The Lag Effect of Long-Term Exposure to PM2.5 on Esophageal Cancer in Urban-Rural Areas Across China

Long-term exposure to PM 2.5 pollution is a significant health concern and increases risks for 20 cancers in China. However, the studies regarding the effect of PM 2.5 and esophageal cancer 21 incidence (ECI) among urban-rural areas are limited. In this study, we examined the sex- and area- 22 specific association between long-term exposure to PM 2.5 and ECI, as well as explored the 23 corresponding lag effects on ECI using a geographical weighted Poisson regression. We found that 24 each 10 ug/m 3 PM 2.5 caused ECI risk increases of 1.22% (95% CI: 1.09%, 1.36%) and 1.90% 25 (95% CI: 1.66%, 2.14%) for males and females after covariates controlled, respectively, during the 26 study period. Moreover, the higher 0.17% and 0.64% incidence risks for males and females in 27 rural areas than urban areas, as well as a larger lag period in rural areas, respectively. In addition, 28 higher risks for both sexes appeared in north, northwestern, and east China. The findings indicated 29 that long-term exposure to PM 2.5 was significantly associated with increased risks for ECI, which 30 reinforce a comprehensive understanding for ECI related to PM 2.5 .


Introduction 34
Esophageal cancer (EC) is one of common and leading cause of cancer in China and worldwide. 35 In 2017, there were new 473,000 cases of EC in world, 49.68% of which occurred in China 36 (Kamangar et al., 2020). Therefore, China has the most cases and heaviest burden. According to a 37 report from National Cancer Center of China, esophageal cancer incidence (ECI) ranks the sixth 38 leading cause and consists of 6.69% of all cancer cases. In 2015, there were 44,067 and 17,667 cases, 39 accounting for 8.64% and 4.28% for males and females of all cancer cases, respectively (Chen et 40 al., 2016). 41 Several studies have showed that considerable proportion of EC were attributable to smoking 42 and alcohol (Ishiguro et al., 2009;Steevens et al., 2010). However, some multi-site cohort studies 43 from China and Iran proved these risks factors contributed little to EC in high-risk areas as they 44 were rarely occurred (Gholipour et al., 2016;Wen et al., 2017). Moreover, the prevalence of smoking 45 and alcohol drinking significantly differed among men and women in high-risk areas, but the ECI 46 was similar (Abedi-Ardekani et al., 2010). These implied that other risk factors play important roles 47 in ECI. Particulate Matter (PM2.5) has been identified as Group I carcinogen by the International 48 Agency for Research on Cancer (IARC), contributing to 4.2 million premature deaths worldwide 49 (Cohen et al., 2017). Furthermore, it has elevated to the fourth leading risk factor of deaths and led 50 to more than 1.1 million deaths in China (Zhou et al., 2019). Therefore, PM2.5 has become a growing 51 concern whose impacts on health need to be explored. 52 Most studies related to the PM2.5 risks in developed countries revealed that the adverse effects 53 of PM2.5 on health was a severe threat to cancers, including EC. For example, a meat-analysis 54 included 5 case-control and 15 prospective cohort studies proved that a significant association 55 between PM2.5 and non-lung cancers (Kim et al., 2020). The studies from European cohorts also 56 found that PM2.5 pollution was associated with lung, breast, stomach and upper aerodigestive tract 57 cancers (Raaschou-Nielsen et al., 2013;Weinmayr et al., 2018). In addition, some studies also 58 reported EC risks related PM2.5 or air pollution. An ecological study from China found a significant 59 association between PM2.5 and EC (Wang et al., 2019). Similar study suggested that some air 60 pollutants were a contributor on EC after controlling other confounders (Huang et al., 2017). A case-61 control study from Europe showed a significant exposure-response relationship of indoor air 62 pollution and EC, and reported an elevated 2.71-fold risk for EC (Sapkota et al., 2013). Moreover, 63 other studies further reinforced the evidence of significant effects of PM2.5 pollution caused by 64 biomass fuel on EC (Okello et al., 2019;Sheikh et al., 2019;Sheikh et al., 2020). However, such 65 studies were still lacked as they were either from developed countries or limited in a small area. 66 PM2.5 remains an elevated level and includes hazardous substances such as polycyclic aromatic 67 hydrocarbons (PAHs) in China for a long time. 500 PAHs and their derivatives have been detected 68 in air where more than 90% PAHs with 4-6 rings or high toxicity distributed in PM2.5, since PAHs 69 tend to appear in PM2.5 compared to coarse particles as well as coal consumption was a common 70 source of PAHs and PM2.5 (Liu et al., 2017;Zhang et al., 2017). Besides, PM2.5 is a carrier of toxic 71 compounds that includes various hazardous substances (i.e. chemical elements, heavy metals, 72 PAHs), leading to the damage of chromosomes, DNA, and other genetic materials (Chu et al., 2015;73 Ghio et al., 2018;Wang et al., 2020). A case-control study in Iran provided the evidence for a causal 74 role for PAHs in EC and suggested a dose-response relationship (Abedi-Ardekani et al., 2010).

