was made; suggesting that this period could indicate the preclinical phase of dementia and may be a factor in the decline of the exercise routines of study participants (reverse causation).
In addition, other factors may also account for these conflicting results. The design of some of the studies are cross sectional (Landi et al., 2007; Spartano, Demissie, et al., 2019), hence the results may not reflect the association between physical activity and cognitive function over time. Other studies relied on the status of physical activity at baseline to determine cognitive function during follow-up with no information on the status of physical activity status during follow-up (Landi et al., 2007; Middleton et al., 2008; Willey et al., 2014; Zhu et al., 2017). Although most of these studies have controlled for demographic characteristics, education level and comorbid conditions, two important condition that have been reported to be associated with cognitive function, namely participation in social activities (Litwin & Stoeckel, 2016; Morowatisharifabad et al., 2019) and depression symptoms (Crocco et al., 2010; Gujral et al., 2017) were not controlled for.
This study examines the association between PE and cognitive function taking into account the shortcomings of previous studies mentioned above. The study has two objectives. First, it examines the association between self-reported PE and decline in cognitive function during follow-up among community-dwelling older adults with normal cognitive function at baseline. Second, it addresses the shortcomings of previous studies mentioned above by taking the following steps. To address the issue of confounding by social activity status and depression symptoms, both variables were included as covariates in the regression analysis. To minimize the possibility of reverse causation, the following approaches were used: (a) only participants with normal cognitive function at baseline were included in the study; (b) to account for differences in baseline cognition test scores among the participants, baseline cognition test score of participants was included in the regression analysis as a covariate; and (c) furthermore, participants were categorized into tertile groups based on their baseline cognition test scores, and the association between PE status and cognitive function was determined among each of the tertile groups separately.
Method
Study Population
Data for this secondary analysis were derived from the National Health Aging Trends Study (NHATS), an epidemiological study in which a nationally representative sample of Medicare beneficiaries >65 years old participated. The NHATS design and recruitment strategies have been described previously (Kasper & Freedman, 2019) but in brief, a three-stage sampling design was used to randomly select study participants of ≥65 years of age living in contiguous regions of the United states. Participants completed a face-to-face interview and assessment at baseline and during annual follow-up visits. This analysis is based on data collected at baseline and these follow-up visits. Demographic characteristics, educational level, and medical comorbidities data collected at baseline; and participation in PE and social activities, presence of depression symptoms and assessment of cognitive function data collected both at baseline and follow-up periods were used in this analysis. The availability of data on PE, cognitive function, social activity, and depression symptoms at baseline and follow-up periods makes this data set suitable to examine the independent association between participation in PE and cognitive function over time. Because the objective of the study was to examine the association between PE and cognitive function over time among community-dwelling older adults with normal cognitive function at baseline, the following participants were excluded from the study: participants who (a) live in residential facilities, (b) were told to have dementia or Alzheimer’s disease by their health care provider at baseline, (c) participants whose cognition test showed impairment in one or more cognitive domains at baseline, and (d) participants who did not complete cognitive function test during at least one follow-up period.
Measures
Cognitive measures
Cognitive function and decline in cognitive function were used for assessment of cognitive performance.
Cognitive function
Four cognition domains, namely memory, executive function, visuo-spatial ability, and orientation were assessed at baseline and during annual follow-up visits. Cognition assessment tools used were immediate and delayed free recall of 10 words, clock drawing test, and orientation to time and current events. The validity and reliability of these assessment tools have been well established in previous studies (Ashford et al., 1989; Beeri et al., 2006; Hankee et al., 2016; Morris et al., 1989; Watson et al., 1993).
Assessment of cognitive function started with orientation to time, followed by immediate recall, clock drawing, naming of president and vice president, and delayed recall in that order.
Orientation to time was assessed using the following question: “Without looking at a calendar, newspaper or watch, please tell me today’s date”; and the year, month, date, and day of the week were recorded. One point was given for each correct answer for a total of four points.
Three lists of 10 nouns were used to assess immediate and delayed recall, and participants were randomly assigned to one of the lists at baseline. List assignment would rotate during follow-up visits so that each participant would be assigned a different list during subsequent visits (Kasper & Freedman, 2019). For immediate recall, a list of 10 nouns was read to respondents at a slow steady rate (one word per 2 s) as they appeared on a computer screen. After all the words were read out, the person was asked to recall as many words as possible in any order. The number of words recalled by the participant made up the immediate recall score (score ranging from 0 to 10).
For clock drawing test, participants were given a sheet of paper and an erasable pencil and provided with the following instructions: “Please draw a clock on this piece of paper. Start by drawing a large circle. Put all of the numbers in the circle and set the hands to show 11:10 (10 past 11).” The participants were given 2 min to complete the activity, and instructions were repeated as needed. The clock drawing test was scored out of five points.
