This systematic review pooled 74 studies and 4.2 million people, following them a median of 10 years, to test how physical activity relates to dementia — and, unusually, broke it down by the type of activity.

High overall activity was associated with less all-cause dementia (relative risk 0.80, 95% CI 0.75–0.86), Alzheimer's disease (0.79) and vascular dementia (0.71). But the direction split by domain: leisure-time (RR 0.76) and household activity (0.85) were protective, while occupational (RR 1.20, 95% CI 1.01–1.42) and commuting activity (1.10) were associated with higher risk. Certainty was low, and heterogeneity high.

The usable refinement is that the advice 'be more active' is too blunt. Recommend leisure-time activity a patient chooses and enjoys — walking, gardening, sport — rather than implying that a physically punishing job counts as protection. The occupational signal probably reflects strenuous, low-control work rather than exercise, and it is a reason not to reassure a manual worker that their job protects their brain.

Leisure-time activity linked to ~24% less all-cause dementia (RR 0.76)
Occupational (RR 1.20) and commuting activity linked to higher risk
Advise chosen, enjoyable leisure activity — not 'your job keeps you active'
Low certainty and high heterogeneity: refines advice, does not prove cause

The statistics, in plain English

A relative risk of 0.76 for leisure activity means those most active in their own time had about a quarter less dementia than the least active; an occupational relative risk of 1.20 means the reverse for work-based activity. Because these are observational associations with high heterogeneity and low certainty, the domain distinction is the durable message — the exact percentages are not.