Long-Term Effect of Temperature Increase on Liver Cancer in Australia: A Bayesian Spatial Analysis

Abstract

Background: While some evidence has potentially linked climate change to carcinogenic factors, the long-term effect of climate change on liver cancer risk largely remains unclear. Objectives: Our objective is to evaluate the long-term relationship between temperature increase and liver cancer incidence in Australia. Methods: We mapped the spatial distribution of liver cancer incidence from 2001 to 2019 in Australia. A Bayesian spatial conditional autoregressive (CAR) model was used to estimate the relationships between the increase in temperature at different lags and liver cancer incidence in Australia, after controlling for chronic hepatitis B prevalence, chronic hepatitis C prevalence, and the Index of Relative Socio-economic Disadvantage. Spatial random effects obtained from the Bayesian CAR model were also mapped. Results: The research showed that the distribution of liver cancer in Australia is spatially clustered, most areas in Northern Territory and Northern Queensland have higher incidence and relative risk. The increase in temperature at the lag of 30 years was found to correlate with the increase in liver cancer incidence in Australia, with a posterior mean of 30.57 [95% Bayesian credible interval (CrI): 0.17, 58.88] for the univariate model and 29.50 (95% CrI: 1.27, 58.95) after controlling for confounders, respectively. The results were not highly credible for other lags. Discussion: Our Bayesian spatial analysis suggested a potential relationship between temperature increase and liver cancer. To our knowledge, this research marks the first attempt to assess the long-term effect of global warming on liver cancer. If the relationship is confirmed by other studies, these findings may inform the development of prevention and mitigation strategies based on climate change projections.

Headnote

Background: While some evidence has potentially linked climate change to carcinogenic factors, the long-term effect of climate change on liver cancer risk largely remains unclear.

Objectives: Our objective is to evaluate the long-term relationship between temperature increase and liver cancer incidence in Australia.

Methods: We mapped the spatial distribution of liver cancer incidence from 2001 to 2019 in Australia. A Bayesian spatial conditional autoregressive (CAR) model was used to estimate the relationships between the increase in temperature at different lags and liver cancer incidence in Australia, after controlling for chronic hepatitis B prevalence, chronic hepatitis C prevalence, and the Index of Relative Socio-economic Disadvantage. Spatial random effects obtained from the Bayesian CAR model were also mapped.

Results: The research showed that the distribution of liver cancer in Australia is spatially clustered, most areas in Northern Territory and Northern Queensland have higher incidence and relative risk. The increase in temperature at the lag of 30 years was found to correlate with the increase in liver cancer incidence in Australia, with a posterior mean of 30.57 [95% Bayesian credible interval (CrI): 0.17, 58.88] for the univariate model and 29.50 (95% CrI: 1.27, 58.95) after controlling for confounders, respectively. The results were not highly credible for other lags.

Discussion: Our Bayesian spatial analysis suggested a potential relationship between temperature increase and liver cancer. To our knowledge, this research marks the first attempt to assess the long-term effect of global warming on liver cancer. If the relationship is confirmed by other studies, these findings may inform the development of prevention and mitigation strategies based on climate change projections. https://doi.org/10.1289/EHP14574

Introduction

Liver cancer is a disease with a heavy burden on patients. Global Burden of Disease Study 2019 (GBD 2019) reported that there were 534,000 new cases and 485,000 deaths from liver cancer in 2019, with the fifth-ranking disability-adjusted life years (DALYs) among all cancers.1 Although the incidence and mortality of liver cancer were relatively stable, the absolute case number showed a significant increase, with a 27.1% and 25.6% increase, respectively.1 The increase is most evident in some countries with a high sociodemographic index (SDI). In Australia, the age-standardized incidence (ASIR) of liver cancer dramatically increased by an estimated 389% from 1971 to 2022 and the age-standardized mortality (ASMR) by 421% from 1971 to 2022.2 Improving medical diagnostic and treatment technologies have not led to an increase in common cancers in Australia, such as lung cancer.2 Therefore, this trend is concerning in light of the general decline in other types of cancer in Australia.

The causes of liver cancer are many, including hepatitis viruses, alcohol, and metabolic disease.3 Effective management of selectedmodifiable factors can reduce liver cancer risk, including alcohol regulation and pricing,4 vaccination for oncogenic viruses,5 and balanced diets.6 Indeed in Australia, positive measures have led to a significant decline in viral hepatitis and alcohol.7'8 This suggests that the increasing incidence of liver cancer cannot be explained by well-understood modifiable factors. More research is needed on other potential causal factors, such as environmental change. The impact of environmental factors on cancer is widely recognized, for instance, lung, breast, and colorectal cancers are associated with air pollution.9

Climate change is a dominant global environmental concern. Climate change refers to the shifts in climate and weather patterns over long periods, in which temperature patterns are one of the most sensitive indicators.10 Compared to the mid-to-late 19th century, global temperature has risen by 1.09 C.11 The latest synthesis report by the Intergovernmental Panel on Climate Change (IPCC) concluded that accumulated greenhouse gas emissions have made the goal of limiting global warming to below 1.5 C within the 21st century unattainable, and 2 C average warming may also be exceeded.11 Without immediate effective actions, the climate will continue to warm in the foreseeable future, and no region in the world will be spared.11

Climate change is known to affect health in several ways, such as through its impacts on infectious diseases.12 The impact of climate change on cancer has sparked discussions among researchers, as it could increase cancer risk through many pathways including water safety, food security, air pollution, and other carcinogenic factors.13 The National Cancer Institute in the United States has also included this research on its agenda and urgently called for more research to focus on climate change and cancer.14 The exploration of the long-term relationship between global warming and cancer is still in its infancy.

Tueller et al. concluded that liver cancer mortality is higher in humid and hot climates.15 However, this is a global scale study that ignores the climatic differences within each country. In addition, our previous study indicated that liver cancer risk in Australia exhibited a higher geographical concentration in equatorial and tropical climate zones, suggesting that climatic factorsmay play a role in elevated liver cancer incidence.16 This exploratory analysis compared the risk of liver cancer among different climate zones representing complex climate gradients but did not assess the long-term effect of temperature increase and liver cancer. The previous study only observed the overlap between the distribution of climate zones and liver cancer to offer initial insights into a potential link between the two. Modeling data and nonparametric analysis were used,16 but the analyses fell short of providing definitive, robust data and models to quantify any long-term impact of temperature increase on liver cancer.

Here, we consider the potential impact of temperature increase over a nearly 30-year period on liver cancer and examine the possible long-term effects of global warming on liver cancer in Australia.

