Exposure to wildfire-related PM2.5 and site-specific cancer mortality in Brazil from 2010 to 2016: A retrospective study

Background Long-term exposure to fine particles ≤2.5 μm in diameter (PM2.5) has been linked to cancer mortality. However, the effect of wildfire-related PM2.5 exposure on cancer mortality risk is unknown. This study evaluates the association between wildfire-related PM2.5 and site-specific cancer mortality in Brazil, from 2010 to 2016. Methods and findings Nationwide cancer death records were collected during 2010–2016 from the Brazilian Mortality Information System. Death records were linked with municipal-level wildfire- and non-wildfire-related PM2.5 concentrations, at a resolution of 2.0° latitude by 2.5° longitude. We applied a variant difference-in-differences approach with quasi-Poisson regression, adjusting for seasonal temperature and gross domestic product (GDP) per capita. Relative risks (RRs) and 95% confidence intervals (CIs) for the exposure for specific cancer sites were estimated. Attributable fractions and cancer deaths were also calculated. In total, 1,332,526 adult cancer deaths (age ≥ 20 years), from 5,565 Brazilian municipalities, covering 136 million adults were included. The mean annual wildfire-related PM2.5 concentration was 2.38 μg/m3, and the annual non-wildfire-related PM2.5 concentration was 8.20 μg/m3. The RR for mortality from all cancers was 1.02 (95% CI 1.01–1.03, p < 0.001) per 1-μg/m3 increase of wildfire-related PM2.5 concentration, which was higher than the RR per 1-μg/m3 increase of non-wildfire-related PM2.5 (1.01 [95% CI 1.00–1.01], p = 0.007, with p for difference = 0.003). Wildfire-related PM2.5 was associated with mortality from cancers of the nasopharynx (1.10 [95% CI 1.04–1.16], p = 0.002), esophagus (1.05 [95% CI 1.01–1.08], p = 0.012), stomach (1.03 [95% CI 1.01–1.06], p = 0.017), colon/rectum (1.08 [95% CI 1.05–1.11], p < 0.001), larynx (1.06 [95% CI 1.02–1.11], p = 0.003), skin (1.06 [95% CI 1.00–1.12], p = 0.003), breast (1.04 [95% CI 1.01–1.06], p = 0.007), prostate (1.03 [95% CI 1.01–1.06], p = 0.019), and testis (1.10 [95% CI 1.03–1.17], p = 0.002). For all cancers combined, the attributable deaths were 37 per 100,000 population and ranged from 18/100,000 in the Northeast Region of Brazil to 71/100,000 in the Central-West Region. Study limitations included a potential lack of assessment of the joint effects of gaseous pollutants, an inability to capture the migration of residents, and an inability to adjust for some potential confounders. Conclusions Exposure to wildfire-related PM2.5 can increase the risks of cancer mortality for many cancer sites, and the effect for wildfire-related PM2.5 was higher than for PM2.5 from non-wildfire sources.

], p = 0.002). For all cancers combined, the attributable deaths were 37 per 100,000 population and ranged from 18/100,000 in the Northeast Region of Brazil to 71/100,000 in the Central-West Region. Study limitations included a potential lack of assessment of the joint effects of gaseous pollutants, an inability to capture the migration of residents, and an inability to adjust for some potential confounders.

Conclusions
Exposure to AU : Pleasecheckthattheeditstothesentence}Exposureto:::}captureyourmeaning:Ifnot; pleasepro wildfire-related PM 2.5 can increase the risks of cancer mortality for many cancer sites, and the effect for wildfire-related PM 2.5 was higher than for PM 2.5 from non-wildfire sources.

Author summary
Why was this study done?
• Cancer is a leading cause of death worldwide, and cancer-related deaths are projected to increase in the future in all countries, including Brazil.
• Given the increasing frequency and duration of wildfires in recent decades, the effects of wildfires on health need to be better understood.
• The association between wildfire PM 2.5 and site-specific cancer mortality remains unclear.
What did the researchers do and find?
• We conducted AU : Pleasecheckthattheeditstothesentence}Weconducted:::}captureyourmeaning:If a retrospective study using data from the Mortality Information System in Brazil to assess whether wildfire-related PM 2.5 exposure was associated with mortality from cancer for common cancer sites in adults.
• Municipality-level wildfire-and non-wildfire-related PM 2.5 concentrations were estimated and linked with the mortality data.
• We foundAU : Pleasecheckthattheeditstothesentence}Wefound:::}captureyourmeaning:Ifnot; pleas that wildfire-related PM 2.5 exposure was associated with cancer mortality for various common cancer sites in adults in Brazil, and higher effects were observed for wildfire-related PM 2.5 than for non-wildfire sources of PM 2.5 .

