Ambient Fine Particulate Matter, Residential Greenness, and Childhood Cancer Risk by Trimester of Exposure in Minnesota 2000–2014: A Case-control Study

Abstract

Background:

Outdoor air pollution is a suspected risk factor for childhood cancer, and there is some evidence that greenness may reduce cancer risk. We examined relationships between prenatal exposure to ambient fine particle air pollution (PM_2.5), greenness, and childhood cancer risk in Minnesota.

Methods:

Cases included individuals born in the state of Minnesota and reported to the Minnesota Cancer Reporting System between 2000–2014 (n=1272, aged 0–14yrs at diagnosis) and birthyear-matched cancer-free controls (n=5245). We used Network Common Data Form to estimate monthly PM_2.5 exposure and the Normalized Difference Vegetation Index (NDVI) for census tract-level greenness. We estimated odds ratios (OR) and 95% confidence intervals (95%CI) between high PM_2.5 (≥12 µg/m³) and NDVI (≥ 0.3) exposure and each childhood cancer per trimester (T1, T2, T3) and full pregnancy (FP) using adjusted logistic regression.

Results:

High PM_2.5 exposure was associated with increased odds of Burkitt lymphoma (OR_T1: 1.93 [1.13–3.30]), lymphoreticular neoplasms (OR_T1: 1.43 [1.01–2.01]), Hodgkin lymphoma (OR_T2: 2.60 [1.52–4.45]), lymphoid leukemias (OR_T3: 1.17 [1.01–1.37]), B-cell leukemia (OR_T3: 1.20 [1.02–1.41]), and intracranial and intraspinal embryonal neoplasms (OR_T3: 1.90 [1.07–3.37]). High residential greenness during early life was associated with reduced risk of malignant epithelial neoplasms and melanomas (OR_T1: 0.34 [0.13–0.91]). Unexpectedly, greenness was also associated with increased risk of renal tumors (OR_T2: 1.52 [1.11–2.08]; OR_FP: 1.30 [0.98–1.72)) and neuroblastoma (OR_T3: 1.44 [1.03–2.03]).

Conclusions:

We observed elevated risk associated with PM_2.5 exposure throughout pregnancy for multiple individual cancers, namely leukemias, lymphomas, and CNS tumors.

Impact:

These findings demonstrate the complex relationship between PM_2.5 and greenness.

Introduction

Outdoor air pollution is a ubiquitous environmental exposure with many components, some of which are known human carcinogens (IARC Group 1)(1) with established links to childhood cancer risk.(2–8) Exposure to ambient fine particulate matter, which is comprised of particles of aerodynamic equivalent diameter of 2.5 µM or less (PM_2.5), is of particular health concern. Ambient PM_2.5 exposure has been associated with elevated risk of multiple types of childhood cancer, including lymphoid leukemia, retinoblastoma, lymphomas, and central nervous system tumors.(9–12)

There is evidence that green space (land covered by vegetation, such as natural forests, urban vegetation, and parks)(13) can potentially mitigate some of the adverse health effects associated with outdoor air pollution.(14–17) In particular, trees and shrubs can reduce ambient air pollution by removing pollutants via dry deposition or gaseous absorption.(18–20) This has implications for reducing the burden of adverse health outcomes associated with exposure to ambient air pollution, including childhood cancer.(21) However, the potential for green space to mitigate childhood cancer risk has only been evaluated in a limited geographic region,(12) and little is known about the potential critical windows of exposure.

The goals of this analysis were to 1) evaluate associations between prenatal PM_2.5 exposure and childhood cancer risk in Minnesota (MN), 2) to examine the potential mitigating factor of greenness in these relationships, and 3) to identify potential critical windows of exposure during pregnancy.

Materials and Methods

Study population

The study included individuals reported to the MN Cancer Reporting System and Vital Records and Certificates of the MN Department of Health (MDH) who were born in the state of MN between January 15, 1976, and December 30, 2014. This pan-cancer analysis included all cancer cases born and diagnosed in Minnesota during the study period. Cancer cases aged 0–14 years at diagnosis were frequency matched by birth year at a ratio of 1 to 4 to controls with no previous diagnosis of cancer from MN Birth Records. Although seasonality influences PM_2.5 concentrations in MN (Supplemental Material, Figure S1), we elected to match on birth year so as not to limit our ability to detect differences in risk due to this exposure variability.

Among 24532 identified subjects (n=3936 cases; n=20596 controls) from the MN cancer and birth registries, 9695 children (n=958 cases; n=8737 controls) were excluded due to missing latitude or longitude for residence at birth, and 293 additional children (n=36 cases; n=257 controls) were excluded because of missing gestational age. As ambient fine particulate matter (PM _2.5 µg/m³), and normalized difference vegetation index (NDVI) were publicly available for 2000–2014 for our study, 8002 children (n=1646 cases; n=6356 controls) whose mother’s pregnancy did not begin and end between 2000 and 2014 (i.e., began in 1999) were excluded. Additionally, one subject was excluded owing to having his/her residence at birth outside MN, and 24 cases with other/unspecified malignant neoplasms or unknown cancer types were not included. After exclusions, 1272 cases and 5245 controls were considered eligible and selected for analysis.

