Exposure to air pollution and risk of haematological malignancies: a systematic review and dose-response meta-analysis of epidemiologic evidence

Quantifying the potential association between air pollutants exposure and haematological malignancies (HM) risk can provide more direction for its prevention. In this systematic review and meta-analysis, case-control and cohort studies looked at the association between air pollution and the risk of HM in the general population were included. PubMed, Web of Science, Wanfang database, and China National Knowledge Infrastructure were searched as of 14 April 2022. The Mantel–Haenszel random effects model was used to calculate the meta-analysis relative risk (meta-RR). A two-stage random-effects dose-response meta-analysis was performed to estimate the degree of the associations and a dose-response curve was fitted using a restricted cubic spline model, as well as assessed publication bias. This review was registered with International Prospective Register of Systematic Reviews (CRD42022325677). The literature search yielded 7260 articles, and 41 studies were included. Benzene exposure significantly increased the risk of leukaemia (meta-RR 1.24, 95% confidence interval (95%CI) 1.01–1.54). The meta-RR of traffic density, nitrogen dioxide (NO2), particulate matter (PM2.5 and PM10) and leukaemia were 1.08 (95%CI 0.99–1.17), 1.02 (95%CI 0.96–1.09), 1.05 (95%CI 0.99–1.11), 1.04 (95%CI 0.69–1.56). 1.07 (95%CI 0.93–1.22), 1.01 (95%CI 0.96–1.06), 1.06 (95%CI 0.98–1.14) were the meta-RR of traffic density, NO2, PM2.5 and lymphoma. The meta-RR of NO2 and multiple myeloma was 1.00 (95%CI 0.92–1.09). Disease subtype, age and region appeared to modify these associations. When residential distance from a main road was less than 300 m, the risk was relatively high and gradually increased with the decrease of the distance; with the increase of NO2 exposure concentration, the risk of acute myeloid leukaemia (AML) gradually increased, increasing rapidly once NO2 concentration reached 40 μg m−3; with increasing benzene exposure concentration, the risk of AML and acute lymphoblastic leukaemia gradually increased, particularly after the concentration reached 3 μg m−3. These findings can be used as epidemiological evidence for the causal relationship between air pollutants and HM.


Introduction
Air pollution is a growing global health issue (Landrigan et al 2018). It is caused by human activities or natural processes that cause substances such as dust, smoke, gases, and particulate matter (PM) to enter the atmosphere. When pollutant concentration and time of exposure reach a certain level, a serious impact on human health may result (Shahrbaf et al 2021). According to the latest report from the World Health Organization, 99% of the world's population is breathing polluted air. Air pollution has become one of the 'greatest environmental threats to human health' , with 13 people dying from air pollution every minute worldwide (WHO 2021). The Lancet Commission on Pollution and Health reported that global deaths from new forms of pollution such as ambient particulate air pollution, ozone pollution, smog, and environmental chemical pollution increased by 55% between 2000 and 2019, and this increase in mortality largely resulted from an increased incidence of air pollution-related noncommunicable diseases (Fuller et al 2022). Air pollution exposure induces respiratory infections and exacerbates asthma in children (Rice et al 2021); and increases the risk of fatal diseases such as cancer and stroke in adults (Guan et al 2018, Xue et al 2021. Air pollution is currently the fourth leading risk factor for global disease mortality after hypertension, smoking, and dietary factors (GBD 2019Risk Factors Collaborators 2020, Hoffmann et al 2021. In 2013, the International Agency for Research on Cancer (IARC) classified outdoor air pollution as a human carcinogen (IARC 2013). A meta-analysis of results from 14 outdoor air pollution studies in North America and Europe showed that each 10 µg m −3 increase in PM 2.5 concentration was associated with a 9% increase in the risk of lung cancer, and each 10 µg m −3 increase in PM 10 was associated with an 8% increase in the risk of lung cancer (Hamra et al 2014, Turner et al 2020. Several studies have also reported a positive association between air pollution and the risk of bladder and kidney cancer (Latifovic et al 2015, Zare et al 2020, upper gastrointestinal tract tumours (Wong et al 2016), breast cancer (Reding et al 2015, Zhang et al 2019, and prostate cancer (Parent et al 2013). Haematological malignancies (HM) comprise diverse malignant diseases that primarily affect the blood, lymph nodes or bone marrow, contributing to the global cancer burden (Sergentanis et al 2018). Between 2000 and 2019 a 34%-40% increase in the rate of new cases and deaths from HM was reported (Parkin 2001, Sung et al 2021. HM are prevalent in children, but are also increasing among adults. Among people aged 20 years or older, new cases and deaths from HM increased from 733 000 and 463 000 in 2000 to 1 278 000 and 676 000, respectively, in 2019 (Parkin 2001, Sung et al 2021. Genetic factors, viral infections, and environmental exposures are known risk factors for HM. Recently, research has increasingly focused on environmental exposures as nontraditional risk factors for HM. Exposure to outdoor air pollutants, such as benzene and nitrogen dioxide (NO 2 ), has been shown to be an important risk factor for leukaemia and lymphoma (Schraufnagel et al 2019, Konstantinoudis et al 2020. The IARC reported a possible strong association of benzene exposure with acute myeloid leukaemia (AML), acute lymphoblastic leukaemia (ALL), and non-Hodgkin's lymphoma (NHL) in children as well as in adults (IARC 2018). These studies represent robust evidence on the association of air pollutant exposure and HM.
However, most studies have focused on the relationship between air pollutant exposure and individual HM, such as leukaemia, with a target population largely restricted to children. A multidimensional analysis of the relationship between global population-wide air pollutant exposure and multiple HM has not been previously reported. The purpose of this study was to systematically review the evidence and the potential associations of air pollutant exposure (including traffic-related metrics) and the risk of HM. This study compared the metaanalysis of the highest versus lowest exposure categories, and conducted a dose-response meta-analysis to quantify the relationship between air pollutant exposure and HM. This research aims to lay the foundation for the risk assessment of air pollutants and the estimation of the attributable burden of disease, and to provide a scientific basis for evaluating the health effects of air pollutants.

