The rate of age-related decline in mental ability shows significant heterogeneity among older adults, resulting in a large spectrum of what we would consider “normal” brain ageing (1). The latest review of dementia and cognitive decline published by Age UK divides brain function into 3 types: automatic functions [for example breathing, digestion and homeostasis]; motor function [encompassing movement, balance and speech]; and cognitive function [thoughts, emotions, learning, memory, comprehension and decision making] (2). Healthy decline of cognitive function in older people is a concept unconsciously but unanimously accepted and is well summarised by Harada et al. (3).

Defining MCI is more challenging due to a lack of standardised diagnostic criteria (4). It is generally accepted as the intermediate state between healthy ageing and dementia whereby a person experiences dementia-like symptoms such as memory loss, but these symptoms are not severe enough to significantly interfere with daily life. Estimating the prevalence, incidence and social burden of MCI is complex (5). However, it is believed 5%–20% of adults over 65 have MCI (6).

Statistics released by the WHO in 2020 approximate that 5%–8% of adults aged 60 and over have dementia, equating to around 50 million people worldwide, 60% of whom live in low- and middle-income countries (7). This number is projected to increase to 82 million by 2030 and 152 million by 2050. Whilst this is largely attributable to the globally ageing population, the fact that dementia is currently ranked at the seventh leading cause of death highlights the need for preventative action (8).

Air pollution is estimated to kill nearly 6.5 million people annually, being the largest environmental factor globally (9), and is a leading risk factor for mortality (10). In 2017, the global burden of disease of air pollution was measured at 147 million DALYs. Exposure to polluted air is a global crisis with over 90% of the world's population still living in areas where air quality levels exceed WHO limits (11). The highest concentrations of air pollution and indeed its burden of disease are recorded in low- and middle-income countries (10), and are also a key factor in terms of social inequalities outcomes in UK (12). Furthermore, older adults and children are thought to be more vulnerable to the effects of noxious pollutants (13).

A number of conceivable biological mechanisms exist to explain how pollutants may damage cognitive health. More is known about the toxic effects air pollution can have on respiratory and cardiovascular health but recently studies suggest they both may facilitate cognitive impairment. Hüls et al. (14) describe how lung function partially mediated visuo-construction performance. Meanwhile, Saito et al. (15) highlight the interrelationship between AD and CVD risk factors and the potential role of β-amyloid. Alternatively, inhaled pollutants, particularly those of smaller aerodynamic diameters, may access the brain directly via the olfactory bulb (16). At a cellular level, Zhang et al. (17) reported PM_2.5 triggered oxidative stress and myelin sheath degradation in murine models. Additionally, Woodward et al. (18) suggested pollutants activate the transmembrane protein toll-like receptor 4, which is involved in inflammatory signalling. Like other cerebrovascular risk factors, the harmful effects of air pollution likely accumulate over a lifetime. Greater exposure to PM_2.5 has been linked to neuroinflammation, altered immune responses and accumulation of amyloid β-42 and α-synuclein in the brains of children and young adults in Mexico City (19), and the likely mechanisms of its neurological effects including neurodegeneration, are reviewed in Olasehinde and Olaniran (20).

A 2019 report released by the Alzheimer's Society outlined the possible link between air pollution and dementia (21). Similarly, a 2017 Lancet commission included air pollution as a potential risk factor for dementia (22). Nevertheless, a direct causal link has yet to be proven and neither air pollution nor pollution in general were mentioned in the most recent global action plan on dementia (23). The length of clinical trial needed to ascertain the effect of air pollution on cognitive function is unattainable; therefore, longitudinal observation studies are likely to provide the best evidence. Ambient polycyclic aromatic hydrocarbons negatively affect different cognitive functions throughout the lifespan, with general memory being mostly affected in older adults (24). Various reviews have summarised the increasing body of evidence that suggests long-term exposure to air pollution is associated with increased risk of dementia and cognitive decline (25–29). However, only one comprising evidence until 2018 conducted a systematic search (26), and did not meta-analyse the data. We conducted a systematic review of the sources published until 2021 and a meta-analysis on the differential impact of specific air pollutants [PM_2.5, PM₁₀, NO₂, NO_x and O₃] on cognitive functions specifically in the elderly. Pollutants have explicit sources (30). Comparing the severity by which each pollutant negatively affects cognitive functions will allow public health bodies to prioritise the most harmful pollutants; and in turn, tackle their sources.