Cancer definition

Cancers were defined using the International Classification of Childhood Cancer, third edition (ICCC-3) and using the IARC (2017) extended recodes based on histologic type ICD-O-.(22) Additionally, we included Langerhans cell histiocytosis (LCH; ICD-O-3 morphology code 9751). Twenty cancers were included in the study: Ia. Lymphoid leukemias, Ib. Acute myeloid leukemia, Other leukemias (Ic. Chronic myeloproliferative diseases, Id. Myelodysplastic syndrome and other myeloproliferative diseases, Ie. Unspecified and other specified leukemias); IIa. Hodgkin lymphoma, IIb. Non-Hodgkin lymphoma, IIc. Burkitt lymphoma, IId. Miscellaneous lymphoreticular neoplasms; IIIa.1 Ependymomas, IIIb. Astrocytomas, IIIc.1 Medulloblastomas, Other intracranial and intraspinal embryonal (IIIc.2 Primitive neuroectodermal tumor (PNET), IIIc.3 Medulloepithelioma, IIIc.4 Atypical teratoid/rhabdoid tumor), Other CNS cancer types (IIId. Other gliomas, IIIe. Other specified intracranial and intraspinal neoplasms, IIIf. Unspecified intracranial and intraspinal neoplasms); IV. Neuroblastoma and other peripheral nervous cell tumors; V. Retinoblastoma; VI. Renal tumors; VII. Hepatic tumors; VIII. Malignant bone tumors; IX. Soft tissue and other extraosseous sarcomas; X. Germ cell tumors, trophoblastic tumors and neoplasms of gonads; XI. Other malignant epithelial neoplasms and malignant melanomas.

Exposure assessment and data sources

The exposure window of interest was between conception and birth. We used the latitude and longitude from the geocoded residential address at birth to identify the geographic location (census tract) for each study subject. Cases and controls were retroactively assigned to the most appropriate census tracts (Supplemental Figure S2). Births occurring in the years 2000–2005 were assigned to 2000 census tracts; births occurring in 2006–2014 were assigned to 2010 census tracts. We then assigned exposure to PM_2.5 and NDVI at the census tract-level in the months and years of maternal pregnancy periods. For each birth, we assigned exposure for each month of pregnancy, from which trimester-specific and full pregnancy estimates were calculated. Pregnancy trimesters were calculated by counting backwards from the birth date and gestational months for each study case and control.

Particulate matter <2.5 µg/m³ (PM_2.5)

Full exposure assessment details can be found in Williams et al., 2024.(12) Briefly, an annual global estimation model was used for PM_2.5 data.(23) Van Donkelaar and colleagues created a ground-truthed PM_2.5 model by combining multiple NASA remote sensing datasets including MODIS, Multi-angle Image SpectroRadiometer (MISR), and Sea-viewing Wide Field-of-view (SeaWiFS). The resulting Network Common Data Form (NetCDF) dataset is a coarser resolution than MODIS at 0.01 decimal degrees or approximately 850 square meters. Monthly PM_2.5 values from the model were used to calculate trimester-specific and full pregnancy exposure estimates for our analysis. Annual estimated PM_2.5 exposures are shown by census tract in Supplemental Figure S3.

Normalized Difference Vegetation Index (NDVI)

Full exposure assessment details can be found in Williams et al., 2024.(12) Briefly, NDVI data, a measure of greenness, was extracted from Google Earth Engine.(24) The NDVI is derived from NASA satellite sensor Moderate Resolution Imaging Spectroradiometer (MODIS) that revisits every place on the globe every 16 days and has a pixel resolution of 250×250 meters (Date Accessed: October 1, 2021). A total of 23 images are collected for every year (aside from 2001, which has 22). The satellite sensor collects red and near-infrared bands which are used for NDVI calculation by subtracting the red from the near infrared and then dividing that result by the sum of the near infrared and red bands.(25) The NDVI ranges from −1.0 to 1.0, corresponding to shrubs and grassland (0.2 to 0.3) and temperate and tropical rainforests (0.6 to 0.8); anything above 0.3 is considered dense vegetation.(26) A monthly NDVI value was assigned by taking the average of two satellite images that are taken each month. Monthly values were then used to calculate trimester-specific and full pregnancy exposure estimates for our analysis. Annual estimated NDVI exposures are shown by census tract in Supplemental Figure S4.

Exposure to PM_2.5 and NDVI were characterized as follows (Supplemental Figure S5A and S5B, respectively): (1) continuous measures of mean values of PM_2.5 and NDVI (2) categorical measures of low and high PM_2.5 (low: <12 µg/m³, high: ≥12 µg/m³) and NDVI (low: <0.3, high: ≥0.3) and (3) average PM_2.5 and NDVI in the full pregnancy.(27) Categorical PM_2.5 cutoffs were chosen in accordance with the U.S. Environmental Protection Agency’s National Ambient Air Quality Standard of 12.0 µg/m³, lowered from 15.0 in 2012.(28) Thus, this value represents a conservative health-protective value for the study time period of 2000–2014. For NDVI, the cutoff of 0.3 was selected because values above 0.3 are considered dense vegetation.(26)

Statistical analysis

Demographic covariates were compared between the cases and controls and among cancer types in the cases, respectively. These were collected on birth certificate records and included maternal age at delivery (18–29 years, 30–34 years, 35+ years), maternal race and ethnicity (Non-Hispanic White, Non-Hispanic Black, Hispanic [all races], Non-Hispanic Asian or Pacific Islander, Non-Hispanic other race), maternal education (some high school or less, high school graduated/GED, some college/Associate’s degree, Bachelor’s degree, Advanced degree), child sex (male, female), year of birth (2000–2003, 2004–2007, 2008–2011, 2012–2014), age at cancer diagnosis (0–4, 5–9, 10–14 years old), mean values and low/high dichotomous values of PM_2.5 and NDVI by trimester, and average PM_2.5 and NDVI in the full pregnancy. Chi-square and Fisher’s tests were used for categorical variables, and the Kruskal-Wallis rank-sum test was applied for continuous variables to evaluate differences in demographic covariates between the cases and controls, as well as among mean monthly PM_2.5 and NDVI values by pregnancy year.