Search strategy and selection criteria
The study followed the Preferred Reporting Items for Systematic Reviews and Meta-Analyses guidelines. According to the PECOS statement (Population, Exposure, Comparison, Outcome, and Study) (Morgan et al 2018), this study conducted a systematic search of PubMed, Web of Science, Wanfang database, and China National Knowledge Infrastructure (CNKI) for all epidemiological studies on HM and air pollutant exposure published before 14 April 2022, without restricting the language of the article (table S1). And a manual search was conducted from references of relevant original studies or reviews to identify any additional articles related to the study. The protocol for this meta-analysis can be found in the International Prospective Register of Systematic Reviews, the registration number is CRD42022325677.
Inclusion criteria were as follows: (a) case-control or cohort studies; (b) general population; (c) outdoor air pollutants, including studies with any type of assessment of traffic exposure; (d) reported risk estimates of illness or death, or reported data allowing the calculation of illness or death risk.
Exclusion criteria: (a) ecological studies; (b) lack of outcome indicators or key information; (c) exposure assessment limited to occupational activities; (d) studies reporting relative risk (RR) estimates for non-study cancer types or multiple different solvent exposures; (e) duplicate publications, conference reports, case reports, reviews, animal studies, and disease mechanism studies.
The literature screening process was first performed independently by two authors (Yan and Lin), and in case of disagreement or missing abstracts, both authors read the full text to determine whether the inclusion criteria were met. Any disagreement was resolved through discussion until a consensus was reached or by consulting a third author (Liu).

Data analysis 2.2.1. Quality assessment
The characteristic and quality of each included study were assessed by the Newcastle-Ottawa Scale (NOS) (Stang 2010) (table S2). This is based on a 'star system' in which a study is judged according to three broad perspectives: the selection of the study groups; the comparability of the groups; the ascertainment of either the exposure or outcome of interest for casecontrol or cohort studies, respectively. High-quality answers to each NOS scale question are identified with an asterisk, and the final NOS score for each study is the sum of the stars, with a high score indicating a high-quality study. The study evaluated the overall certainty of the evidence according to the GRADE approach (Atkins et al 2004), which takes in account issues related to both downgraded and upgraded the certainty of evidence based on five and three GRADE domains respectively.

Data collation 2.2.2.1. Highest versus lowest exposure meta-analysis
Case-control and cohort studies were included in this study and basic information such as region, cancer type, and age were extracted. To perform the highest and lowest meta-analysis, odds ratios and 95% confidence intervals (95%CI) in case-control study, rate ratios or hazard ratios and 95%CI in cohort study were extracted.

