Physical activity pre- and post-dementia: English Longitudinal Study of Ageing

ABSTRACT Background: To inform public health interventions, further investigation is needed to identify: (1) frequency/intensity of everyday physical activity (PA) needed to reduce dementia risk; (2) whether post-diagnosis reduction in PA is associated with cognitive outcomes in people with dementia. Methods: Data from 11,391 men and women (aged ≥50) were obtained from the English Longitudinal Study of Ageing cohort. Assessments were carried out at baseline (2002–2003) and at biannual follow-ups (2004–2013). Results: Older adults who carried out moderate to vigorous activity at least once per week had a 34%–50% lower risk for cognitive decline and dementia over an 8–10 year follow-up period. From pre- to post-dementia diagnosis, those who decreased PA levels had a larger decrease in immediate recall scores, compared to those who maintained or increased PA levels (analyses were adjusted for changes in physical function). Conclusion: PA was associated with cognitive outcomes in a dose-dependent manner. Reduction in PA after diagnosis was associated with accelerated cognitive decline and maintaining PA may reduce symptom progression in dementia.


Introduction
Dementia is neurodegenerative condition characterised by progressive and marked decline in memory and other cognitive functions. Based on UK data, in 2013 dementia prevalence was 1.3% in the total population and 7.1% in those over the age of 65 (Prince et al., 2014). Without significant public health interventions, this was projected to increase by 40% over the next 12 years, due to population ageing. Comparatively, current worldwide prevalence is estimated at 5.2% in those above age 60 and expected to double within 20 years, making dementia one of the leading causes of disability and dependence globally (Prince et al., 2015). Present global economic cost of dementia including medical care (20%), social care (40%) and unpaid care, e.g. family/friends (40%) is estimated at US$818 billion (Prince et al., 2015). Understanding modifiable lifestyle factors that could prevent or delay disease onset will therefore have significant implications for these economic costs and the quality of life for those at risk for dementia.
Observational studies and randomised controlled trials provide evidence for the benefits of regular physical activity (PA) as a behavioral approach to reduce dementia risk (Blondell, Hammersley-Mather, & Veerman, 2014;Clifford, Bandelow, & Hogervorst, 2010;€ Ohman, Savikko, Strandberg, & Pitk€ al€ a, 2014). PA refers to any bodily movement produced by skeletal muscles that requires energy expenditure (Denkinger, Nikolaus, Denkinger, & Lukas, 2012) and can include everyday activities (e.g. walking, gardening, recreational sports) or targeted exercise training (e.g. gym workouts). Several neurophysiological mechanisms may elucidate the benefits of physical activity on cognitive health. These include increased cerebral blood flow (Rogers, Meyer, & Mortel, 1990), reduced age-related loss of brain tissue (Colcombe et al., 2006) and increased levels of neurotrophins (particularly brain-derived neurotrophic factor; BDNF) associated with synaptic plasticity and neuronal survival (Kramer & Erickson, 2007). PA has also been associated with reduced risk for age-related co-morbidities including cardiovascular disease, stroke, diabetes mellitus and depression, which are also risk factors for dementia (Hogervorst, 2012).
While evidence for the benefits of PA on cognitive health is robust, limitations in interpreting these studies have been consistently noted. For instance, a number of studies do not adjust for relevant confounders of PA, such as psychological factors, cardiovascular health and chronic disease (Blondell et al., 2014;Hogervorst, 2012;Miller, Taler, Davidson, & Messier, 2012;Snowden et al., 2011), potentially exaggerating effect size and reducing reliability of findings. Retrospective self-report measures often do not report details of frequency and intensity of PA, limiting their clinical applicability (Blondell et al., 2014;Miller et al., 2012). Studies offering a followup of at least 10 years are few, leading to issues with reverse causality, where undiagnosed symptoms of dementia may lead to less physical activity rather than the opposite (Blondell et al., 2014). Finally, the association between pre-to postdiagnosis changes in PA and cognition has not been previously investigated, but would be valuable to inform public health interventions.
The present study investigated the association between PA and dementia risk over a 10-year follow-up in the English Longitudinal Study of Ageing (ELSA).