75
Similar studies from China, Brazil and South African all proved the roles of PAHs in EC 76 pathogenesis (Abedi-Ardekani et al., 2010;Abnet et al., 2018;Ferndale et al., 2020). 77 PM2.5 pollution still is a severe health concern as PM2.5 concentration and carcinogenic 78 constitutes remain a marked urban-rural difference across China. Thus, it is essential to understand 79 the health effects of PM2.5 on EC in view of geographical variations and sex differences. In this 80 study, we firstly examined the association between long-term exposure to PM2.5 and ECI using 81 yearly incidence and the annual PM2.5 concentration of 2007-2015. Then, we explored the sex-and 82 area-specific ECI risks for long-term exposure to PM2.5 by applying a geographic weighted Poisson 83 regression. Finally, we quantified the lag effects of PM2.5 with successive 9 years before ECI for 84 sexes and urban-rural differences across China. oxygen demand (CODMn) to evaluate water quality as they were available and consecutive. The 124 indicators were examined by methods of gas-phase molecular adsorption spectrometry and fast 125 digestion spectrometry, respectively (http://www.cnemc.cn/jcgf/shj/). We collected NH3-N and 126 CODMn concentrations from all national water quality surveillance points (WQSPs) in 2007-2015. 127 For example, there were 245 WQSPs that distributed over 31 provincial administrative regions 128 in mainland China in 2015. Subsequently, we estimated the NH3-N and CODMn concentrations 129 of missing areas through Kring interpolation and zonal statistics (Belkhiri et al., 2020).

130
Socioeconomic covariates included nightlight level (NLL) and average education year. NLL 131 is a comprehensive proxy of urbanization, population density, GDP and economic activities, 132 which has been wildly applied in related health concerns (Nadybal et al., 2020;Wu et al., 2020).

133
In this study, NLL layers with a spatial resolution of 1 km were downloaded from the Data Center 134 for Resources and Environmental Sciences, Chinese Academy of Sciences (RESDC) 135 (http://www.resdc.cn). Its light intensity ranges from 0 to 63, the larger values mean the higher 136 socioeconomic levels. Because the NLL data were available in 2000-2013, we estimated the 137 corresponding data in 2014-2015 by average variation rate. Moreover, average education year 138 was defined as the average year among population aged 6 years or older in each area, derived 139 from China Population Census in 2010.