Logistic regression was used to estimate odds ratios (OR) and 95% confidence intervals (95% CI) between exposure to PM_2.5, NDVI, and each type of childhood cancer in each of the three trimesters and full pregnancy. When considering exposure to PM_2.5 in each trimester and full pregnancy, cancer-specific ORs (95% CI) were produced per PM_2.5 IQR change in controls (1^st trimester: 2.50 µg/m³; 2^nd trimester: 2.46 µg/m³; 3^rd trimester: 2.50 µg/m³; full pregnancy: 1.88 µg/m³). The following a priori confounders were evaluated for inclusion in our models: sex, maternal age at delivery, maternal race and ethnicity, maternal educational attainment, age at diagnosis, and year of birth (the matching variable).(29–39) These were selected based upon associations with pediatric cancer risk observed in the literature. As a sensitivity analysis, we also included census tract-level Yost index as a measure of community-level socio-economic status.

The final models included the following covariates: maternal age at delivery (18–29 years ref. group), maternal race and ethnicity (Non-Hispanic White ref. group), maternal education (some high school or less ref. group), child sex (female ref. group), year of birth (2000–2003 ref. group; the matching variable). Models were also mutually adjusted for PM_2.5 exposures in the other two trimesters (low PM_2.5 ref. group) to isolate the impacts of trimester-specific exposure. The same strategy was applied with NDVI as the main exposure variable. NDVI and PM_2.5 were both modeled independently of the other exposure (i.e., PM_2.5 was not adjusted for NDVI exposure and vice versa) and mutually adjusted for the other exposure. The main PM_2.5 model was not mutually adjusted for NDVI, while the main NDVI model was mutually adjusted for PM_2.5.

We used SAS version 9.2 (SAS Institute, Cary, NC) to conduct all logistic regression analysis and R version 4.2.2 to implement other statistical analysis. A two-sided alpha of lower than 0.05, and 95% confidence intervals that excluded the null value of 1 were utilized for identification of statistically significant effect estimates.(40) Further, the false discovery rate (FDR) method (Benjamini and Hochberg) was applied to adjust p-values generated from multiple comparisons in the multivariable logistic regression models to control the proportion of false positives in the significant results.(41) The FDR adjusted p-values lower than threshold determination (0.05) were considered statistically significant.

Results

Population characteristics

The study population included 20 cancer subtypes, the most common of which was lymphoid leukemias (n=304, 24.0%; Table 1), followed by neuroblastomas (150, 12.0%) and astrocytomas (134, 11.0%). The case and control populations differed with respect to two birth characteristics, namely race and ethnicity (p=<0.01) and sex (p=0.027). The case population included a larger proportion of non-Hispanic White (79.0% vs. 76.0%) and a lower proportion of non-Hispanic Black (5.1% vs. 8.2%) children than the control population. The case population also had an overrepresentation of children assigned male at birth (55.0% vs. 52.0%) as compared to controls.

Table 1.

Distribution of characteristics between cases and controls from the Minnesota birth and cancer registry linkage (2000–2014).

Characteristic	Controls, N = 5,245a		Cases, N = 1,272a		p-valueb
	N	%	N	%
Maternal age at delivery					0.09
18–29 years	2,964	57.0	689	54.0
30–34 years	1,451	28.0	356	28.0
35+ years	783	15.0	220	17.0
Missing	47		7
Maternal race/ethnicity					<0.01
Non-Hispanic White	3,974	76.0	1,003	79.0
Non-Hispanic Black	428	8.2	65	5.1
Hispanic (All races)	370	7.1	93	7.3
Non-Hispanic Asian or Pacific Islander	332	6.4	83	6.6
Non-Hispanic other race	113	2.2	23	1.8
Missing	28		5
Maternal education					0.05
Some high school or less	587	11.0	112	9.0
High school graduated/GED	1,265	25.0	288	23.0
Some college/associate degree	1,399	27.0	356	29.0
Bachelor’s degree	1,195	23.0	319	26.0
Advanced degree	701	14.0	173	14.0
Missing	98		24
Child sex					0.03
Female	2,531	48.0	570	45.0
Male	2,714	52.0	702	55.0
Year of birth					>0.9
2000–2003	1,759	34.0	418	33.0
2004–2007	1,846	35.0	452	36.0
2008–2011	1,185	23.0	291	23.0
2012–2014	455	8.7	111	8.7
Age at cancer diagnosis (years)					>0.9
0–4	0	NA	907	71.3
5–9	0	NA	287	22.6
10–14	0	NA	78	6.1
Missing	5,245		0
Cancer diagnosis					NA
Lymphoid leukemias	0	NA	304	24.0
Mature B-cell leukemias	0	NA	268	92.7c
Mature T-cell leukemias	0	NA	21	7.3c
Acute myeloid leukemias	0	NA	59	4.6
Other leukemias d	0	NA	34	2.7
Hodgkin lymphomas	0	NA	18	1.4
Non-Hodgkin lymphomas (except Burkitt lymphoma)	0	NA	28	2.2
Burkitt lymphoma	0	NA	18	1.4
Miscellaneous lymphoreticular neoplasms	0	NA	53	4.2
Ependymomas	0	NA	41	3.2
Astrocytomas	0	NA	134	11.0
Medulloblastomas	0	NA	37	2.9
Other intracranial and intraspinal embryonal e	0	NA	20	1.6
Other CNS cancer types f	0	NA	59	4.6
Neuroblastoma and other peripheral nervous cell tumors	0	NA	150	12.0
Retinoblastoma	0	NA	45	3.5
Renal tumors	0	NA	109	8.6
Hepatic tumors	0	NA	26	2.0
Malignant bone tumors	0	NA	19	1.5
Soft tissue and other extraosseous sarcomas	0	NA	56	4.4
Germ cell tumors, trophoblastic tumors and neoplasms of gonads	0	NA	43	3.4
Other malignant epithelial neoplasms and malignant melanomas	0	NA	19	1.5

^aFrequency (percentage).

^bPearson’s Chi-squared test; Fisher’s exact test; Wilcoxon rank sum test.