The databases Web of Science, PubMed and Embase were searched from inception up to the 6th of February 2021. Each database was subject to three different search strategies:  1.

(air pollution) AND inflamm* AND brain

Primary studies with:  •

Assessment of exposure to air pollution explained in the Methods section;

Data were extracted using standard extraction tables. Information collected included: sample size, study characteristics, gender ratio, mean age, age range, cardiovascular comorbidities, air pollutant component, pollutant modelling method, exposure time and cognitive-related outcome measure.

Each individual study included was assessed for risk of bias across four parameters: generalisabilty, exposure measurement, outcome measurement, and adjustment for confounders.

Studies were firstly divided based on the air pollutant they reported on, and then scrutinised further depending on the cognitive domain they assessed. Cognitive domains/functions for this review were ascertained after analysing the reported data based on authors' own nomenclature. Any air pollutant or cognitive domain that was reported by fewer than five studies was governed insufficient to be included in the meta-analysis. HR, OR and β with 95% CI were used for the meta-analysis, which were conducted using MedCalc (https://www.medcalc.org/). Data is presented in tables and forest plot graphs to visualise the difference in reported effect sizes between studies. In instances where data was reported with a standard error [SE] measurement as opposed to a CI; the formula CI = mean ± SE × 1.96 was used in line with Altman et al. (33). For articles reporting β, or which provided data to calculate it, fixed effects were determined using a Fisher Z transformation of the β coefficients. The random effects model (34) general estimate was given after analysing heterogeneity using the I² and Cochran's Q statistics (35). Results from each meta-analysis are provided in the Supplementary Material.

A total of 1,996 records were identified using three separate search strategies (See Supplementary data spreadsheet). Removing duplicates left 866 unique studies. From these, 832 papers were excluded based on inclusion and exclusion criteria. The full-text screening of the remaining papers against the inclusion and exclusion criteria listed above yielded a total of 26 papers. A further paper was identified from individual records. Figure 1 shows the flow chart for study inclusion.

Study characteristics are shown in Table 1. In total, over 2.4 million subjects, 53.8% of which are female, were included in the 26 studies. However, the geographical area covered is limited given the overlap in the samples. The sample from two studies (42, 43) was from Northern Manhattan, four studies (44, 46, 50, 51) used data from a USA-wide study on women's health, two studies analysed data from London (56, 57), three studies (37, 39, 40) analysed data from Ruhr and adjacent areas in Germany, three studies used data from the Taiwanese National Registry (47, 52, 60), one study used data from a multicentre ageing Korean study (36), and three studies used data from a municipality in northern Sweden (59, 61, 62). Nineteen papers included both male and female participants in their cohort. The age of participants ranged from 45 to 100 years with an estimated mean age of 70 years. Cardiovascular comorbidity data was extracted due to the well-described association between air pollution and vascular conditions (63). Median exposure time estimated across the sample was 10 years, with only 4/26 studies with relatively small samples analysing exposure for 20 years or more. These were Schikowski et al. (37) (789 individuals, 22-years exposure to NO₂, NO_x, PM_2.5 and PM₁₀), Ranft et al. (39) (399 individuals, 20-years exposure to PM₁₀), Ailshire et al. (51) (779 individuals, approximately 23-years exposure to PM_2.5), and Russ et al. (54) [572 individuals exposed to CO, NH₃, NMVOC, NO_x, PM_2.5, and SO_x for most of their lifetime (∼60 years)]. Only the studies from USA included ethnicity as covariate in the analyses or restricted the analysis to a certain ethnic group. For example, Paul et al. (45) only included latino (mainly Mexican-Americans) individuals, and Ailshire et al. (51) only included pure black and white individuals and not those with mixed race/ethnicity.

Risk of bias for the studies included can be seen in Table 2. In general the quality of the studies included was good and the risk of bias, as per the categories defined above, was low with only five studies been considered of moderate risk, mainly owed to the relatively small sample sizes and the use of not widely used tools for assessing the outcome, limiting the reproducibility of their findings.

