Objective To investigate whether the incidence of dementia is related to residential levels of air and noise pollution in London.
Design Retrospective cohort study using primary care data.
Setting 75 Greater London practices.
Participants 130 978 adults aged 50–79 years registered with their general practices on 1 January 2005, with no recorded history of dementia or care home residence.
Primary and secondary outcome measures A first recorded diagnosis of dementia and, where specified, subgroups of Alzheimer’s disease and vascular dementia during 2005–2013. The average annual concentrations during 2004 of nitrogen dioxide (NO2), particulate matter with a median aerodynamic diameter ≤2.5 µm (PM2.5) and ozone (O3) were estimated at 20×20 m resolution from dispersion models. Traffic intensity, distance from major road and night-time noise levels (Lnight) were estimated at the postcode level. All exposure measures were linked anonymously to clinical data via residential postcode. HRs from Cox models were adjusted for age, sex, ethnicity, smoking and body mass index, with further adjustments explored for area deprivation and comorbidity.
Results 2181 subjects (1.7%) received an incident diagnosis of dementia (39% mentioning Alzheimer’s disease, 29% vascular dementia). There was a positive exposure response relationship between dementia and all measures of air pollution except O3, which was not readily explained by further adjustment. Adults living in areas with the highest fifth of NO2 concentration (>41.5 µg/m3) versus the lowest fifth (<31.9 µg/m3) were at a higher risk of dementia (HR=1.40, 95% CI 1.12 to 1.74). Increases in dementia risk were also observed with PM2.5, PM2.5 specifically from primary traffic sources only and Lnight, but only NO2 and PM2.5 remained statistically significant in multipollutant models. Associations were more consistent for Alzheimer’s disease than vascular dementia.
Conclusions We have found evidence of a positive association between residential levels of air pollution across London and being diagnosed with dementia, which is unexplained by known confounding factors.
- air pollution
- primary care
- alzheimer’s disease
This is an open access article distributed in accordance with the Creative Commons Attribution Non Commercial (CC BY-NC 4.0) license, which permits others to distribute, remix, adapt, build upon this work non-commercially, and license their derivative works on different terms, provided the original work is properly cited, appropriate credit is given, any changes made indicated, and the use is non-commercial. See: http://creativecommons.org/licenses/by-nc/4.0/.
Statistics from Altmetric.com
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.
Strengths and limitations of this study
Annual concentrations during 2004 for air and noise pollution exposure were modelled at a fine resolution, including near-road estimates of traffic pollution.
Pollution data were anonymously linked to electronic health records of over 100 000 older adults registered with 75 general practices in Greater London during 2005–2013.
Incident dementia diagnoses were identified, including Alzheimer’s disease and vascular dementia.
The robustness of the results was tested for by adjusting for area deprivation and other comorbidities.
Accuracy and completeness of primary care diagnoses of dementia and subdiagnoses are a known issue.
Dementia, encompassing both vascular dementia and Alzheimer’s disease, is now reported as the leading cause of death in England and Wales, accounting for 12% of all registered deaths.1 While temporal changes in recording may have influenced how the underlying cause is now determined, more important are increases in longevity among the population at older ages, caused by declining trends in deaths from cardiovascular and cerebrovascular disease.2 In terms of years of life lost, the Global Burden of Disease in 2013 ranked all dementia as the fifth leading cause,2 noting their increasing importance as a cause of death despite little change in age-standardised rates. Therefore, primary prevention of all dementia is a major global public health concern for the coming decades.3 For Alzheimer’s disease for example, while it has been estimated that small delays in its onset and progression could significantly reduce its estimated future burden,4 research has primarily focused on lifestyle factors, where a large systematic review estimated that about a third of Alzheimer’s disease may be attributable to potentially modifiable risk factors such as smoking and physical inactivity.5
More recently research has also extended to the role of environmental risk factors and dementia, where a large systematic review identified moderate evidence for an association with eight different factors including air pollution.6 While air pollution is a well-established risk factor for cardiovascular and respiratory disease,7 its role in relation to dementia is less well considered and understood.8 A recent systematic review of the epidemiological evidence linking air pollution to dementia-related outcomes identified 18 studies, with most reporting adverse associations.9 However, there was a significant variation in the size and quality of the studies involved, reiterating the noted lack of robust longitudinal or population-based studies.6 8 Subsequently, a large population-based study in Ontario, Canada, reported that living close to major roads was associated with a higher incidence of dementia,10 with a further analysis revealing corresponding associations with modelled levels of nitrogen dioxide (NO2) and mass of fine particulate matter with a median aerodynamic diameter ≤2.5 µm (PM2.5).11 These findings raise questions around the mechanisms for the early development of neuroinflammation and neurodegeneration,12 and require further exploration and replication in other large population cohorts with different exposure models, including traffic noise which has been linked to cognitive decline in adults.13
In this paper, we use modelled estimates at a fine spatial scale for modelled estimates of air and noise pollution to investigate relationships with the incidence of dementia across Greater London.
