Interactive effects of air pollutants and temperature on incidence of dementia: a prospective cohort study

Mounting evidence has linked air pollution with dementia and temperature modifies the association of air pollution with other disease. However, their interactions on dementia are unclear. We used a prospective cohort study (the UK Biobank) included 498 660 adults without cognitive impairment or dementia at baseline and followed up for 11.50 years (5734 907 person-years). We applied Cox proportional hazards regression with time-varying exposures to examine the effects of air pollutants [particulate matter (PM2.5, PM10), nitrogen oxides (NO2, NO X ) and sulphur dioxide (SO2)], the mean and variability of seasonal temperature, and their interactions on dementia. During the follow-up time, we ascertained 4119 cases of dementia. We observed a higher hazard of incident dementia for 1 μg m−3 increase in SO2 (hazard ratio [HR] = 1.11; 95% confidence interval [CI]: 1.08, 1.14), NO2 (HR [95% CI] = 1.02 [1.01, 1.02]), NO X (HR [95% CI] = 1.01 [1.00, 1.01]), PM2.5 (HR [95% CI] = 1.03 [1.02, 1.05]), and PM10 (HR HR [95% CI] = 1.02 [1.00, 1.03]). A lower risk of dementia in summertime temperature variability (HR for 1 °C increment above 1.27 °C = 0.61; 95% CI: 0.51, 0.72) was found. We observed a nonlinear relationship between higher risk of dementia and higher summer temperatures, and strong U-shaped relation of both wintertime temperature and wintertime temperature variability with dementia. We found the significantly synergistic effect between SO2 and summertime temperature (p < 0.001), the antagonistic effect between NO2 (p = 0.043), NO X (p = 0.026) and summertime temperature variability. Participants in a lower social economic position dominated susceptibility in temperature-air pollution interaction on dementia. In conclusion, some evidence of interactive effects between summer temperature and air pollutants was found, but no consistent interaction could be identified during the winter. Our study added weight to the evidence of air pollutants, temperature and their interaction on the onset of dementia.


Introduction
Dementia is a chronic neurodegenerative disease commonly seen in the elderly [1]. Global costs for dementia were estimated at $1.3 trillion in 2019 [2]. It is estimated that there are currently over 57.4 million dementia cases worldwide, and this number will rise to 152.8 million in 2050 from the Global Burden of Disease (GBD) Study of 2019 [3]. Currently, dementia is considered as an irreversible disease with no curative treatment available yet, therefore, it is important to identify modifiable risk and protective factors.
With the continuous global warming caused by greenhouse gas emissions, the frequency and degree of abnormal temperature exposure of the population will be relatively increased [12]. Experiments indicated the potential link between temperature changes and the onset of dementia [13,14]. Evidence from human beings also suggested that climate change could impact human health at all ages, especially infants and the elderly [15]. A cohort study from the UK during 2001 and 2011 showed that temperatures associated with the risk of dementia [16]. A crosssectional study with a large sample size of the elderly showed a negative association between low temperatures and cognitive ability in traditionally warmer regions of the United States [17]. Besides, there is limited evidence on the exposure-response relationship between ambient temperature and dementia.
It is worth noting that some studies have shown the interaction between temperature and air pollution have adverse effects on health, including mortality, respiratory and circulatory systems [18][19][20][21][22][23], with no related evidence on dementia. It is suggested that climate change will modify the concentrations and dynamics of air pollutants [24]. Therefore, it is necessary to explore the interaction between ambient temperature and air pollutants for the development of dementia.
The purpose of this study was to investigate the impact of temperature and air pollutants on the occurrence of dementia, and whether there is a potential interaction between air pollution and temperature by using the data from a large population-based cohort (UK Biobank), so as to provide scientific and beneficial suggestions for the etiology and prevention of dementia.

Study population
This study applied data from the UK Biobank cohort [25], a prospective cohort study based on a large-scale population. More information is available online (www.ukbiobank.ac.uk).
The UK Biobank cohort has recruited more than 500 000 participants aged 37-73 from 22 assessment centers distributed in England, Scotland, and Wales [25] across 2006-2010. We included the individuals with information on air pollutants and temperature at their residence from 2006 to 2019 while without dementia when they entered the cohort. Participants who self-reported prevalent cognitive impairment, Alzheimer's disease (AD), dementia, chronic degenerative neurological problems were excluded (figure 1). Participants were followed up until loss to follow-up, death, incidental dementia, or the end of follow-up on 30 November 2020, whichever came first.

