Residential proximity to intensive animal agriculture and risk of lymphohematopoietic cancers in the Agricultural Health Study

Abstract

Background:

Although occupational animal exposure has been associated with lymphohematopoietic malignancies, no studies have evaluated cancer risks associated with living near intensive animal agriculture.

Methods:

We linked participants in the prospective Agricultural Health Study to a database of permitted animal feeding operations in Iowa. We created metrics to reflect the intensity of animal exposures within 2 and 5km of the participants’ residences, enumerating both total and inverse distance-weighted (IDW) animal units (AUs), which are standardized by animal size and manure production. We estimated risk of lymphohematopoietic malignancies and subtypes (hazard ratio, HR; 95% confidence interval, 95%CI), adjusting for demographic and farming-related factors, including occupational pesticide exposure. We examined associations by animal type and animal-related work activities.

Results:

We observed 519 and 211 cases (1993-2015) among 32,635 pesticide applicators and 19,743 spouses, respectively. Among applicators, no associations were evident with AUs within 2km, but risk of any lymphohematopoietic cancer was elevated across IDW AU quintiles within 5km. Risk of non-Hodgkin lymphoma (NHL) was elevated for Q2 (HR=1.52; 95%CI=1.09-2.10), Q3 (HR=1.57; 95%CI=1.13-2.17) and Q4 (HR=1.74; 95%CI=1.26-2.40) of IDW AUs within 5km (p_trend=0.52) and increased with dairy cattle AUs (p_trend=0.04). We found positive trends for leukemia and some NHL subtypes with increasing number of both beef and dairy cattle (p_trend<0.05). Risks did not vary by animal-related work (p_interaction=0.61). Associations using the total and IDW metrics were similar. Associations were inconsistent among spouses.

Conclusion:

Residential proximity to intensive animal agriculture may influence the risk of NHL and leukemia, even after consideration of occupational animal contact and pesticide exposures.

INTRODUCTION

Animal feeding operations (AFOs) are facilities where farmed animals, their wastes, and other by-products of their production are stored over small land areas¹. These operations can be major sources of hazardous emissions into the surrounding environment, including microorganisms, endotoxins, nitrate, pesticides, antibiotics, and hormones, as well as particulate matter and gaseous emissions of ammonia, hydrogen sulfide, and other odorous gases^2–6. These byproducts are transported through multiple environmental media and can consequently reach nearby areas through air, soil/sediment, ground- and surface-water contamination^7,8.

Occupational studies have demonstrated links between agricultural work and lymphohematopoietic cancers, particularly non-Hodgkin lymphoma (NHL), mostly among male^9–16, but also female¹⁷ farmers. Many of these studies specifically examined pesticides as the exposure of interest; however, several of these investigations and others suggest that animal exposures may independently influence lymphohematopoietic cancer risk. These include studies of NHL among workers in non-farming occupations such as among veterinarians and meat inspectors^18,19, case-control studies of leukemia and NHL using county-level animal inventories^20,21, and cohort studies where positive associations between NHL and occupational animal exposures remained significant after adjusting for pesticide use^16,22. Studies examining proximal animal exposures may elucidate whether direct animal contact or exposure to area-level environmental emissions from AFOs could be responsible for these associations.

The Agricultural Health Study (AHS) is a prospective cohort of licensed pesticide applicators (mostly farmers) and their spouses in Iowa and North Carolina, with elevated incidence and mortality of several lymphohematopoietic cancers compared to the general U.S. population^23–25. Several pesticides have been linked to an increased risk of lymphohematopoietic cancers in this cohort^26–28, and farming characteristics such as raising certain types of animals have been identified as independent risk factors for NHL¹⁶. In this study, we used a database of AFOs to evaluate whether the intensity of animal agriculture near the residence influences the risk of lymphohematopoietic cancers in AHS participants in Iowa. We examined these associations while considering animal-related farm activities, occupational contact with animals, pesticide use, and various lymphohematopoietic cancer risk factors.

METHODS

Study Population

The AHS study design has been described elsewhere²⁹; key components are highlighted here. This prospective cohort includes 52,394 licensed private pesticide applicators residing in Iowa and North Carolina, 32,345 of their spouses, and 4,916 commercial pesticide applicators in Iowa.

Applicators and spouses completed questionnaires during enrollment (1993-1997) to report their use of specific pesticides and health and lifestyle factors such as smoking, alcohol consumption, selected medical conditions, and demographic information. Applicators were asked detailed information on pesticide application and farming activities, including types of animals raised and numbers of livestock and poultry on the farm. Spouses provided more limited information on farming activities but were asked explicitly about their direct contact with cattle, swine, poultry and sheep in the last 12 months. Since the study initiation, applicators and spouses have been re-interviewed about pesticide use and lifestyle and other characteristics in follow-up surveys conducted by phone (1999-2005 and 2005-2010).