Methods

Data Collection

Data on liver cancer cases from 2001 to 2019 were collected from the Australian Cancer Database (ACD) released by the Australian Government Department of Health (AIHW).17 Cancer is a notifiable disease in Australia. Australasian Association of Cancer Registries received cancer information from relevant institutions (e.g., hospitals, pathology laboratories, and registries of births, deaths, and marriages). Cancer registries provide AIHW with demographic information, cancer site, date of diagnosis, date of death and other relevant data annually.17 AIHW then performs data cleansing and quality control processes, including coding adjustments, error correction, and de-duplication, to generate the ACD.17 ACD includes cancer site, count, age-standardized rate, and the International Classification of Diseases 10th version (ICD-10). Liver cancer was identified using ICD-10 code C22 (malignant neoplasm of liver and intrahepatic bile ducts). Australian liver cancer data were aggregated at the 2016 Statistical Area Level 3 (S A3) level. S A3 areas are typically designed for populations in the range of 30,000 to 130,000 people, and predominantly represent functional areas of regional towns and cities or clusters of suburbs within major urban areas.18 There are a total of 358 SA3 regions across Australia, with 18 nonspatial special codes present. Due to potential identifiability issues associated with small data cells, the data of Australian Capital Territory and Tasmania were not available in this study, but the remaining six Australian states and territories were included, totaling 311 S A3 regions.

Population data from 2001 to 2019 at SA3 level were obtained from the Australian Bureau of Statistics (ABS).19 We used the ABS recommended standard to calculate the ASIR of liver cancer based on 30 June 2001.20 This approach provided a consistent and reliable baseline for age standardization across all study years, enabling us to analyze liver cancer trends over time effectively, free from the influence of demographic changes. Covariates included the Index of Relative Socio-economic Disadvantage (IRSD), chronic hepatitis B (CHB), and chronic hepatitis C (CHC) prevalence, which were included because nearly 70% of liver cancer cases could be attributed to these two types of hepatitis.3 To control for the effect of socioeconomic factors, the Index of Relative Socio-economic Disadvantage from the Socio-Economic Indexes for Areas (SEIFA) in 201621 was employed in the model examining the relationship between temperature and liver cancer incidence. The Index of Relative Socio-economic Disadvantage for each SA3 was derived from the average Index of Relative Socio-economic Disadvantage of the Statistical Area Level 2 (SA2), which are designed to represent functional areas with populations typically between 3,000 and 25,000 people, including regional towns, urban suburbs, and rural localities.22 The Index of Relative Socio-economic Disadvantage serves to comprehensively represent the socioeconomic status across different regions, encompassing factors such as income,housing, education, and employment.21 Additionally, it has the potential to mirror key health indicators, including accessibility to health care services.23 Due to significant variations in the Index of Relative Socio-economic Disadvantage values across different regions, we standardized the index using Z scores.

Due to data accessibility, data were only available for chronic hepatitis B and chronic hepatitis C at SA3 level in 2016. Chronic hepatitis B and chronic hepatitis C data were collected from the Australasian Society for HIV, Viral Hepatitis, and Sexual Health Medicine (ASHM).24'25 Particulate matter less than 2.5 um in aerodynamic diameter (PM2.5) data at SA3 level from 2004 to 2019 was obtained from European Centre for Medium-Range Weather Forecasts (ECMWF) Atmospheric Composition Reanalysis 4 (EAC4).26

Temperature data ( C) were sourced from 1,541 weather stations from 1971 to 2009 from the Bureau of Meteorology (BOM), Australia.27 These stations are distributed throughout Australia. Except for a few SA3 regions with very small geographical areas, the majority of SA3 regions have 1-37 stations. The distance between the nearest two stations is about 0-379.5 kilometers. Temperature data are derived from station observations, with daily maximum and minimum temperatures. The mean temperature was determined by taking the mean of these maximum and minimum temperatures, and the annual mean temperature was calculated by averaging the daily mean temperature throughout the year. Then the mean temperature derived from stations was converted to the S A3 level. Historically, Australian weather stations predominantly utilized liquid-in-glass thermometers installed on Stevenson screens for observation, and since the 1990s, resistance temperature detectors have been used for automatic weather stations to record temperatures. These observation stations undergo routine inspections for performance evaluations, maintenance, and training initiatives. Additionally, BOM implements a comprehensive quality control regimen, with procedures to check data consistency, flat-lining, outliers, etc. of temperature data.28

Among 311 SA3 areas, we excluded a very small amount of missing data (0.64%) and the presence of geographic islands because they made it impossible to create spatial weights and conduct spatial analysis. Ultimately, we included 304 SA3 for spatial analysis. For these areas, data on liver cancer, temperature, and Index of Relative Socio-economic Disadvantage were fully available. S A3s missing chronic hepatitis B and chronic hepatitis C data were interpolated using the data from their respective Primary Health Networks. All publicly available data sources and links are shown in Table S1.

The study was approved by the University Human Research Ethics Committee of Queensland University of Technology (approved number 5913). The requirement for human subject informed consent was waived for this study because the individual data has been grouped to be less specific to minimize the risk of reidentification to enable sufficient protection of privacy. Thus, the release of these data for this analysis is not prohibited by law and is permitted in the applicable legislation.

Statistical Analysis

Average ASIR from 2001 to 2019 at the SA3 level was calculated to explore the spatial distribution of liver cancer in Australiausing the following equation: ASIR= Ym = i (w) x^'- Where O;is the number of incident liver cancer cases observed in age group i, Nt is the total population in age group i, Pt is the proportion of the standard population in the corresponding age group i, and n is the number of age groups.

We also applied global Moran's / to test the global spatial autocorrelation of liver cancer. Moran's / is a rational numbernormalized to a value between -1.0 and 1.0. The closer Moran's / is to 1, the more significant the spatial correlation is. Local Moran's / (LISA) map is used to identify the spatial clusters of liver cancer incidence in Australia. LISA calculates local spatial autocorrelation for each area based on the global Moran's / model to test the significance of clustering areas, thereby identifying spatial hotspots or coldspots.29 Clusters detected by LISA map are usually categorized into five types, as follows: the area consistently showing high values along with its neighboring areas (high-high cluster); the areas with high values surrounded by neighboring areas with low values (high-low outlier); the areas with low values surrounded by neighboring areas with high values (low-high outlier); the areas with consistently low values along with its neighboring areas (low-low cluster); no statistically significant clustering detected (not significant). To visualize the spatial clustering of liver cancer from a statistical significance perspective, retrospective discrete Poisson space-time scan statistics were conducted to report the distribution of relative risk (RR) of liver cancer in Australia. These statistics evaluate the statistical significance of liver cancer incidence in each region, allowing for the accurate identification of areas where the risk of liver cancer is significantly higher than average.30 Ordinary Kriging method was employed to interpolate average ASIRs and RRs on a continuous spatial field. This method is widely utilized in disease spatial interpolation, aiming to mitigate area-related biases and ensure data smoothing.31 To calculate the average temperature of the SA3 regions from 1971 to 2009, we employed spatial ordinary Kriging interpolation31 to smooth the temperature data obtained from all of the stations in each year and converted them into a continuous spatial raster surface. We utilized the spatial zonal statistics function32 in ArcGIS to morph the continuous spatial raster surface into polygons at the SA3 level, completing the transformation of temperature data from point data to polygon data at the SA3 level.