The Clinical Practice Research Datalink (CPRD) is a large, validated primary care database that has been collecting anonymous patient data from participating UK general practices since 1987.14 It has been shown to be broadly representative of the UK population, with about 7% of all people in the UK actively registered on it in 2011.14 Approximately three-quarters of contributing CPRD practices in England have consented to their data being linked to external sources. This is facilitated by a ‘trusted third party’ to CPRD, ensuring that researchers have no access to geographical identifiers such as residential postcode. Key variables which have been linked to the practice data include the Index of Multiple Deprivation (IMD), a composite small-area (approximately 1500 people) measure used in England for allocation of resources,15 and the Office for National Statistics death registration data, which allows underlying cause of death to be identified. For this study, we selected 75 (linked) practices that lay within the study area bounded geographically by the orbital M25 motorway around Greater London (60 were in outer London boroughs, 15 were in inner London boroughs) that were continually recording data between 2004 and 2010.16
Estimates of air and noise pollution exposure
The following measures of exposure, estimated annually between 2004 and 2010, were linked to CPRD: (1) air pollution concentrations, (2) traffic intensity or distance measures, and (3) road traffic noise levels. A priori we chose to focus on annual concentrations estimated in 2004 as this was the earliest year linked. Modelled estimates in other years (2005–2010) were highly correlated (r>0.95), and repeating the analysis using alternative years produced identical results (data not shown).
Modelled annual concentrations for air pollutants were estimated using the KCLurban dispersion modelling system at a resolution of 20×20 m.17 It incorporates hourly meteorological measurements, empirically derived concentrations of NO-NO2-O3 and derived PM (particulate matter), using information on source emissions from the London Atmospheric Emissions Inventory.18 For this analysis, we focused on annual concentrations in 2004 for NO2, PM2.5 (fine particulate matter with a median aerodynamic diameter ≤2.5 µm) and O3 (ozone). Additionally, we present an estimate for primary PM2.5 attributable to road traffic sources estimated from the sum of contributions from the following emission sources: tyre, brake, exhaust, surface wear and resuspension.17 Air pollution concentration estimates were derived for each of the 190 115 London postcodes address centroids based on interpolation from the closest 20×20 m point of the dispersion model.18 This resulted in patients residing in the same postcodes, which typically average about 15 households nationally, being assigned the same exposure levels. Additionally, some postcodes covering smaller geographical areas may also be assigned to the same 20×20 m point. The exposure concentrations were linked to CPRD by the ‘trusted third party’, who subsequently remove the postcode, ensuring we had no direct access to any geographical identifiers.
Traffic proximity measures were developed relating to ‘heavy’ vehicle density, which was defined as light goods vehicles, heavy goods vehicles (rigid and articulated trucks/lorries), buses and coaches. We included a distance measure (in metres) from the postcode centroid to the nearest road classified in the top quartile of heavy vehicle intensity. Traffic intensity was estimated as total vehicle kilometre driven (heavy vehicles only) in each year for all major roads that fell within a 50 m and 50–100 m radius of the postcode address centroid. We used an arbitrary cut-off of >100 000 km driven to define ‘high volume’ in the analyses.