Then participants are asked the following question: “Who is the president of the United States?” followed by “Who is the vice president of the United States?” Participants are expected to say the first and last names of the president and vice president and one point is given for each correct answer for a total of four points. The score for orientation was added to the score for naming of the president and vice president, for a total score of eight points for this domain.
This was followed by delayed free recall. Participants were asked to recall the words from the list that was read earlier. Each correct delayed word recall was given one point, with scores ranging between 0 and 10. The total score for the memory domain was obtained by adding scores for immediate recall and delayed recall.
Total cognition test score at baseline was obtained by adding individual test score results in each of the cognition domains. In previous studies, a cutoff point of 1.5 standard deviation below the population mean score of cognition test scores has been used to indicate impairment (Morris, 2012), and this criteria has been used by NHATS to determine cognitive impairment among its participants previously (Kasper et al., 2013). Based on this criteria, a score of ≤3 is considered to indicate impairment in orientation and memory tests, and a score of ≤1 considered to indicate impairment for clock drawing test. These cut-off points were adapted in this study and used to identify participants who have cognitive impairment.
Decline in cognitive function
For the purpose of this study, participants whose assessment of cognitive function during follow-up indicated impairment in two or more cognitive domains (i.e., cognition test score ≤3 for memory and orientation tests and ≤1 for clock drawing test) are considered to have decline in cognitive function.
PE measure
The following two questions were used to assess self-reported PE at baseline and during follow-up. “In the last month, did you ever go walking for exercise?” and “In the last month, did you ever spend time on vigorous activities that increased your heart rate and made your breath harder? (This includes things like working out, swimming, running or biking, or playing a sport.)” Based on responses to these questions, three PE groups were created: (a) “No PE” if participants responded “No” to both questions; (b) “low-intensity PE” if participants responded “Yes” to the first question and “No” to the second question; and (c) moderate-to-high intensity PE if participants responded “Yes” to the second question.
Demographic characteristics
Demographic variables included in the analysis are age in years, gender, race/ethnicity (“White,” “non-Hispanic,” “Black, non-Hispanic”; “Other” which included Hispanic, American Indian, Asian, Native Hawaii, and mixed race), and marital status (currently married or not married). Educational level was determined by asking participants the highest degree or level of school completed, and four categories were created: “high school not completed,” “high school completed,” “associate degree or vocational training after high school,” and “bachelor’s or higher degree.”
Chronic diseases
Data on chronic diseases, which have been reported to be risk factors for cognitive impairment such as hypertension, diabetes mellitus, heart disease and stroke were obtained at baseline using the following question: “Please tell me if a doctor ever told you that you have high blood pressure; diabetes; heart disease; heart attack; stroke.”
Social activity
Participation in organized social activity was assessed at baseline and during annual follow-up visits with the following question: “In the last month, besides religious services, did you ever participate in clubs, classes, or other organized activities?” and participants provided a “Yes” or “No” or “Don’t know” response. For the purpose of this study, participants who provided a “Don’t know” response were included in the “No” category.
Depression symptoms
Patient Health Questionnaire-2 (PHQ-2) was used to assess the presence of depression symptoms and this assessment was done at baseline and at annual follow-up visits. Participants with a score of ≥3 were considered to have depression symptoms. (Kroenke et al., 2003)
Analyses
Chi-square statistics and one-way analysis of variance were used to describe the characteristics of study participants by PE status at baseline for categorical and continuous variables, respectively.
Data were restructured into long format with individual participant indicating a cluster and visit year indicating time unit within a cluster. Generalized estimating equations (GEE) method for panel data with auto-regressive correlation structure was used to determine the association between exercise status (exposure variable) and decline in cognitive function (outcome variable). This method of regression analysis provides average estimate of the outcome for the group as a whole (population average) and addresses the correlations between repeated measures of exposure and outcome variables. Both time-invariant variables (age, gender, marital status, race/ethnicity, education level, and chronic diseases) and time-variant variables (social activity and depression symptoms) were included as covariates. Interaction terms which include PE × Race, PE × Sex, and PE × Education Level were included in the model and retained if interaction terms showed significant association.
Although only participants with normal cognitive function at baseline were included in the study, baseline cognition test score of participants who reported engaging in moderate-to-high intensity exercise was significantly higher in comparison to participants in the other groups. To control for the difference in baseline cognition scores, two approaches were taken. The first approach was to include baseline cognition score as a covariate in the regression model. In the second approach, participants were stratified into tertiles groups based on their baseline cognition test score, and separate regression analysis was performed in each of the tertile groups.