^cPercentages represent proportion of the total number of lymphoid leukemias.

^dChronic myeloproliferative diseases, Myelodysplastic syndrome, Unspecified and other specified leukemias.

^ePrimitive neuroectodermal tumor (PNET), Medulloepithelioma, Atypical teratoid/rhabdoid tumor.

^fOther gliomas, other specified intracranial and intraspinal neoplasms, unspecified intracranial and intraspinal neoplasms. 1st Trimester included 1st, 2nd and 3rd month of a pregnancy; 2nd Trimester included 4th, 5th and 6th month of a pregnancy; 3rd Trimester included 7th, 8th, 9th month of a pregnancy. Bold values indicated P-value less than 0.05.

PM_2.5 exposure tended to be low on average (<12.0 µg/m³) in both the case and control populations (Table 2, Figure 1, Table S1). There were no significant differences between the case and control populations with respect to average air pollution exposure and average residential greenness in trimesters or in the full pregnancy (all p-values >0.05). Mean annual PM_2.5 and NDVI as well as the total number of study cases and controls by census tract are visualized in the maps of Minnesota from 2000 to 2014 yearly (Supplemental Figures S2, S3, and S4, respectively).

Table 2.

PM_2.5 and NDVI exposure distribution between cases and controls per pregnancy period.

Characteristic	Controls, N = 5,245a		Cases, N = 1,272a		p-valueb
	Mean (SD)	(Min, Max)	Mean (SD)	(Min, Max)
Mean NDVI exposure
1st trimester	0.36(0.20)	(0.00, 0.85)	0.37(0.20)	(0.00, 0.83)	0.12
2nd trimester	0.37(0.18)	(0.00, 0.82)	0.37(0.19)	(0.00, 0.80)	0.70
3rd trimester	0.38(0.20)	(0.00, 0.85)	0.38(0.20)	(0.00, 0.80)	0.40
Full pregnancy	0.37(0.08)	(0.06, 0.65)	0.37(0.08)	(0.09, 0.60)	0.68
Mean PM2.5 exposure
1st trimester	8.71(2.01)	(3.95, 17.27)	8.63(2.00)	(3.90, 17.47)	0.12
2nd trimester	8.62(1.94)	(3.66, 17.11)	8.54(1.85)	(4.15, 16.62)	0.40
3rd trimester	8.58(1.97)	(3.81, 17.31)	8.61(1.97)	(4.08, 16.45)	0.60
Full pregnancy	8.63(1.48)	(4.43, 13.23)	8.58(1.44)	(4.76, 13.11)	0.27

Abbreviation: PM2.5: Particulate matter <2.5 mg/m³; NDVI: Normalized Difference Vegetation Index.

^aMean (SD), (Min, Max); Frequency (percentage).

^bPearson’s Chi-squared test; Fisher’s exact test; Wilcoxon rank sum test. 1st Trimester included 1st, 2nd and 3rd month of a pregnancy; 2nd Trimester included 4th, 5th and 6th month of a pregnancy; 3rd Trimester included 7th, 8th, 9th month of a pregnancy. Bold values indicated P-value less than 0.05.

Figure 1. Average PM_2.5 and NDVI exposure during the full pregnancy, by cancer subtype.

For PM_2.5, low is defined as <12.0 µg/m³; high ≥12.0 µg/m³. For NDVI, low is defined as <0.3; high ≥0.3.

PM_2.5 exposure and childhood cancer risk

There was no association between PM_2.5 exposure and risk of all cancers combined for either trimester-specific or full pregnancy estimates (Table 3). However, we observed elevated effect estimates associated with each interquartile range increase in exposure throughout the pregnancy period for multiple individual cancers, namely leukemias, lymphomas, and CNS tumors. Increasing PM_2.5 exposure during the first trimester was associated with increased odds of Burkitt lymphoma (OR: 1.93, 95% CI: 1.13–3.30) and miscellaneous lymphoreticular neoplasms (OR: 1.43, 95% CI: 1.01–2.01). Notably for Burkitt, effect estimates were non-significantly elevated in all trimesters and for the full pregnancy estimate, though confidence intervals were wide. Increasing exposure in the second trimester was associated with significantly increased odds of Hodgkin lymphoma (OR: 2.60, 95% CI: 1.52–4.45). Increasing exposure during the third trimester was associated with significantly increased odds of lymphoid leukemias (OR: 1.17, 95% CI: 1.01–1.37), specifically B-cell leukemia (OR: 1.20, 95% CI: 1.02–1.41), as well as intracranial and intraspinal embryonal neoplasms (OR: 1.90, 95% CI: 1.07–3.37). Elevated effect estimates associated with increases in exposure in the third trimester was also observed for T-cell leukemias and at a greater magnitude, though it did not reach significance (OR: 1.64, 95% CI: 0.97–2.76). After adjusting for multiple comparisons, only the association with Hodgkin lymphoma retained significance (FDR p<0.01).

Table 3.

Odds ratios (OR) and 95% confidence intervals (CI) for the association between an interquartile range increase in PM_2.5 and childhood cancer risk, per pregnancy period.

