Poulsen et al. Environmental Health (2020) 19:81 https://doi.org/10.1186/s12940-020-00631-9 RESEARCH Open Access Intracranial tumors of the central nervous system and air pollution ? a nationwide case-control study from Denmark Aslak Harbo Poulsen1* , Ulla Arthur Hvidtfeldt1, Mette S?rensen1,2, Robin Puett1,3, Matthias Ketzel4,5, J?rgen Brandt4, Camilla Geels4, Jesper H. Christensen4 and Ole Raaschou-Nielsen1,4 Abstract Background: Inconclusive evidence has suggested a possible link between air pollution and central nervous system (CNS) tumors. We investigated a range of air pollutants in relation to types of CNS tumors. Methods: We identified all (n = 21,057) intracranial tumors in brain, meninges and cranial nerves diagnosed in Denmark between 1989 and 2014 and matched controls on age, sex and year of birth. We established personal 10- year mean residential outdoor exposure to particulate matter < 2.5 ?m (PM2.5), nitrous oxides (NOX), primary emitted black carbon (BC) and ozone. We used conditional logistic regression to calculate odds ratios (OR) linearly (per interquartile range (IQR)) and categorically. We accounted for personal income, employment, marital status, use of medication as well as socio-demographic conditions at area level. Results: Malignant tumors of the intracranial CNS was associated with BC (OR: 1.034, 95%CI: 1.005?1.065 per IQR. For NOx the OR per IQR was 1.026 (95%CI: 0.998?1.056). For malignant non-glioma tumors of the brain we found associations with PM2.5 (OR: 1.267, 95%CI: 1.053?1.524 per IQR), BC (OR: 1.049, 95%CI: 0.996?1.106) and NOx (OR: 1.051, 95% CI: 0.996?1.110). Conclusion: Our results suggest that air pollution is associated with malignant intracranial CNS tumors and malignant non-glioma of the brain. However, additional studies are needed. Keywords: Epidemiology, Air pollution, Register study, CNS-tumors Background only established exogenous risk factor is ionizing radi- Intracranial tumors of the central nervous system (CNS) ation, mounting evidence points toward a protective as- are a heterogeneous group of tumors of primarily the sociation with allergic and atopic conditions [1, 3, 4]. A meninges, the cranial nerves and the brain. Benign tu- range of medications has been investigated for potential mors are most often located in the meninges, and glioma associations [5?7] and some studies have suggested a is the most common malignant tumor type. The inci- link with exposures to pesticides or fertilizers [8?11], dence of intracranial CNS tumor types differ by age, sex but the results are inconclusive. At present, no occupa- and race [1, 2]. Around 5% of these tumors are attribut- tional or environmental risk factors for CNS tumors able to a range of hereditary syndromes [1]. While the have been conclusively established [2, 12]. One candi- date is, however, air pollution, which has been classified as ?carcinogenic to humans? by the International Agency * Correspondence: Aslak@Cancer.DK 1Danish Cancer Society Research Center, Strandboulevarden 49, DK-2100 for Research on Cancer [13, 14] based primarily on Copenhagen ?, Denmark mechanistic studies and epidemiological research Full list of author information is available at the end of the article ? The Author(s). 2020 Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated in a credit line to the data. Poulsen et al. Environmental Health (2020) 19:81 Page 2 of 12 demonstrating associations with lung cancer. Studies found a suggestive evidence of an association between have indicated that ultrafine particles may reach the malignant tumors and long-term exposure to PM2.5 ab- brain via the olfactory nerve or by crossing from the sorbance, a quantity closely related to elemental carbon lungs to the blood stream and then passing the blood- and proposed by the authors to be a proxy for traffic re- brain barrier [15, 16]. Air pollution exposures have been lated UFP [34]. associated with stroke and negative cognitive effects In summary, some studies have found indications of [17?20] and neuroimaging has shown links with reduced an association between one or more air pollutants and total brain and brain white matter volume [21]. one or more tumor types, however, the results are in- Ambient air pollution consists of a complex array of consistent. Possible explanations may relate to the rela- substances that may be mutagenic/carcinogenic either tively low number of cases but also to differences in directly or