To assess the impact of temperature increase on liver cancer incidence in Australia, we first employed Joinpoint regression to determine the average annual percentage changes (AAPC) in liver cancer incidence within each S A3 region from 2001 to 2019 to represent these average trends during the period.33 Subsequently, for each region, we performed linear regression models where annual average temperatures served as the dependent variable, and time (year) was the independent variable (model 1). This analysis was conducted across predefined lag periods to account for the potential lag effects of temperature changes on liver cancer in the Bayesian conditional autoregressive (CAR) model (described below). We extracted the temperature slopes for each SA3 region to calculate the estimated annual percentage change (EAPC) for the corresponding area from each regression model (model 2). The formula is given by the following:in which, P0 is the intercept of regression model, P: is the slope (coefficient), and is the error term. These EAPC values represent indicators of temperature increase, which were utilized in spatial models (described below) to explore the association between the extent of temperature increase and liver cancer incidence across different regions in Australia.

Our study included lag 30 (1971 to 1989), lag 20 (1981 to 1999), lag 15 (1986 to 2004), and lag 10 (1991 to 2009), aligning with the established latency window of 10-30 years before diagnosis for environmental factors' effects on cancer.34 This range was chosen to capture significant environmental impacts on liver cancer within a timeframe supported by most cohort studies,34 while also considering the quality and reliability of historical environmental data. Furthermore, for lags extending beyond 20 years, we opted to include only an additional 10-year interval (i.e., lag 30) to investigate potential impacts over a more extended timeframe. Excessively long lags (over 30 years), alongside the robustness of historical environmental data, may introduce more uncontrollable variables35 (e.g., changes in socioeconomics, corresponding policies, and characteristics of the population) that would impact the stability of the analysis.

We then used the AAPC of liver cancer as the dependent variable and the EAPC of temperature at different lag periods as the independent variable, while adjusting for chronic hepatitis B, chronic hepatitis C, and Index of Relative Socio-economic Disadvantage detected by directed acyclic graph (DAG)36 (Figure S1) to construct a Bayesian spatial conditional autoregressive model with Gaussian family link.

The Bayesian CAR model enhances linear regression models to address spatial autocorrelation issues by incorporating CAR priors after considering the effects of covariates on regression.37 Markov chain Monte-Carlo (MCMC) simulation was used for the inference of models, including means, standard deviations, and credible intervals (Crls). The Bayesian framework is conducted by selecting a prior distribution of fixed and random effects. In our study, we utilized the default prior settings in the CARBayes package, specified as c (1, 0.01).37 The package has been widely used in spatial epidemiology, especially in cancer research, where the choice of priors is reliable and robust for model outcomes.38 The formulation is given by in which \ik is AAPCs of liver cancer incidence in location k; v2 is a scale parameter for Gaussian family; $x is the coefficient for EAPC of temperature, P2 is the coefficient for chronic hepatitis B, P3 is the coefficient for chronic hepatitis C, and P4 is the coefficient for the Index of Relative Socio-economic Disadvantage; (pk is the spatial random effect term; Wis the spatial weight matrix; x2 is the variance parameter of random effects; and p is a spatial dependence parameter. The posterior mean, corresponding to 95% Crls and deviance information criterion (DIC) were calculated by the Bayesian CAR model. Similar to interpreting coefficients in fre-quentist approaches, the posterior mean in the Bayesian CAR model serves as the estimator that assesses the relationship between independent and dependent variables. A negative posterior mean suggests a negative relationship, whereas a positive one indicates a positive relationship. This is considered highly credible only when the 95% CrI of the posterior mean does not contain zero.39 For instance, if the posterior mean is 0.05, it implies that for each 1% increase in the magnitude of temperature change, the magnitude of change in liver cancer incidence would increase by 0.05%. In addition, we extracted the posterior mean for every SA3 region for mapping to demonstrate the risk of liver cancer incidence after controlling for variables in the Bayesian CAR model.

We also mapped the spatial random effects and spatial clusters of spatial random effects by Anselin local Moran's / statistic (LISA) obtained from the Bayesian CAR model. Spatial random effects are typically divided into two categories.40 Structured spatial random effects reflect the correlation among observed data due to the proximity of their geographic locations. Unstructured spatial random effects capture the random variations attributed to factors or variables not explicitly accounted for in the model.

Sensitivity Analysis

ABS releases SEIFA every 5 years. It underwent a significant change in its geographic framework after 2011.21 We adapted to changes in the SEIFA geographic classification by using the 2016 Index of Relative Socio-economic Disadvantage for precise matching with liver cancer incidence data. This adaptation was necessary due to the shift from the Australian Standard Geographical Classification (ASGC) to the Australian Statistical Geography Standard (ASGS), which introduced and then altered SA3 regions between the 2011 and 2016 SEIFA editions. To ensure the robustness of results, we conducted a sensitivity analysis by fuzzy-matching the 2011 Index of Relative Socio-economic Disadvantage data to the 2016 S A3 definitions and incorporating it into the Bayesian CAR model.

In addition, although still controversial, several studies found that PM2.5 may be associated with liver cancer incidence.41 To control for the potential bias of PM2.5 on the relationship between temperature increase and liver cancer, average PM2.5 from 2004 to 2019 at the SA3 level was included in the Bayesian CAR model. Due to the unavailability of earlier air pollution data, a lag effect could not be considered. Therefore, the results are shown as a sensitivity analysis to evaluate the robustness of the main results.

The ASIR and EAPC were calculated by R (version 4.3.1; R Development Core Team). The process of Kriging interpolation, zonal function analysis, spatial autocorrelation testing, and map generation were carried out using ArcGIS Pro version 3.1.2. Space-time scan statistics were conducted by SaTScan. Joinpoint regression analysis was executed using Joinpoint Regression Program version 5.0.2. The Bayesian CAR model was performed by R version 4.3.1 using the CARBayes package.

