Air Pollution and Dementia: A Systematic Review

Air pollution

According to a recent Lancet commission on pollution and health, pollution is the largest environmental cause of disease and premature death in the world today, responsible for an estimated 16% of all deaths worldwide and associated with a much wider range of disease than was previously thought [1]. Air pollution in particular is at highest concentration in Low and Middle-Income Countries (LMIC) but can disperse globally and has a disproportionally greater effect on the vulnerable, children and older adults [1].

Dementia

The risk of dementia, and the cognitive decline that precedes it, rise with increasing age. The globally ageing population means that the absolute numbers of those living with dementia continue to increase with an estimated new case every three seconds [12]. The rise in dementia cases is global but due to differing patterns in risk factor exposure and healthcare access, the rise is greater in LMIC [12].

Air pollution and dementia

Exposure to air pollution, especially fine particulate matter, is thought to increase risk of hypertension, raised lipids, atherosclerosis, oxidative stress, insulin resistance, endothelial dysfunction, enhanced propensity toward coagulation, inflammation, and stroke, all of which also raise risk of cognitive decline and dementia [1–4, 13–17].

The 2017 Lancet commission on dementia prevention, intervention and care included air pollution in a list of potential risk factors for dementia [18]; the 2018 Lancet commission on pollution states that the evidence of causation is building, in particular for fine particulate matter and dementia in the elderly, and it calls for research to explore emerging causal links [1]. Given that air pollution is known to have a negative effect on human health, a clinical trial of the length needed to evaluate effect on cognitive function is unlikely and the best evidence to demonstrate a causal link will come from longitudinal observational studies. Recent interest in this area has led to the publication of several such studies examining air pollution exposure and incident cognitive decline or dementia [6, 11].

Our aim was to systematically review the evidence base with respect to the relationship between air pollution and incident cognitive decline and dementia in adult populations and to update our earlier review in this area [11]. The protocol for this review is registered with the International prospective register of systematic reviews (http://www.crd.york.ac.uk/prospero/) no. CRD42018094299 and is an update of an earlier review CRD42014007582 [12]

Inclusion criteria

Longitudinal studies with evidence of some assessment of exposure to air pollution (aggregate assessment or constituent parts);

Exclusion criteria

Studies reporting only occupational exposure to pollutants;

The selection of longitudinal studies with assessment of exposure to air pollution, formal assessment of cognitive function and reports of cognitive decline (i.e., a change in cognitive function) or incident dementia were used to ensure the inclusion of the most robust data with regard to evaluation of causality. Data were extracted using standard extraction tables and information was collected on: the region where the study took place, the size and composition of the study population, the duration of follow up, the assessment of cognitive function or incident dementia, the measurement of exposure to air pollutants, types of pollutant, the analyses (principle summary measures include hazard ratios and odds ratios), results, and reported co-variates. In order to be as conservative as possible, results following adjustment for confounding were preferred for inclusion in the table.

Each included paper was also assessed for validity. Formal scoring was not used as existing instruments have poor discriminative ability when assessing quality. Instead each paper was assessed against key criteria based on the Critical Appraisal Skills Programme (CASP^©) checklists [20]. Potential sources of bias in each study were tabulated.

Study characteristics

Four studies reported results from populations in the United States of America [21, 29], two from Canada [8, 9], two from Taiwan [27, 28], one from Sweden [26, 31], and two from the United Kingdom [23, 24] The samples from the UK both included populations from London but one reported on cognitive function [23] and the other on incident dementia [24]. The samples from Taiwan both selected participants from the National Health Insurance Research Database but selected differing subgroups of the population and presented results for different pollutants: for Jung et al. [27], particulate matter 2.5 (PM_2.5); and for Chang et al. [28], nitrogen dioxide (NO₂) and carbon monoxide (CO). The samples from Canada both selected residents of Ontario but also selected differing sub groups and reported on different measures of pollution; for Chen et al. [8], PM_2.5; and for Chen et al. [9], residential proximity to a major roadway. There were three articles reporting on the Swedish Betula study, one on NO_x and incident dementia [26], one on NO_x and episodic memory [30], and one on PM_2.5 and incident dementia [31]. Sample size ranged from 1,469 [30] to over two million [9], and two studies recruited only women (participants in the Nurses Health Study) [21] and the Women’s Health Initiative Memory Study (WHIMS) [29]. All studies were longitudinal but follow up was reported inconsistently. It varied from one year [22] to ∼5–10 years [23, 30] in studies with cognitive outcomes, and from ∼7 [8, 28] to ∼15 years [26] in studies with incident dementia outcomes. See Table 1 for study characteristics.

Exposure assessment

The most commonly examined pollutant was PM_2.5, reported in nine articles [8, 31]. One study used distance to a major roadway as the main outcome with additional adjustment for PM_2.5 and NO₂ exposure in sensitivity analyses [9]. Four studies reported on NO₂ [8, 28] and one on NO_x [30]. See Supplementary Table 1.