140
Past literature showed that PM2.5 was influenced by temperature, precipitation, and normalized 141 vegetation index (NDVI) (Wei et al., 2019). We adjusted the corresponding confounders of annual 142 PM2.5 in each county. Moreover, spatial location (longitude and latitude) and a dummy variable 143 for also were adjusted. In this study, these indicators were derived from website 144 (http://www.resdc.cn). The involved variables and covariates were presented in Fig. S2. 145

146
We used geographically weighted Poisson regression function (GWPRF) with temporal and 147 spatial terms to examine the effect of PM2.5 on ECI. We examined the effect from global view 148 where the spatial term was smooth the spatial disparities by a natural cubic spline function. 149 However, given the spatial non-stationarity and heterogeneity of health outcome, we calibrated 150 GWPRF through introducing local terms. Moreover, an adaptive bi-square kernel and golden 151 selection search were used to determine spatial weights and bandwidth, respectively. The 152 function framework was presented below, and the other detailed information was elaborated in 153 published literature (Cheng et al., 2011). 154 = 0 ( , ) + ∑ =1 ( , ) + = 1,2, … ,

155
Where represents the ECI at county . , marks the geographic coordinate of county 156 , is the coefficients for variable at different counties. 0 and are the intercept and 157 stochastic variables, respectively. 158 We constructed three models based on GWPRF to estimate the effect. Model 1 only included 159 the annual PM2.5 and time, location and urban-rural dummy variables. Model 2 further adjusted 160 for water quality (i.e., NH3-N and CODMn). Model 3 further adjusted for socioeconomic factors 161 based on Model 2. Notably, to examine the lag effect of long-term exposure to PM2.5 on ECI, we 162 established the models with single-year lags (same year (lag0), 1 year prior (lag1) … 8 year prior 163 (lag8)). Given that single-year lag structures might underestimate the lag effects, we also 164 evaluated the cumulative ECI risks using moving average-year lags (lag01, lag02 … lag08) 165 before baseline incidence. 166 Moreover, two sensitive analysis were implemented to examine the robustness of the results. 167 Firstly, we applied Model 3 in three economic regions (East, Middle and West China) to examine 168 the results in different economic levels. Secondly, other confounding factors (i.e., temperature, 169 precipitation and NDVI) were further included based on Model 3, as well as exposure windows 170 were also adjusted to check whether the effects of PM2.5 on ECI would be modified. All the results 171 were examined at 95% confidence interval (CI). All analysis was processed in GWR 4.0 (GWR4 172 Development Team) and the results were visualized in ArcGIS 10.6 (ESRI, Redlands, CA, USA). 173

Descriptive statistics 176
The mean ECI for males was 19.48±17.73 per 100,000 persons, ranging from 0.48 per 177 100,000 persons in Xinning to 113.56 per 100,000 persons in Cixian. In parallel, the mean ECI for 178 females was 7.50±11.06 per 100,000 persons, ranging from 0.00 per 100,000 persons in Xiajiang, 179 Furong, Linwu, Wuzhishan, Shimian and Chengjiang to 63.36 per 100,000 persons in Cixian. 180 These suggested a great variation among areas. Moreover, there was a marked distinction of ECI 181 between rural areas and urban areas (Table 1).

182
During the study period, the mean PM2.5 concentration was 50.32 ug/m 3 that was greatly 183 higher than the guideline value (10 ug/m 3 ) of WHO. Furthermore, PM2.5 showed a marked 184 difference varying from 4.22 ug/m 3 to 102 ug/m 3 . Similarly, the differences for NH3-N and 185 CODMn were 13.15-fold and 6.62-fold among areas. The detailed information of other covariates 186 was presented in Table 1 and Figure S2.  We found a significantly positive association between PM2.5 and ECI, as well as a stronger 196 association for females than males. Model 1 showed that the greatest risk for males and females was 197 1.32% (95% CI: 1.20%, 1.45%) and 2.70% (95% CI: 2.49%, 2.92%) at lag4, respectively. When 198 water quality was included in Model 2, the greatest risk for males was 1.17% (95% CI: 1.03%, 199 1.31%) occurring at lag7, the corresponding risk for females was 2.09% (95% CI: 1.87%, 2.30%) 200 at lag4. Model 3 was further adjusted for socioeconomic covariates based on Model 2, presenting 201 the maximum magnitude of risks was 1.44% (95% CI: 1.30%, 1.59%) and 2.42% (95% CI: 2.17%, 202 2.66%) for males and females at lag7 and lag4, respectively ( Figure 1). 203 When stratified by urban-rural dummy variables, the three models all showed the stronger 204 association in rural areas than urban areas, suggesting rural residents were more sensitive to PM2.5. 205 Moreover, a shorter lag period occurred in urban than rural areas due to the higher PM2.5. 206 concentrations. Figure 3 (a, b) showed that the geographical variation of the relative risks (RRs) of 207 PM2.5 for males and females, respectively. The higher RRs appeared in north, northwest, and east 208 of China; meanwhile, a higher RRs only for females also appeared in south areas of Northeastern 209 China.