Results

Table 1 presents summary statistics for included variables. ASIR of liver cancer in Australia from 2001 to 2019 was 6.71 per 100,000 persons. Average temperatures range from 17.80 C for the years 1971 to 1989 up to 18.36 C from 1986 to 2004. In 2016, the prevalence of chronic hepatitis B and chronic hepatitis C was 0.93% and 0.99%, respectively. The average Index of Relative Socio-economic Disadvantage for the years 2011 and 2016 were similar, at 997.58 and 996.44, respectively. Additionally, the concentration of PM2.5 was7.90 ug/m3 for the period 2004 to 2019.

Liver Cancer Incidence in Australia

Figure 1 shows the trend of liver cancer incidence and the spatial distribution of average ASIR by ordinary Kriging interpolation from 2001 to 2019. During this period, liver cancer incidence has a persistent upward trend, escalating from 4.58 to 8.04 per 100,000 persons. Specifically, the rate for males surged from 6.56 to 11.22 per 100,000 persons, while for females it rose from 2.67 to 4.82 per 100,000 persons. There is geographic variation in the distribution of liver cancer in Australia, with high incidence concentrated in the Northern Territory and North Queensland, especially in the Northern Territory where almost all SA3s have incidence rates ranging from 14.56 to 31.27 per 100,000 persons. Of the nine SA3 areas in the Northern Territory, most regions of the four SA3s, East Arnhem, Daly Tiwi-West Arnhem, Katherine, and Barkly, had ASIR above 20.72 per 100,000 persons. Global spatial autoregression analyses reflected the spatial clustering of ASIR in Australia, with Moran's / value of 0.28 (Z score was 13.62, p-value< 0.001). The results of space-time scan statistics were similar to the distribution of ASIR, with the Northern Territory and Northern Queensland high-risk areas from 2001 to 2019. In the eastern part of New South Wales adjacent to Sydney, the RRs of some SA3 ranged from 1.22 to 4.57, and some high-risk areas were found in Victoria (Figure 2). Moreover, Figure S2 shows that the regions with the highest increase in liver cancer incidence in Australia from 2001 to 2019 were mainly located in Northern Territory, northern Queensland, and parts of Western Australia, with AAPC ranging from 54.70% to 82.20%.

Temperature in Australia

As shown in Figure 3, the distribution of average temperatures over the four lag periods in Australia was similar. Across all lag periods, higher average temperatures were found in northern Australia, mainly in northern Western Australia, the Northern Territory, and North Queensland. The variations in the maximum and minimum average temperatures across the four lag periods were minimal. Specifically, the lowest average temperatures ranged from 10.46 C to 11.3FC, mainly in New South Wales and Victoria, and the highest average temperatures ranged from 26.13 C to 28.29 C. In addition, Figure S3 illustrates that during all lag periods, temperatures in most parts of Australia showed an increase (EAPC >0), but the areas with the largest overall increase gradually shifted from the northern to the southern part.

Bayesian CAR Models

Table 2 indicated that using the Bayesian CAR model, a positive correlation was found between the average temperature increase with a lag of 30 years, from 1971 to 1989, and the increase in liver cancer incidence between 2001 and 2019. The posterior mean between the EAPC of temperature over the lag 30 years and the AAPC of liver cancer incidence was 30.57, with a 95% CrI of 0.17 to 58.88, excluding 0, suggesting a credible result. After adjusting for the effects of chronic hepatitis B, chronic hepatitis C, and Index of Relative Socio-economic Disadvantage (posterior mean: 29.50; 95% CrI: 1.27, 58.95), this result remained credible, while exhibiting a smaller DIC in comparison to the univariate model (DIC = 3,161.93). However, this association was not found at the other lags, and although the posterior mean was 7.40 for univariate model and 11.12 for multivariate model at lag 20, it was not a highly credible result because the 95% CrI included 0. In sensitive analyses (Table S2), results were still stable when the Index of Relative Socio-economic Disadvantage 2011 and PM2.5 were included as covariables.

The posterior mean of liver cancer incidence calculated by the Bayesian CAR model is shown in Figure 4. Different from the distribution of ASIR, the higher posterior mean primarily spanned eastern Queensland and eastern New South Wales, ranging from 41.99 to 49.56. Figure 5 highlights that the distribution of total spatial random effects in the model is predominantly concentrated in eastern New South Wales, Western Australia, and eastern Queensland. The LISA map of spatial random effects, shown in Figure 6, also indicated high-high clustering in 16 SA3 regions in New South Wales and seven SA3 regions in southern Western Australia.

Discussion

In this study, we observed spatial clustering in the distribution of liver cancer incidence in Australia, showing some consistency with temperature distribution. Particularly in the northern regions of Australia (such as the Northern Territory and North Queensland), areas with higher cancer incidence corresponded to higher temperatures. As the lag period decreased, we observed a shift in the highest temperature increases from Australia's northern to southern regions. The spatial distribution of temperature changes at lag 30 aligned with the regional trends observed in the rise of liver cancer incidence. Further utilization of Bayesian CAR model quantified the relationship between increases in temperature at different lag periods and increases in incidence rates. The results indicated a positive correlation between temperature increase at a lag of 30 years and the increase in incidence. This study found that, in terms of incidence or relative risks, the northern regions of Australia have more areas with high-risk areas. This aligns with previous findings conducted at the state and territory levels in Australia, where the Northern Territory stands out as having the highest incidence of liver cancer in the country.42 However, finer spatial resolutions had not been applied in previous research, limiting the ability to precisely identify high-risk communities and populations.

The northern regions of Australia mainly fall within the equatorial and tropical zones. The higher incidence of liver cancer in this area suggested a potential association between temperature and liver cancer, with a positive correlation between an increase in temperature and liver cancer. Our results support a study by Tian et al. based on GLOBOCAN data reported an association between the incidence of liver cancer and a hot climate.43 In addition, no credible association with liver cancer incidence was observed at lag 20, 15, and 10 years, which corresponds to the latency period for liver cancer development. It is widely accepted that there is a latency period of several years to several decades between environmental exposure and cancer development,44 with inconsistent latency duration based on different environmental factors. Our findings imply there may be a long-term cumulative effect of temperature increases on the development of liver cancer. This extended latency period could be attributed to the indirect effects of climate change on liver cancer through impacts on factors such as behaviors, diet, and carcinogens, which then affect the incidence of liver cancer. Although few studies explicitly determined the impact of climate change on these risk factors, it is anticipated they also take years to affect cancer outcomes.45 Moreover, although we did not find a highly credible association between temperature increase and liver cancer in shorter lag periods [i.e., posterior mean at lag of 10 years (95% CrI): -18.43 (-37.57, 0.13)], a trend of gradually decreasing posterior mean and negative effect values was observed (Table 2). Also, in sensitive analysis, a negative association was shown at lag 10 when PM2.5 was introduced to the model as a confounder. Like the paradox caused by the reduction of susceptible populations in the harvesting effect of short-term environmental exposures,46 cancer occurrence is affected by the long-term cumulative effect. Prior to reaching the onset threshold, the population boasts a higher percentage of healthier individuals, which contributes to a trend of decreasing posterior mean and even the paradox of protective associations.