Because we were unable to identify external AFO data sources of comparable quality in North Carolina, we limited these analyses to the Iowa participants (N=58,563; 36,792 applicators and 21,771 spouses). We assessed cancer incidence by linkage to the Iowa Cancer Registry from enrollment through December 31, 2015 and determined vital status by matching to the Iowa mortality files and the National Death Index. Follow-up time was censored at the date of any cancer diagnosis, date of death, migration out of state or the date of last follow-up or contact, whichever was earlier. Subtypes of lymphohematopoietic malignancies were defined according to the Surveillance, Epidemiology, and End Results Program Lymphoma Subtype Recodes³⁰. We evaluated subtypes of leukemia and NHL in which there were ≥15 total cases and ≥3 cases in each exposure category. The only leukemia subtype that met these criteria was acute myeloid leukemia (AML). NHL subtypes included total B-cell, chronic lymphocytic leukemia/ small lymphocytic lymphoma (CLL/SLL), diffuse large B-cell lymphoma (DLBCL), follicular lymphoma (FL), and multiple myeloma (MM).

Animal Feeding Operations (AFOs) Database

We obtained a database of 11,395 current and historical AFOs maintained by the Iowa Department of Natural Resources (IDNR; hereafter “AFOs database”) in March 2017³¹. Operations were a combination of confinements and open feedlots. Permitted animal inventories for swine, dairy cattle, beef cattle, chickens, turkeys, horses, sheep/lamb/goats, and other animals were available as animal units (AU), a metric standardized according to animal weight and manure production (1 AU=1 mature cow [1000-pounds]). The database included both large permitted operations required based on animal volumes under federal and state laws to submit manure management plans (MMPs; >500 AUs for confinements, >1000 AUs for open feedlots) that describe how wastes will be managed on site³², as well as some smaller facilities that are reported on a voluntary basis. The AFOs database also included the geographic location, date of first approval of the MMP (i.e., operation initiation) and some ancillary information (e.g., business name).

Locations in the AFOs database (latitude and longitude) included two indicators of location quality as determined by IDNR: the geocoding method and a measure of the assumed accuracy (in meters). Most of the methods reflected classifications typical of commercial geocoding packages, including locations confirmed at the address-level or through aerial imagery. However, indicative of the rural location of most of these operations, many were located to the centroid of public land survey sections, from quarter section (1/4 sq mi) to half section (1/2 sq mi). Seven facilities with missing geocoding method or location accuracy >1500m and 25 duplicates (same name, address, operation type, and geocode) were excluded prior to linkage with the cohort. Of the 11,363 remaining facilities, 4.9% were missing total AU counts (1.9% of permitted and 11.6% of voluntarily reported). Because there was no indication that these facilities were non-operational, we replaced missing AUs with the median AU value from all permitted or volunteer facilities in the database, respectively.

Exposure metrics

Of the Iowa participants, 91.8% had a well-geocoded (i.e., matched to a street address) enrollment address suitable for linkage to the AFOs database. Applicators and spouses primarily resided together at enrollment; therefore, these addresses reflected N=32,772 unique locations from which to estimate AFO exposures.

We used the AFOs database to derive location-and animal-specific AFO-related exposure metrics for each participant. We first generated individual-level proximity-based exposure estimates reflecting the density of all AFOs within a 2 and 5km radius of each participant’s residence, distances selected based on hypothetical transport of ambient air pollutants¹. Exposure metrics incorporated the AU counts for each AFO in one of two ways: 1) the summation of all AUs (total and animal-specific) within 2 or 5km; and 2) an inverse distance-weighted (IDW) sum of the AUs within 2 or 5km, weighted by the continuous linear distance from the AFO to the residence.

Statistical analysis

We excluded participants with poor geocoded enrollment addresses (N=3,429 applicators; 1,335 spouses) or who self-reported a diagnosis of cancer prior to enrollment (N=728 applicators; 693 spouses). After these exclusions, N=32,635 applicators and N=19,743 spouses remained eligible for analyses and were evaluated separately. Participants retained for analysis were similar to those excluded in regard to demographic and lifestyle-related characteristics.

We used Cox proportional hazards regression to estimate hazard ratios (HR) and 95% confidence intervals (95%CI) for associations between AU metrics and lymphohematopoietic cancer risks overall and for specific sites and their subtypes. Models compared risks among tertiles or quintiles of total and IDW AUs and animal-specific AUs to referent groups of those in the lowest tertile/quintile. Where the proportion of participants with no animal exposure exceeded the cutpoint of the lowest quintile (i.e., >20%) or tertile (i.e., >33%), associations were estimated among quartiles or by categories split according to the median value (< and ≥ median) among the exposed and compared to the group with no exposure. The median of each exposure category was parameterized as a continuous variable for tests of linear trend. We used p<0.05 as the criterion for statistical significance. Models of animal-specific AUs were mutually adjusted for other animal types.