Cumulative ECI risks 215
During the study period, the cumulative risks tended to be generally stable with fluctuation in 216 different exposure windows. We calculated the cumulative ECI risks of exposure to PM2.5 in 217 different models, the results of which supported the association of PM2.5 and ECI as well. Model 1 218 showed that each 10ug/m 3 PM2.5 caused the maximum elevation of 1.28% (95% CI: 1.16%, 1.41%) 219 and 2.49% (95% CI: 2.29%, 2.70%) for ECI risks among males and females in average 5-year 220 exposure (lag05), respectively (Figure 2a). When all covariates were controlled, the cumulative risks 221 were 1.22% (95% CI: 1.09%, 1.36%) and 1.90% (95% CI: 1.66%, 2.14 %) for males and females, 222 respectively ( Figure 2g). In addition, the higher cumulative risks of exposure to PM2.5 in rural than 223 urban areas. 224 Figures 3 (c, d) represents that cumulative risks geographically varied among areas. During the 225 study period, the areas of higher RRs did not change dramatically. Although the more risk areas 226 were in males, the risk magnitude was lower than that of females. in 388 areas (a, b denote the single risks for males and females, respectively; c, d denote the 233 cumulative risks for males and females, respectively) 234 235 3.4 Sensitive analysis 236 The results of sensitive analysis were showed in Figure S3, stratified by economic regions, the 237 positive association between PM2.5 and ECI was unaffected. When other confounding factors were 238 controlled, the association between PM2.5 and ECI still remained significant ( Figure S4). These 239 suggested that our analysis was stable and robust. 240 241

Discussion 242
This study analyzed the lag effect of long-term exposure on PM2.5 and ECI in 388 areas across 243 China. Moreover, we also explored the disparities of ECI risks by urban-rural areas and sexes using 244 GWPRF to construct different models. Our findings provide valuable information for prevention 245 and control strategies in China, and experience for alike countries. 246 We found that long-term exposure to PM2.5 was significantly associated with the increased ECI, 247 which was consistent with our previous findings (Li et al., 2020). Similarly, a cross-sectional study 248 from China observed multiple air pollutants were an contributor for ECI after controlling other 249 confounders ( They also observed that PM2.5 was significantly associated with all-cause mortality in rural areas 273 (Garcia et al., 2016). Although air pollution was heavier in urban areas, the incidence risk was higher 274 in rural areas as this might be influenced by outdoor activity patterns, socioeconomic status, anti-275 pollution behaviors. One explanation lied in disparities of domestic fuel in urban-rural areas. Prior 276 study showed that biomass fuel still dominated in rural residents as they still extensively use 277 traditional stove (Shen et al., 2013 (Liang et al., 2020). Meanwhile, DNA damage 287 depended on the concentrations of PM2.5-bound PAHs and relied on their oxidation and 288 bioavailability (Turap et al., 2018). Another explanation considered that social infrastructure and 289 health service were limited, as well as environmental protection sources was not accessible in rural 290 areas. Generally, rural residents conducted more outdoor activities, resulting in higher PM2.5 291 exposure (Han et al., 2021). They less use anti-PM2.5 air filters and take special face masks to reduce 292 exposure to PM2.5, which increases the exposure risks in rural areas (Zhao et al., 2021). In addition, 293 our sex-specific model showed women are more sensitive to PM2.5, which attributes to biological, 294 demographical, and behavioral differences. Women generally have higher life expectancy and less-295 educated attainment than men in China, so sex might modify the effect of PM2.5 on incidence 296 ( Beelen et al., 2015;Shan et al., 2020). Biologically, women are lack of effective protections against 297 environmental toxicants and reactive oxygen species, because their mtDNA content and expression 298 levels of respiratory electron chain genes are more sensitive to air pollution compared to men 299 (Winckelmans et al., 2017). Furthermore, Women have smaller lung size and airway diameter, 300 which might increase women's airway reactivity and exacerbate particulate deposition (Abed Al 301 Ahad et al., 2020).