	First Trimester (IQR: 2.50) a					2nd Trimester (IQR: 2.46) b				Third Trimester (IQR: 2.50) c				Full Pregnancy (IQR: 1.88) d
Cancer type	N, 5069	OR	95% CI		FDR P	OR	95% CI		FDR P	OR	95% CI		FDR P	OR	95% CI		FDR P
All cancer combined	1236	0.96	0.88	1.04	0.78	0.93	0.86	1.02	0.63	1.04	0.95	1.13	0.77	0.95	0.87	1.04	0.62
Lymphoid leukemias	297	0.99	0.85	1.16	0.98	0.92	0.79	1.09	0.75	1.17	1.01	1.37	0.27	1.03	0.87	1.23	0.94
Mature B-cell leukemias	263	1.01	0.86	1.20	0.98	0.96	0.81	1.14	0.86	1.20	1.02	1.41	0.27	1.07	0.90	1.28	0.69
Mature T-cell leukemias	20	0.91	0.50	1.65	0.98	1.06	0.57	1.97	0.95	1.64	0.97	2.76	0.33	1.34	0.70	2.56	0.62
<1 year	17	0.62	0.30	1.27	0.68	0.57	0.27	1.17	0.63	1.20	0.64	2.25	0.83	0.65	0.33	1.29	0.62
1–5 years	230	1.05	0.88	1.25	0.97	0.93	0.78	1.12	0.75	1.13	0.95	1.34	0.45	1.04	0.86	1.26	0.83
Acute myeloid leukemias	58	0.95	0.66	1.37	0.98	0.84	0.58	1.23	0.75	1.14	0.81	1.62	0.80	0.98	0.67	1.43	0.95
Other leukemias e	32	0.76	0.47	1.23	0.78	0.80	0.49	1.29	0.75	1.01	0.64	1.59	0.98	0.77	0.48	1.23	0.62
Hodgkin lymphomas	16	0.98	0.53	1.83	0.98	2.60	1.52	4.45	0.01	0.89	0.45	1.77	0.88	1.92	0.95	3.87	0.43
Non-Hodgkin lymphomas	27	0.51	0.29	0.90	0.17	0.80	0.47	1.34	0.75	0.67	0.40	1.14	0.45	0.62	0.38	1.00	0.43
Burkitt lymphoma	18	1.93	1.13	3.30	0.17	1.24	0.68	2.28	0.76	1.38	0.74	2.54	0.70	1.72	0.86	3.44	0.62
Miscellaneous lymphoreticular neoplasms	52	1.43	1.01	2.01	0.27	1.05	0.72	1.51	0.95	1.09	0.76	1.58	0.83	1.32	0.86	2.01	0.62
Ependymomas	40	1.23	0.82	1.84	0.78	0.83	0.53	1.30	0.75	0.95	0.60	1.49	0.89	0.97	0.62	1.51	0.95
Astrocytomas	130	0.74	0.58	0.95	0.17	1.01	0.80	1.28	0.95	0.84	0.66	1.07	0.45	0.80	0.63	1.01	0.43
Medulloblastomas	36	0.67	0.42	1.08	0.42	0.90	0.57	1.43	0.87	0.86	0.54	1.36	0.81	0.73	0.47	1.13	0.62
Other intracranial and intraspinal embryonal f	19	1.13	0.61	2.06	0.98	0.66	0.33	1.31	0.65	1.90	1.07	3.37	0.27	1.17	0.59	2.32	0.86
Other CNS cancer types g	57	1.01	0.72	1.41	0.98	0.79	0.55	1.13	0.63	1.30	0.94	1.79	0.45	0.96	0.67	1.37	0.95
Neuroblastoma and other peripheral nervous cell tumors	148	0.93	0.75	1.16	0.97	0.79	0.63	0.99	0.48	0.79	0.63	0.99	0.27	0.76	0.61	0.96	0.43
Retinoblastoma	43	1.10	0.73	1.65	0.98	1.02	0.67	1.55	0.95	1.21	0.80	1.82	0.75	1.17	0.74	1.84	0.83
Renal tumors	107	0.93	0.72	1.20	0.97	0.94	0.72	1.21	0.86	1.07	0.83	1.38	0.83	0.97	0.75	1.27	0.95
Hepatic tumors	24	1.18	0.72	1.94	0.97	0.99	0.58	1.68	0.97	0.98	0.57	1.67	0.97	1.14	0.66	1.95	0.86
Malignant bone tumors	19	1.29	0.75	2.22	0.81	1.44	0.82	2.51	0.63	0.93	0.50	1.73	0.89	1.39	0.75	2.59	0.62
Soft tissue and other extraosseous sarcomas	54	0.97	0.67	1.41	0.98	0.77	0.52	1.15	0.63	1.28	0.89	1.84	0.45	0.99	0.68	1.46	0.97
Germ cell tumors, trophoblastic tumors and neoplasms of gonads	40	0.67	0.43	1.05	0.41	1.03	0.69	1.55	0.95	1.09	0.73	1.61	0.85	0.85	0.55	1.32	0.83
Other malignant epithelial neoplasms and malignant melanomas	19	0.98	0.54	1.76	0.98	1.50	0.82	2.73	0.63	0.78	0.39	1.56	0.80	1.16	0.62	2.17	0.86

Abbreviation: PM_2.5: Particulate matter <2.5 mg/m³; FDR: False Discovery Rate. 1st Trimester included 1st, 2nd and 3rd month of a pregnancy; 2nd Trimester included 4th, 5th and 6th month of a pregnancy; 3rd Trimester included 7th, 8th, 9th month of a pregnancy. Bold values indicated P-value less than 0.05.

^aAdjusted for trimester 1 PM2.5 for each control IQR increase from trimester 1, maternal age at delivery, maternal race/ethnicity, maternal education, child sex, year of birth, trimester 2 PM2.5 (high/low), trimester 3 PM2.5 (high/low).

^bAdjusted for trimester 2 PM2.5 for each control IQR increase from trimester 2, maternal age at delivery, maternal race/ethnicity, maternal education, child sex, year of birth, trimester 1 PM2.5 (high/low), trimester 3 PM2.5 (high/low).

^cAdjusted for trimester 3 PM2.5 for each control IQR increase from trimester 3, maternal age at delivery, maternal race/ethnicity, maternal education, child sex, year of birth, trimester 1 PM2.5 (high/low), trimester 2 PM2.5 (high/low).

^dAdjusted for PM2.5 for each control IQR increase from full pregnancy, maternal age at delivery, maternal race/ethnicity, maternal education, child sex, year of birth.