Exposure
Exposures of interest included both air pollutants and temperature. The air pollutant concentration for each year from 2006 to 2019 was regarded as the annual exposure concentration for the individual, which were provided by the public UK Air information resource of the Department for Environment Food & Rural Affairs (https://uk-air.defra.gov.uk/ data/pcm-data). The individual exposure data from 2006 to 2019 were extracted using the ArcGIS 10.7 software. We assigned the respective exposures of the 1 km × 1 km grid maps to the participants' residential addresses (both temperature and air pollutants). High or low level of air pollutants was dichotomously defined as higher or lower than the median of annual concentration.
We obtained monthly mean temperature data of 2006-2019 from the Meteorology (Met) Office of UK [26,27] (www.metoffice.gov.uk/research/ climate/maps-and-data/data/haduk-grid/datasets). We then created, for each year, temperature-related exposures for the mean monthly summertime (June-August) temperature (Tsum) and the monthly mean winter (December-February) temperature (Twin). Then, we calculated the standard deviation (SD) of monthly mean summer (June-August) temperature (Tsv, summertime temperature variability) and the SD of monthly mean winter (December-February) temperature (Twv, wintertime temperature variability) for each participant's residential address. We stratified the temperature variables as low (lowest quartile), medium (second and third quartile), or high (fourth quartile).

Dementia diagnosis
The incident of dementia and its subtypes that occurred during follow-up were derived from EHR of the National Health Service which is linked to UK Biobank. Participants with dementia were identified using the International Classification of Disease, 10th Revision, (ICD-10) coding system, and we subtyped dementia as AD, vascular dementia (VaD), frontotemporal dementia (FTD) and other dementia (table S1).

Covariates
Adjustment of covariates took into account of three aspects according to well-known risk factors of dementia [4]: socio-demographics, individual lifestyle, and disease status. Socio-demographics included age; sex; qualification (categorized as higher or professional (college/university degree, NVQ, or other professional qualification), upper secondary (A-levels or equivalent), lower secondary (O-levels, CSEs or equivalent), none of the above); Townsend deprivation index is an area index (Townsend deprivation index [28]

Statistical analysis
Firstly, we summarized the baseline characteristics of participants by dementia status, and calculated the incidence rates per 1000 person-years. Categorical variables were presented as percentages, and continuous variables with skew distribution were represented by median and inter-quartile range (IQR). Pearson chi-square tests were performed to compare distributions of exposures.
The Cox proportional hazard model was used to test the effect of air pollutants and temperature, and their interactions on the incidence of dementia. The proportional hazards assumption was tested by Schoenfeld residuals, with P < 0.05 indicating a violation of this assumption. In order to separate the independent effects of interested exposures, we included yearly summertime and wintertime temperature and their variability simultaneously in the model and treated them as time-varying exposures, and separately included yearly pollutant as a timevarying exposure into the model in order to avoid multicollinearity. We also applied three models with increasing strength of confounder adjustment. Model 1 included sociodemographic covariates. Model 2 further adjusted the individual lifestyles, and model 3 additionally included disease status based on model 2. There was no consensus on whether the disease status (pre-existing comorbidities) was the confounding factor or the effect modifier [11,30]. Effect estimates from model 2 and 3 should be interpreted with caution as these results may represent overcontrol and the potential endogeneity problem [31]. Therefore, we considered model 1 as the main model. Moreover, we also used restricted cubic splines with four knots at the 5th, 35th, 65th and 95th centiles to flexibly model the association of exposures with dementia, and nonlinearity were checked visually and tested using likelihood ratio tests. In addition, the interaction between air pollutants (low, high) and each temperature index was tested using the likelihood ratio test to compare the models with and without the interactive term, with p for interaction <0.05 being considered as statistically significant, and analyses were stratified by air pollutants.
Subgroup analyses were performed by stratifying according to age (<60 years and ⩾60 years), sex (female and male), Townsend deprivation index (1, 2-4 and 5), so as to identify the subgroups who were more susceptible to the effects of air pollutant, temperature and their interactive effects.
We conducted several sensitivity analyses to examine the robustness of the findings. Firstly, we explored the associations between exposures and dementia after excluding participants with changes in baseline residential address within three years of the start of follow-up. Secondly, to control for competing diseases, we tested effect of air pollutants and temperature on dementia after excluding participants with a diagnosis of dementia or dead within three years of entering the cohort. Thirdly, we incorporated 5 year or 12 year strata of follow-up time in the Cox model to account for time trends in incidence. In addition, we also considered two variables indicating (i) the 22 cities of assessment centers (ii) calendar years of follow-up (2006-2020), and additionally adjusted them (dummy region variables and dummy calendar year variables) in the main model to further mitigate some residual or unmeasured spatial and temporal effects and check the robustness.
All statistical analyses were performed using R software (version 4.1.2) with 'rms' and 'survival' packages and Stata (version 15). P values were twosided and the statistical significance was set to less than 0.05.