Multiple imputation by chained equations method was performed to account for missing data and 10 imputed data sets were generated. Regression analysis using generalized estimating equations method was repeated using imputed data. Stata statistical software (version 14, College Station, TX) was used and statistical significance was determined at p value ≤.05.
Results
Flow of participants is shown in Figure 1. A total of 8,245 participants completed the NHATS at baseline. Among these, the following participants were excluded from the study: (a) participants who resided in a residential facility (n = 1,048); (b) those who were told by their health care provider to have dementia or Alzheimer’s disease (n = 402); (c) participants who did not have complete data on cognition tests at baseline (n = 55); and (d) those whose cognition test result indicated impairment in one or more of the cognitive domains, leaving 5,278 participants.
Flow chart of study participants in this study.
Of the 5,278 participants with normal cognitive function, data on cognitive function during follow-up was not available for 1,051 participants and these were excluded from the study. Comparing those with and without cognitive function data during follow-up years (n = 4,227 and n = 1,051, respectively), participants who did not have data on cognitive function were more likely to be older (p = .032), non-White race (p = .025), have lower education level (p < .001), and less likely to report engaging in social activity (p < .001) and PE (p < .001). There was no significant difference in sex, marital status, depression score, and history of chronic diseases between the two groups.
Participants with data on cognitive function at baseline and during follow-up (n = 4,227) were included in the final analysis. The mean (SD) age of these participants was 75.7 (7.1) years with a range of 65–102 years. Women accounted for 58% and participants reported their race as White (n = 3,095; 73%); Black (n = 818; 19%); and Hispanic, Asian, American Indian or Mixed (n = 314; 8%). Table 1 depicts demographic and clinical characteristics of study participants by PE status at baseline. As shown, the proportion of participants who engaged in moderate-to-high intensity exercise was higher among those who reported White race, higher level of education, participation in social activity, and those who scored <3 in PHQ-2 test. The group who reported moderate-to-high intensity exercise also had a significantly higher cognition test scores.
Distribution of Demographic and Selected Clinical Characteristics of Study Participants by PE Status.
Note. PE = physical exercise; PHQ-2 = Patient Health Questionnaire-2.
p < .001.
Of these 4,227 participants, data on cognitive function were available for three follow-up periods in 2,417 participants (57%), two follow-up periods in 972 participants (23%), and one follow-up period in 838 participants (20%). Figure 2 shows the pattern of total cognition score by PE status during follow-up periods. As shown, the mean total cognition score was highest for participants who reported engaging in moderate-to-high intensity exercise. Overall, decline in cognitive function in two or more domains was observed in 102 participants (2.6%), 94 participants (2.8%), and 157 participants (5.7%) at first, second, and third follow-up years, respectively. Table 2 shows the proportion of participants with decline in cognitive function at follow-up by PE status. As shown, the proportion of participants with decline in cognitive function was highest among those who did not engage in any exercise and lowest among those who participated in moderate-to-high intensity exercise.
Mean cognition score by physical exercise (PE) status (BL = baseline; FUY = follow-up year).
Proportion of Study Participants with Decline in Cognitive Function by PE Status.
Note. PE = physical exercise.
Table 3 shows results of two regression analyses with the first one (Regression Model A) without including baseline cognition test score in the model, and the second one (Regression model B) with baseline cognition test score included in the model, and with decline in cognitive function in two or more domains during follow-up as the dependent variable. In both models, participation in PE was associated with lower risk of cognitive decline independent of demographic and educational characteristics, cardiovascular risk factors and diseases, depression symptoms, social activity as well as baseline cognitive function. It is notable that social activity was also associated with lower risk of decline in cognitive function in both models, while depression symptoms was positively associated with decline in cognitive function. Demographic characteristics and education level of participants no longer showed significant association with decline in cognitive function after baseline cognition status was included in the model. There was no significant interaction between PE status and sex, race or education level (p > .1).
Results of Generalized Estimating Equations Analysis with and without Baseline Cognition Score in the Model (Regression Model B and Regression Model A, Respectively) (Dependent Variable: Decline in Cognitive Function).
Note. OR [CI] = odds ratio [confidence interval]; PE = physical exercise; PHQ-2 = Patient Health Questionnaire-2.
Associate degree includes vocational school completion after high school.