What do these findings mean?
• Our findings suggest a high wildfire-related PM 2.5 attributable cancer burden, among adults in Brazil.
• The potentially higher risk of wildfire-related PM 2.5 compared with non-wildfire-related PM 2.5 for all cancers combined suggests that the wildfire control and systemic prevention strategies are warranted, to reduce cancer mortality risk in Brazil. This could be a health co-benefit of measures to preserve the Amazon rainforest and limit climate change.

Introduction
Wildfires have become more frequent under climate change in recent years and pose a serious threat to human health. Even those living many kilometers away from wildfires are exposed to their smoke; thus, the health impacts of wildfires on the general population are a concern.
Wildfires emit high concentrations of air pollutants and hazardous substances, including fine particles �2.5 μm in diameter (PM 2.5 ), which are regarded as the fire tracer in epidemiological studies.
It has been estimated that 0.62% of all-cause deaths are annually attributable to the acute impacts of wildfire-related PM 2.5 exposure globally [1]. Apart from death, increased risks of morbidity from respiratory diseases, cardiovascular diseases, low birth weight and preterm birth, and influenza were observed after acute short-term wildfire smoke exposure [2]. Only a few studies have reported the long-term effect of wildfire-related PM 2.5 (e.g., on general health and lung capacity) [3][4][5][6]. Currently, there are research gaps regarding the potential health impacts of long-term wildfire smoke exposure, including on the risk of cancer.
Occupational studies investigating the risk of cancer in firefighters found that firefighters who experienced a high degree of wildfire smoke exposure had higher risks of cancer compared to firefighters exposed to limited wildfire smoke [7][8][9]. Wildfire-related particles were suggested to have smaller sizes and to contain more oxidative and proinflammatory components than urban sources of particles [2,10]. Thus, AU : Pleasecheckthattheeditstothesentence}Thus; :::} wildfire-related PM 2.5 exposure could also increase cancer mortality in the general population, and the effect may be higher than for nonwildfire PM 2.5 sources.
The majority of the Amazon rainforest, which represents over half of the planet's rainforests, is contained within Brazil [11]. The current unprecedented scale of wildfires means AU : IchangedU many Brazilian people are exposed to fire smoke. Given that toxic smoke from wildfires travels long distances with wind, the assessment of the health effects of wildfires should not be limited to firefighters. If the association between wildfire-related PM 2.5 and cancer mortality is higher than that for non-wildfire PM 2.5 , cancer could be an important consideration when making public health allocation strategies, especially in Brazil. Assessment of the impact of exposure to wildfire-related PM 2.5 upon mortality from all types of cancer would also inform public health measures to improve cancer survival.
To address this important issue, in this study, we analyze the associations between wildfirerelated PM 2.5 and cancer-specific mortality, using national mortality data spanning 2010-2016 in Brazil. This study also AU : Pleasecheckthattheeditstothesentence}Thisstudyalso:::}captureyourmean compares the impacts of non-wildfire-related and wildfire-related PM 2.5 on cancer mortality. Finally, we estimate regional cancer death counts attributable to wildfire-related PM 2.5 .

Methods
This study is reported as per the REporting of studies Conducted using Observational Routinely-collected Data (RECORD) statement (S1 RECORD Checklist).

Protocol
This research was conducted using data from the Brazil Mortality Information System. Our study did not employ a prospective protocol. Analyses were first planned and performed in August 2021. During peer review, we added a figure of sensitivity analysis and a table comparing the effect on cancer mortality and drowning (as the negative control). Changes to the paper were also made at the request of peer reviewers.