Dose-response meta-analysis
In dose-response meta-analysis, for each exposure level of air pollutants in the original studies, the mean or median concentration was extracted. When the exposure level was an interval, the mean or median was not directly displayed, so '(upper limit + lower limit)/2' was used to calculate the concentration. When the level was 'open' , we used 20% below the upper limit or 20% above the lower limit as the exposure concentration for that level (supplementary material S1).
The methods of missing value (e.g., missing the number of cases in each exposure level or missing total population in each level) imputation in doseresponse was based on previous studies (Bekkering et al 2008, Xu et al 2015.
When the total population in all levels is known, the number of cases in each exposure level can be estimated by the number of cases in all exposure levels, the total population in each level and the corresponding RR: (1) C is the number of cases in all exposure levels and C i is the number of cases in exposure level i; T is the total population in all levels and T i is the total population in exposure level i; RR i is the risk ratio at exposure level i. When the total population in all levels is known, averaged them (assuming the number in each level is equal) to get the total population in each level:

Statistical analysis 2.2.3.1. Highest versus lowest exposure meta-analysis
In the highest versus lowest exposure meta-analysis, a random effects model was used to calculate the metaanalysis relative risk (meta-RR) and 95%CI of HM associated with air pollutants (Filippini et al 2019).

Dose-response meta-analysis
In this study, a two-stage random-effects model (Greenland et al 1992, Orsini et al 2012 was used to conduct a dose-response meta-analysis to further characterise the relationship between air pollution and HM. The construction of the model is based on a 'two-stage' approach, where the first stage is to fit a cubic spline function to each original study, and the second is to weight and combine the parameters of these functions, the final dose-response meta-analysis of air pollution and HM was obtained (supplementary material S1).

Sensitivity analysis and publication bias
In the random-effects model, statistical heterogeneity among studies was assessed using the I 2 statistic. This study reported the effect size for each analysis, using less than 50% as the node for heterogeneity size assessment, defined as low heterogeneity.
With regard to sources of heterogeneity, we used directed acyclic graphs (Textor et al 2016) to screen and identify potential confounders, and ultimately used the minimal sufficient adjustment sets for estimating the effect of air pollution on HM: disease subtype, age, and region (figure S1). Subgroup analyses were performed including: disease subtype (leukaemia subtype: AML, ALL, acute nonlymphocytic leukaemia, chronic myeloid leukaemia, chronic lymphocytic leukaemia; lymphoma subtype: Hodgkin lymphoma (HL), NHL), age (⩽6 years, 7-17 years, ⩾18 years), and region.
In addition, sensitivity analyses were conducted to explore whether the study results were influenced by individual studies, and potential publication bias was assessed using funnel plots.
All data analyses were performed using Stata software (release 15.1; Stata Corp.), and P values were two-sided. P < 0.05 indicated a statistically significant result.

Description of included studies
The literature screening process and results are shown in figure 1. We identified a total of 7260 papers by searching PubMed, Web of Science, Wanfang database, and CNKI. Following removal off duplicate items, 6822 articles were filtered according to their titles and abstracts. Following initial screening, 58 papers were reviewed in full and assessed for eligibility. We excluded 20 articles that did not match the inclusion criteria, and three articles were included by manual search from the references of relevant original studies or relevant reviews, resulting in the inclusion of 41 articles. The 41 articles were in the English-language and all performed highest versus lowest exposure meta-analyses of air pollutants and HM. A dose-response meta-analysis was conducted in 19 of the studies (table S3). Furthermore, the mean NOS scale scores of the included studies were above 8, indicating the high quality of the included studies and low risk of bias (table S4).

