In this secondary analysis of the Canadian Study of Health and Aging, we evaluated the impact of exercise on cognitive change in elderly people over five years. By mapping the transitions in cognitive states in relation to exercise, and adjusting for age and education, we found that people who participated in high levels of physical activity had a greater chance of stabilizing or improving their cognition compared with people who participated in low/no exercise, especially as baseline cognition worsened. Interestingly, the chance of cognitive decline was more similar between exercise groups. High exercise was also associated with a lower risk of death but this effect was greatest when cognition was highest, and was attenuated as cognition worsened.
Our data must be interpreted with caution. The follow-up period was only 5 years and the effects of exercise on cognitive transitions need to be examined for longer follow-up periods. Although the CSHA is a large, representative sample, 9.6% were lost to follow up. Those lost to follow-up were younger, more often female, less educated, and less likely to exercise than participants.. These differences may have affected the association between exercise and cognition. Misreporting of physical activity is also of concern. However, by using broad groups of activity levels, we minimize the impact of this bias. Misclassification between exercise groups is possible, but unlikely to have been systematic in relation to cognitive testing in that the self-assessed risk factor questionnaire, where the exercise/exposure data were collected, was completed independently of the 3MS.
Because our model is new, the results must be evaluated with care, and skepticism is likely. Even so, the model is well grounded in several earlier papers regarding cognition and frailty. 
Here, we have included adjustment for three covariates, which allowed us to compare risks between groups of exercisers and adjust for two confounders, age and education. By using the output of the model to calculate probabilities of cognitive events, we were better able to describe the dynamic changes of the brain that occur with ageing. 
Furthermore, we were able to present a unified model of relationship between exercise and cognition, taking into account baseline cognition, exercise levels, and mortality. Nevertheless, the model is not perfect. Because the fit of the model to the cognitive transitions is sacrificed as more covariates are incorporated, we only adjusted for two confounders (age and education). As a result, this data should not be interpreted in isolation but in light of the past and future studies of different designs, in diverse populations, controlling for numerous confounders which have concluded that exercise is independently associated with cognitive performance.
According to both previous epidemiological studies using the CSHA sample 
and those of other groups, 
exercise is associated with a reduced risk of cognitive decline. Additionally, clinical trials 
have suggested that people who exercise can have improved cognition. The difference between an improved chance of cognitive improvement and a reduced chance of cognitive decline is subtle. When only survivors are included, as was the case in previous regression analyses, 
the probability of stability or improvement is the complement to the probability of cognitive decline. However, when considering three or more possible outcomes, cognitive stability/improvement, cognitive decline, and mortality, this is not the case. Not all cognitive outcomes need be altered to the same degree. Here, cognitive decline, slightly higher in risk but lower in magnitude in high exercisers, was present in about the same proportion across exercise groups, By contrast, the chance of stability/improvement was much higher amongst people who exercise. Optimistically, this suggests that even people with cognitive impairment may be able to improve their cognition by starting an exercise program. It also reinforces the contention that, rather than viewing changes with age as a picture of gradual loss, a more dynamic understanding of what is happening in the older brain needs to be considered.
Such a consideration might well explore practical means of not just attenuating decline, but of augmenting improvement, as is illustrated here for exercise.
It is possible that the association between exercise and cognition is not a causal relationship. Low exercise at baseline might reflect early cognitive impairment; this has been a criticism of papers that have evaluated exercise in relation to diagnostic states.
In reply to such criticism, the relationship has been found to hold when other potential confounders – such as vascular risk or functional impairment – were included in the adjustment. The evaluation of cognitive states is likely less susceptible to problems in the direction of the association, i.e. given that transitions were considered from each cognitive state and because each cognitive state only included a 3 point range of scores on the 3MS, the variation in cognition at baseline between exercise groups in each cognitive state is unlikely to show such a tight gradient. Even if it were confounded, the closely graded nature of the link – i.e. that people with a shade more cognitive errors were a shade less likely to exercise - would be of interest.
Since exercise is associated with increased longevity, 
there is recurring concern that exercise may delay cognitive impairment but still increase the duration of impairment, giving the paradoxical result of more cognitive impairment on a population basis.
Our study provides some evidence against this paradox being true. The analyses suggest that most of the extension of life occurs in the least impaired states, so that exercise may extend longevity without prolonging time in impaired states.
Our analyses better quantify the nature of the apparent benefit conferred by exercise, and draw attention to the possibility of improvement in cognition with age. The analyses suggest too that most of the extended life occurs in the least impaired states. That the increased likelihood of improving was consistent for all stages of cognitive decline suggests that it is neither age nor cognitive impairment alone are reasons not to exercise.