Results
We assessed 502 490 participants at baseline. After excluding participants who self-reported prevalent cognitive impairment or AD or dementia (n = 124), had chronic degenerative neurological problems at baseline (n = 173), without basic follow-up data (n = 590), and had no exposure information (n = 2943), 498 660 participants were included in the analysis. The percentages of covariates missing at baseline were summarized in table S2 and Figure 2 showed that there were no apparent longterm trends for yearly mean air pollution and seasonal mean temperatures. During the whole study period, the distribution of air pollutants, Tsum, and Twv was positive skewed, while the distribution of Twin and Tsv was negative skewed. The median (IQR) estimates of exposures were shown in table 1.
The exposure concentrations of PM 2.5 (median [IQR]: 9.97 [2.24]) and PM 10 (14.53 [3.09]) of all participants were below the European Union (EU) limit value (20 µg m −3 for PM 2.5 , 40 µg m −3 for PM 10 ), but most participants were exposed to concentrations above the WHO guideline limit (5 µg m −3 for PM 2.5 , 15 µg m −3 for PM 10 ). For SO 2 (1.99 [1.04]), all participants exposed to concentrations significantly lower than the UK limit value of 20 µg m −3 . For NO 2 (17.52 [7.40]), most of the participants were exposed to concentrations below the EU limit value of 40 µg m −3 but above the WHO guideline limit of 10 µg m −3 . Nearly half of the participants were exposed to NO x (25.49 [12.53]) concentration higher than the UK limit value of 30 µg m −3 . The descriptive statistics of average temperature and air pollutants across the UK from 2006 to 2019 were shown in table S3. The Pearson correlation coefficients among the 11 exposures were shown in figure S1. Table 2 showed the correlations between exposures (including air pollutants and temperature) and dementia. In the basic models adjusted for age, sex, qualification, and Townsend deprivation index, the ambient air pollutants, Tsum, Twv, low level of Tsv and Twin were associated with increased risks of incident dementia, while most of those association were not statistically significant in model 3 adjusted for disease status. The associations between SO 2 with incident dementia were all statistically significant. The negative correlation between high level of Tsv and dementia were observed in models after adjusting for Tsum, Twin, Twv, socio-demographics, and individual lifestyle covariates.

Ambient air pollution and seasonal temperature with incident dementia
During the first three years after entering the cohort, about 20.27% of participants changed their residential address, and 1.03% were diagnosed with dementia or dead. Excluding those participants separately from the analysis did not change our main model results (table S4). Sensitivity analyses additionally including 5 year or 12 year strata of follow-up time to adjust for time trends also gave similar results (table S5). Besides, it should be pointed out that the estimated HRs for all exposures were consistent with our main results after adjusting for the potential city and time fixed effects, except for Twv with consideration of dummy year variable (table S6). We also find some of the estimated HRs in our main model (SO 2 , Twin) were slightly larger than the results of models additionally adjusted dummy region and dummy year, which may imply the possibility of potential omitted variable bias by some unmeasured covariates that covary with various cities and time in the outcome and exposure. Although there may still be potential omitted variable problem, our main results are still stable based on the above sensitivity analysis results.
In figures 3 and 4, we used restricted cubic splines to flexibly model and visualize the relation of ambient air pollution and temperature with dementia. As depicted by figure 3, the exposure-response curve for the association between NO 2 , PM 2.5 , PM 10 and the incidence of dementia were linear, and the curve of SO 2 and NO X can be considered as linear due to the low sample size in the tail of curve. Figure 4 showed that none of the four temperature variables had a linear relationship with dementia. The risk of dementia was relatively flat until around 16.11 • C of Tsum and then started to increase rapidly afterwards (P for nonlinearity = 0.002). Above 16.11 • C, the HR per 1 • C  higher Tsum was 1.14 (1.07-1.22). About the curve of Tsv (figure 4(C)), the risk of dementia was lower when