indirectly via induced inflammation and oxi- tumor definitions, exposure assessment and the covari- dative stress [16, 22]. As ultrafine particles (UFP) may ates adjusted for. We, therefore, conducted a large reach the brain and due to their large number, small size register-based case-control study of all intracranial CNS and large surface area, scientific focus with regard to tumors, diagnosed in Denmark over the years 1989 to brain tumors has been on UFP and in a recent cohort 2014 with an aim to investigate the different subtypes of study from Canada UFP exposure level was associated tumors in relation to a range of air pollutants from a with risk of malignant brain tumors [23]. However, con- state-of-the-art exposure model. centrations of airborne UFP are difficult to model. In Denmark, nitrogen oxides show good temporal correl- Materials and methods ation with UFP in street canyons [24, 25] and has,been The study was conducted in Denmark (population dur- used in epidemiological studies [26?28]. ing the study period, approx. 5.4 million), were all Several studies have investigated air pollution in rela- citizens since 1968 have had a unique personal identifi- tion to brain tumors with inconsistent results. An eco- cation number, which allows individuals to be identified logical study from the US found volatile organic and tracked across all health and administrative registers compounds to be associated with county-level incidence in Denmark [35, 36]. of tumors of the CNS [29]. Two studies of a cohort from the US found no association between a range of air pol- Case ascertainment lutants (particulate matter with a diameter less than From the Danish Cancer Register, which holds nearly 2.5 ?m (PM2.5) and 10 ?m (PM10), sulphur dioxide, ni- complete records of all cancer diagnoses in Denmark trogen dioxide, carbon monoxide or ozone (O3)) and since 1943 [37, 38], we identified all Danes, aged 20 years mortality from malignant tumors of the brain [30, 31]. or above, with a primary intracranial tumor (i.e. a tumor In an exploratory analysis of 54,304 members of the Da- of the meninges, the cranial nerves, or the brain) in the nish Diet Cancer and Health cohort, mean outdoor resi- period 1989 to 2014 Cases were not allowed to have dential NOx levels since 1971 were associated with an other cancer diagnoses, except non-melanoma skin can- incidence rate ratio of 2.28 (95% confidence interval cer, prior to their brain tumor. (CI): 1.25?4.19, cases = 95) per 100 ?g/m3 increase, for tumors of the brain and central nervous system [26]. In Sampling of controls a nationwide Danish study of 4183 cases diagnosed over For each case we sampled two random controls, alive the years 2000?2009, the odds ratio (OR) for a tumor and without a cancer diagnosis at the date of diagnosis (benign or malignant) of the brain was 1.11 (95% 0.84? for their matched case (index-date), from the Danish 1.4) per 100 ?g/m3 increase in NOx and the correspond- Civil Registration System [39]. Controls were matched ing OR for non-glioma tumors of the brain was 1.53 on sex and birth year and month. (95% CI 1.02?2.29) [32]. The study indicated that increased brain tumor risk may only be detectable at Exclusion criteria high NO levels (> 80 ?g/m3x ). The Danish Nurses Health Cases and controls were excluded if they had: 1) No re- Cohort study investigated intracranial CNS tumors (n = corded address in Denmark (excluding Greenland and 121) and reported suggestions of a weak association with the Faroe Islands) at index-date. 2) Less than 80% geo- mean outdoor concentrations of particulate matter < codable address history for the ten-year period preceding 2.5 ?m, PM < 10 ?m, NO2 and NOx over past 3 years. index-date. 3) Missing information on marital status, Restricting the analysis to meningioma or benign tumors employment status, household disposable income or area increased risk estimates, but still with wide confidence of residence (parish). Information for some of these ex- intervals spanning the null [33]. The same tumor end- clusions was available only after initial matching. There- points were investigated in the ESCAPE project that fore, cases ending without controls or vice versa after combined cohorts from six European countries, which full matching and exclusions were excluded. Poulsen et al. Environmental Health (2020) 19:81 Page 3 of 12 Exposure assessment B01AC30, N02BA01, N02BA51), non-aspirin NSAID We extracted address