Highest versus lowest exposure meta-analysis 3.2.1. Leukaemia
In 21 studies that used traffic density to assess air pollution exposure and leukaemia risk, the relative risk comparing the highest versus lowest exposure categories was slightly higher at 1.08 (95%CI 0.99-1.17). Fourteen studies reported a meta-RR of 1.02 (95%CI 0.96-1.09) for NO 2 and leukaemia risk. Eight studies reported a meta-RR of 1.24 (95%CI 1.01-1.54) for benzene and leukaemia risk. Ten studies reported PM 2.5 and leukaemia (meta-RR 1.05, 95%CI 0.99-1.11) and three reported PM 10 and leukaemia (meta-RR 1.04, 95%CI 0.69-1.56). The meta-RR of O 3 and leukaemia risk estimated from the three studies was 0.97 (95%CI 0.91-1.04). One study provided a meta-RR of 0.99 (95%CI 0.87-1.14) for CO and leukaemia risk (table 1, figure 3).
When we restricted inclusion to the AML studies only, the meta-RR for benzene and AML was 1.84 (95%CI 1.31-2.59). CO exposure was not associated with a meta-RR of 0.91 (95%CI 0.79-1.05) for AML, although it was associated with an increased risk of ALL, and the meta-RR was 1.05 (95%CI 1.01-1.10). For traffic indicators, NO 2 , PM and O 3 , no excessive risks were found in the meta-RRs with leukaemia subtypes. For the pollutants above, the risk of leukaemia was higher (meta-RR 1.24, 95%CI 1.01-1.54) after applying the age limitation of ⩽6 years. The meta-RRs associated with benzene exposure were more pronounced in this subgroup than in the total population, with meta-RRs of 3.21 (95%CI 1.39-7.42) and 1.19 (95%CI 1.00-1.40) for AML and ALL, respectively, according to disease subgroup. After restricting the age to adolescents aged 7-17 years and adults aged ⩾18 years, the meta-RR after grouping according to disease subtype were similar to those observed the population. For other air pollution exposure indicators, the association was slightly stronger for ⩽6 years children (table 1, figures S2-S14).
For traffic indicators, exposure to traffic pollutants was associated with leukaemia risk in Europe based on ten studies, and the meta-RR was 1.23 (95%CI 1.02-1.49); NO 2 exposure in Asia was associated with leukaemia risk (meta-RR 2.29, 95%CI 1.44-3.64); based on three studies, benzene exposure was associated with leukaemia risk in Southern Africa with a meta-RR of 1.21 (95%CI 1.04-1.41) (table 1, figure 4).
The results of the meta-analysis by disease subtype showed that the meta-RR differed significantly only for benzene. When limiting studies to include only NHL subtypes, the estimated meta-RR was 0.70 (95%CI 0.29-1.71). When limiting inclusion to HL subtype studies only, the meta-RR was 4.30 (95%CI 1.50-12.36), benzene exposure was associated with increased HL risk. For the above pollutants, after limiting the analysis to children aged ⩽6 years, adolescents aged 7-17 years, and adults aged ⩾18 years, we found PM 2.5 exposure in adolescents aged 7-17 years was associated with lymphoma risk, and there was a strong association for NHL (meta-RR 2.11, 95%CI 1.11-4.03) (table 2, figures S15-22).

Dose-response meta-analysis
The results of a dose-response meta-analysis of traffic density exposure and leukaemia showed that an increase in the number of vehicles per day was associated with an increased risk of leukaemia (P = 0.151) ( figure 5(A)). There was also an increase relationship (P = 0.769) between the residential distance from a main road and leukaemia risk. Signs of a higher meta-RR appearing at a distance less than 300 m from the residence to the nearby main road, and increasing progressively with decreasing distance ( figure 5(B)). There was no association between road density around homes and the leukaemia risk (P = 0.153) (figure 5(C)).
A dose-response meta-analysis of NO 2 exposure and leukaemia risk showed that an increase in Figure 2. Characteristics of included studies according to air pollution assessment, and other characteristics. Distance, distance between the residence and a major road; NO2, nitrogen dioxide; PM, particulate matter; road density, sum of the length of roads within a defined area around the residence; traffic count, estimated number of vehicles per day in the roads within a defined distance from the residence. NO 2 concentration was not associated with a significant increase in the risk for leukaemia (P = 0.657). Subgroup analysis according to disease type showed that the estimated risks for AML (P = 0.230) and ALL (P = 0.508) gradually increased at higher exposure levels (>40 µg m −3 ) ( figure 6).
The results of a dose-response meta-analysis of benzene exposure and leukaemia risk showed that the risk for leukaemia increased with increasing benzene concentrations. Subgroup analysis also showed that the risk of AML (P = 0.646) and ALL (P = 0.661) increased with increasing benzene concentration, and the trend of risk increase was larger than that of leukaemia, especially at 5 µg m −3 (figure 7).

Sensitivity analyses and publication bias
After systematically excluding each study in turn from the meta-analysis, this study repeated all analyses with stable final values for each effect (figures S24-S30).
In the dose-response meta-analysis, alternative estimates were used for the highest versus lowest exposure categories with unknown mean/median values, i.e., the ±15% value associated with the nearest boundary was entered instead of the ±20% value, and   the final results were robust (figures S31-S33). The funnel plots based on different exposure assessment methods showed a slightly asymmetric distribution (figures S34-S36). The GRADE assessment showed high-certainty of the evidence for an enhanced leukaemia cancer incidence and mortality by benzene exposure. Other air pollutants and the risk of HM had moderate-certainty or low-certainty of the evidence (table S5).