Method
Participants ELSA is an ongoing epidemiological study of older men and women (50 years of age). The sample was drawn from those who participated in the Health Survey for England (HSE) between 1998-2000. The primary form data collection in ELSA is a computer assisted personal interview (CAPI) carried out at the person's home or residence. Details of data collection are described elsewhere (Steptoe, Breeze, Banks, & Nazroo, 2013). The core ELSA cohort included 11,391 individuals (46% male, 54% female) who began at baseline (wave 1; 2002-2003) and had biannual follow up-interviews (waves 2-6; 2004-2013). Participants with dementia in wave 1 were excluded; therefore incident cases of dementia were recorded over waves 2-6. Participants were also excluded if they were diagnosed with Parkinson's disease or psychiatric ocnditions (e.g. hallucinations, schizophrenia, psychosis), which could significantly affect physical or cognitive function. The final sample included 11,289 participants. All participants gave informed consent before taking part and the London Multi-Centre Research Ethics Committee approved this study.

Physical activity (PA) definition
Self-reported PA data were collected using questionnaires administered at each wave. Respondents were given examples of PA at mild intensity (e.g. laundry, home repairs), moderate intensity (e.g. cleaning the car, walking at a moderate pace) and vigorous intensity (e.g. digging with a spade, cycling, aerobics). They were then asked to indicate their frequency of participation in mild, moderate and vigorous PA (hardly ever, one to three times per month, once per week or more than once per week). Items on the ELSA physical activity questionnaire were modified from the Whitehall II Health questionnaire (Marmot et al., 1991). The chosen examples of PA at different intensities were selected as they were most commonly reported in two UK-based population studies, and carried out in a similar age group, namely the Ely Diabetes study and the UK sample from the European Prospective Investigation into Cancer (EPIC) cohort (McMunn, Hyde, Janevic, & Kumari, 2004). Different PA intensities were categorised based the activity's metabolic equivalent (MET) score (Ainsworth et al., 2000). MET scores between 2 and 3.5 corresponded to mild PA, scores 3.5-6 defined moderate PA and a score of greater than 6 was equivalent to vigorous PA.
Self-reported data were recorded at each wave (where available). A proxy interview was pursued if an eligible respondent was physically or cognitively impaired; in hospital; or temporarily in care. Those who refused to take part in person but agreed that someone else could do the interview on their behalf could also have a proxy interview. Interviewers were asked to identify a proxy informant as any responsible adult (aged 16 years) who knew enough about the respondent's circumstances to be able to provide information about them.
For the purpose of this study we further categorised PA into four groups coded with numerical values: (1) PA less than once per week (<1£/wk); (2) mild PA once per week (1£/wk); (3) moderate or vigorous PA once per week (1£/wk); (4) moderate or vigorous PA twice per week or more (>1£/wk). For those without a dementia diagnosis, 'average PA' was defined as the mean level of PA across waves 1-6. For those with incident dementia, average PA was defined as the mean level of PA using only the waves prior to diagnosis.

Dementia assessment
Dementia was defined as a physician diagnosis of dementia or Alzheimer's disease reported by the participant in the CAPI. If the individual was unable to take part in the CAPI for any reason, a family member or long-term carer was able to complete the Informant Questionnaire on Cognitive Decline in the Elderly (IQCODE) (Jorm & Jacomb, 1989). The IQCODE consists of 26 questions, which ask the informant to rate how their friend or relative is in situations now compared to 10 years ago (e.g. 'recognising faces of family and friends', 'remembering things that have happened recently'). Scores range on a five-point scale from '1 = much improved' to '5 = much worse'. Scores for each question are summed and divided by 26 to give a final score between 1 and 5. Scores above 3.5 were a cut-off for identifying dementia cases. The IQCODE has been shown to be a valid predictor of dementia (Jorm, 2004) performing as well as other commonly used screening methods (e.g. Mini-Mental State Examination (MMSE) (Folstein, Folstein, & McHugh, 1975)).