The interactive effects of seasonal temperature and air pollution
We examined associations of temperature and temperature variability with incident dementia in the main models stratified by ambient air pollution (low, high), and the potential interactions between temperature and air pollutant. No significant difference was observed by stratified air pollution, except for SO 2 with Tsum (the HR (95% CI) of interaction term = 1.15 (1.09, 1.22), p for interaction <0.001), and NO 2 (the HR (95% CI) of interaction term = 0.89 (0.79, 1.00), p for interaction = 0.043), NO X (the HR (95% CI) of interaction term = 0.87 (0.78, 0.98), p for interaction = 0.026) with Tsv (tables 3 and S7). Those with lower NO 2 , NO X pollution had lower HRs when exposed to Tsv (table 3). Then, the multiplying term of seasonal temperature (as ordered categorical variables) and air pollutants (as binary variables) were separately incorporated into the main model (figures [5][6][7][8]. Compared with those with low SO 2 (<1.99 µg m −3 ) and low Tsum (Q1: 11.13 • C-12.06 • C), individuals with higher Tsum and SO 2 had greater risks of dementia ( figure 5). For those in high level of air pollutants, lower Tsum they exposed to, higher dementia risks they had. For Twin, participants exposed to high/low levels of air pollutants showed the same trend (

Subgroup analyses
The risk of dementia due to the combined effects of Tsum/Tsv and air pollutants exposure stratified by age, sex and Townsend deprivation index is illustrated in table S8 and S9. We outlined that dementia risk of both NO 2 and NO X exposure with Tsv (p for both ⩽0.05) was significant among old participants, while no significant risk was observed among younger (<60 years) (

Discussion
This study explored the impact of temperature and air pollution on dementia among the respondents in UK Biobank. We found a significant association between dementia and long-term exposure to low levels of air pollution (SO 2 , NO 2 , NO X , PM 2.5 , PM 10 ). Furthermore, there was a nonlinear relationship between dementia and higher summer temperatures, and strong U-shaped relation between both wintertime temperature and wintertime temperature variability and dementia. A lower risk of dementia in summertime temperature variability was found. We newly observed significant interactions of SO 2 -summertime temperature and NO 2summertime temperature variability in relation to dementia risk. Subgroup analysis showed the combined effects of NO 2 -summer temperature variability were stronger in older and participants in a lower social economic position, and participants in a lower social economic position dominated susceptibility in SO 2 -summer temperature synergy.

Discussion on the effects of air pollutants
To date, emerging evidence relates air pollution to dementia [11,[32][33][34][35], our findings added weight to this evidence of lower levels and their exposureresponse curve, especially in SO 2 . The majority of individual exposure concentration of SO 2 in our study was far below the current UK limit (20 µg m −3 ), but we still found the statistical positive correlation between SO 2 and dementia. A cohort study of 398 889 participants showed that higher levels of SO 2 was associated with greater decreases in cognitive functions in older individuals [36], another cohort study in rural and suburban China had similar results [37]. A case control study demonstrated that SO 2  exposure significantly affected AD cognitive deterioration at both high and low concentrations [8]. In a word, we were able to establish harmful effects at both high and low levels, but it is important to note that EU and WHO do not have regulations on long-term SO 2 exposure limits. Although WHO gave recommendation of short-term (24 hour) air quality guideline (AQG) level (40 µg m −3 ) of SO 2 , it is only based on causal research for respiratory effects and mortality, which means lacking studies of other systems and incident of diseases in which long-term chronic exposure may be impaired. In addition, considering the efforts of air pollution controlling, we are more likely to long-term exposure to low-level of SO 2 , our study provide evidence of long-term AQG for SO 2 .

Discussion on the effects of temperature variables
Our results of seasonal temperature and variability were basically in line with previous studies, and the U-shaped exposure-response curve of winter temperature and variability added weight to this evidence of impact of winter temperature on dementia. Firstly, we found that greater summer temperature variability (Tsv > 1.27 • C) was associated with a 61% reduction in dementia risk for each 1 • C increase. In a cohort study from 2001 to 2011, they examined the effect of seasonal mean temperature and seasonal temperature variability during summer and winter on dementia, and found that warmer-than-average temperatures may decrease the risk of dementia-associated hospital admissions [16]. Experimental research also found that increasing ambient temperature prevented the onset and progression of dementia by reducing the production of amyloid-beta peptides [13].
Our results for summer temperature variability were in line with theirs, because participants in our study with greater Tsv are more likely to experience warmerthan-average temperatures in summer from 2006 to 2019 (87% of those with high Tsv experienced it). Besides, our study showed a significant increase in dementia risk when summer temperatures were higher than 16.11 • C, which is consistent with the result of the above study and another study which pointed each 1 • C increase in temperature above 17 • C was associated with a 4.5% increase in the risk of dementia admission [38]. This may be related to high temperatures in hot months may trigger other adverse health status, such as first-ever strokes [39], acute myocardial infarction and congestive heart failure [40], which could be a key factor in the onset of dementia.