Even after considering the impact of various factors in the model, there remains a significant spatial heterogeneity in the posterior mean distribution of liver cancer, primarily concentrated in the eastern parts of Queensland and New South Wales. This result was further supported by the distribution and clustering patterns of spatial random effects. Although the Bayesian CAR model accounted for the effects of important factors such as temperature increase, chronic hepatitis B, chronic hepatitis C, and Index of Relative Socio-economic Disadvantage, the distribution patterns of the posterior mean and spatial random effects suggested that there remain unmeasured or unknown factors affecting the heterogeneity of liver cancer distribution. These factors may include variables linked to liver cancer risk, such as alcohol consumption,47 obesity,48 and nonalcoholic fatty liver disease (NAFLD).49 The LISA map detected significant clusters of spatial random effects in eastern New South Wales, indicating potential unmeasured risk factors in this region. A cohort study conducted in New South Wales, Australia, involving 226,162 participants showed that alcohol use increased the risk of liver cancer by 22%, ranking highest among all cancers.50 In addition, the Primary Health Network area with the highest obesity rate in Australia is also located in New South Wales.51 These may potentially explain the clustering of spatial random effects within this area. Unfortunately, due to data limitations, the impact of these factors cannot be thoroughly assessed within the model. More research is needed to identify these unknown risk factors.

Several mechanisms could explain the relationship between temperature increase and liver cancer, including the impact of climate change on metabolic diseases related to liver cancer. Climate change may increase the risk of obesity, diabetes, and other metabolic diseases.52 Climate change also could influence dietary patterns by affecting the supply and storage of food systems, leading to changes in the quality of nutrient intake and impacting body metabolism.53 These metabolic diseases or factors have been proven to be significantly associated with the risk of liver cancer. Regrettably, few studies quantify and confirm the impact of climate change on the aforementioned important metabolic diseases, and more research is needed to focus on this issue to accurately evaluate these carcinogenic pathways.

Aflatoxin may play a crucial role. Aflatoxin is produced by Aspergillus flavus and mainly contaminates crops such as corn, peanuts, and rice, affecting the entire process of crop growth and storage.54 The International Agency for Research on Cancer classifies aflatoxin as a group 1 carcinogen, with the liver being its primary target organ.55 It is estimated that up to 28.2% of hepatocellular carcinoma can be attributed to aflatoxin.56 The synergistic effects of aflatoxin with major drivers (i.e., HBV and hepatitis C) are greater than the independent effect of each factor57; such additional exposure to aflatoxin in patients with viral hepatitis will significantly increase the risk of liver cancer. Aflatoxin contamination varies with humidity at different stages, with preharv-est drought exacerbating contamination and postharvest storage in wet conditions even more dangerous.58 The temperature range suitable for the growth of aflatoxin-producing Aspergillus is consistent with a minimum temperature of 25 C,59 such as found in subtropical and tropical regions. Climate change is expanding the areas suitable for the growth of Aspergillus flavus, leading to the emergence of this fungus in new areas.60 With global warming, the impact of aflatoxin and other mycotoxins on crops has reached 60%-80%.61 This kind of impact poses a potential threat to health, and although it may be alleviated through some measures, due to the stability of fungal toxins, it is difficult to eliminate them in food or feed processing.61 The distribution of aflatoxin in Australia is typically centered in the tropical and subtropical regions of the north and east.62 However, with the expected shift in the optimal area for Aspergillus flavus from tropical to temperate regions due to climate change,62 there may be further effects of aflatoxin on liver cancer.56 Based on current evidence, it is challenging to draw positive conclusions about the global or regional levels of aflatoxin, therefore making it uncertain whether routine food safety inspection measures, such as food sampling, can effectively mitigate changes in aflatoxin levels under future climatic changes. However, this should also serve as a wake-up call for the importance of safeguarding our food. Equally crucial is public health education about the myco-toxin, alongside the ongoing enforcement of rigorous food safety regulations.

Strengths and Limitations

Our study used the latest liver cancer data up to 2019 in Australia. Due to the well-established cancer reporting system in Australia, the data used in this study is of high quality and representative and is at a finer scale than previous studies.16 To our knowledge, this is the first study to quantify the long-term effects of temperature increase, an important consequence of climate change, and liver cancer in Australia. For the exploration of this potential association, we employed a robust and reliable model, the Bayesian CAR model, which showed a potential association after accounting for relevant confounders.

This analysis has some limitations. First, due to the lack of data on other risk factors at a finer spatial resolution, such as alcohol consumption, obesity, etc., this study could not fully consider all uncertainties within cluster random effects (e.g., Sydney). Nonetheless, from a causal inference perspective, these factors should be considered as mediating factors rather than confounders. Although alcohol and tobacco (cultivation and production) may also contribute to climate change, liver cancer is more directly linked to consumption. Future research should focus on mediators to evaluate their mediation effects or effects modification. Chronic hepatitis B and C are responsible for more than half of liver cancer, suggesting that a substantial portion of the confounding effect was controlled for in the Bayesian CAR model. Therefore, it is possible that our findings reflect a potential correlation between temperature and liver cancer. In addition, certain chemical pollutants, such as vinyl chloride, pesticides, and heavy metals, can also increase the risk of liver cancer.63-65 Climate change may exacerbate human exposure to these chemicals,66 further elevating liver cancer risks. Future research should comprehensively explore the interaction and mediation effects between these chemical pollutants and climate change on liver cancer incidence from multiple perspectives, including both ecological and individual levels.

Second, we used the 2016 Index of Relative Socio-economic Disadvantage to control for the potential effects of socioeconomic factors, which may have changed in some regions during the study period. Due to updates in geographical zoning standards, we could not use the Index of Relative Socio-economic Disadvantage covering the entire study period. Results remained stable when using the Index of Relative Socio-economic Disadvantage in 2011 for sensitivity analyses.

Furthermore, although our study was intended to focus on assessing the effects of climate change on cancer, climate change is a comprehensive and complex phenomenon that encompasses long-term shifts in various environmental factors, such as rainfall, drought, evaporation, solar radiation, and humidity. However, our study only assessed the effects of temperature increase over certain periods, we did not control confounders such as humidity because it may affect the interpretation of the effects of temperature on health,67 due to potential interactive effects.68 Such measures should be used with caution, and as an alternative, the effect modification of humidity should be analyzed rather than simply controlled for68; however, our study was limited by the type of data (merged data) and model which are not ideal for that type of analysis.69 Future studies should adopt a more holistic approach to examine the aggregate effects of changing weather variables, including interactions between variables. Given the well-understood shortcomings of ecological research, future studies should use individual-level data to confirm the observed relationships.