Covariates
Several baseline factors were accounted for as possible covariates that could affect cognition, PA or health status (Hogervorst, 2012, Hamer, Lavoie, & Bacon, 2014. Demographic factors included age (years) and gender (male/female). Socioeconomic variables were derived for education (degree level/ below degree/no qualification) and wealth (3rd quintile or less [least wealthy]/4th quintile or above [most wealthy]). Wealth was calculated as net of debt and included the value of housing (excluding mortgage), financial assets (e.g. savings, business assets) and physical wealth (e.g. artwork, jewelry). Presences of several self-reported doctor diagnosed cardiovascular disease factors were accounted for, including high blood pressure, diabetes, stroke and heart attack (binary yes/ no). In sensitivity analysis, binary variables for presence of any self-reported doctor-diagnosed heart condition (including the latter cardiovascular factors plus any other reported) or chronic disease (lung disease, asthma, osteoporosis, cancer, neurological or psychiatric problems) were included. Depressive symptoms were assessed using an eight-tem Epidemiological Studies Depression Scale (CES-D) validated for use in older adults (Karim, Weisz, Bibi, & Rehman, 2014). Here, a CES-D score of 4 or higher identified elevated symptoms of major depression. Other health related covariates included smoking status (non-smoker/previous smoker/current smoker) and frequency of alcohol consumption (more than once daily/daily/one to two times per week/less than once per week). Validated neuropsychological measures of cognition (Hamer et al., 2014) included immediate recall (of 10 words in a list) and verbal fluency (naming category items in 1 min, i.e. animals). Finally, physical function was measured using gait speed, defined as the time in seconds taken to walk a course of 2.44 meters (averaged across two trials).

Statistical analysis
Descriptive analyses were performed on data stratified by incidence of dementia. Chi-squared statistics were computed to assess baseline group differences for categorical data and independent samples t-statistics were used for continuous data.
The association between PA and dementia risk was estimated using Cox proportional hazards regression, calculating multivariate-adjusted hazards ratios (HR's) with 95% confidence intervals (CI's). The timescale variable was the number of completed assessment waves and the censor variable was whether a dementia diagnosis was reported (yes/no). Separate models were used to predict dementia risk from (i) baseline PA and (ii) average PA levels across the follow-up period. Covariates were adjusted in a step-wise approach. In the first step, models contained variables for age, gender, education, wealth, baseline cognitive function and physical function. The second step adjusted for additional variables including alcohol consumption, smoking status, depression, mood problems, cardiovascular factors and chronic disease. In sensitivity analysis exclusion criteria were manipulated to assess the possibility of reverse causality. Here, participants were excluded if they had a dementia diagnosis in waves 1-2 (i.e. up to two years since baseline). This examined robustness of findings and reduced the likelihood that undiagnosed prodromal symptoms of dementia contributed to decreased PA, rather than the opposite. Imputation methods were not performed on missing data.
Subgroup analysis was carried out in those with dementia, who carried out an average of moderate/vigorous at least once per week before diagnosis. Multivariate regression was used to investigate whether reduction in PA after diagnosis were associated with cognitive outcomes (immediate recall and verbal fluency). Here, delta change (d) scores were calculated as post-diagnosis minus pre-diagnosis score. Age, gender, education, and changes in physical function (pre-to post-diagnosis) were also adjusted for.
In all analyses, two-sided p values are reported and values 0.05 were considered statistically significant. Analyses were performed using IBM SPSS Statistics (Version 22).

Results
Of the core-interviewed participants (N = 11,391), 2576 participants died and 3156 were lost over the follow-up period (e.g. moved out of UK, became institutionalised and were unproductive, were unable/unwilling to continue with study) (see Figure 1).
Compared to those who left the study, participants with data available in final wave (N = 5659) tended to be younger, female, more physically active, wealthier, have a higher educational level and score higher on measures of cognitive and physical function at baseline. They also consumed more alcohol, but were less likely to smoke or have reported a psychiatric condition, cardiovascular condition or chronic disease at baseline.

Baseline PA and incident dementia
For the included sample (N = 11,289), the cohort was divided into those who received a dementia diagnosis (across waves 2-6) versus those that did not. Approximately 3.5% of the sample received a dementia diagnosis. Baseline characteristics for the sample are outlined in Table 1.