^eChronic myeloproliferative diseases, Myelodysplastic syndrome, Unspecified and other specified leukemias.

^fPrimitive neuroectodermal tumor (PNET), Medulloepithelioma, Atypical teratoid/rhabdoid tumor.

^gOther gliomas, Other specified intracranial and intraspinal neoplasms, Unspecified intracranial and intraspinal neoplasms.

Conversely, there was a consistent inverse relationship between increasing PM_2.5 exposure during pregnancy and risk of neuroblastoma and other peripheral nervous cell tumors. PM_2.5 exposure in the second and third trimesters was associated with 21% decreased odds of neuroblastoma (OR: 0.79, 95% CI: 0.63–0.99). Similarly, PM_2.5 exposure throughout the pregnancy was associated with 24% decreased odds of neuroblastoma (OR: 0.76, 95% CI: 0.61–0.96).

Mutual adjustment of the association between cancer risk and PM_2.5 exposure by residential greenness changed only a few associations (Supplemental Table S2). First, for mature T-cell leukemias, the magnitude of the observed association increased significantly, from 1.64 (95% CI: 0.97–2.76) to 2.02 (95% CI: 1.10–3.74) for exposure during the third trimester. Second, the association between PM_2.5 exposure and Hodgkin lymphoma during the second trimester was attenuated from 2.60 (95% CI: 1.52–4.45) to 2.39 (95% CI: 1.36–4.19), though the odds remain significantly increased. Sensitivity analyses adjusting for community-level socio-economic status yielded no substantial changes, though some cancers saw nominal changes to statistical significance (Supplemental Table S3).

Residential greenness and childhood cancer risk

There was no association between increasing residential greenness and risk of all cancers combined for either trimester-specific or full pregnancy estimates (Table 4), adjusted for low and high PM_2.5 exposure. Increasing greenness during the first trimester was associated with 66% decreased risk of malignant epithelial neoplasms and melanomas (OR: 0.34, 95% CI: 0.13–0.91). Increasing second trimester exposure was associated with a decreased risk of other leukemias (i.e., chronic myeloproliferative diseases, myelodysplastic syndrome, unspecified and other specified leukemias; OR: 0.53, 95% CI: 0.29–0.97) and astrocytomas (OR: 0.71, 95% CI: 0.53–0.95). All associations had an FDR-adjusted p>0.05. Unexpectedly, increasing greenness was also associated with renal tumors (second trimester OR: 1.52, 95% CI 1.11–2.08; full pregnancy OR: 1.30, 95% CI: 0.98–1.72) and neuroblastoma and other peripheral tumors (third trimester OR: 1.44, 95% CI: 1.03–2.03). Notably, neuroblastoma risk also exhibited an unexpected relationship with PM_2.5 exposure.

Table 4.

Odds ratios (OR) and 95% confidence intervals (CI) for the association between an interquartile range increase in the average normalized difference vegetation index (NDVI) and childhood cancer risk, per pregnancy period, with adjustment for exposure to PM_2.5 (high/low).

	First Trimester (IQR: 0.32) a					Second Trimester (IQR: 0.29) b				Third Trimester (IQR: 0.33) c				Full Pregnancy (IQR: 0.11) d
Cancer type	N, 5059	OR	95% CI		FDR P	OR	95% CI		FDR P	OR	95% CI		FDR P	OR	95% CI		FDR P
All cancer combined	1231	1.11	0.97	1.26	0.69	0.97	0.88	1.08	0.86	0.98	0.86	1.11	0.99	1.00	0.92	1.09	0.99
Lymphoid leukemias	297	1.14	0.90	1.45	0.69	0.98	0.81	1.18	0.86	1.03	0.81	1.32	0.99	1.03	0.87	1.22	0.99
Mature B-cell leukemias	263	1.16	0.90	1.50	0.69	0.98	0.80	1.19	0.86	0.98	0.76	1.28	0.99	1.03	0.86	1.23	0.99
Mature T-cell leukemias	20	0.90	0.36	2.25	0.87	1.32	0.64	2.72	0.74	2.01	0.79	5.12	0.88	1.20	0.63	2.27	0.99
<1 year	17	1.52	0.59	3.94	0.69	0.70	0.33	1.49	0.73	1.03	0.39	2.70	0.99	0.88	0.45	1.71	0.99
1–5 years	230	1.12	0.85	1.46	0.73	1.04	0.84	1.29	0.86	1.08	0.82	1.42	0.99	1.07	0.88	1.29	0.99
Acute myeloid leukemias	58	1.06	0.62	1.79	0.87	1.06	0.69	1.63	0.86	0.77	0.45	1.32	0.88	0.98	0.68	1.42	0.99
Other leukemias e	31	0.88	0.43	1.81	0.87	0.53	0.29	0.97	0.32	0.91	0.44	1.89	0.99	0.63	0.38	1.03	0.82
Hodgkin lymphomas	16	0.73	0.25	2.09	0.87	0.63	0.26	1.50	0.73	0.57	0.21	1.60	0.88	0.57	0.27	1.17	0.90
Non-Hodgkin lymphomas	27	1.62	0.74	3.58	0.69	0.99	0.53	1.84	0.97	1.72	0.79	3.73	0.88	1.42	0.83	2.44	0.90
Burkitt lymphoma	18	0.43	0.16	1.14	0.69	0.65	0.29	1.44	0.73	0.99	0.37	2.64	0.99	0.67	0.34	1.33	0.90
Miscellaneous lymphoreticular neoplasms	52	1.08	0.62	1.87	0.87	1.11	0.72	1.73	0.86	0.72	0.41	1.27	0.88	0.99	0.67	1.46	0.99
Ependymomas	40	1.18	0.62	2.23	0.87	0.91	0.54	1.51	0.86	0.77	0.40	1.48	0.92	0.89	0.57	1.39	0.99
Astrocytomas	130	1.19	0.83	1.70	0.69	0.71	0.53	0.95	0.25	1.08	0.75	1.54	0.99	0.94	0.73	1.21	0.99
Medulloblastomas	36	1.64	0.83	3.24	0.69	0.75	0.44	1.27	0.73	1.00	0.51	1.94	0.99	0.99	0.62	1.58	0.99
Other intracranial and intraspinal embryonal f	19	1.65	0.64	4.23	0.69	0.67	0.32	1.40	0.73	0.85	0.33	2.15	0.99	0.93	0.49	1.77	0.99
Other CNS cancer types g	55	1.04	0.61	1.79	0.89	1.19	0.77	1.84	0.74	0.79	0.45	1.37	0.89	1.05	0.72	1.54	0.99
Neuroblastoma and other peripheral nervous cell tumors	146	1.26	0.90	1.75	0.69	1.14	0.88	1.48	0.73	1.44	1.03	2.03	0.87	1.15	0.91	1.46	0.90
Retinoblastoma	43	1.47	0.80	2.73	0.69	1.06	0.65	1.73	0.86	0.81	0.44	1.52	0.99	1.03	0.67	1.59	0.99
Renal tumors	107	1.19	0.80	1.77	0.69	1.52	1.11	2.08	0.21	0.96	0.64	1.43	0.99	1.30	0.98	1.72	0.82
Hepatic tumors	24	1.09	0.49	2.45	0.87	0.58	0.29	1.17	0.64	0.67	0.29	1.57	0.88	0.68	0.38	1.20	0.90
Malignant bone tumors	19	0.61	0.23	1.60	0.69	1.62	0.73	3.61	0.73	0.47	0.18	1.20	0.88	0.85	0.44	1.64	0.99
Soft tissue and other extraosseous sarcomas	54	0.92	0.53	1.59	0.87	1.21	0.78	1.89	0.74	0.77	0.44	1.35	0.88	0.99	0.68	1.46	0.99
Germ cell tumors, trophoblastic tumors and neoplasms of gonads	40	1.10	0.59	2.06	0.87	0.66	0.40	1.10	0.64	1.01	0.53	1.91	0.99	0.85	0.55	1.32	0.99
Other malignant epithelial neoplasms and malignant melanomas	19	0.34	0.13	0.91	0.69	1.14	0.53	2.47	0.86	1.60	0.60	4.27	0.88	0.98	0.51	1.87	0.99