The selected studies adopted a variety of modelling approaches, to obtain high resolution (to residential address level) exposure estimates for their populations. The methodologies varied from relatively simple interpolation approaches from selected monitoring sites within the study domain [28], to more refined approaches, exploiting Land Use Regression approaches and satellite data to improve predictions at locations remote from air pollution monitoring sites [8, 22]. Three studies employed an emissions approach with dispersion modelling, incorporating annual meteorology [23, 25], and atmospheric chemistry [25]. Two studies attempted to split the PM modelled estimates in those derived from vehicle tailpipes [23, 31] and PM derived from residential wood burning [31]. While most studies employed single models to estimate exposures to a range of pollutants, several studies employed different approaches for different pollutants, such as O₃ and PM_2.5 [8]. In most cases, studies presented some form of model evaluation or provided reference to an external source relating to model performance. Only one study employed road distance as their primary (proxy) measure for exposure to traffic related air pollutants [9], but this employed modelled pollutant estimates in their subsequent sensitivity analysis. For one study [28], the exposure measures used in the analyses were unclear. The period for evaluating associations between exposure to pollution and cognitive decline or incident dementia, ranged from days to weeks for the cognitive assessments to months to years for dementia. For dementia in particular, various lag or aggregated exposure periods were also used. It should be stated that there is no clear consensus as to what the most informative exposure period is to assess the neurological impacts of air pollution, but as modelling approaches look further back in time within the available cohorts the modelling uncertainty increases as the available measurement data become sparser. As with all modelling approaches, there is likely to be significant exposure misclassification, as modelling estimates at a point are unlikely to represent the true exposure of a mobile population. Here the view is that this degree of misclassification will be greater for pollutants with a high degree of spatial variation, such as NO₂, and less marked with pollutants with a more uniform distribution such as PM_2.5. These issues were reflected in the discussion of most of the papers cited, as was the difficulty in disaggregating the effect of independent pollutants that were highly correlated within the models.

Assessment of outcomes, dementia, and cognitive decline

Seven articles reported incident dementia cases [8, 31], one focused only on incident Alzheimer’s disease (AD) [27]. Six articles used varied measures of cognitive change [21–23, 30]. See Supplementary Table 1.

Association between air pollution exposure and cognitive outcomes

Overall, the evidence pointed to an association between greater pollution exposure and increased risk of dementia regardless of pollutant measure (see Fig. 2). The evidence relating to cognitive decline was equivocal. There was no clear pattern by region of recruitment or concentration of pollutant. Variation in statistical methods and the frequent use of quantiles for pollutant exposure prevented meta-analysis. See Table 2 for main results.

OPEN IN VIEWER

Fig.2

Number of studies investigating relationship between exposure to pollutants and cognitive function or dementia.

For PM_2.5, three studies [21, 29] reported an association between PM_2.5 and decline in cognitive performance (i.e., higher exposure associated with higher risk), with the WHIMS study additionally reporting a dose dependent relationship between apolipoprotein E4 (APOE4) allele and PM_2.5, such that the lowest decline was in those with lowest exposure and without an APOE4 allele [29]. In the Whitehall study, the association between PM_2.5 and decline in cognitive performance was seen only for memory performance with a four-year time lag (average exposure over four years prior to second cognitive assessment) but not in other cognitive domains or with other time lag periods [23]. Two further studies found no relationship between PM_2.5 and decline in cognitive performance [22, 25], although one reported a dose response relationship for the interaction between presence of APOE4, PM_2.5, and cognitive decline [25] Greater exposure to PM_2.5 was also associated with an increased risk of dementia [8, 31] and AD [24, 27], in UK, Canadian, Swedish, United States, and Taiwanese populations with the WHIMS study also reporting a dose dependent relationship for APOE4, PM_2.5, and dementia risk [29]. For NO₂/NO_x, greater exposure was consistently associated with an increased risk of dementia [8, 28] and AD [24]. There was one analysis of NO_x and cognitive decline that reported no relationship between NO_x and decline in episodic memory [30]. Four studies also examined ozone as a pollutant [8, 27]. One found no relationship with incident dementia [8], one reported a decreased risk of dementia and AD [24], one found greater ozone exposure to be associated with increased risk of incident AD [27], and one reported only a dose dependent relationship between APOE4, ozone interaction, and cognitive decline [25], i.e., the lowest decline in those with lowest exposure and without an APOE4 allele [25]. A single study looked at carbon monoxide and found an association between greater exposure and increased risk of dementia [28].

Taking a different approach, Chen et al. and Carey et al. used a proxy measure of pollution exposure looking at the association between place of residence and distance to the nearest major roadway [9, 24]. This has been shown to have a cross sectional association with poorer cognitive scores in a population in Germany [47], but has not been examined with incident dementia. The results for Chen et al. showed a statistically significant dose response such that the closer the residence to a major roadway the greater the risk of incident dementia [9]. See Supplementary Table 2.