Pollution exposure
Daily all-source PM 2.5 and wildfire-related PM 2.5 were estimated during the study period; the details of model development and validation have been described in our previous work [1,15,16]. In summary, fire-induced change in PM 2.5 was predicted by the chemical transport model GEOS-Chem (version 12.0.0) as the difference in PM 2.5 from simulations with and without fire emissions. The anthropogenic emissions from 5 fire sources (boreal forest fires; tropical forest fires; savanna, grassland, and shrubland fires; temperate forest fires; agricultural waste burning) were from the EDGAR v4.2 inventory (http://edgar.jrc.ec.europa.eu/). The allsource PM 2.5 was then downscaled from the original resolution of 2.0˚latitude × 2.5˚longitude to a higher resolution of 0.25˚× 0.25˚using a random forest model, taking into account the impacts of meteorology on PM 2.5 in the fine grid cells. The downscaled all-source PM 2.5 from GEOS-Chem was validated against ground-level PM 2.5 monitored at 6,882 global sites, with a high coefficient of determination of up to 0.865 [1]. Then wildfire-related PM 2.5 was derived as the product of all-source PM 2.5 and the wildfire-to-all ratio calculated by the GEOS-Chem model. Annual mean non-wildfire-and wildfire-related PM 2.5 were calculated from daily nonwildfire PM 2.5 and wildfire-related PM 2.5 during 2000AU : Pleasecheckthat2000iscorrectherein2000 À -2016. The official geographical boundaries of municipalities were downloaded from the website of the Brazilian Institute of Geography and Statistics (BIGS; https://www.ibge.gov.br/pt/inicio.html).

Other covariates
Daily mean temperatures were calculated from hourly temperature records from the European Centre for Medium-Range Weather Forecasts Reanalysis v5 (ERA5) dataset, with a 0.25˚× 0.25˚(approximately 28 km × 28 km) spatial resolution. This dataset has global coverage and is comparable to weather station observations in evaluating temperature-mortality associations [17]. The municipality-level temperature was represented by the temperature of the grid at the geographical center of each municipality.
Municipality-level population size and gross domestic product (GDP) per capita for every year during the study period were downloaded from BIGS and then adjusted to United States dollars at the current price, according to the consumer price index during 2008-2020 and the average exchange rate in 2020 [18,19]. All variables were linked to death cases according to municipality and year.

Statistical analysis: PM 2.5 -cancer mortality association
A variant difference-in-differences (DID) approach with quasi-Poisson regression was applied to examine the associations between exposure and all cancers and site-specific cancer mortality. The essence of the variant DID design is that the difference in temporal concentrations (wildfire-and non-wildfire-related PM 2.5 in this study) is related to the difference in cancer mortality rates in each location during the study period [20]. Factors that keep stable during the study time and time trends in confounders that changed similarly across locations are controlled. Confounders AU : Pleasecheckthattheeditstothesentence}Confoundersthatcorrelate:::}capturey that correlate with the wildfire-or non-wildfire-related PM 2.5 concentrations and that change differently across regions by time should be adjusted in the model. The parameters of the variables are defined based on previous studies [21][22][23]. Temperature has been demonstrated to be associated with cancer mortality and thus is fitted in the main model [24,25]. Socioeconomic factors are represented by GDP per capita. Cancer-specific mortality associations were evaluated using the following model: where Y s,t represents the number of cases in municipality s, year t; I s is a dummy variable for municipality s; I t is a dummy variable for year t; PM 2.5s,t is the average wildfire-or non-wildfire-related PM 2.5 in municipality s, year t; β s is the intercept or slope for the linear terms; ln (Pop s,t ) is an offset term representing the natural log of the population in municipality s, year t; and Temp values are the means of summer and winter temperatures and their standard deviations (SDs).
We also performed subgroup analyses by age group (20-59 years versus 60 years or above) and sex. We used fixed-effects meta-analyses to compare the effect estimates between sex and age groups. All results are expressed as relative risks (RRs) and 95% confidence intervals (95% CIs) per 1-μg/m 3 increase in annual average PM 2.5 concentration. Several sensitivity analyses were performed-adding gas pollutants (CO, NO 2 , O 3 , SO 2 ), Normalized Difference Vegetation Index (NDVI), and nighttime light (NTL); modeling the summer and winter temperatures using natural cubic splines with 2 or 3 degrees of freedom; and removing GDP per capita from the main model-to check the robustness of the main findings.
R software (version 3.4.3; https://www.r-project.org/) was used to perform all data analyses. The "gnm" package was used to perform the conditional Poisson regression model. The "mvmeta" package was used to compare the subgroup differences. Statistical significance was defined as a 2-sided p-value < 0.05. This study was approved by the Monash University Human Research Ethics Committee. The Brazilian Ministry of Health did not require ethics approval or informed consent for secondary analysis of aggregated anonymized data from the Mortality Information System.