Discussion
The study systematically and quantitatively assessed the potential association between global air pollution exposure and the risk of various HM, and attempted to determine the exposure populations, regions, and concentration thresholds. In the population, benzene exposure significantly increased the risk of leukaemia (meta-RR 1. The meta-RR of NO 2 and multiple myeloma was 1.00 (95%CI 0.92-1.09). Subgroup analysis showed that the risk of leukaemia was higher in ⩽6 years children exposed to benzene, especially for AML; traffic indicators were more strongly associated with ALL in children aged ⩽6 years; PM 2.5 exposure in adolescents aged 7-17 years was associated with lymphoma risk, particularly for NHL; benzene exposure was more strongly associated with HL; CO exposure was associated with ALL; traffic pollutants exposure in Europe, NO 2 exposure in Asia, and benzene exposure in North America were found to be associated with leukaemia risk in the population across these regions and in the population aged 7-17 years. Dose-response analysis showed that traffic indicators, NO 2 , and benzene exposure were linearly associated with leukaemia risk. When residential distance from a main road was less than 300 m, signs of higher leukaemia risk appeared and gradually increased with decreasing distance; with increasing NO 2 concentration, the risk of AML gradually increased, with a significant increase above a concentration of 40 µg m −3 ; with increasing benzene concentration, the risk of AML and ALL gradually increased, particularly after the concentration reached 3 µg m −3 . The findings of this meta-analysis can be used as epidemiological evidence for the causal relationship between air pollutants and HM.
The following limitations should be noted. The 41 studies identified in this study focused on three geographic regions and it was not possible to assess air pollutants and HM in the remaining four continents. Dose-response requires data from more than three studies and multiple subgroups for analysis. Some air pollutants and subgroups contain fewer studies, and this study was not able to explore the doseresponse relationship between air pollutant exposure and lymphoma and multiple myeloma. The occurrence of publication bias cannot be completely excluded. Quality assessment using the NOS may be subjective and is limited, which may affect the credibility of this study. In this study, sources of heterogeneity in the included literature may not have been fully discussed due to the limitations of the data required, for example, a subgroup analysis of smoking versus non-smoking was not possible when analysing the relationship between air pollution and HM.
Benzene exposure was significantly and positively associated leukaemia risk. When the concentration exceeded approximately 3 µg m −3 , the upward trend of the meta-RR became obvious, and the highest risk of leukaemia observed with concentrations of approximately 5 µg m −3 . Thereafter, the meta-RR of leukaemia plateaued, while the risk of AML and ALL continued to gradually increase. Recently studies have consistently reported a strong and positive association between benzene exposure and leukaemia (Carlos-Wallace et al 2016, Wei et al 2021). Biologically this relationship is plausible, for example, animal and in vitro experiments have found that the target organ of benzene is the bone marrow and its toxic metabolites can attack haematopoietic stem cells in a variety of ways, causing haematotoxicity (Recio Bauer et al 2005, Chow et al 2015. Benzene metabolites can cause haematopoietic stem cell damage through various pathways, causing the conversion of normal cells to clonal leukaemic cells and promoting the development of leukaemia (Zhou et al 2022). In addition, the meta-RR of leukaemia risk with benzene exposure in children ⩽6 years of age (meta-RR 1.39, 95%CI 1.03-1.87) was higher than that in other groups, indicating that young children may be more susceptible to benzene exposure. The RRs of leukaemia obtained from other studies were similar to those in the present study, being 1.39 (95%CI 1.03-1.87) and 1.45 (95%CI 0.77-2.73) (Janitz et al 2017, Filippini et al 2019. Studies have shown that benzene was genotoxic and can induce DNA damage and chromosomal alterations, leading to impaired haematopoiesis and the development of leukaemia (Spatari et al 2021, Sun et al 2022. Furthermore, considering the specificity of children's behavioural patterns in case-control or cohort studies, the decreased Table 2. Summary RRs for the association between air pollution and lymphoma, stratified by age, lymphoma subtype and region.     