histories for all cases and controls (ATC: M01A except: M01AX), and antidiabetics (ATC: since 1979 from the Danish Civil Registration System A10A, A10B). We identified all subjects who within a [39] since air pollution modeling was possible from this year redeemed at least two separate prescriptions for year onwards. We geocoded all addresses, and front door each drug to increase the likelihood that they were actu- air-pollutant concentrations at 2 m height were esti- ally using the drug (NSAIDs and HRT) and/or that the mated for each of these addresses using the Danish indication was diabetes, asthma or allergy. DEHM/UBM/AirGIS modelling system [40, 41]. This in- tegrated air pollution model system incorporates detailed time-varying information based on three air pollution Statistical methods contributions: 1) regional background, modelled with We used conditional logistic regression to calculate ORs the DEHM model [42] on 150 km ? 150 km scale cover- for all intracranial CNS tumors (International Classifica- ing the northern hemisphere and increasing resolution tion of Diseases, version 10 (ICD10): C70.0, C71.0-C71.9, towards a 5.6 km ? 5.6 km scale over Denmark, based on C72.2, C72.5, D32.0,D33.0-D33.3,D42.0,D43.0-D43.3) and historical international and national emission data (in- for the following subgroups: malignant tumors (prefix C), cluding natural emissions); 2) urban background mod- non-malignant tumors (prefix D), cranial nerves (C72.2, elled with the UBM model [43] on 1 km ? 1 km scale, C72.5,D33.3,D43.3), meningioma (C70.0,D32.0,D42.0), gli- based on high resolution emission data for Denmark for oma (morphology codes 9380/0?9480/9 within ICD10: all emission sectors, land-use data and building heights; C71.0-C71.9, D33.0-D33.3, D43.0-D43.2), malignant non- and 3) address-level local air pollution calculated with glioma brain tumors (all other morphologies within the OSPM model [44], based on street-level data on traf- ICD10: C71.0-C71.9) and non-malignant non-glioma fic type and intensity combined with emission factors brain tumors (all other morphologies within ICD10: and taking into account meteorology as well as street D33.0-D33.3, D43.0-D43.2). All pollutants were analyzed and building configuration. The model system is de- categorically as well as linearly since previous studies had scribed in detail elsewhere [45, 46]. The model system suggested that effects may only be apparent among the provided yearly mean values of PM2.5, black carbon (BC highly exposed. In the categorical analyses, we used cat- ? (a primary component of particulate matter), as well egories based on percentiles of exposure among controls: as the gases: O3, and NOx. The level of the latter corre- <50th percentile (reference), 50-94th percentile, 95-99th lates well with ultrafine particulate matter and has previ- percentile and > 99th percentile. These categories pro- ously been utilized as a proxy for these [26?28] . vided a separate risk estimate for the few people with very For each individual, we calculated a time-weighted high exposure, since a previous study found associations average (TWA) concentration over residential addresses only in this group [32]. For all linear analyses, we tested during the 10 years preceding the index-date. for deviations from linearity by likelihood ratio testing comparing our model with a model including also the Covariates second-degree polynomial of the pollutant and with a From Statistics Denmark, we obtained yearly individual- model dividing the pollutant in 20 equal sized groups. We level information on disposable income, marital status, found no consistent signs of deviations (Supplement employment status and country of origin. For all Danish Table 1). parishes we obtained yearly information on percentage We analyzed data in three models: a crude model, only of the adult population: in lowest income quartile, un- taking into account the matching factors; a model add- employed, retired, doing manual labor, owning their itionally adjusting for individual-level covariates, and a own dwelling, living in social housing, being of Danish final, main model, which also included parish-level co- origin, previously convicted (theft, robbery, vandalism or variates. Individual-level covariates included personal in- violence), single parent families and having basic educa- come in deciles (calculated annually based on the tion as highest attained education level. In 1996, a