Results
A total of 1,332,526 adult cancer death records from 5,565 municipalities, with municipality areas ranging from 3.56 to 159,533 km 2 , covering almost the total population of Brazil from 2010 to 2016 were included in the main analyses. Cancer death counts from common cancer sites are presented in S1 Table. Of all records included, death counts varied from 0 to 123,571 within municipalities. Mean annual wildfire-related PM 2.5 was 2.38 μg/m 3 (ranging from 0.60 to 12.49 μg/m 3 ), with regional variability ( Table 1). The distribution of wildfire-related PM 2.5 showed a radial pattern from municipalities in the Central-West Region and surrounding areas (Fig 1). The proportion of wildfire-related PM 2.5 of all-source PM 2.5 is shown in S1 Fig  The associations between a 1-μg/m 3 increase of wildfire-related PM 2.5 concentration and cancer mortality risks for single lag years and moving average lag years are shown in S3

PLOS MEDICINE
relationships between wildfire-related PM 2.5 and total cancer mortality modeled by natural cubic splines with 1-4 degrees of freedom were similar, and linear analysis had the lowest QBIC (Quasi-Bayesian Information Criteria)AU : PleasespelloutordefinetheabbreviationQBIC: , indicating linear associations between wildfirerelated PM 2.5 and total cancer mortality (S4 Fig). Compared with non-wildfire PM 2.5 , people were more vulnerable to wildfire-related PM 2.5 (Fig 2).
The RR for mortality for all cancers combined per 1-μg/m 3 increase of wildfire-related PM 2.5 concentration was 1.02 (95% CI 1.01-1.03, p < 0.001). Cancer mortality was higher for wildfire-related PM 2.5 than for other sources of PM 2 (Fig 3). However, no significant association with lung cancer mortality was observed (1.00 [95% CI 0.98-1.01], p = 0.503). RRs associated with wildfire-related PM 2.5 were greater than RRs for non-wildfire PM 2.5 for colorectal (1.03 [95% CI 1.02-1.04], p = 0.001) and testis The solid lines represent the RR, and the shaded areas represent the 95% CI. The model, AU : InthelegendtoFig2 : Pleasecheckthattheedi by its design, controlled for factors that were stable across the study period or had similar trend across geographical locations, and also adjusted for spatial-temporal factors including seasonal temperature and GDP per capita. CI, confidence interval; RR, relative risk.  The RR for mortality from skin cancers was more pronounced in males, while a higher RR for nasopharyngeal cancer mortality was observed in females. NotablyAU : Pleasecheckthattheeditstothesentence}Notably; :::}captureyou , higher risks were observed for cancers of the colon/rectum, skin, and prostate among people aged 20-59 years, and higher RR for testicular cancer mortality was observed among people aged 60 years and older (Fig 4). Subgroup analyses for other cancer sites are shown in S6 Fig. Lung  Wildfire-related PM 2.5 attributable cancer deaths ranged from 0 to 822/100,000 population for municipalities during the study period, assuming the association is causal (Fig 5). In total, there were 53,135 cancer deaths (95% CI 30,743-75,322)AU : Pleaseindicatewhatthevaluesinparenthese attributable to 2-year average wildfire-related PM 2.5 exposure from 2010 to 2016. Though the highest cancer mortality rate was in the South Region, the number of attributable cancer deaths per 100,000 population was higher in the Central-West Region (75/100,000). Males and people 60 years or older experienced a higher cancer burden. Along with AU : Pleasecheckthattheeditstothesentence}Alongwith:::}cap increased cancer cases and higher wildfire-related PM 2.5 exposure, the number of attributable cancer deaths was more pronounced in 2015 and 2016, which is consistent with the high wildfire-related PM 2.5 exposure in those years ( Table 2). State-level attributable cancer deaths are presented in S2 Table. Sensitivity analyses showed that our estimation was robust. The estimations using 2-year average (lag0-1) and three-year average (lag0-2) wildfire-related PM 2.5 concentrations were similar (S3 Fig). We also compared associations estimated by adding other air pollutants (CO, NO 2 , O 3 , SO 2 ), NDVI, and nighttime light; changing the degrees of freedom of temperature; and removing GDP per capita from the model (S3 Table). Unintentional AU : Pleasecheckthattheeditsto drowning was used as the outcome for a negative control analysis: Drowning was not significantly associated with wildfire-related PM 2.5 exposure (S4 Table).