Discussion on interactions
To our knowledge, no previous study has investigated the associations between the combination of temperature and air pollutants and the incidence of dementia, but we found the significantly synergistic effect between SO 2 and summertime temperature, the antagonistic effect between NO 2 , NO X and summertime temperature variability. Most of the studies had examined whether temperature and air pollution interact to mortality, including total non-accidental, cardiovascular, and respiratory deaths etc [41][42][43], but they provided some controversial findings. The results of our study regarding the synergistic effect of SO 2 -summertime temperature are consistent with most studies about mortality. A study in 2017 indicated that SO 2 exhibited larger adverse effects on mortality in high temperature level days than in low temperature level [42]. Another study also found a strong synergistic effect of high temperature and season on mortality associated with SO 2 (mean (IQR); 5.4 (2.9)) in Korea [44]. This synergistic effect may be explained by an increase in concentration of sulphates by enhanced photochemical reaction of SO 2 in hot summer. Furthermore, a time series study from 2008 to 2014 in Hefei, China observed stronger associations between air pollutants (both PM 10 and NO 2 ) and mortality at high temperatures than at medium temperatures. There was no statistical difference for interaction of both PM 10 and NO 2 with summer temperature in our study, possibly due to the low concentration of pollutants in the UK, different atmospheric pressure, humidity [39], etc. Besides, participants living in cooler summer/ colder winter and high-level air pollution had a significantly higher risk of incident dementia compared with participants living in higher temperature and high-level air pollution, which could be attributed to thermal inversions and accumulation of air pollutants on the ground caused by colder temperatures.
We also found those in a lower social economic position were more vulnerable to summer temperature-air pollution interaction on dementia, which may be related to poor living condition (less air-conditioning and air purifiers, poor nutritional intake, etc.), poor baseline health status, and more difficult of accessing to good health care [45,46]. We observed the stronger combined effects of NO 2 with summer temperature variability among the old, and its evidence from the physiology study suggests that older people are hard to thermoregulate and acclimate to high temperature [47], and suggesting that they also may be less resilient to significant changes in temperature [48].
There are few studies on the effect of temperature on dementia and the potential modifiers, it is inevitable that there are other uncertainties, the temporal and spatial variation of temperature and meteorological conditions must be taken into account, such as air conditioning use, clothing thickness, humidity and wind speed, etc. Therefore, more epidemiology and experiments in this area are needed to verify their interactions in the future.

Strengths
The present study is based on the large-scale population cohort of UK Biobank, which makes it possible to study the combination effect of air pollutants and temperature variables on dementia. Furthermore, we observed nonlinear relationships and exposureresponse curves of both seasonal temperature and temperature variability with dementia risk, and calculated HR in different exposure ranges. In addition, we found the significant interactions between air pollutants and summertime temperature on the onset of dementia and the susceptible population, which is likely to be a new study direction of reducing the occurrence of dementia by explore the potential biological mechanism of outdoor temperature and air pollution, especially considering the emergency condition of dementia in the future.

Conclusion
Despite there are several limitations, such as exposure misclassification, volunteer bias, the omitted variable bias and the possibility of unmeasurable confounding, this is only study to date describing the non-linear relationship between seasonal temperature and variability on dementia, and exploring temperature-air pollution interaction on dementia. This prospective cohort study still provided novel evidence for increased risk of dementia associated with air pollution and seasonal temperature and wintertime temperature variability exposure, decreased risk of dementia associated with summertime temperature variability. We observed significant interactions of SO 2 with summertime temperature and NO 2 with summertime temperature variability on dementia, particularly for those participants in a lower social economic position. This emphasized the importance of controlling air pollution at lower concentrations and the possibility of reducing the risk of dementia by exploring the potential mechanism.

Data availability statement
No new data were created or analyzed in this study. Province (Grant Number 2021JJ40805). The funding source was not involved in the study design; in the collection, analysis, and interpretation of data; in the writing of the report; and in the decision to submit the article for publication. The authors acknowledge Dr Shenxin Li from School of Geosciences and Info-Physics, Central South University for providing guidance and suggestions on analytical methods for this study.

Credit authorship contribution statement
F X developed the research question, and W D revised the manuscript. J W and Y L consulted the literature. Y G and Y S collected and sorted out the data. J W analyzed the data and wrote the manuscript. X S and I W interpreted the results. S Y assessed the risk of bias. All authors approved the final version of publication.