Road traffic noise levels were estimated using the TRAffic Noise EXposure (TRANEX) model.19 This uses information on road traffic flows and speeds, road geography, land cover, and building heights to estimate average sound pressure level (LAeq) in decibels (dB) over different time periods. Evaluations of TRANEX in two other English cities have shown high correlation (r≥0.85) between modelled and measured 1-hour LAeq.19 In our analysis, we focused on the average annual Lnight recorded overnight between 23:00 and 07:00, as this period represents when most of our study subjects would be at their residence. Alternative analyses using daytime noise (LAeq16) produced identical results (data not shown) due to the extremely high correlation with night noise (r=0.999).20 For the linkage of the noise model to CPRD, the geometric centroids of the address locations in each postcode were directly used.
Among the 75 practices, a total of 555 385 patients were actively registered on 1 January 2005, representing about 7% of the Greater London population at that time. From these, we selected 139 718 adults aged 50–79 years who had been registered for ≥1 year continuously with their practice. From this group, 131 869 (94%) were successfully linked to our pollution exposures. Non-linkage was mainly due to a few practices being near the study area boundary, so many of their patients’ individual postcodes were not eligible. Patients were followed until the earliest of (1) date of first diagnosis of dementia, (2) date of death or deregistration from practice, (3) date when the practice ceased contributing to CPRD and (4) 31 December 2013.
We searched the primary care records for the date of first dementia diagnosis, using the Read codes for dementia within the Quality and Outcomes Framework (QOF).21 Although most Read codes are non-specific, we identified subgroups that classified the dementia as Alzheimer’s disease or vascular dementia. Using the death records, we also used the International Classification of Diseases 10th Revision codes to identify patients with dementia listed as the primary cause of death (if not coded as such on the general practitioner (GP) record), and to further classify them based on specific mention of Alzheimer’s disease or vascular dementia anywhere on the death record. We also extracted from the GP record information on ethnicity, smoking, body mass index (BMI) and alcohol consumption, using the last measurement before baseline when available, or during the study if that was the only one available. A missing category was retained where no information was available. In addition, we extracted information on comorbidity recorded at baseline, based on a list of conditions we have previously shown to be independently predictive of mortality.22
Patients (n=391) with an existing diagnosis of dementia by 1 January 2005 were excluded. Additionally, we sought to exclude patients (n=423) when there was evidence on their medical record that they were living in a residential or care home as of 1 January 2005. We did this through a combination of Read codes, or where four or more patients aged 65–99 were recorded living at the same address. We also excluded a small number of patients (n=77) where no IMD could be assigned. This resulted in 130 978 eligible patients for our analyses.
We used Cox proportional hazards models to investigate the associations between all exposure measures estimated in the year before baseline (2004) and subsequent dementia incidence. To account for the clustering by practice (homogeneity between patients from the same practice), we fitted models with a shared frailty (at practice level), which are the survival data equivalent to random effects.23 We adjusted cumulatively for (1) age, sex, ethnicity, smoking and BMI (recorded alcohol consumption was not independently predictive and was dropped from the model), (2) IMD decile (based on national ranking across England) and (3) comorbidities (ischaemic heart disease (IHD), stroke, heart failure, diabetes) that were independently predictive of dementia. For model (2) we explored the effect of adjusting for other pollutants (air pollution for noise and vice versa). In sensitivity analyses, we explored within-practice and between-practice effects by fitting different models (a non-frailty model stratified on practice, practice fitted as a fixed effect, fitting practice mean exposure level in addition to individual level). For all pollutants, we fitted a continuous measure based on IQR to enable comparison between estimates, and quintiles to describe the shape of the association. All analyses were carried out in Stata V.13.
Patient and public involvement
No patients were involved in developing the research question, outcome measures and overall design of the study. Due to patient anonymity, we are unable to disseminate the results of the research directly to study participants.
During the study follow-up period (mean=6.9 years), a total of 2181 patients (1.7%, incidence rate 2.4 per 1000 per year) received a first diagnosis of dementia during follow-up (table 1). Among those patients diagnosed, 848 (38.9%) had a mention of Alzheimer’s disease, 634 (29.1%) mentioned vascular dementia, while 747 (34.3%) received a non-specific Read code. A total of 48 patients (2.2%) received diagnoses for both Alzheimer’s disease and vascular dementia. While crude incidence rates for dementia were lowest in smokers, later adjustment for age and other covariates explained this association; however, the lowest dementia risk seen in obese subjects persisted (data not shown).