total distribution among controls), marital status (currently of 2160 parishes existed with a median number of 1032 living together, formerly married or never married) and inhabitants (range 33?35,979) and a mean area of 16.2 employment status (retired, unemployed, blue collar, km2 (range 0.1?126.2). From the Danish National Pre- low-level white collar, high-level white collar). The main, scription Register, which holds information on all pre- final model also included percentages of adult parish scription drugs redeemed in Denmark since 1995 [47], population in the lowest income quartile, unemployed, we obtained information on prescriptions for asthma or retired, doing manual labor, owning their own dwelling, allergy conditions (ATC: R03, R06A) [47], prescriptions living in social housing, being of Danish origin, previ- for hormone replacement therapy (HRT) (ATC: G03C, ously convicted of a crime, single parent families and G03D, G03F, G03HB01), NSAIDs (ATC: B01AC06, having basic education as highest attained education Poulsen et al. Environmental Health (2020) 19:81 Page 4 of 12 level. To minimize potential effects of prodromal symp- Table 1 Descriptive characteristics of intracranial CNS tumor toms, we assessed covariates 1 year before index-date. cases and matched controls. Denmark, 1989?2014 We also conducted sensitivity analyses using the TWA CASES CONTROLS exposure over 1 year and over 5 years. Finally, in an TOTAL 21,057 37,368 analysis restricting to cases diagnosed after 1995, we ad- Individual level factors: justed for ever-use of HRT [1, 6] NSAID and non- FEMALE 53% 54% aspirin NSAID [5], which some studies have associated a with CNS tumors and Danish administrative data sug- Age at index date gest higher usage in more urbanized regions. We also median 62 63 adjusted for use of antidiabetics, and medication for 10th pctl 39 39 asthma or allergy conditions since studies have sug- 90th pctl 80 80 gested that both diabetes and allergy could be associ- Region of origin 0 ated with a decreased risk for types of CNS tumors Denmark 96% 95% and some studies also associate these conditions with air pollution [48, 49]. Non-Western 2% 2% Statistical analysis was performed in SAS 9.3 (SAS In- Western 3% 3% stitute Inc., Cary, NC, USA). Marital status Living together 61% 59% Previously married 25% 26% Results Never married 14% 14% We identified 22,454 adult cases diagnosed with a primary intracranial CNS tumor in the period 1989 to Occupational status 2014 and 44,908 matched controls. We excluded cases Unemployed 4% 4% (45 / 0.2%) and controls (3567 / 7.9%) not living in Low skill level 23% 22% Denmark at the time of diagnosis, as well as cases (865 / Medium skill level 16% 16% 3.9%) and controls (2318 / 5.2%) who had less than 80% High skill level 8% 8% geocodable address history in Denmark for the period Retired 49% 50% 10 years before index-date. We also excluded 42 cases (0.2%) and 126 controls (0.3%) due to missing data on Disposable income (dKK) one or more covariates. Finally, we excluded 445 cases Median 125,757 122,812 (2.0%) with no remaining controls and 1529 controls 10th pctl 61,311 58,995 (3.4%) with no case, after the above exclusions. The 90th pctl 254,411 254,053 resulting population comprised 21,057 cases and 37,368 At least 2 prescriptions within a year (data only available since controls. Of these cases, 7465 were glioma and 5657 1996) were meningioma. Malignant tumors comprised 46% of Non-aspirin NSAID 28% 27% the cases. Aspirin 12% 13% Table 1 shows virtually no differences between cases Antidiabetics 4% 4% and controls for any covariate. Similarly, Table 2 shows similar distributions of the air pollutants for cases and HRT 13% 12% controls below the 99th percentile. However, the max- Allergy medication 8% 8% imum observed concentrations of BC, PM2.5 NO2 and Parish level factors: NOx were higher among cases. The correlation coeffi- % of population with only basic education cients between air pollutants were in the range between Median 27 27 ? 0.98 and 0.94 (Table 3). 10th pctl 15 15 In linear analysis, the ORs for association with total intracranial CNS tumors was 0.967 (95%CI: 0.934? 90th pctl 41 42 1.002) for O3, 1.011 (95%CI: 0.992?1.030) for NOx 1.031 % of population in manual labor (95%ci: 0.997?1.066) for NO2 and 1.016 (95%CI: 0.996? Median 29 29 1.037) for BC. For BC, the categorical results showed an 10th pctl 18 18 exposure response pattern. 