In summary, this exploratory analysis suggested a potential relationship between global warming and liver cancer. Our findings indicated that there is a certain parallel in the distribution of liver cancer and temperature in Australia. Furthermore, temperature increase had a positive correlation with incidence of liver cancer over longer lag periods, while short lag periods found noassociation. In the future, more studies based on large sample data are needed to quantify the association between climate change and liver cancer using more precise study designs and considering relevant carcinogenic pathways. Acknowledgments We thank the Australian Institute of Health and Welfare and the population-based cancer registries of New South Wales, Victoria, Queensland, Western Australia, South Australia, Tasmania, the Australian Capital Territory, and the Northern Territory for the provision of data from the Australian Cancer Database. We also thank the Bureau of Meteorology for providing weather observation data and the Australian Bureau of Statistics for accessing population and socioeconomic data. Ting Gan is supported by the China Scholarship Council Postgraduate Scholarship (CSC), the Queensland University of Technology Higher Degree Research Tuition Fee Sponsorship and Queensland University of Technology CSC Top-Up Scholarship.

References

References 1. Kocarnik JM, Compton K, Dean FE, Fu W, Gaw BL, Harvey JD, et al. 2022. Cancer incidence, mortality, years of life lost, years lived with disability, and disability-adjusted life years for 29 cancer groups from 2010 to 2019: a systematic analysis for the global burden of disease study 2019. JAMA Oncol 8(3):420-444, PMID: 34967848, https://doi.org/10.1001/jamaoncol.2021.6987. 2. AIHW (Australian Institute of Health and Welfare). 2021. Cancer in Australia 2021. https://www.aihw.gov.au/reports/cancer/cancer-in-australia-2021/summary [accessed 3 September 2023].

3. Liu Y, Zheng J, Hao J, Wang RR, Liu X, Gu P, et al. 2022. Global burden of primary liver cancer by five etiologies and global prediction by 2035 based on global burden of disease study 2019. Cancer Med 11(5):1310 1323, PMID: 35118819, https://doi.org/10.1002/cam4.4551.

4. Jiang H, Livingston M, Room R, Gan Y, English D, Chenhall R. 2019. Can public health policies on alcohol and tobacco reduce a cancer epidemic? Australia's experience. BMC Med 17(1)213, https://doi.org/10.1186/s12916-019-1453-z. 5. Finn 0J. 2018. The dawn of vaccines for cancer prevention. Nat Rev Immunol 18(3):183-194, PMID: 29279613, https://doi.org/10.1038/nri.2017.140.

6. Bradbury KE, Murphy N, Key TJ. 2020. Diet and colorectal cancer in UKbiobank: a prospective study. Int J Epidemiol 49(1)246-258, PMID: 30993317, https://doi.org/ 10.1093/ije/dyz064. 7. Kirby Institute. 2018. HIV, Viral Hepatitis and Sexually Transmissible Infections in Australia: Annual Surveillance Report 2018. https://www.kirby.unsw.edu.au/ research/reports/asr2018 [accessed 1 May 2020].

8. AIHW. 2024. Alcohol, Tobacco and Other Drugs in Australia, https://www.aihw. gov.au/reports/alcohol/alcohol-tobacco-other-drugs-australia/conte nts/data-by-region/alcohol-consumption [accessed 1 June 2022]. 9. Turner MC, Andersen ZJ, Baccarelli A, Diver WR, Gapstur SM, Pope CA, et al. 2020. Outdoor air pollution and cancer: an overview of the current evidence and public health recommendations. CA: A Cancer Journal for Clinicians 70(6):460-479, https://doi.org/10.3322/caac.21632.

10. NASA (National Aeronautics and Space Administration). 2024. What Is Climate Change? https://science.nasa.gov/climate-change/what-is-climate-change/ [accessed 26 March 2024]. 11. IPCC (Intergovernmental Panel on Climate Change). 2023. Synthesis Report of the IPCC Sixth Assessment Report (AR6). https://www.ipcc.ch/report/ar6/syr/ [accessed 1 March 2023].

12. MoraC, McKenzieT, Gaw IM, Dean JM, von Hammerstein H, KnudsonTA, etal. 2022. Over half of known human pathogenic diseases can be aggravated by climate change. Nat Clim Chang 12(9):869-875, PMID: 35968032, https://doi.org/10. 103S/S41558-022-01426-1. 13. Hiatt RA, Beyeler N. 2020. Cancer and climate change. Lancet Oncol 21(11): e519-e527, PMID: 33152311, https://doi.org/10.1016/S1470-2045(20)30448-4.

14. Winstead E. 2023. Cancer and Climate Change: the Health Threats of Unnatural Disasters, https://www.cancer.gov/news-events/cancer-currents-blog/2023/ cancer-climate-change-impact#:~:text=The%20biggest%20climate%20change %2Drelated,DellaValle [accessed 5 December 2023]. 15. Tueller G, Kerry R, Young SG. 2023. Spatial investigation of the links between aflatoxins legislation, climate, and liver cancer at the global scale. Spat Spatiotemporal Epidemiol 46:100592, PMID: 37500231, https://doi.Org/10.1016/j. sste.2023.100592.

16. GanT, Bam brie kH, Ebi KL, Hu W. 2023. Does global warming increase the risk of liver cancer in Australia? Perspectives based on spatial variability. Sci

Total Environ 859(pt 2):160412, PMID: 36427742, https://doi.Org/10.1016/j. scitotenv.2022.160412.

17. AIHW. 2024. Australian Cancer Database (ACD). https://www.aihw.gov.au/ about-our-data/our-data-collections/australian-cancer-database [accessed 6 December 2023].

18. ABS (Australian Bureau of Statistics). 2021. Statistical Area Level 3. https:// www.abs.gov.au/statistics/standards/australian-statistical-geography-standard-asgs-edition-3/jul2021-jun2026/main-structure-and-greater-capital-city-statistical-areas/statistical-area-level-3 [accessed 4 July 2024].

19. ABS. 2024. Regional Population by Age and Sex. https://www.abs.gov.au/statistics/ people/population/regional-population-age-and-sex/latest-release#data-downloads [accessed 28 March 2024].

20. ABS. 2013. Which Population to Use for Age Standardisation? https://www.abs. gov.au/ausstats/[email protected]/Lookup/3101 .OFeature+Articlel Mar%202013 [accessed 28 March 2024].

21. ABS. 2023. Socio-Economic Indexes for Areas (SEIFA/, Australia. https://www. abs.gov.au/statistics/people/people-and-communities/socio-economic-indexes-areas-seifa-australia/latest-release [accessed 15 March 2024].