The summary statistics for the modelled air and noise pollutants in 2004 are shown in table 2. All air pollutants were strongly positively related to each other (r>0.9), except for O3 which was negatively correlated (r=−0.9 or greater) with both NO2 and PM2.5. Night noise (Lnight) was positively related to NO2 and PM2.5, but associations were less in magnitude (r=0.3–0.4) to all air pollutants. A different pattern for noise was also observed when intraclass correlations (ICCs) were calculated by practice. While most variation by noise was observed within practices (ICC=0.05), the opposite was true for air pollutants where most variation was between practice (ICC>0.7). When mean concentrations were calculated by traffic distance and intensity (online supplementary table S1), all pollutants declined with increasing distance from major roads, except O3 which was higher with further distance. While patients residing in postcodes closest to major roads (0–50 m) had much higher night noise levels than those farthest away (60.4 dB vs 50.9 dB), the difference in modelled air pollution concentrations, especially PM2.5, was much less.
Table 3 summarises a series of adjusted HRs for an incident diagnosis of dementia associated with comparable interquartile changes in different pollutant exposures. The strongest positive associations were seen for NO2, where a +7.5 µg/m3 change produced an HR of 1.16 (95% CI 1.05 to 1.28), adjusting for IMD and other confounders (HR2 in table 3). Further adjustment for related comorbidities (HR3=1.16) did not explain this association. Corresponding associations were smaller with other measures (PM2.5 HR=1.07, PM2.5 traffic HR=1.08, Lnight HR=1.02, distance from road HR=1.02) or negative (O3 HR=0.84). We explored different approaches in estimating within-practice and between-practice estimates (online supplementary table S2). For NO2 and PM2.5the HRs were >1 for both estimates, but the CIs were wide. For night noise (Lnight), there was stronger evidence suggesting between-practice association (HR=1.42, 95% CI 1.03 to 1.96). To investigate the shape of the association, figure 1 plots the adjusted HR (HR3 in table 3) by air and noise pollution fifths, and road distance and traffic intensity 50 m categories. The corresponding HRs are given in full in online supplementary table S3. For NO2 and PM2.5 the increase in dementia risk was noticeably higher in the top fifth of exposure. Patients with an assigned annual exposure of NO2 of >41.5 µg/m3 had a marked increase in risk (HR=1.40, 95% CI 1.12 to 1.74) compared with those in the bottom fifth (<31.9 µg/m3). For other measures (noise, distance, intensity), there was less evidence of any trend, while for O3 the risk was highest in the lowest fifth of exposure (<34.7 µg/m3).
The associations between dementia and an interquartile change (+7.5 µg/m3) in NO2 are explored further in figure 2, which plots the adjusted HRs (HR2 in table 3) from Cox models stratified on a series of risk factors. Generally, there was little evidence of any effect modification across these factors, with all categories producing an HR >1. Associations between NO2 and dementia were still observed when restricted to patients registered for their practice continually for more than 10 years (HR=1.13, 95% CI 1.01 to 1.26), or to patients without IHD, stroke, diabetes or heart failure at baseline (HR=1.21, 95% CI 1.08 to 1.34).
We repeated the analysis, now subclassifying dementia diagnoses recorded as Alzheimer’s disease, vascular dementia or non-specific where no further information was available (table 4). The positive associations with NO2 and PM2.5 were more consistent for Alzheimer’s disease and non-specific diagnoses. For example, patients in the top fifth exposure category of NO2 (>41.5 µg/m3) were at a higher risk of receiving an Alzheimer’s disease diagnosis than patients in the bottom fifth (HR=1.50, 95% CI 1.08 to 2.08). For vascular dementia, there was less evidence of consistent effects with air or noise pollution.
In a sample of 75 general practices across Greater London, the recording of new dementia diagnoses was positively associated with measures of NO2 and PM2.5 assigned at residential address at the beginning of the incident period. The association could not be explained by confounding and was consistent within subgroups. When we restricted to specific diagnoses, associations were still observed with Alzheimer’s disease but not vascular dementia.