302
The spatial distribution of RRs was not consistent with the distribution of ECI, partly attributing 303 to spatial heterogeneities of covariates. Generally, location with higher PM2.5 or higher prevalence 304 of covariates (e. g. NH3-N and CODMn) tended to be accompanied with higher RRs. For instance, 305 PM2.5 was higher in some areas of south and southwestern and east China where RRs was also 306 greater, but the ECI was not higher. This suggested that socioeconomic factors possibly modified 307 the association and PM2.5 and ECI. Prior studies from China, Iran, India and America all proved a 308 significantly inverse association between EC and socioeconomic status or urbanization (Dar et al., 309 2013;Gao et al., 2018;Wong et al., 2018). In this study, we also detected that socioeconomic 310 covariate can reduce ECI risks related to PM2.5 (Model 3). Moreover, some areas with higher ECI 311 burdens also had higher PM2.5 and RRs than the adjacent areas. These suggested that PM2.5 caused 312 risk for ECI, but other covariates can modify their relationship and thus showed spatial 313 heterogeneities of RRs. In addition, the differences of activity patterns and PM2.5 components might 314 influence the exposure risks in different areas, further modified the association between them (Chen 315 et al., 2020). More evidence is needed to provide in the future epidemiological studies to verify this 316 assumption.

317
The effects of PM2.5 on ECI increased with the exposure extension of average-year lags, but 318 finally levelled off. The stronger effects occurred in level-off window before incidence, partly 319 because long-term exposure to lower PM2.5 concentration elevated the cumulative risks of 320 carcinogens, further lead to a slow and subtle change of reactor for PM2.5. A cohort study from 321 Taiwan of China indicated the long-term exposure to low concentrations of PM2.5 was associated 322 with the increased hazard risks of 1.09 (95% CI: on health using different lag structures and models (Kim and Lee, 2019). In addition, the shorter lag 328 window in urban than rural might result from the higher PM2.5 concentrations in urban areas. Long-329 term exposure higher dose of PM2.5 could reduce response time, leading to a prolonged lag window 330 in rural areas. This finding stressed that the incidence risks were not only related with exposure dose 331 of PM2.5 but were influenced by exposure time.