22. ABS. 2021. Statistical Area Level 2. https://www.abs.gov.au/statistics/standards/ australian-statistical-geography-standard-asgs-edition-3/jul2021-jun2026/main-structure-and-greater-capital-city-statistical-areas/statistical-area-level-2 [accessed 4 July 2024]. 23. AIHW. 2024. Health Across Socioeconomic Groups, https://www.aihw.gov.au/ reports/australias-health/social-determinants-of-health [accessed 15 March 2024].

24. MacLachlan J, Thomas L, Cowie B, Allard N. 2016. Hepatitis B Mapping Project: Estimates of Geographic Diversity in Chronic Hepatitis B Prevalence, Diagnosis, Monitoring and Treatment- National Report 2016. ASHM (Australasian Society for HIV, Viral Hepatitis and Sexual Health Medicine), https://ashm.org.au/resources/ viral-hepatitis-mapping-project/ [accessed 7 September 2024]. 25. MacLachlan J, Thomas L, Cowie B, Allard N. 2016. Hepatitis C Mapping Project: Estimates of Geographic Diversity in Chronic Hepatitis C Prevalence, Diagnosis, Monitoring and Treatment - National Report 2016. ASHM (Australasian Society for HIV, Viral Hepatitis and Sexual Health Medicine), https://ashm.org.au/resources/ viral-hepatitis -mapping-project/[accessed 7 September 2024].

26. Inness A, Ades M, Agusti-Panareda A, Barre J, Benedictow A, Blechschmidt A-M, et al. 2019. The CAMS reanalysis of atmospheric composition. Atmos Chem Phys 19(6):3515-3556, https://doi.org/10.5194/acp-19-3515-2019. 27. BOM (Bureau of Meteorology). 2024. Climate Data Online, http://www.bom.gov. au/climate/data/ [accessed 28 March 2024].

28. BOM. 2018. Observation of Air Temperature, https://reg.bom.gov.au/climate/ cdo/abou1/airtemp-measure.shtml [accessed 13 March 2024]. 29. Anselin L 1995. Local indicators of spatial association-LISA. Geographical Analysis 27(2):93-115, https://doi.Org/10.1111/j.1538-4632.1995.tb00338.x.

30. Prates M0, Kulldorff M, Assungao RM. 2014. Relative risk estimates from spatial and space-time scan statistics: are they biased? Stat Med 33(15)2634-2644, PMID: 24639031, https://doi.org/10.1002/sim.6143. 31. Berke 0.2004. Exploratory disease mapping: kriging the spatial riskfunction from regional count data. Int J Health Geogr 3(1):18, PMID: 15333131, https://doi.org/ 10.1186/1476-072X-3-18.

32. Liu L, Lin Y, Liu J, Wang L, Wang D, Shui T, et al. 2017. Analysis of local-scale urban heat island characteristics using an integrated method of mobile measurement and GIS-based spatial interpolation. Building and Environment 117:191-207, https://doi.Org/10.1016/j.buildenv.2017.03.013. 33. NCI (National Cancer Institute). Average Annual Percent Change (AAPC) and Confidence Interval, https://surveillance.cancer.gov/help/joinpoint/setting-parameters/method-and-parameters-tab/apc-aapc-tau-confidence-intervals/ average-annual-percent-change-aapc [accessed 28 March 2024].

34. VoPhamT. 2019. Environmental riskfactors for liver cancer and nonalcoholic fatty liver disease. Curr Epidemiol Rep 6(1):50-66, PMID: 31080703, https://doi.org/10. 1007/S40471-019-0183-2. 35. Randolph B, Tice A. 2017. Relocating disadvantage in five Australian cities: socio-spatial polarisation under neo-liberalism. Urban Policy and Research 35(2):103-121, https://doi.org/10.1080/08111146.2016.1221337.

36. Textor J, van der Zander B, Gilthorpe MS, Liskiewicz M, Ellison GT. 2016. Robust causal inference using directed acyclic graphs: the R package 'dagitty*. Int J Epidemiol 45(6):1887-1894, PMID: 28089956, https://doi.org/10.1093/ije/dyw341. 37. Lee D.2013. CARBayes:an R package for Bayesian spatial modeling with conditional autoregressive priors. J Stat Soft 55(13):1 24, https://doi.org/10.18637/jss.v055.i13.

38. Duncan EW, Cramb SM, Aitken JF, Mengersen KL, Baade PD. 2019. Development of the Australian cancer atlas: spatial modelling, visualisation, and reporting of estimates. Int J Health Geogr 18(1)21, PMID: 31570101, https://doi.org/10.1186/ S12942-019-0185-9. 39. Hespanhol L, Vallio CS, Costa LM, Saragiotto BT. 2019. Understanding and interpreting confidence and credible intervals around effect estimates. Braz J Phys Ther 23(4)290-301, PMID: 30638956, https://doi.Org/10.1016/j.bjpt.2018.12.006.

40. Duncan EW, White NM, Mengersen K. 2017. Spatial smoothing in Bayesian models: a comparison of weights matrix specifications and their impact on inference. Int J Health Geogr 16(1):47, PMID: 29246157, https://doi.org/10.1186/ S12942-017-0120-X.

41. GanT, Bambrick H,Tong S, Hu W. 2023. Air pollution and liver cancer: a systematic review. J Environ Sci (China) 126:817-826, PMID: 36503807, https://doi.org/ 10.1016/j .jes.2022.05.037. 42. Cocker F, Chien Yee K, Palmer AJ, de Graaff B. 2019. Increasing incidence and mortality related to liver cancer in Australia: time to turn the tide. Aust N Z J Public Health 43(3)267-273, PMID: 30958629, https://doi.org/10.1111/1753-6405. 12889.

43. Tian Y, Huang Q, Jiang X, Chen G, Zheng C, Yang X. 2021. Geographical distribution of liver cancer incidence based on big data analysis. J Phys: Conf Ser 1976(1):012029, https://doi.org/10.1088/1742-6596/1976/V012029. 44. Little MP, Eidemiiller M, Kaiser JC,Apostoaei Al. 2024. Minimum latency effects for cancer associated with exposures to radiation or other carcinogens. Br J Cancer 130(5):819-829, PMID: 38212483, https://doi.org/10.1038/s41416-023-02544-z.

45. Zhang Y-B, Pan X-F, Chen J, Cao A, Zhang Y-G, Xia L, et al. 2020. Combined lifestyle factors, incident cancer, and cancer mortality: a systematic review and meta-analysis of prospective cohort studies. Br J Cancer 122(7): 1085 1093, PMID: 32037402, https://doi.org/10.1038/s41416-020-0741-x. 46. Gasparrini A, Leone M. 2014. Attributable risk from distributed lag models. BMC Med Res Methodol 14(1):55, PMID: 24758509, https://doi.org/10.1186/1471-2288-14-55.

