36 Million People, and the Authors Graded Their Own Result Two Stars

A 2025 systematic review in Nature Aging pooled 28 studies covering roughly 36 million people to estimate how long-term exposure to fine particulate matter relates to dementia risk.

The headline figure: at an average PM2.5 exposure of 15 micrograms per cubic metre, dementia risk was approximately 37% higher than at near-clean air levels. That is a large number attached to an exposure almost nobody chooses and most people cannot easily escape, which is why it travelled widely.

The part that rarely travels with it is that the authors graded their own finding, using a Burden of Proof framework designed to force explicit statements about evidence quality. All-cause dementia received two stars out of five — a rating indicating weak or inconsistent evidence.

The subtype breakdown is where it gets genuinely interesting. Alzheimer's disease specifically scored higher than all-cause dementia. Vascular dementia showed no significant association at all.

That last result cuts against the most intuitive explanation. If inhaled particulate drives dementia primarily by damaging blood vessels — the mechanism most people would reach for first, given what PM2.5 does to cardiovascular risk — then vascular dementia should show the strongest association of any subtype. It showed the weakest. Either the vascular pathway is not the dominant mechanism, or the subtype classifications in the underlying studies are too noisy to detect it. Both are plausible, and the data cannot currently distinguish them.

The general difficulties with this literature are worth being explicit about. Air pollution exposure correlates strongly with income, education, occupational hazard, noise exposure, green space access and healthcare quality, all of which independently predict dementia risk. Statistical adjustment reduces this confounding but cannot eliminate it. Exposure is typically estimated from residential address and monitoring networks, which ignores where people actually spend their time. And dementia diagnosis timing varies enormously by healthcare access — itself correlated with pollution exposure.

None of this makes the finding wrong. Air pollution is firmly established as harmful across cardiovascular and respiratory outcomes, and biological plausibility for a neurological effect exists: ultrafine particles can reach the brain via the olfactory route, and systemic inflammation is a recognised contributor to neurodegeneration.

The accurate summary is that this is a real signal, of uncertain size, with a mechanism that does not yet fit the obvious story. Both halves of that sentence are load-bearing, and coverage that keeps only the 37% is not reporting the paper.

Why this matters for tracking: air quality is one of the few meaningful health exposures that is measured continuously, publicly and for free almost everywhere. If it is worth logging, log the long-run average for where you actually live and work rather than reacting to daily spikes — the exposure this literature concerns is measured in years, and the interventions that move it, such as indoor filtration or where you choose to live, operate on that same timescale.