Strengths and weaknesses
While we were able to link pollution exposures to the primary care record to obtain diagnoses of dementia, there are concerns around the variability of dementia diagnoses in UK primary care,24 and a recent review has concluded that dementia diagnoses on primary care databases may not be an accurate reflection of the true prevalence.25 Under-recording is thought to be a common issue, as the diagnosis is associated with a stigma for many, and GPs may be reluctant to diagnose dementia unless highly certain.26 A recent study across 23 London practices increased the prevalence on their QOF dementia registers by 9% by a simple coding review.26 To account for under-recording at baseline, a priori we decided to exclude all patients identified as living in care homes at the beginning of follow-up regardless of dementia diagnosis. During follow-up, we observed broadly similar number of dementia subtypes being newly diagnosed, and since it is expected that about two-thirds of dementia is Alzheimer’s disease,27 this suggests the under-recording of Alzheimer’s disease in particular may be an issue. Under-recording in our study could be problematic if it was related to key practice characteristics such as deprivation, as dementia recording has been shown to be lower among more affluent practices.24 28 In our study we observed a wide range of incident rates by practice (0.2%–8.4%), and since the majority of air pollution variation was between practices we cannot discount unmeasured practice characteristics as a possible explanation for our findings. There are also known variations in the prevalence and diagnosis rates of dementia across England,24 with London being among the reported lowest,29 so we also have to acknowledge that the associations we observed may be specific within London and may not extend nationally.
Another weakness of the study is the lack of historical data surrounding exposure. Most large epidemiological studies of long-term exposure to pollution will have difficulty capturing an accurate picture of lifetime or cumulative exposure. This may be pertinent for Alzheimer’s disease where the pathogenesis of the disease may take place over many years.30 We did not have any information relating to previous address or location, and the London population is thought to be mobile and dynamic over time.31 Thus, we are making an assumption that an annual estimate for a single year (2004) represents long-term exposure, based on the last known address for the patient at that practice. We tested this in two ways: (1) sensitivity analyses based on patients who had been continually registered at their practice for a long time (>10 years) produced similar findings; (2) where we did have other modelled years available during follow-up (2005–2010), these were highly correlated over time (r>0.95), so alternative analyses using them made no discernible difference. However, we cannot discount historical factors as an explanation for our findings. For example, it could be that recent exposure levels are acting as a proxy for other historical environmental factors linked to pollution, such as lead from petrol,32 where cumulative exposure has been linked to cognitive decline in later life.33
A novel aspect of our analysis was the ability to simultaneously study the modelled effects of air and noise pollution on dementia, overcoming acknowledged limitations from other studies.11 34 Previous validation of the pollution models used in this study had shown low roadside correlation between them, suggesting that the independent effects of traffic pollution and road noise can be investigated.20 However, a potential limitation is that by being based within Greater London, our air pollution exposure estimates may be somewhat homogeneous, lacking the variability we would expect to see nationally when more rural geographical areas are included. Within London, the contribution of regional (background) PM2.5 and O3 to overall levels tends to dominate.17 However, we were able to make use of a dispersion model with exceptionally fine resolution (20×20 m) to estimate significant changes in exposure of air pollution such as NO2, between major roads and suburban background locations.17 Despite this, the reality was that subtle roadside changes predicted by the model were small in comparison with larger differences estimated between the areas represented by the GP practices (ICCs>0.7 for all air pollutants), suggesting most modelled air pollution variation was between (practice) areas.16 While this limited statistical power to test for any within-practice effects in the study, we did not find evidence to suggest that the overall associations with NO2 and PM2.5 were entirely explained by between-practice differences in modelled exposures.
Finally, another limitation was the incomplete information on key confounders and reliance on an area-based measure (IMD) for socioeconomic status. While mid-life obesity is a risk factor for Alzheimer’s disease,5 the BMI measures we extracted around baseline showed that the risk declined with obesity in later life, a finding which has been observed elsewhere.35 However, further adjustment for IHD, stroke, diabetes and heart failure, which would be associated with earlier unmeasured risk factors, including individual socioeconomic status, did not explain our findings.