Baseline total cognition test scores for the tertile groups ranged between 10 and 17 points (mean [SD] = 15.1 [1.7]) for the lowest tertile group (n = 1,438); between 18 and 21 points (mean [SD] = 19.5 [1.1]) for the middle tertile group (n = 1,559); and between 22 and 31 points (mean [SD] = 23.7 [1.8]) for the highest tertile group (n = 1,230). As shown below, the risk of cognitive decline during follow-up was lower among study participants who reported engaging in PE in each of the tertile groups. The odds ratio (OR) confidence interval [CI] for low-intensity exercise and moderate-to-high intensity exercise in the tertile groups were: lowest tertile group (OR [CI] = 0.50 [0.33, 0.76], p = .001 and OR [CI] = 0.53 [0.34, 0.83], p = .006, respectively); middle tertile group (OR [CI] = 1.23 [0.69, 2.21], p = .475 and OR [CI] = 0.38 [0.15, 0.97], p = .042, respectively); and highest tertile group (OR [CI] = 0.17 [0.04, 0.76], p = .020 and 0.16 [0.05, 0.057], p = .005, respectively). These findings suggest that the association between self-reported PE and lower risk of decline in cognitive function is independent of baseline cognition test scores.
A regression analysis was repeated after imputation of data, and similar results were obtained, low-intensity PE: OR [CI] = 0.70 [0.54, 0.91], p = .005 and moderate-to-high intensity PE: OR [CI] = 0.38 [0.27, 0.53], p < .001.
Discussion
The association between participating in PE and lower risk of decline in cognitive function may indicate the positive impact of PE in maintaining cognitive function among community-dwelling older adults. It is noteworthy that this association was independent of confounding factors which included not only demographic characteristics, education level, and vascular comorbidities, but also participation in social activity and depression symptoms. These results are in agreement with previous studies that have reported positive association between PE and cognitive function (Engeroff et al., 2018; Landi et al., 2007; Middleton et al., 2008; Willey et al., 2014; Zhu et al., 2017). Biochemical and structural changes associated with PE have been described above (Leritz et al., 2011; Liu & Nusslock, 2018; Mandolesi et al., 2018; Pattyn et al., 2013) and these changes may explain the association between PE and lower risk of cognitive decline reported in this study. Previous epidemiological studies (Loprinzi et al., 2018) have also reported dose–response relationship between self-reported PE and cognitive function among older adults, which may suggest a causal relationship. In this study, the questions used to assess PE status did not include details about frequency and duration of PE and for this reason dose–response relationship is not reported.
Another notable finding in this study is the association between participation in social activity and lower risk of decline in cognitive function. This finding is in agreement with previous studies that have reported positive association between social activity and cognitive function (Brown et al., 2016; Fratiglioni et al., 2000). It is postulated that participation in social activities may stimulate cognitive activities in different domains of cognition and improve overall cognitive performance (Brown et al., 2016).
The risk of cognitive decline was significantly higher among study participants who reported depression symptoms. Impairment in memory and executive function among individuals with major mood disorder has been previously reported, and both neurochemical and neuroanatomical changes associated with mood disorder have been forwarded to explain the association between depression and cognitive impairment (Crocco et al., 2010; Kwak et al., 2016). The result of this study agrees with these previous reports.
This study has several strengths. Older adults with normal cognitive function at baseline were followed up for up to 3 years and this allowed us to examine the pattern of decline in cognitive function over time. Assessment of multiple cognitive domains was done using tools with established validity and reliability (Ashford et al., 1989; Beeri et al., 2006; Hankee et al., 2016; Morris et al., 1989; Watson et al., 1993). Both PE status and cognitive function were assessed at baseline and during follow-up periods, and hence possible changes in the status these variables during follow-up were accounted for. Similarly, data on social activity and depression symptoms were also available at baseline and during follow-up years. Furthermore, the possibility of reverse causation was minimized in the analysis.
However, the study also has its limitations. Data on PE were self-reported and may be liable to self-report bias. Another limitation may be that PE status was determined by a single question and did not include frequency and duration of exercise, hence a dose–response relationship between PE and cognitive function could not be determined. However, it is notable that a single question to determine PE status, similar to the one used in this study, has been shown to have good reliability and validity (Milton et al., 2011; Schechtman et al., 1991). Finally, participants were followed for only 3 years and this may have resulted in under-estimation of the extent of cognitive decline.
In summary, results of this study show an independent association between participation in PE and lower risk of decline in cognitive function among older adults with normal cognitive function at baseline. These findings reinforce results of previous studies that have reported positive association between PE and cognitive function. Future long-term epidemiological research projects that examine the association between PE and cognitive function should put emphasis on the type, frequency, and duration of PE, such that a detailed dose–response relationship can be determined.
Footnotes
Acknowledgements
The authors gratefully acknowledge the support provided by the Department of Medicine at Morehouse School of Medicine and Department of Neurology at Emory University School of Medicine.
Declaration of Conflicting Interests
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: The National Health and Aging Trends Study (NHATS) is sponsored by the National Institute of Aging (Grant NIA U01AG032947) through a cooperative agreement with the Johns Hopkins Bloomberg School of Public Health.
Ethical Approval
The ethical approval of this study was obtained from Morehouse School of Medicine IRB [1385694-2].