Abbreviation: PM2.5: Particulate matter <2.5 mg/m3; NDVI: Normalized Difference Vegetation Index; FDR: False Discovery Rate.

^aAdjusted for trimester 1 NDVI for each control IQR increase from trimester 1, maternal age at delivery, maternal race/ethnicity, maternal education, child sex, year of birth, trimester 2 NDVI (high/low), trimester 3 NDVI (high/low), trimester 1 PM2.5 (high/low), trimester 2 PM2.5 (high/low), trimester 3 PM2.5 (high/low).

^bAdjusted for trimester 2 NDVI for each control IQR increase from trimester 2, maternal age at delivery, maternal race/ethnicity, maternal education, child sex, year of birth, trimester 1 NDVI (high/low), trimester 3 NDVI (high/low), trimester 1 PM2.5 (high/low), trimester 2 PM2.5 (high/low), trimester 3 PM2.5 (high/low).

^cAdjusted for trimester 3 NDVI for each control IQR increase from trimester 3, maternal age at delivery, maternal race/ethnicity, maternal education, child sex, year of birth, trimester 1 NDVI (high/low), trimester 2 NDVI (high/low), trimester 1 PM2.5 (high/low), trimester 2 PM2.5 (high/low), trimester 3 PM2.5 (high/low).

^dAdjusted for NDVI for each control IQR increase from full pregnancy, maternal age at delivery, maternal race/ethnicity, maternal education, child sex, year of birth, trimester 1 PM2.5 (high/low), trimester 2 PM2.5 (high/low), trimester 3 PM2.5 (high/low).

^eChronic myeloproliferative diseases, Myelodysplastic syndrome, Unspecified and other specified leukemias.

^fPrimitive neuroectodermal tumor (PNET), Medulloepithelioma, Atypical teratoid/rhabdoid tumor.

^gOther gliomas, Other specified intracranial and intraspinal neoplasms, unspecified intracranial and intraspinal neoplasms.

With respect to unadjusted models, mutual adjustment for PM_2.5 significantly attenuated the previously observed protective association between exposure to residential greenness during the full pregnancy and risk of hepatic tumors (unadjusted OR: 0.55, 95% CI: 0.33–0.92; mutually adjusted OR: 0.68, 95% CI: 0.38–1.20; Supplemental Table S4). There were no other substantial changes between unadjusted and adjusted models, nor in the sensitivity analyses including Yost index (Supplemental Table S5).

Discussion

In our study of 1,236 Minnesota children with cancer, there were no associations between each interquartile range increase in ambient PM_2.5 exposure during pregnancy for all cancers combined. Among single cancers, exposure to high levels of PM_2.5 during pregnancy was associated with increased risk of 17% and 20% for lymphoid and B-cell leukemias, 93% and 43% for Burkitt and Hodgkin lymphomas, and 90% for intracranial and intraspinal embryonal neoplasms. Some effects were trimester-specific. Increasing residential greenness was associated with 66% reduced risk of skin cancers including malignant epithelial neoplasms and melanomas, and an unexpected increased risk of 52% and 44% for renal tumors and neuroblastoma after adjusting for ambient air pollution exposure.