The established body of epidemiological evidence linking long-term concentrations of air pollution to adverse health effects has mainly focused on cardiovascular disease.36 The Global Burden of Disease studies, which have described the worldwide impact of air pollution, considered a wide range of outcomes (IHD, stroke, lung cancer, chronic obstructive pulmonary disease) but did not consider neurodegenerative outcomes.7 Research linking air pollution exposure to neurocognitive function has gradually increased from observational findings in 2002 from dogs in Mexico City,37 to larger studies which assessed cognitive decline,38 39 and large population cohorts that specifically investigated the association in relation to diagnoses of dementia.10 11 40–43
A 2015 review on the effect of long-term exposure to outdoor air pollution (15 studies) and noise (8 studies) on cognitive and psychological functions in adults showed that both exposures were separately shown to be associated with one or several measures of global cognitive function, but no study considered both exposures simultaneously, which they highlighted as a need for further research.34 The same authors followed with data from the Heinz Nixdorf Recall cohort study13 on 4086 adults using an additively calculated global cognitive score. They concluded ‘air pollution and road traffic noise might act synergistically on cognitive function in adults’. Our study could consider both measures (air pollution and night noise), and while both showed independent associations with dementia, in a combined model any associations with noise were diminished and of borderline statistical significance.
The largest cohort studies to date investigating dementia and long-term exposure to air pollution are from North America.10 11 43 Chen et al 10 11 used a large Canadian population-based cohort of over 2 million adults aged 55–85 years to ascertain approximately 250 000 incident dementia cases during 2001–2013. In their first analysis,10 the authors found that the risk of dementia increased with nearness to a major road (adjusted HR=1.07, 95% CI 1.06 to 1.08, for people living <50 m from a major traffic road versus >300 m). Associations were stronger among urban residents, especially those who lived in major cities. In a subsequent analysis,11 they used land-use regression models to estimate associations between incident dementia and air pollution, finding significant positive associations with both PM2.5 and NO2, and smaller negative associations with O3. An IQR increase in PM2.5 (4.8μg/m3) was associated with an HR of 1.04 (95% CI 1.03 to 1.05), while for NO2 (14.2 parts per billion) HR was estimated to be 1.10 (95% CI 1.08 to 1.12). By comparison, our estimated distribution of the same pollutants within Greater London was much less spread, with IQRs approximately one-quarter in size. Therefore, comparative HRs for similar unit changes in our study are much greater (eg, for a 1 µg/m3 change in PM2.5 the HR would be 1.07 compared with 1.01 from the Canadian study, for NO2 this would be 1.02 vs 1.00). Our estimate for PM2.5 was more in line what was found in a large US study of Medicare enrollees for first-ever hospitalisation for dementia during 1999–2010 (HR=1.08, 95% CI 1.05 to 1.11, for a 1 µg/m3 change in PM2.5). Chen et al 11 speculate that the stronger associations observed with NO2 may be in part due to it better capturing fine-scale variability in traffic-related air pollution, whereas PM2.5 and O3 have larger regional components. However the resolution of their air pollution models was coarser (1×1 km resolution) than in our study (20×20 m) and may not capture primary emissions from road traffic. While our models were able to estimate traffic-specific components of PM2.5,17 effect estimates remained higher for NO2.