47. Rumgay H, Murphy N, Ferrari P, Soerjomataram 1.2021. Alcohol and cancer: epidemiology and biological mechanisms. Nutrients 13(9):3173, PMID: 34579050, https://doi.org/10.3390/nu13093173. 48. Sohn W, Lee HW, Lee S, Lim JH, Lee MW, Park CH, et al. 2021. Obesity and the risk of primary liver cancer: a systematic review and meta-analysis. Clin Mol Hepatol 27(1):157-174, PMID: 33238333, https://doi.org/10.3350/cmh.2020.0176.

49. Huang DQ, Singal AG, Kono Y, Tan DJ, El-Serag HB, Loomba R. 2022. Changing global epidemiology of liver cancer from 2010 to 2019: NASH is the fastest growing cause of liver cancer. Cell Metab 34(7):969-977, PMID: 35793659, https://doi.Org/10.1016/j.cmet.2022.05.003.

50. Sarich P, Canfell K, Egger S, Banks E, Joshy G, Grogan P, et al. 2021. Alcohol consumption, drinking patterns and cancer incidence in an Australian cohort of 226,162 participants aged 45 years and over. Br J Cancer 124(2):513-523, PMID: 33041337, https://doi.org/10.1038/s41416-020-01101-2.

51. AIHW. 2024. Overweight and obesity, https://www.aihw.gov.au/reports/overweight-obesity/overweight-and-obesity/contents/overweight-and-obesity [accessed 6 Dec ember 2023]. 52. Al-Shihabi F, Moore A, Chowdhury TA. 2023. Diabetes and climate change. DiabetMed40(3):e14971, PMID: 36209378, https://doi.org/10.1111/dme.14971.

53. Swinburn BA, Kraak VI, Allender S, Atkins VJ, Baker PI, Bogard JR, et al. 2019. The global syndemic of obesity, undernutrition, and climate change: The Lancet Commission report. Lancet 393(10173):791-846, PMID: 30700377, https://doi.org/ 10.1016/S 0140-6736( 18)32822-8.

54. Ismail A, Gongalves BL, de Neeff DV, Ponzilacqua B, Coppa CFSC, Hintzsche H, et al. 2018. Aflatoxin in foodstuffs: occurrence and recent advances in decontamination. Food Res Int 113:74-85, PMID: 30195548, https://doi.Org/10.1016/j. foodres.2018.06.067.

55. IARC (International Agency for Research on Cancer). 2021. Chemical Agents and Related Occupations. https://www.ncbi.nlm.nih.gov/books/NBK304416/ [accessed 24 March 2023].

56. Liu Y, Wu F. 2010. Global burden of aflatoxin-induced hepatocellular carcinoma: a risk assessment. Environ Health Perspect 118(6):818-824, PMID: 20172840, https://doi.org/10.1289/ehp.0901388. 57. Liu Y, Chang C-CH, Marsh GM,WuF. 2012. Population attributable risk of aflatoxin-related liver cancer: systematic review and meta-analysis. Eur J Cancer 48(14):2125-2136, PMID: 22405700, https://doi.Org/10.1016/j.ejca.2012.02.009.

58. Magan N, Medina A, Aldred D. 2011. Possible climate-change effects on myco-toxin contamination of food crops pre-and postharvest. Plant Pathology 60(1):150-163, https://doi.Org/10.1111/j.1365-3059.2010.02412.x. 59. Klich MA. 2002. Biogeography of Aspergillus species in soil and litter. Mycologia 94(1):21-27, PMID: 21156474.

60. Baranyi N, Kocsube S, Varga J. 2015. Aflatoxins: climate change and biodegra-dation. Current Opinion in Food Science 5:60-66, https://doi.org/10.1016/jxofs. 2015.09.002. 61. Eskola M, Kos G, Elliott CT, Hajslova J, Mayar S, Krska R. 2020. Worldwide contamination of food-crops with mycotoxins: validity of the widely cited 'FA0 estimate' of 25%. Crit Rev Food Sci Nutr 60(16):2773-2789, PMID: 31478403, https://doi.org/10.1080/10408398.2019.1658570.

62. Haerani H, Apan A, Basnet B. 2020. The climate-induced alteration of future geographic distribution of aflatoxin in peanut crops and its adaptation options. Mitig Adapt Strateg Glob Change 25(6):1149-1175, https://doi.org/10.1007/ s11027-020-09927-0. 63. Edwards D,Voronina A, Attwood K, Grand'Maison A. 2021. Association between occupational exposures and sarcoma incidence and mortality: systematic review and meta-analysis. Syst Rev 10(1):231, PMID: 34389054, https://doi.org/10. 11S6/S13643-021-01769-4.

64. Zhang H, Zhang R,ZengX, Wang X, Wang D,Jia H,etal. 2022. Exposure to neoni-cotinoid insecticides and their characteristic metabolites: association with human liver cancer. Environ Res 208:112703, PMID: 35016862, https://doi.org/10. 1016/j.envres.2022.112703. 65. LeroyerA, Gomajee H, LeroyR, Mazzuca M, Leleu B, Nisse C. 2022. Cancer mortality and chemical exposure in a retrospective zinc and lead smelter cohort: a 48-year follow-up. Int J Hyg Environ Health 242:113955, PMID: 35385816, https://doi.Org/10.1016/j.ijheh.2022.113955.

66. Naidu R, Biswas B, Willett IR, Cribb J, Kumar Singh B, Paul Nathanail C, et al. 2021. Chemical pollution: a growing peril and potential catastrophic risk to humanity. Environ Int 156:106616, PMID: 33989840, https://doi.Org/10.1016/j. envint.2021.106616.

67. Armstrong B, Sera F, Vicedo-Cabrera AM, Abrutzky R, Astrom DO, Bell ML, et al. 2019. The role of humidity in associations of high temperature with mortality: a multicountry, multicity study. Environ Health Perspect 127(9):097007, PMID: 31553655, https://doi.org/10.1289/EHP5430.

68. Baldwin JW, Benmarhnia T, Ebi KL, Jay 0, Lutsko NJ, Vanos JK. 2023. Humidity's role in heat-related health outcomes: a heated debate. Environ Health Perspect 131(5):055001, PMID: 37255302, https://doi.org/10.1289/EHP11807.

69. Greenland S, Morgenstern H. 1989. Ecological bias, confounding, and effect modification. Int J Epidemiol 18(1)269-274, PMID: 2656561, https://doi.org/10. 1093/ije/18.1269.