Average PA levels (across follow-up) and incident dementia
Average PA levels before diagnosis in those with incident dementia (2.63 § 0.05; i.e. mild PA 1£/wk) were lower than the average PA levels of those with no diagnosis (3.17 § 0.01; i.e. moderate/vigorous PA 1£/wk) (p < 0.001). The trajectory of PA levels across all waves (stratified by diagnosis) is illustrated in the Online Supplementary Figure S1. In Cox proportional hazards analyses, higher average PA levels across the follow-up period were associated with a lower HR for  Table 3).

Subgroup analysis
In those with dementia, mean PA levels, immediate recall score and verbal fluency score were all shown to decline from pre-to post-diagnosis (p 0.01) (see Online Supplementary  Table S1). This was the case even though physical function appeared to increase (i.e. quicker walking speed). Those who were active before diagnosis (i.e. averaged moderate/vigorous PA 1£/wk) but reduced their PA levels after diagnosis (N = 282), had a larger drop in immediate recall score from pre-to post-diagnosis (ß = 0.41, p < 0.001), in comparison to For categorical data crosstabs indicate percentages for each category. For continuous data mean ( § SEM) is included. The cohort is divided into those with and without dementia diagnosis between ELSA waves 2-6. Step 1: Adjusted for baseline factors including PA, age, gender, education, wealth, cognitive function, physical function.
Step 2: Adjusted for baseline factors including PA, age, gender, education, wealth, cognitive function, physical function, smoking status, alcohol consumption, depression, mood problems, cardiovascular conditions and chronic disease. Table 3. Cox proportional hazards regression to predict dementia risk from average PA levels across the follow-up (N = 10,667).
Step 1 Step 2  those who maintained or increased their PA levels (see Table 4 and Figure 2). The decrease in PA was proportionate to a decrease in cognitive score. This association was independent of changes in physical function, age, gender and education.
Although not statistically significant, there was a similar trend between PA and verbal fluency.

Sensitivity analysis
Excluding anyone with a dementia diagnosis at waves 1-2 reduced the possibility of reverse causality. After exclusion, there were 271 remaining dementia cases (total sample N = 8564), which decreased the power to detect significant effects. In basic and fully adjusted models, lower levels of PA independently increased the HR for dementia. However, the effect of moderate/vigorous PA became marginally significant (multivariate adjusted HR = 0.69, 95%CI = 0.46-1.04, p = 0.07).