Some smaller studies have separated Alzheimer’s disease from dementia. In Europe, a 15-year longitudinal study in northern Swedish city found evidence of positive associations with both vascular dementia and Alzheimer’s disease and nitrogen oxide using a land-use regression model with a spatial resolution of 50×50 m.41 Comparison between participants in the highest quartile of residential exposure at baseline, versus those in the lowest, produced similar estimates for Alzheimer’s disease (HR=1.38) and vascular dementia (HR=1.47). There have been recent cohort studies from Taiwan: Jung et al 40 showed long-term exposure to O3 and PM2.5 was shown to increase the risk of Alzheimer’s disease, while Chang et al 42 found associations between dementia and NO2 and carbon monoxide. A smaller case–control study by Wu et al 44 linked PM10 and O3 to an increased risk of Alzheimer’s disease and (vascular) dementia. In our study, lower O3 was negatively associated with risk of dementia, primarily as a result of the strong negative correlation with the other modelled air pollutants.45
The implications of linking exposure to air pollution such as NO2 to the development of dementia, specifically Alzheimer’s disease, raise many questions.12 The cause of these neurodegenerative diseases is still largely unknown and may be multifactorial.8 While toxicants from air pollution have several plausible pathways to reach the brain, how and when they may influence neurodegeneration remains speculative.8 30 46 Traffic-related air pollution has been linked to poorer cognitive development in young children,47 and continued significant exposure may produce neuroinflammation and altered brain innate immune responses in early adulthood.48 In later life, the risk for accelerated cognitive decline may involve gene–environment interactions, such as that with apolipoprotein E,49 where evidence comes from findings in neurotoxicological experiments with mice.50
Our observation of an association of air pollution with new dementia diagnoses among older adults living in Greater London is in contrast to an earlier analysis on these data which failed to show consistent associations between air pollution and cardiorespiratory outcomes.16 These suggest there may be a geographical pattern specific to dementia, and potentially Alzheimer’s disease, which requires further exploration nationally. In the Ontario cohort, Chen et al estimated that 6.1% of their total dementia cases were attributable to elevated air pollution exposure.11 In our study, a theoretical shift of all patients to the bottom 20% of NO2 exposure produces an attributable fraction of 7% (data not shown). While this would be smaller than previous population attributable fraction (PAF) estimates for dementia in the UK for a range of independent risk factors such as for hypertension or obesity,5 even a small PAF for dementia would be impactful, where environmental risk factors such as air pollution can be more easily modified at the population level.30 There would be significant public health gains even if the impact was only to delay the progression of dementia.4
With the future global burden of dementia likely to be substantial,3 further epidemiological work is urgently needed to confirm and understand better recent findings linking air pollution to dementia.8 30 Our results suggest both regional and urban background pollutants may be as important as near-traffic pollutants. Future large-scale studies will need to rely on improved recording and linkage of dementia diagnoses across electronic systems, particularly Alzheimer’s disease, where multiple sources can improve diagnostic accuracy.25 Since exposure is lifelong, and most cases are diagnosed in later life, historical data are also ideally required to better estimate cumulative exposure over preceding decades. In conclusion, our findings add to a growing evidence base linking air pollution and neurodegeneration and should encourage further research in this area.
Contributors HRA, RWA, JG, SDB and FJK contributed to the wider study conception and design. HRA, SDB, JG, DD and FJK contributed to exposure assessment. IMC, RWA and DGC acquired linked health data. IMC conceived the specific study design, conducted the data analyses and drafted the initial report. HRA, RWA, DGC and DPS all contributed to the data analysis plan. All authors contributed to interpreting the analyses and to critically revising the article, and approved the final draft. IMC is the guarantor of the work.
Funding This work was supported by the UK Natural Environment Research Council, Medical Research Council, Economic and Social Research Council, Department for Environment, Food and Rural Affairs, and Department of Health (NE/I007806/1; NE/I008039/1; NE/I00789X/1) through the Environmental Exposure & Health Initiative . The research was also part funded by the National Institute for Health Research Health Protection Research Unit (NIHR HPRU) in Health Impact of Environmental Hazards at King’s College London in partnership with Public Health England (PHE) and Imperial College London. The views expressed in this paper are those of the authors and not do not reflect the official policy or position of any of the following: the NHS, the NIHR, the Department of Health, Public Health England or the Medicines and Healthcare Products Regulatory Agency (MHRA). Clinical Practice Research Datalink is owned by the Secretary of State of the UK Department of Health and operates within the MHRA. Clinical Practice Research Datalink has received funding from the MHRA, Wellcome Trust, Medical Research Council, NIHR Health Technology Assessment Programme, Innovative Medicines Initiative, UK Department of Health, Technology Strategy Board, Seventh Framework Programme EU, various universities, contract research organisations, and pharmaceutical companies.
Competing interests None declared.
Patient consent Not required.
Ethics approval This study (protocol number 12_026AR) was approved by the Independent Scientific Advisory Committee evaluation of joint protocols of research involving CPRD data in September 2017.
Provenance and peer review Not commissioned; externally peer reviewed.
Data sharing statement Due to data restrictions, we are unable to share any aspect of the data.