Wilker · BMJ (Clinical research ed.) 2023 · systematic review and meta-analysis of longitudinal observational studies · n=51 studies

Ambient air pollution and clinical dementia: systematic review and meta-analysis.

Cited 183 times in the scientific literature.

Level 3 - non-randomized controlled study

Systematic review and meta-analysis of observational longitudinal cohort studies.

PubMed 37019461 · doi:10.1136/bmj-2022-071620 · record verified 2026-08-31

What was done

Authors conducted a systematic review and meta-analysis across EMBASE, PubMed, Web of Science, Psycinfo, and OVID Medline through July 2022. They included longitudinal studies of adults (≥18 years) assessing associations between clinical dementia and EPA criteria air pollutants or traffic proxies averaged over at least one year. Study risk of bias was assessed with the ROBINS-E tool, and random-effects meta-analyses using Knapp-Hartung standard errors were performed when at least three studies reported comparable metrics.

What was found

From 2,080 identified records, 51 studies met inclusion criteria. For particulate matter <2.5 µm (PM2.5), the pooled hazard ratio (HR) per 2 μg/m³ was 1.04 (95% CI 0.99 to 1.09; 14 studies). When stratified by outcome ascertainment method, the HR was 1.42 (95% CI 1.00 to 2.02) across 7 studies using active case ascertainment versus 1.03 (95% CI 0.98 to 1.07) across 7 studies using passive ascertainment. The pooled HR per 10 μg/m³ nitrogen dioxide was 1.02 (95% CI 0.98 to 1.06; 9 studies) and per 10 μg/m³ nitrogen oxide was 1.05 (95% CI 0.98 to 1.13; 5 studies). Ozone showed no clear association (HR 1.00, 95% CI 0.98 to 1.05 per 5 μg/m³; 4 studies).

Why it matters

This review provides quantitative estimates linking ambient particulate and traffic-related air pollution to incident dementia for regulatory and burden-of-disease models, while demonstrating that passive case registries likely underestimate effect sizes relative to active diagnostic screening.

Limits

Most included studies were judged to have a high risk of bias (frequently biasing toward the null). Exposure metrics were geographic proxies rather than direct individual measurements, critical life-course exposure windows were largely unexamined, and data for non-PM2.5 pollutants were sparse.

Cited by