likelihood of children having changed address in the past, and the likelihood that young children spend more time at home than older children, the higher meta-RR can be explained, in part, by lower benzene exposure misclassification, consistent with other studies (Vinceti et al 2012, Heck et al 2014, Janitz et al 2017. At the same time, for disease subtypes, this study found that the risk between AML and benzene exposure was higher, and this risk was most obvious in children ⩽6 years, consistent with previous studies (Puett et al 2020, Wei et al 2021. Therefore, it is important to focus on the association of air pollutant exposure with leukaemia and its subtypes in the paediatric population. The IARC and previous meta-analyses focused on the analysis of the association between air pollutants and leukaemia, largely disregarding other HM. In this study, benzene exposure was found to be associated with HL with a high RR of 4.30 (95%CI 1.50-12.36). Benzene is a known haematotoxin and benzene toxicity is characterised by a persistent decrease in the leukocyte count, which is strongly associated with an increased risk of malignancies of the lymphohaematopoietic system (Vermeulen et al 2022). Significantly, benzene exposure was not associated with NHL risk in this study. However, a strong positive association was reported by Rana et al which may be explained by the fact that the study included populations from the petroleum industry, oil refineries, and solvent exposure (2021). Instead, our study mainly focused on the general population, and exposure concentrations as well as the RRs may be lower. This study found that PM 2.5 and NO 2 were associated with lymphoma, and particularly NHL. Looking at different populations, PM 2.5 exposure was positively associated with lymphoma and was associated with a higher RR of NHL in adolescents aged 7-17 years and also among subjects aged ⩾18 years. This evidence supports further research on the relationship between air pollutants and other HM and their subtypes.
Compared to Asia and Europe, benzene exposure was found to be more strongly associated with leukaemia risk in North American populations, as well as in population aged 7-17 years. North American countries such as the U.S. have a well-developed oil industry, high oil consumption, and higher rates of oil refining and processing than some European countries. As an example, in 2020, North American oil supplies accounted for 25.6% of total global supply, while Europe supplies only accounted for 5.6% (the U.S. tops the list of countries with approximately 20.0% of world oil consumption) (Shi et al 2020, Jiang 2021, Zhang 2021). Furthermore, this study found a stronger association between traffic indicators and the risk of leukaemia in Europe compared to North America and Asia (meta-RR 1.23, 95%CI 1.02-1.49). The differences between these regions may also be due to the misclassification of exposures caused by the increased mobility of European residents (WEF 2020).
In the highest versus lowest meta-analysis, no significant association between traffic pollutants and HM was found in this study. Due to the inconsistent results of current epidemiological studies, traffic indicators have not been consistently associated with leukaemia, and exposure as assessed by traffic density including traffic counts, residential distance from a major road, and road density traffic indicators has limited relevance to leukaemia. This discrepancy suggests that traffic-related metrics may be less sensitive and underestimate exposure as a measure of outdoor air pollutants compared to simulated levels of carcinogenic pollutants such as benzene. The discrepancy could explain the smaller impact estimates observed when using traffic-related metrics as a proxy exposure measure. The dose-response meta-analysis showed no significant threshold changes in the traffic count and road density studies. However, a higher risk was observed for households within 300 m of the distance between the residence and the main road. This suggests that policymakers or decision makers should consider the distance from main roads when planning homes, schools, or other facilities for children.
With the increase of NO 2 exposure concentration, the risk of AML gradually increased, and an association was observed with a concentration above 10 µg m −3 . When the concentration threshold reached 40 µg m −3 , the risk tended to increase significantly. This concentration was consistent with the reference level of NO 2 in the 2005 Global Air Quality Guidelines (Zhu et al 2022), however, the reference level of NO 2 in 2021 has been reduced to an annual average value of 10 µg m −3 (WHO 2021). There has been a recent increase in the risks associated with NO 2 exposure and the changes in NO 2 reference levels are consistent with the findings observed in this study. The dose-response meta-analysis also showed a linear relationship between traffic indicators, benzene and NO 2 exposure and leukaemia risk, although further studies are needed to confirm these findings.

Conclusion
Our systematic evaluation and dose-response metaanalysis supported a positive association between outdoor air pollution and the risk of leukaemia and lymphoma. Further evidence is required to confirm the relationship with multiple myeloma. Disease subtype, age and region appeared to modify these associations. An analysis of the relationship between leukaemia and residential distance from main roads, NO 2 and benzene exposure provide evidence of a threshold. Therefore, policies aimed at reducing air pollution exposure and protecting special populations are necessary to further reduce risks due to air pollutants. The results of this study can be used as a reference for future studies on the evaluation the role of air pollutants in cancer and particularly, HM.

Data availability statement
All data that support the findings of this study are included within the article (and any supplementary files).

Data sharing
All data relevant to the study are included in the article or uploaded as supplementary information. The data that support the findings of this study are openly available.