Discussion
Using prospective longitudinal data from the ELSA cohort, the present study examined the association between selfreported PA and dementia over a 10-year follow-up period. In the present study, PA was associated with a lowered risk for dementia in a dose-dependent manner. Here, moderate/vigorous PA >1£/wk appeared to be the most risk reducing, followed by moderate/vigorous PA 1£/wk and then mild PA 1£/wk. These findings were replicated in models adjusted for lifestyle and clinical covariates. Results also remained consistent in sensitivity analysis, which reduced the possibility of reverse causality, although the effect of moderate/vigorous PA 1£/wk became marginally significant, possibly due to the lower sample size and reduced power to detect effects. The present results support previous analysis within this cohort, which found that higher levels of PA predicted healthy ageing, including survival from major chronic disease, depressive symptoms, physical or cognitive impairment (Hamer et al., 2014). This study also validates a number of longitudinal studies, which have found an association between higher levels of PA and reduced risk for dementia and cognitive dysfunction (Blondell et al., 2014). A novel finding was observed in those who were active before dementia diagnosis (i.e. carried out moderate/vigorous activity 1£/wk), but then reduced their PA levels after diagnosis. In this subgroup, those who reduced PA levels after diagnosis had a larger drop in immediate recall score (pre-to post-diagnosis), compared to those who maintained or increased their PA levels. Furthermore, those with larger decreases in PA after diagnosis had a correspondingly larger decrease in recall score. This did not appear to be explained by changes in physical function. It is possible that reduction in PA levels after dementia diagnosis is due to loss of confidence or anxiety associated with cognitive symptoms of dementia (i.e. confusion, memory loss, disorientation). On the other hand, it could also be caused by the stigma of having dementia or perceived dependency on carers for day-to-day activities (Whitehead, Drummond, Walker, & Parry, 2013). Further research with a larger sample size is needed to investigate the reasons why those who were active before diagnosis may become less active after diagnosis, as this will provide insight into public health intervention adherence. Even still, these results suggest that maintenance of PA after dementia diagnosis may be beneficial for episodic memory.
Neuroimaging studies and randomised controlled trials provide biological plausibility for the neuroprotective effects of regular PA. Regular PA is associated with increased hippocampal, prefrontal cortex and basal ganglia brain volumes, increased white matter integrity, as well as increased brain function and connectivity between frontal and hippocampal brain regions (Erickson, Hillman, & Kramer, 2015). Additionally, both animal and human studies suggest that exercise induced increases in BDNF may promote neurite outgrowth even in older adults (Aguiar et al., 2011;Leckie et al., 2014). It is theorised that these neuroprotective factors may all contribute to an individuals cognitive reserve; meaning that regular PA may increase a persons ability to recruit compensatory cognitive processing in order to reduce the effects of brain pathology (Stern, 2012). Indeed, mechanisms contributing to neurodegeneration in dementia include synaptic loss, neuronal death, inflammation and accumulation of toxic b-amyloid plaques and intracellular neurofibrillary tangles. Regular PA has been  shown to alleviate or delay pathogenesis of dementia (Phillips, Baktir, Das, Lin, & Salehi, 2015).
Limitations of the present study are acknowledged. Firstly, self-reported measures of PA have been criticised, as adults may not adequately recall PA, particularly low-intensity activities (Middleton et al., 2011). Therefore, despite the large effect sizes in this stidy, the associations between PA and cognitive outcomes in this study may have been underestimated due to regression dilution bias (Blondell et al., 2014). Secondly, within the present sample, 3.5% of participants were classified as having dementia. Given the average age of our sample at baseline (62 § 0.1), dementia incidence is slightly lower than reported national prevalence rates (Prince et al., 2014). There is evidence that approximately half of the cases of dementia in the UK remain undiagnosed (Prince et al., 2014) meaning that physician diagnosis may have underestimated dementia prevalence rates in our sample. Physician diagnosis of dementia reported in the CAPI was not formally validated and so results must be interpreted with this consideration in mind. Moreover, those with dementia or experiencing symptoms of dementia may have chosen not to participate in this study or may have been excluded in the baseline wave leading to further attrition bias, which may have led to an underestimation of effect size.
Finally, in the present study, only complete cases were included in regression analyses. Imputation methods were not used and missing data (due to death or loss to follow-up) may have impacted the study findings. Those with available data throughout the study follow-up period (waves 1-6) were more likely to be younger, female, more physically active, wealthier, have a higher educational level and score higher on measures of cognitive and physical function at baseline; they also consumed more alcohol, but were less likely to smoke or have reported a psychiatric condition, cardiovascular condition or chronic disease at baseline. Therefore results of this study must be interpreted in the context of this missing data. For instance, participants who had the lowest cognitive or physical ability may have been less likely to be represented in the final sample. Nevertheless, in clinical practice, PA lifestyle interventions would be designed to target populations prior to disease onset or cases were disease activity is in its earliest stages where disease progression is limited. Therefore, although the final sample may be biased towards a healthier population, these participants would probably also be those who would be most likely to carry out and benefit from a PA intervention designed to maintain cognitive health in older age.
This study also has a number of important strengths, including the ability to utilise longitudinal data from a large cohort. Sensitivity analysis in a recent meta-analysis (Blondell et al., 2014), found that only 7 of 26 reviewed cohort studies had a follow-up period of greater than ten years and the majority of these investigated risk for cognitive decline rather than dementia. Furthermore, studies tended to look at the relationship between baseline PA and incident dementia (many years later) without consideration of intermediary time-points. The present data is rare in that repeated measures of PA were taken at a number of regular intervals across the follow-up period and in synchrony with recorded dementia diagnosis. Furthermore, a definition of PA was used that outlined both intensity and frequency of activity, and this is relevant for increasing the clinical applicability of these findings. To inform intervention guidelines, the present study found that average PA levels in those without cognitive impairment or dementia remained above moderate to vigorous PA 1£/wk. Moderate PA includes any type of activity with a MET score of 3.5-6 and vigorous PA includes activity with a MET score above 6.
Overall, maintaining higher levels of PA in older age was associated with decreased risk for dementia. For those who were active before diagnosis, reducing PA levels after diagnosis may be detrimental for dementia prognosis, however further research is needed to investigate this possibility.