Discussion
We did a national analysis of the association between wildfire-related PM 2.5 exposure and cancer mortality. We found that wildfire-related PM 2.5 was significantly associated with an increased risk of all-cause cancer death in Brazil during 2010-2016. Increased risks were detected for cancers of the nasopharynx, esophagus, stomach, colon/rectum, larynx, AU : Pleasecheckth skin, breast, prostate, and testis. Notably, we found that people may be more vulnerable to wildfire smoke than non-wildfire PM 2.5 sources, especially for esophageal, colorectal, and testicular cancer. To our best knowledgeAU : Pleasecheckthattheeditstothesentence}Toourbest:::}captureyourmea , this study is the first to specifically focus on associations between wildfire-related PM 2.5 and site-specific cancer mortality. The disease burden attributable to wildfire may be higher than previous estimates based on respiratory and cardiovascular diseases.
Cancer mortalityAU : Pleasecheckthattheeditstothesentence}Cancermortality:::}captureyourmeanin is an outcome reflecting both the incidence of cancer and survival after diagnosis [26]. The association between wildfire-related PM 2.5 and cancer mortality may be explained by increased cancer incidence and shortened survival. PM 2.5 is classified by the

Estimated RRs (95% CIs) for the association between a 1-μg/m 3 increase in 2-year average (lag0-1) wildfirerelated PM 2.5 and cancer mortality, from 2010-2016, by sex and age. AU : PleasecheckthattheeditstothetitleforFig4captureyourm
The vertical dashed line represents the reference line for RR = 1, helping to compare the effect estimates with the null hypothesis; the error bars represent 95% CIs. The model, by its design, controlled for factors that were stable across the study period or had similar trend across geographical locations, and also adjusted for spatial-temporal factors including seasonal temperature and GDP per capita. The p-values for differences were estimated by fixed-effects meta-analysis with no statistical adjustment, because models were based on the same sample. CI, confidence interval; RR, relative risk. https://doi.org/10.1371/journal.pmed.1004103.g004 International Agency for Research on Cancer (IARC) as a Group 1 carcinogen for sufficient evidence in increasing lung cancer risk [27]. Assuming AU : Pleasecheckthattheeditstothesentence}Assu a potentially significant wildfire-cancer incidence association, cancer patients with weakened immune systems may be more vulnerable to wildfire-related PM 2.5 , resulting in a significant mortality association [28,29]. Although the mechanisms are not clear, some other studies have provided evidence for related cancers. According to previous studies, all-source PM 2.5 exposure may shorten cancer survival [30,31]. The potential AU : Pleasecheckthattheeditstothesentence}Thepotentialmechanism:::}captureyourmeani mechanism for shortened survival may be accelerated cancer progression due to PM 2.5 inducing oxidative stress, genotoxicity, and/or inflammation [32][33][34][35][36]. PM 2.5 may enter AU : Pleasecheckthattheeditstothesentence}PM2:5mayenter:::}captureyourmeaning:Ifnot; pleaseprovid the digestive tract, altering the immune response and the gut microbiota and epithelial cells [37]. Moreover, PM 2.5 binds chemicals with endocrine-disrupting properties [38] that may be associated with cancer development and progression of hormone-sensitive cancers, such as breast and testicular cancer. Findings from an epidemiological study of breast cancer [39] and mechanistic studies of testicular cancer [40] also support the hypothesisAU : Ichangedsupporttheassump . Therefore, it is biologically plausible that exposure to wildfire-related PM 2.5 might increase mortality for cancers at different sites.
Wildfire particles are smaller than those from urban sources, and particles reaching miles away may have greater oxidative potential [41,42]. These characteristics of wildfire particles pose a significant health risk to individuals. The health impact of short-term exposure to

