This study using data from a cohort of community-dwelling participants older than 72 years of age found an inverse J-shaped association between household/transportation sub-score and risk of dementia. The greatest risk reduction (about 70%) was observed in participants with household/transportation sub-scores of about 2. In the Voorrips questionnaire, this activity level can be attained by doing some light and heavy household work with sometimes assistance, preparing warm meals 3-5 times per week, walking up 1-5 flights of stairs per day, and walking for shopping 2-4 times per week (see Supplementary Table S1). When the household/transportation sub-score was subdivided in three classes in multi-adjusted models, the risk was decreased by 45% for the moderate level (1.6-2.0 points) and by 38% for the highest level (>2.0 points). Similar results were obtained when the analysis was restricted to AD. A reduction by 48% of the risk of dementia was also observed in the sub-sample of ≥80-year-old participants for the moderate household/transportation activity class. On the other hand, despite the decreased risk indicated by the cubic spline curve, the leisure/sport activity sub-score was not significantly associated with the disease risk. These findings suggest that in older people, maintaining a moderately high PA in everyday life is related to a lower 5-year risk of dementia.
Our study has some limitations. First, in our aged population (median = 80 years) the level of leisure and sport activities was relatively low, and 42% of them did not perform any of these activities. This may preclude detecting significant PA effects on the risk of dementia. However, this finding is representative of people belonging to this age class who are seldom involved in intense PA. When the analyses were stratified according to age, the relatively small number of dementia events (38 incident cases) also may have limited the statistical power in the 72-80 years group. Second, the low PA level could be a consequence of comorbidities that are more common at this age, and this might confound the association with dementia. However, the effect of moderate household/transportation activities persisted in the different models adjusted for a large range of covariates, including diabetes and CVD. A residual confounding may also remain because we did not have information on participants who were not confined but who could have gait and balance problems or need assistance to walk. Third, the follow-up duration was relatively short (5 years), and low PA could be an early manifestation of dementia rather than a pre-morbid risk factor. To limit this reverse causality issue, we excluded dementia at the first follow-up visit (i.e., 2 or 3 years after the baseline) or participants with IADL/ADL limitations at baseline, with unchanged results. However, our results might still partially reflect very early dementia-related behavioral changes. Finally, PA was assessed with the Voorrips questionnaire. Although self-reported PA is susceptible to information bias, it is considered a reliable method to collect data on PA for epidemiologic studies . Objective methods, such as accelerometers, have known a rapid development. However, data collection and processing criteria are not always well validated, particularly in elderly people. Moreover, these methods do not give information about the activity types .
The strengths of this study include its multi-center longitudinal design, the sample size (n=1550 elderly participants from the general population), the low attrition rate of the cohort, and the high quality of dementia screening/diagnosis. Few studies have examined the link between PA and dementia risk in very old people as we did here. PA was assessed using the Voorrips questionnaire that was validated in elderly populations and explores different types of activity. To be more clinically relevant, two sub-scores specific for two PA domains were generated, and sitting unloaded activities were excluded because they do not involve physical capacities and are mainly cognitive. To reduce the information loss these scores were analyzed as continuous measures; the exposure-response curve shape and the optimal location of the PA level thresholds were determined by a data driven method.
The literature on the effect of PA on dementia risk is abundant, but not always conclusive. Several meta-analyses have shown a protective effect of PA on dementia onset. The risk reduction for high level of PA ranges from 21% to 28% for all-cause dementia and from 26% to 45% for AD [5, 18-20]. However, results are heterogeneous, notably due to the PA type. For instance, in a systematic review that included 24 studies (among them 5 in middle-aged populations), Stephen et al  reported a beneficial effect of PA in 18 of them. While the association with AD risk reduction was clear for leisure time activities, it was less obvious for occupational and commuting PA. In a meta-analysis on 44 studies (6 in middle-aged populations), Lee  found that high and moderate amounts of PA were related to reduced risks of all-cause, AD and vascular dementia. The author identified various activities, such as leisure time PA, regular exercise and gardening, that had a protective effect on the dementia risk. Conversely, walking was not significantly associated with the dementia risk (four studies on people aged 60 or 65 years and over). Among the 27 studies in older people, four did not find any positive effect of different PA types (swimming, walking, dancing,…) [21-23], and another one  detected a positive effect only for productive activities (gardening, household, …). It should be noted that sometimes, PA included also reading, watching TV, and listening to music. None of the studies clearly excluded sitting recreational activities, as we did. More studies are needed with more details on the PA type, intensity, and duration .
Another possible source of heterogeneity among studies is the differential subgroup effect. A more robust effect of PA on the dementia risk was found in men than in women . For APOE e4 the results are conflicting with a more pronounced effect either in e4 carriers  or in non-carriers . In our study, we did not find any significant interaction with APOE e4 and sex. The activity level was higher in women for the household activities and preparing meals but not for shopping. However, overall the benefic effect is the same for both men and women.
Little is known about the dose-response relationship between PA and dementia. In a meta-analysis that included 15 prospective studies (three in middle-aged populations), Xu et al  analyzed the categorical and continuous effects of PA. First, they found that the risk of all-cause and AD dementia was reduced by 27% and 26%, respectively, in the group with the highest PA activity level compared with the group with the lowest level. Then, they observed in four studies an inverse linear dose-response relationship. Specifically, an increase in leisure time PA by 500 kcal/week was associated with a decrease of about 10% in dementia risk. Other studies examined the risk of dementia across PA quintiles and showed either a linear reduction of the risk , or poor evidence for a dose-response relationship [19, 27, 29]. In our study, the exposure-response curve was non-linear, and a slight risk rebound was observed for the highest household/transportation activity sub-scores. Likewise, the risk reduction was significant for the second class but the p-value crossed the 0.05 threshold for the third class (Table 2, model 2b). This inverse J-shaped association was already found for outcomes such as mortality  and stroke . One explanation is that in some individuals, a high-intensity physical activity could cause sudden increase in blood pressure and hemorrhagic strokes , which are known to be associated with an increased dementia risk , but this remains to be explored. More accurate statistical methods to evaluate the continuous effect may help to determine the optimal PA level threshold. Therefore, it is difficult to formulate recommendations on PA frequency, duration and intensity because of the huge variations in PA definitions and assessment periods and methods.
Several mechanisms have been proposed to explain the role of PA in dementia. Oxidative stress has been recognized as a contributing factor to aging and the progression of multiple neurodegenerative diseases, including AD. Exercise reduces oxidative stress in the brain by inducing antioxidant enzymes. Moreover, regular PA increases the cerebral blood flow and cerebral metabolism . Neurotrophic activity plays a role in modulating the brain synaptic plasticity, angiogenesis, and adult hippocampal neurogenesis through the release of neurotrophic factors (brain-derived neurotrophic factor, insulin-like growth factor, and vascular endothelial-derived growth factor) that is enhanced by exercise . Animal studies have shown that exercise contributes to lowering the accumulation of beta amyloid and tau protein (two AD hallmarks) in brain and cerebrospinal fluid . All these pathways support the biological plausibility of the association between PA and reduced dementia risk.