90th pctl 38 38 Malignant tumors of the intracranial CNS was associ- % of population retired ated with BC (OR: 1.034, 95%CI: 1.005?1.065 per IQR), NO (OR: 1.042, 95%CI: 0.992?1.095 per IQR) and NO Median 6 62 x (OR: 1.026, 95%CI: 0.998?1.056 per IQR). Except for 10th pctl 3 3 Poulsen et al. Environmental Health (2020) 19:81 Page 5 of 12 Table 1 Descriptive characteristics of intracranial CNS tumor For malignant non-glioma tumors of the brain we cases and matched controls. Denmark, 1989?2014 (Continued) found an association with PM2.5 (OR: 1.267, 95%CI: CASES CONTROLS 1.053?1.524 per IQR), and all PM2.5 exposure categories 90th pctl 11 11 above the median were associated with elevated risks % of population unemployed with some suggestion of exposure-response. For BC, NO2 and NOx the corresponding ORs in the linear ana-median 4 4 lyses were 1.049 (95%CI: 0.996?1.106), 1.267(95%CI: 10th pctl 1 1 0.983?1.194) and 1.051 (95% CI: 0.996?1.110), respect- 90th pctl 8 8 ively (Table 4). % of population in 1st income quartile There was no consistent evidence of any associations Median 10 10 in either the linear or the categorical analyses for sub- 10th pctl 5 5 groups not mentioned above (Table 4). Adjusting for area-level covariates generally attenuated 90th pctl 18 18 risk estimates (Table 5). This was particularly notable % of population in social housing for meningioma, where all pollutants were significantly Median 13 13 associated in the crude and individual level adjusted 10th pctl 0 0 models but only the associations with NO2 and O3 90th pctl 43 43 remained statistically significant after additional adjust- % of population owning own dwelling ment for parish level covariates. An exception from this pattern was malignant non-glioma tumors of the brain, Median 65 66 for which the risk estimates increased both when includ- 10th pctl 29 30 ing individual and area level factors in the model. 90th pctl 92 92 We also analyzed air pollutant exposure averaged over % single parent families 1 year and 5 years prior to diagnosis (Supplement Median 5 5 Table 2). For malignant tumors, there were indications 10th pctl 3 3 that the risk estimates were lower in association with shorter averaging periods. 90th pctl 7 7 Adjusting for use of NSAIDs, HRT, antidiabetic medi- % of population of Danish origin cation and allergy medication left ORs virtually un- Median 94 95 changed in a sensitivity analysis of cases diagnosed after 10th pctl 84 85 1995 (Supplement Table 3). 90th pctl 98 98 % of population previously convicted Discussion In this nationwide study, the largest to date, with more Median 0 0 than 20,000 intracranial CNS tumor cases, we found 10th pctl 0 0 PM2.5 air pollution NOx and BC to be associated with 90th pctl 1 1 malignant non-glioma tumors of the brain. BC and NO2 a: index date = date of diagnosis of matched case were weakly associated with increased risk for malignant intracranial CNS tumors and O3 was inversely associated with risk for meningioma. Total intracranial CNS tu- NO2, the highest risk estimates were observed for the mors were associated with BC air pollution in an 95-99th percentile of exposures in the categorical ana- exposure-response manner. ORs for benign tumors were lysis (Table 4). sensitive to adjustment. Meningioma was inversely associated with O3 (OR: Some previous studies on air pollutants and tumors of 0.923, 95%CI: 0.862?0.988), which was also indicated by the brain or CNS found positive associations, whereas the categorical analysis. For NO2 there was a positive others did not. When accounting for the difference of linear association (OR: 1.083, 95%CI: 1.016?1.154) this scale, the results have confidence intervals that overlap, was not apparent in categorical analysis. For the other indicating that they are compatible. In general, this is pollutants, there was no evidence of association in linear also the case when comparing the present study with analysis. Neither was there evidence of associations in previous studies (For the reader?s convenience, results categorical analysis, except for the highest percentile of from our previous studies, rescaled to the IQR of the PM2.5 exposure where the OR was 0.65 (95%CI: 0.43? present study, can be found as supplement Table 4). 