PLOS MEDICINE
wildfire smoke has been well documented for all-cause mortality [1,43]. In comparison with the short-term effects of wildfire smoke, far fewer studies have included longer term health impacts, and no study to our knowledge has shown an association between wildfire particles and cancer. Some studies have reported an association between all-source PM 2.5 and cancer risks [44]. Our resultsAU : Pleasecheckthattheeditstothesentence}Ourresults:::}captureyourmeaning:Ifn show a higher risk for wildfire-related PM 2.5 than for PM 2.5 from nonwildfire source, for deaths from all cancers combined. Furthermore, we might expect that cancer at different sites would show different responses to wildfire-related PM 2.5 exposure.
In addition to the short-term effect, other effects of wildfire smoke exposure on cancer risk are still unknown in the general population, but higher cancer risks were observed among wildfire firefighters who were most exposed to wildfire smoke. The American Cancer Society (ACS) Cancer Prevention Study II demonstrated that wildland firefighters have an increased risk of lung cancer mortality [45]. Studies from Australia AU : Pleasecheckthattheeditstothesentence}Stu showed possible increased risks of colorectal and prostate cancers in paid male landscape firefighters, colorectal and kidney cancer in male firefighters, and all malignancies in female landscape firefighters [46][47][48]. However, no increased cancer mortality risks were observed, which may be due to the "healthy volunteer" effect [46][47][48]. No other studies of cancer among wildfire firefighters were identified in the literature. The risks of cancer in firefighters, not limited to wildfires, are still uncertain. In line with our results, increased risks of some respiratory system, digestive tract, skin, and male reproductive cancers were reported in firefighters, but some protective associations were also observed, which again may be due to the healthy volunteer effect [49][50][51][52][53][54]. Consistent with previous studies, lung cancer was not observed to have a higher mortality risk associated with wildfire-related PM 2.5 exposure, while statistically significant associations were observed in female and older populations with non-wildfire PM 2.5 in the current study [46][47][48][49][50][51][52][53][54]. However, our previous study conducted in the same population and over a similar study period showed significant associations between all-source PM 2.5 and lung cancer mortality for all subgroups (both sexes and age groups) [55]. The increased risk of lung cancer mortality may be explained by non-wildfire sources of PM 2.5 exposure, such as industry and transportation emissions, but not wildfires. Further studies on different sources of PM 2.5 and histological subtype-specific estimation are warranted.
Another kind of exposure similar to wildfire smoke is the emission from biomass burning. The literature suggests that indoor biomass burning is associated with lung cancer risk [56,57]. Robust evidence has been provided that biomass smoke from household cooking and heating is also associated with higher risks of gastrointestinal cancers [58][59][60]. Furthermore, indoor wood-burning stoves were suggested to be associated with a modestly increased risk of breast cancer in the Sister Study from the US [61]. An increased risk of hypopharyngeal cancer was also observed among lifelong users of wood in a case-control study from India [62]. Most studies were not able to assess the effect of biomass burning separately, as participants using fossil fuels included biomass and coal as cooking and heating fuels. However, in vitro studies have demonstrated that particles emitted by wood fires have mutagenic and endocrine-disrupting capacity, providing biological support for the possible role of wildfires in the pathogenesis of hormone-sensitive cancers [63,64].
Although both AU : Pleasecheckthattheeditstothesentence}Althoughboth:::}captureyourmeaning:Ifn firefighter smoke exposure and indoor biomass burning pollution may not be directly comparable with wildfire-related PM 2.5 exposure in the general population, this literature sheds some light on cancer-specific risks. The current analysis establishes a higher risk for cancer, which means better control of wildfires is essential for cancer prevention in Brazil. In addition, individual-, community-, and national-level strategies should be considered to minimize fire exposure. However, AU : Pleasecheckthattheeditstothesentence}However:::}captureyourm the effectiveness of personal actions such as relocation, staying indoors, or wearing masks is still controversial [2]. The investigation of wildfire prediction models applied to Brazil is ongoing [65]. Warning systems for extremely hot weather implemented in many countries may also provide a warning for increased wildfire risk [66]. Systemic strategies and guidance are warranted.
Major strengths of this study include that it is the first study to estimate the association between wildfire-related PM 2.5 and cancer-specific mortality, to our best knowledge. Also, this study is based on national death records, and the large sample size allowed the estimation of associations representative for the total Brazilian population. Lastly, the variant difference-indifferences approach could adjust for most of the unmeasured confounders stable during the study time and those that changed similarly across regions.