Study quality

Overall, all studies had reasonable clarity in their research questions, used adequate methodology and standard clinical assessments (although not always the gold standard) for cognitive outcomes, and employed a range of modelling approaches to estimate exposures that employed some form of statistical or dispersion modeling, with some form of prior evaluation. Further caution is required regarding interpreting the data relating to dementia risk and residential distance from a major roadway [9, 24] as this was not additionally adjusted for regional impact of wind conditions. Five studies had a greater potential for bias in measurement of outcome in the form of incident dementia, primarily due to the use of health records for the identification of cases [8, 28]. The use of health records rather than a rigorous assessment of all study participants is pragmatic for large sample sizes but may bring bias. Health records often rely on a level of self-referral for assessment and have the potential for missed cases, diagnoses made later in the disease course, and higher rates of case finding in those with comorbid conditions and are likely to have less systematic recording of potential confounders. Four studies used populations that restrict generalizability; the Nurses Health Study recruited only female nurses [21], the WHIMS included only women [29], the Whitehall study recruited predominantly male civil servants [23], and Cleary et al selected participants from an ongoing University of Washington National Alzheimer Coordinating Center [25]. All studies adjusted for a series of relevant confounders (see Supplementary Table 2). Overall, the majority of the studies were at low or low to moderate risk of bias (Supplementary Table 3).

Strengths and limitations

The systematic nature of our updated review and selected inclusion of only longitudinal studies with incident dementia or cognitive change provides the most rigorous filter with which to examine the evidence relating to the association between air pollution and incident cognitive decline or dementia. Furthermore, the risk of bias in the included studies was low to moderate. However, there are limitations. Studies were drawn from just five countries. The assessment of pollution, although geocoded, may not reflect the true local variation or exposure in a mobile population; for example, if, as shown, risk varies within 300 m of a major roadway, there is the potential for a huge variety of risk within even a small geographical area, potentially even more so when taking account of prevailing wind patterns [9]. This is further limited by the use of varied methods for the assessment of exposure to air pollution in the included articles and the data were too disparate to be combined in a meta-analysis. The use of incident dementia is robust but relies on health records where diagnostic rigor may be weaker and cases may be missed. Conversely, case finding bias may be prompted by other health concerns also stemming from exposure to air pollution. Furthermore, although this is in contrast to studies where specific assessment of cognitive function is required for all participants as part of the study procedures a measure of cognitive decline by itself does not necessarily indicate an ongoing degenerative process. As in all dementia risk factor evidence, there is also the question of adequate assessment of confounding, in particular where there may be an interaction with presence of APOE4. Furthermore, although many co-variates have been accounted for there remains, for air pollution in particular, the possibility of a role for both individual and parental socioeconomic status, living conditions, and pollution exposure through the life-course. This is particularly relevant considering that associations between air pollution and poorer cognitive performance have been shown in childhood [6, 7]. Finally, of course, there may be an emerging publication bias as this area expands and we could not assess this, we did not review the grey literature, nor could we combine the evidence we have in a useful meta-analysis.

Although the evidence base examining the association between air pollution and cognitive decline or dementia is smaller and less convincing than the equivalent evidence linking air pollution to increased risk of cardiovascular disease [1, 2], it is growing quickly. All of the articles that we identified had been published in the last five years, and 11 of the 13 we identified had been published since our last systematic review which searched until 1 November 2013 [11]. Our updated review, examining longitudinal evidence with incident decline, adds confirmatory evidence reducing uncertainty as to the likelihood of an association. Furthermore, the growing evidence base is reporting increasingly consistent results (at least for dementia outcomes), dose response relationships, and biological plausibility particularly for exposure to PM_2.5. A detailed examination of the growing literature on potential mechanisms is beyond the scope of this review; however, for example, see Heusinkveld et al., Mumaw et al., Aragon et al., and Thompson [62–65] for more details.

Our review has drawn together and presented the existing evidence for exposure to air pollution and incident cognitive decline or dementia. Our goal now should be to strengthen the rigor and extent of the research in this area to allow specific recommendations to be made. This could be achieved by the use of an individual participant data meta-analysis but to do this, we need to examine a number of factors in more depth. These include: 1) the role of exposure duration; 2) the role of different pollutants and different combinations of pollutants using more sophisticated adjustment and modelling of exposure, e.g., including adjustment for presence of multiple pollutants, taking account of current and prior residential and other exposures such as school yards or workplaces; 3) the role of exposure in different populations in different geographical regions, such as low and middle income countries; 4) the role of modifying factors such as APOE4; 5) the potential variation in the association of air pollution with different cognitive domains; 6) the need to collect repeat imaging measures to allow insight into pathways and mechanisms; and 7) the potential for ameliorating the effects of exposure.

Conclusion

Air pollution, in common with the majority of established risk factors for dementia, does not influence cognition alone. Rather, it increases the risk of multiple non-communicable diseases, one of which is dementia. However, unlike the majority of the established dementia risk factors, the opportunity for personal control over exposure to risk from air pollution is low. Air pollution is pervasive, global, life-long, and bad for health. Further regulation and reduction of exposure has huge potential for health benefit and cost saving including potentially reducing dementia risk. At present, the evidence suggests that greater exposure to air pollution may increase risk of cognitive decline and dementia, and further research is needed to support robust recommendations.