0.99). In linear analysis, the corresponding OR was 1.088 In a Danish cohort, cancer of the brain was associated (95%CI: 0.955?1.239) (Table 4). with NOx (rescaled to IQR of the present study: HR Poulsen et al. Environmental Health (2020) 19:81 Page 6 of 12 Table 2 Descriptive data on air pollutants among intracranial CNS tumor cases and controls in Denmark, 1989?2014 10 year time-weighted min 1st pctl 5th pctl 10th pctl 25th pctl median 75th pctl 90th pctl 95th pctl 99th pctl max Inter quartile mean exposure range BC (?g/m3) Controls 0.29 0.37 0.43 0.47 0.55 0.72 0.94 1.21 1.49 2.42 8.00 0.39 Cases 0.29 0.38 0.43 0.47 0.56 0.73 0.94 1.21 1.50 2.43 13.64 0.38 PM2.5 (?g/m 3) Controls 9.91 11.83 12.81 13.41 14.67 16.87 20.07 22.45 23.53 26.30 37.09 5.39 Cases 9.93 11.82 12.83 13.40 14.63 16.76 19.95 22.43 23.52 26.06 41.46 5.31 NO2 (?g/m 3) Controls 5.68 8.12 10.35 11.79 14.57 25.35 19.18 31.06 35.72 48.33 73.86 10.78 Cases 5.66 8.20 10.39 11.85 14.79 25.60 19.35 31.22 35.74 48.26 79.00 10.81 NOx (?g/m 3) Controls 6.13 8.84 11.42 13.17 17.06 24.83 35.92 51.01 67.33 116.69 270.31 18.86 Cases 6.15 8.93 11.48 13.26 17.34 25.15 36.35 51.35 67.22 115.37 330.91 19.01 O3 (?g/m 3) Controls 15.82 37.5 47.04 50.70 55.58 60.92 65.30 68.44 70.26 73.81 77.45 9.72 Cases 15.72 37.05 47.08 50.63 55.43 60.78 65.12 68.38 70.33 73.56 77.68 9.69 1.17, 95% CI: 1.04?1.31 per 18.88 ?g/m3) [26]. The In a Danish register-based investigation of tumors of present study could not confirm such an association for the brain, non-glioma tumors were associated with NOx the wider group of intracranial CNS tumors. The present (rescaled to the IQR of the present study: OR 1.08, study, however, suggested that NO2 and BC may be as- 95%CI: 1.004?1.169 per 18.88 ?g/m 3) [32]. The cases of sociated with a small increased risk for malignant intra- that study were also part of the present study, and we cranial CNS tumors. No other study has investigated BC found risk estimates of similar magnitude although not in relation to intracranial CNS tumors. However, in the statistically significant for NOx and BC. For PM2.5 we multinational ESCAPE study, the authors found suggest- found a stronger association. ive evidence that PM2.5 absorbance, a proxy similar to We found inverse associations between ozone and BC, was associated with malignant CNS tumors, al- meningioma. We cannot provide a plausible explanation though with a wide confidence interval spanning the null for a causal inverse association and it could reflect the [34]. For NO2 the HR in that study were near identical inverse relationship between O3 and other pollutants, al- to the ORs of the present study. A recent Canadian though chance is also a possible explanation. Two previ- study found no evidence of an association between ous studies, conducted on subsets of the cases in the PM2.5 or NO2 and malignant brain tumors [23]. Turner present study, have found some indication of associa- et al. [30] reported on malignant tumors of the brain, tions between other air pollutants and benign brain tu- adjusted for a comprehensive array of personal and area- mors. Small sample size and lack of confounder level confounders, and found no significant association adjustment (in one study) mean that these observations with PM2.5, NO2 or O3. When accounting for the differ- may result from chance or confounding [32, 33]. ent unit scales, the HRs in that study were very similar Use of the comprehensive and virtually complete regis- to those observed in our study. ters on the Danish population [35, 36] was a major Table 3 Pearson correlations between pollutants 10 year mean 10 year mean 10 year mean 10 year mean 10 year mean BC (?g/m3) PM (?g/m3) NO (?g/m3) NO (?g/m3) O (?g/m32.5 X 2 3 ) Air pollutants 10 year mean O (?g/m33 ) ?0.85 ?0.62 ?0.88 ?0.98 1.00 10 year mean NO2 (?g/m3) 0.88 0.57 0.92 1.00 10 year mean NOX (?g/m 3) 0.94 0.55 1.00 10 year mean PM2.5 (?g/m 3) 0.54 1.00 10 year mean BC (?g/m3) 1.00 Poulsen et al. Environmental Health (2020) 19:81 Page 7 of 12 Table 4 Fully adjusted, categorical and linear associations between time-weighted average air pollution (10 years before index date) and risk of intracranial CNS tumors, Denmark 1989?2014 Exposure NOX NO2 BC PM2.5 O3 levels (a) Cntrls Cases OR (95%CI) Cntrls Cases OR Cntrls Cases OR (95%CI) Cntrls Cases OR (95%CI) Cntrls Cases OR (95%CI) Intracranial CNS tumors =99pct 50 28 1.057 (0.884?1.265) 47 34 1.267 (0.774?2.075) 52 38 1.374 (0.867?2.177) 49 33 1.833 (1.093?3.073) 35 20 1.054 (0.587?1.893) per IQR 1.049 (0.996?1.106) 1.084 (0.983?1.194) 1.051 (0.996?1.110) 1.267 (1.053?1.524) 0.937 (0.846?1.037) Non-malignant Non-glioma tumors of the brain