Table 3 shows the pollutant modelling method, pollution component, sources, geographical areas covered, and exposure time from the 27 included studies. PM_2.5 was included in 19 studies, PM₁₀ in 11, NO₂ in 12, NO_x in eight, and O₃ in eight. Four of the studies included reported effects of other pollutants, but each of these (i.e., individually) were reported in less than three studies and ruled insufficient in data to be further meta-analysed. The included studies used a range of pollutant modelling methods, with refined land-use regression model being the prevalent (in 9/26 studies). Interpolation of data from local monitoring stations was used by four studies. Four other studies reported exposure assessment by constructing spatiotemporal models using Bayesian Maximum Entropy method. The remaining publications used less popular methods of exposure assessment. Exposure time ranged 3–60 years.

Regarding pollution sources, only one study (51) analysed neighbourhood-stress-related pollution sources [i.e., neighbourhood upkeep, presence (or not) of trash, and deteriorating buildings vs. not], and another (38) focused on exposure to black carbon. Traffic as pollution source was considered in all studies, either directly or indirectly (i.e., considered within geocoded data).

Table 4 displays the cognitive-related outcome measurements each paper used to assess dementia, executive function, cognitive function, memory and language. Seven articles reported incident dementia cases; one focused solely on incident AD; one used dementia hospitalisations and another combined both incident dementia and dementia hospitalisations. Seven of the papers used an internationally recognised definition of dementia: five referring to an edition (ninth or tenth) of the ICD and two implementing the DSM-IV. Cognitive functions were assessed using a variety of methods. For general cognition, three papers included the CERAD neuropsychological battery; four incorporated MMSE and three used results of assessment of other domains to form a cognition composite. Less consistency was seen in the assessment of the other three cognitive domains (i.e., executive function, memory and language).

Statistically significant associations were mainly obtained from the two studies that used data from 5,330 residents in the northern area of Manhattan (42, 43), in measurements that combine a comprehensive battery of cognitive tests (Table 4), adjusted for individual (age, education, sex, race/ethnicity), neighbourhood sociodemographic variables (Census-based socioeconomic status z-score), and a cohort indicator to adjust for secular trends.

The two articles that used a sample from northern Manhattan also reported significant negative associations between the language domain and all three of PM_2.5, PM₁₀ and NO₂ (42, 43) (Table 5).

All similar reviews tend to have broader inclusion criteria, possibly as a result of the small body of evidence currently available, or focus on the effect of different confounds, for example, the income level in the affected areas (29). Only one, with differing methodology and inclusion/exclusion criteria, conducted a systematic search (26). This is the first systematic review and meta-analysis to look specifically at long-term air pollution exposure and cognition in older adults. Additionally, strict inclusion criteria combined with a thorough risk of bias analysis ensured only the most relevant and robust studies were included. Also, all outcome measures were standardised for meta-analysing the data, facilitating comparison and an objective assessment of the effects of the pollutants analysed in cognition in the elderly.

Aside from the publications that analyse dementia risk, which does indicate cognitive decline from an unknown baseline, most of the cognitive function measures are a conflation of differences in an unknown initial level and an unknown amount of age-related cognitive decline, the admixture of which is age moderated within a population. As age was adjusted for in all models from the publications analysed, it will be difficult to ascertain whether or not older adults are more susceptible to a larger effect/difference in cognitive ability due to exposure to the different pollutants. Despite the strengths of the present work mentioned above, the search strategy itself was somewhat limited due to time constraints and so although three popular databases were systematically searched, it is possible that other relevant publications not indexed in these databases were missed. Furthermore, assessing precise air pollution exposure, even when geocoded, is not an exact science and gives rise to limitations. For example, most pollutants (except ozone) will be highly spatially correlated and often implausible to separate effects. Therefore, our results must be interpreted with caution. Moreover, the sheer complexity in defining cognitive function and its domains combined with the plethora of tests that were used to examine them exposed incomparable data. Standardising increments of pollutant level as well as exploring the relationship between length of exposure and cognitive domains would make for more easily interpretable data.

The publications included are not representative of the most polluted countries and areas on the planet. Most publications were from the USA, followed by specific regions in Germany, Sweden, Spain, UK, and the isle of Taiwan. Highly populated countries and areas like mainland China, India, Mexico City with an increasing ageing population are not represented in the analysis. Entire continents like South America, Africa and Australia, are not represented, and neither are areas affected by ecological catastrophes that cause environmental pollution (e.g., wild fires). Global policies that foster research in these areas are, therefore, needed.