Some study limitations should also be recognized. First, we considered only PM 2.5 exposure in the analyses, and potential joint effects of gaseous pollutants were not estimated. Our estimation of wildfire-related PM 2.5 could not capture the complex mixture of environmental pollutants released during wildfires, and thus further studies are needed to refine the exposure metrics. Additionally, some confounders related to wildfires could not be adjusted for in the model due to data unavailability, including data on firefighting foam composition. For example, AU : Pleasecheckthattheeditstothesentence}Forexample:::}captureyourmeaning:Ifnot; pleaseprovid associations between firefighting chemicals and cancer at various cancer sites have been observed in previous studies: Firefighting foam containing per-and polyfluoroalkyl substances was suggested to be related to increased risk of breast cancer, bladder cancer, etc. [67]. Second, PM 2.5 concentrations were estimated at a global scale, and regional validation was not available due to limited ground-level monitors in Brazil. Third, municipality-level exposures were used in the analyses because individual exposure data were not available. The use of aggregated data may lead to some exposure misclassification, including the inability to capture the migration of residents between municipalities. However, potential effects of migration may be limited, as more than 96% of adults had an uninterrupted time of residence in a municipality for at least 2 years according, according to the 2010 census results published by BIGS [68]. Finally, the use of registry data, rather than individual survey data, may have led to some misclassification of residential address and main cause of death, which may lead to bias in the association estimations. Some potentiallyAU : Pleasecheckthattheeditstothesentence}Somepotentiallychanging:::}capturey changing confounders that correlated with both PM 2.5 exposure and cancer mortality were ignored in the analyses, due to the limited availability of individual lifestyle data. Further, the data used in this study did not allow an assessment of competing risks, as only the primary cause of death was recorded. Cancer patients who died from other causes (e.g., heart attack) could not be included. Also, our use of 2-year average PM 2.5 concentration as the exposure may be not appropriate for cancers with a longer survival time. Overall, further cohort studies are warranted to give a more accurate risk estimate.
In summary, this study provides the first quantitative estimate of the association between wildfire-related PM 2.5 and cancer-specific mortality across Brazil. The potentially AU : Pleasecheckthatt higher risk of wildfire-related PM 2.5 compared with non-wildfire-related PM 2.5 for all cancers combined suggests that the wildfire control and systemic prevention strategies are warranted to reduce cancer mortality risk in Brazil. This could be a health co-benefit of measures to preserve the Amazon rainforest and limit climate change.
Supporting information S1 RECORD Checklist. The RECORD statement checklist of items, extended from the STROBE statement, that should be reported in observational studies using routinely collected health data. The solid lines represent the RR, and the shaded areas represent the 95% CI. The model, by its design, controlled for factors that were stable across the study period or had similar trend across geographical locations, and also adjusted for spatial-temporal factors including seasonal temperature and GDP per capita. CI, confidence interval; RR, relative risk. Estimated RRs and 95% CIs for the association between a 1-μg/m 3 increase in 2-year average (lag0−1) wildfire-related PM 2.5 and non-wildfire-related PM 2.5 and mortality from all cancers and site-specific cancers from 2010-2016. The vertical dashed line represents the reference line for RR = 1, helping to compare the effect estimates with the null hypothesis; the error bars represent 95% CIs. The model, by its design, controlled for factors that were stable across the study period or had similar trend across geographical locations, and also adjusted for spatial-temporal factors including seasonal temperature and GDP per capita. p-Values for differences were estimated by fixed-effects meta-analysis with no statistical adjustment, because models were based on the same sample. CI, confidence interval; RR, relative risk. (TIF) Estimated RRs (95% CIs) for the associations between a 1-μg/m 3 increase in 2-year average (lag0-1) wildfire-related PM 2.5 and mortality from all cancers and site-specific cancers from 2010-2016, by sex and age. The horizontal dashed line represents the reference line for RR = 1, helping to compare the effect estimates with the null hypothesis; the vertical error bars represent 95% CIs. The model, by its design, controlled for factors that were stable across the study period or had similar trend across geographical locations, and also adjusted for spatial-temporal factors including seasonal temperature and GDP per capita. CI, confidence interval; RR, relative risk. (TIF)

S7 Fig. Comparison of the estimated RRs (95% CIs) for the association between a 1-μg/m 3 increase in 2-
year average (lag0−1) wildfire-and non-wildfire-related PM 2.5 and lung cancer mortality from 2010-2016, by sex and age. The horizontal dashed line represents the reference line for RR = 1, helping to compare the effect estimates with the null hypothesis; the vertical error bars represent 95% CIs. The model, by its design, controlled for factors that were stable across the study period or had similar trend across geographical locations, and also adjusted for spatial-temporal factors including seasonal temperature and GDP per capita. CI, confidence interval; RR, relative risk. (TIF) S1