The results observed in this study largely agree with the literature on air pollution exposure and childhood cancer risk, with PM_2.5 exposure being associated with increased risk for leukemias,(9,42–47) lymphomas,(10,48) and CNS tumors.(3,49–51) In a previous study in Texas, increasing residential PM_2.5 exposure was associated with a 10% increase in risk of all childhood cancers combined, which we did not observe here.(12) However, similar increases in risk were observed for lymphoid leukemias (15%) and Hodgkin (27%) lymphomas. For leukemias in particular, our results also agree with the literature supporting that prenatal environmental exposures (e.g., pesticides, air pollution) may provide the “first hit” in the two-hit hypothesis of leukemia development.(2,52–57) Contrary to existing literature, we also observed some indication of reduction in risk associated with increasing PM_2.5 exposure during pregnancy for neuroblastoma and other peripheral nervous cell tumors. The potential mechanism behind this is unclear. We also did not observe the increased retinoblastoma risk seen by Heck et al.(9) Given the inconsistencies with existing literature, it is possible that these results could be explained by live birth bias or be due to chance.

With respect to greenness, the few studies that have examined the potential mediating effect of residential greenness on air pollution and health outcomes have presented mixed results,(12,16,58–60) indicating that the relationship is likely complex. The majority of studies of greenness and cancer have been in adults, and have generally seen a decreased risk in cancers including lung, prostate, skin, bladder, and breast.(61–63) One study of childhood leukemia found reduced risk associated with residential greenness exposure at diagnosis,(64) but we did not see this effect in our study. The prior study of PM_2.5 and greenness in Texas(12) reported that an increase in malignant melanoma risk of 75% was associated with increasing residential greenness, where we observed a 66% reduction in risk. The prior study postulated that this could be due to residential greenness being associated with increased outdoor activity and UV exposure.(65) However, residential greenness could also provide shade from UV exposure, resulting in a protective effect. Notably, Minnesota has a higher incidence of UV-attributable melanoma among non-Hispanic White individuals than Texas.(66) Finally, we observed an unexpected increase in risk for renal tumors and neuroblastoma associated with residential greenness. It is possible that this could be related to pesticide exposure if a high proportion of the residential greenness is agricultural.(67)

This study has several strengths. It is one of the largest population-based observational studies examining how exposure to PM_2.5 and NDVI during the pregnancy impacted the risks of major pediatric cancer types in the United States to date. The sample sizes of our study enabled us to provide more precise estimates of the relationships among PM_2.5, NDVI and the risks of pediatric cancer, including rare childhood cancer types, and allowed us to conduct subgroup analyses to assess the different risks of pediatric cancer subtypes, such as B- or T- subtypes of ALL, thereby strengthening the reliability of our conclusions. Another strength of our study was exploring the influence of air pollution and greenness on childhood cancer risks specifically during each trimester of pregnancy. The three trimesters of pregnancy are critical windows of embryonic and fetal growth and development, and environmental exposures during these periods, particularly the first trimester, have been linked to risks of pediatric cancer and other childhood health conditions. Our study results uncovered variations in whether and to what extent PM_2.5 and NDVI were associated with the risks of pediatric cancer during the first, second, and third trimesters and the full span of pregnancy. This generated more robust results in our study on how environmental exposures interacted with fetal growth in different time spans of pregnancy, highlighting the serious implications of these environmental issues.

This study has several notable limitations. First, we excluded a number of cases and controls whose births either fell outside of the bounds of our study window or who lacked geocoded residential information necessary to conduct exposure assessment. These individuals were more likely to be non-Hispanic White and have an older age at diagnosis than the included individuals, which could introduce bias into our study. However, these proportions were similar between excluded cases and controls. When we adjust for multiple comparisons, many of the previously statistically significant associations are reduced, with only the associations with Hodgkin lymphoma retaining significance. Some cancer sizes have very small sample sizes due to rarity, limiting our ability to draw inference. However, we felt it important to retain these groups for two reasons. First, many are understudied in the epidemiologic literature (e.g., sarcomas). Second, these small subgroups can be used by us or other researchers in future meta-analyses. Exposure was assigned at the census tract level due to the varying resolutions of PM_2.5 and NDVI data available, which may have masked some variation that may be detectable on a finer scale. While our NDVI exposure assessment method is well-defined, we used a single source of data, which may not capture all possible variability. While our time-varying exposure assessment methods should adequately capture seasonal variability in our exposures of interest, we are unable to adjust for potentially seasonally variable co-exposures that may impact cancer risk, such as infection and UV exposure. Additionally, residential mobility during pregnancy could introduce exposure misclassification if the address listed on the birth certificate was not where the birthing parent lived during pregnancy.(68,69) However, exposure misclassification from residential mobility is generally assumed to be nondifferential,(70) which would bias any observed associations toward the null. Additionally, we did not have access to residential history after birth to estimate exposure from birth to diagnosis. However, prior research in a childhood cancer population has demonstrated a minimal effect of residential mobility between birth and cancer diagnosis on exposure assessment for air pollution.(49)

In our study of Minnesota children with cancer, we observed elevated risk associated with PM_2.5 exposure during pregnancy for leukemias, lymphomas, and CNS tumors. Increasing residential greenness was associated with reduction in risk for skin cancers and an unexpected increased risk for renal tumors and neuroblastoma, after adjusting for ambient air pollution exposure. As PM_2.5 exposure is also associated with significantly poorer survival for children suffering cancer,(71) identifying ways to reduce exposure and mitigate risk remains of critical importance. These findings demonstrate the complexities in the relationship between PM_2.5 and NDVI in childhood cancer etiology and emphasize the need for more studies of these relationships.

Data Availability

Case and control data can be requested from the Minnesota Department of Health. All air pollution and greenness data are publicly available as described in the Methods (PM_2.5: https://sites.wustl.edu/acag/datasets/surface-pm2-5/; NDVI: https://developers.google.com/earth-engine/datasets/catalog/MODIS_061_MOD13Q1/#dois).