332
Our findings have several public health implications. Although the level of PM2.5 has 333 dramatically decreased in China, the level still exceeded the mortality cut-off value (10 ug/m 3 , WHO) 334 and the first stage of WHO transition period (35 ug/m 3 ). Long-term exposure to PM2.5 over 35 ug/m 3 335 would cause 15% increase in mortality risks (WHO, 2006). Therefore, current control policies of 336 PM2.5 still need to be continued. At the same time, preferred to reduce dependence of economic 337 advancement on coal fuels, cut back the number of coal-fired power plants and set a coal 338 consumption cap on pollutants' emission, upgrade energy structure and formulate stricter emission 339 guidelines to improve air quality in China. Furthermore, governments should plan specific strategies 340 concerning air pollutants in residents' health, because components vary from different areas. 341 Secondly, given the vulnerabilities in rural areas, specific public health policies and interventions, 342 such as installing chimneys or using natural gas and electricity, could effectively reduce exposure 343 risks for rural residents. Thirdly, higher economic and educational levels could offset the adverse 344 effects of PM2.5 to some extent, especially in rural areas. The increase in economic and educational 345 investments, improvement of infrastructures and health services and formulation of assistance 346 system for serious diseases could reduce risks and prevent residents from vicious circle of poverty. 347 Lastly, we encourage the use of anti-PM2.5 instruments and face-mask to reduce exposure risks, 348 especially in heavily polluted days. 349 The present study has several strengths. In contrast to other studies without stratifications, our 350 study not only estimated the effects of PM2.5 by urban-rural areas but examined the corresponding 351 effects by sexes, adding more knowledge and experience for other developing countries regarding 352 health-risk assessment. Secondly, the application of GWPR function allows us to evaluate ECI risks 353 related to PM2.5 at national and county scales, which is pivotal for policy marker to target priority 354 intervention areas and allocate health sources, more precisely. Thirdly, we firstly introduced NLL, 355 NH3-N and CODMn into ECI risks study related to PM2.5, which not only could effectively avoid 356 multicollinearity caused by multivariable but control the interference of water pollution for results 357 through adjusting for NH3-N and CODMn. Lastly, we examined the lag effects of long-term exposure 358 to PM2.5 on ECI through various lag structures, which was important to select appropriate exposure 359 window for future studies. 360 Several limitations should be discussed in our study. Firstly, we did not distinguish the subtypes 361 of EC: esophageal squamous cell carcinoma (ESCC) and esophageal adenocarcinoma (EAC) due 362 to inaccessible data. Nevertheless, ESCC is a dominate subtype that accounts for over 87% of total 363 EC in China. Meanwhile, we not considered exposure risks in different age groups, which might 364 underestimate the adverse effects of PM2.5. Secondly, we did not include other potential confounders 365 such as smoking, alcohol, diet and genetics, etc. These might affect the pattern of ECI and bias our 366 estimation. Thirdly, the 9-year (single and moving-average) lag structures constructed in the study 367 might be still too short to estimate the lag effect of PM2.5 on ECI. If all variables and covariates 368 become available, this limitation would be addressed in the future studies. Fourthly, because of 369 population density and economic disparities, the various PM2.5 components in China and limited 370 CRIs of Western China might have bias in the association between PM2.5 and ECI. However, the 371 existing analysis still contains some risk signal related to PM2.5. Finally, this is an ecological study 372 just explore the related PM2.5 risks at county-level but not involved in individual exposure. County-373 level exposure may cause exposure bias which might misestimate the real associations and even 374 cause non-significant results, if it existed. However, this study still is useful and provide valuable 375 information for policy maker to acknowledge ECI associated with PM2.5 and tailor efficient policies 376 to reduce incidence risks. 377 378

Conclusion 379
Long-term exposure to PM2.5 is significantly associated with increased ECI risks for both 380 sexes in China where the estimated adverse effects are higher in rural areas and females. 381 Furthermore, there is a larger exposure window and higher lag effects of PM2.5 in rural areas, 382 while socioeconomic covariates can mitigate the risks for population. These findings are critical to 383 prioritize intervention areas and target risk population, as well as carry out stricter guidelines and 384 formulate early warning system of PM2.5 to reduce exposure risks. In addition, although the 385 estimated effects of PM2.5 on ECI was so small to be clinically negligible, from a public health 386 perspective, a slight change of ECI induced by PM2.5 may contribute to a great burden. 387 Furthermore, the control and mitigation of PM2.5 pollution not only can reduce the risks for 388 population exposures but also decline the burden of other diseases. 389 390 Declarations:

392
Ethics approval and consent to participate 393 The study used publicly available data and participate consent is not required. the study has been 394 approved by the institutional Review Board at Xi'an Jiaotong University. 395 396 Not applicable 397                Spatial variation of the association between PM2.5 and ECI among males and females in 388 areas (a, b denote the single risks for males and females, respectively; c, d denote the cumulative risks for males and females, respectively) Note: The designations employed and the presentation of the material on this map do not imply the expression of any opinion whatsoever on the part of Research Square concerning the legal status of any country, territory, city or area or of its authorities, or concerning the delimitation of its frontiers or boundaries. This map has been provided by the authors.

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download. renamed35a45.docx