We evaluated potential confounding based on data collected at enrollment. All models were adjusted for age at baseline (continuous), smoking history (never, tertiles of pack-years among former smokers, tertiles of pack-years among current smokers), body mass index (BMI; <25, 25 to <30, ≥30), alcohol use (none, one drink a week or less, 2 or more drinks a week), education (high school or less, greater than high school), applicator type (private vs commercial), and family history of cancer. Missing covariate values were included as separate categories. We evaluated additional potential confounders by assessing >10% change on risk estimates. These included factors linked with lymphohematopoietic cancer in other AHS evaluations (e.g., doctor diagnosis of allergy, grew up on a farm) or relevant to animal exposure (e.g., raising animals on the farm, work in swine or poultry confinements, grinding animal feed, perform veterinary procedures). We also implemented models with and without adjustment for occupational use of pesticides previously associated with lymphohematopoietic cancers in the AHS (lindane, DDT, permethrin, terbufos, diazinon, glyphosate, metolachlor). This was parameterized among applicators as quartiles of intensity-weighted lifetime use³³ and for spouses, by whether the pesticide was ever used.

We assessed multiplicative interactions between AFO exposures and animal-related work or self-reported direct contact with animals on lymphohematopoietic risk. For applicators, animal-related work was defined as either working in swine or poultry confinement or by raising cattle, swine, poultry as a major source of income on the farm. For spouses, animal contact was defined as direct contact with beef/dairy cattle, swine, poultry or sheep in the last 12 months. We examined interactions via stratified associations compared to a common reference group comprised of the lowest quintile of AUs and the non-exposed group for the potential effect modifier (e.g., no animal-related work, not raising animals on the farm, not working in confinements, or no direct animal contact). P-heterogeneity was derived from likelihood ratio tests comparing fit between models including and excluding product interaction terms.

Several sensitivity analyses were conducted to evaluate the robustness of observed associations. We computed associations for AU exposure metrics after excluding voluntarily-reported facilities and restricting to facilities known to be in operation prior to 2005, the mid-point of the follow-up period. Model adjustment for sex and race was precluded by limited variability in these factors, thus we ran models restricted to male applicators, female spouses, and whites.

RESULTS

A total of 519 and 211 incident cases of lymphohematopoietic cancers (424 and 182 cases of NHL, 72 and 21 cases of leukemia) were observed among applicators and spouses, respectively over 22 years of follow-up. Approximately 52% of participants lived within 2km of at least 1 AFO and 91% within 5km. Applicators in the highest quintile of total AUs within 5km of the residence compared to those in the lowest quintile of total AUs were more likely to hold a private pesticide license (92.1 vs. 77.8%), raise animals on their farm (70.3 vs. 54.3%), perform animal-related farm activities (92.6 to 79.1%), as well as raise higher numbers of livestock and poultry (Table 1). Compared to other quintiles, more applicators in the lowest quintile of total AUs within 5km were missing the self-reported number of livestock and poultry on the farm. In contrast, smoking status, alcohol use, and education did not vary by AU quintiles.

Table 1.

Descriptive characteristics of applicators, overall and by quintiles of total animal units (AUs) for AFOs within 5km of the enrollment residence

		Quintile of AUs from AFOs within 5kma
	Applicators	Q1	Q2	Q3	Q4	Q5
	N=32635	N=6552	N=6502	N=6528	N=6525	N=6528
Age at enrollment (mean ± sd; yr)	44.7 ± 12.3	44.9 ± 12.8	45.0 ± 12.4	44.4 ± 12.2	44.6 ± 12.2	44.6 ± 12.1
Gender (% Male)	98.2	97.6	97.9	98.3	98.7	98.7
Applicator Type (%)
Private	87.3	77.8	86.6	89.3	90.7	92.1
Commercial	12.7	22.2	13.4	10.7	9.2	7.9
BMI Category (%)
>10 to 25	18.8	19.3	19.4	18.7	19.4	17.2
>25 to 30	38.6	36.2	38.1	38.8	39.4	40.6
>30	17.3	16.9	17.0	17.2	17.4	18.0
Missing	25.3	27.6	25.6	25.3	23.9	24.2
Smoking Status (%)
Never	58.2	55.1	58.8	59.7	59.2	58.4
Former	25.7	26.4	25.5	25.1	26.2	25.3
Current	13.0	15.4	12.7	12.2	11.6	13.2
Missing	3.1	3.1	3.1	3.2	3.0	3.2
Drinking Category (%)
Never	20.2	20.3	21.6	19.4	18.9	20.6
1 a week or less	51.1	50.8	50.8	51.6	51.3	51.0
>1 a week	26.0	26.3	24.9	26.1	26.9	25.7
Missing	2.8	2.7	2.7	2.8	2.9	2.7
Highest Schooling Achieved (%)
Less than HS or Equivalent	53.2	50.1	53.3	53.7	54.9	53.8
Post-HS Education	44.3	46.9	44.2	43.7	42.6	44.0
Missing	2.6	3.0	2.5	2.6	2.5	2.1
Marital Status (%)
Married	84.5	82.8	84.5	84.6	84.7	85.7
Other	15.5	17.2	15.5	15.5	15.3	14.3
Animals raised on farm (%)
Beef Cattle	40.6	40.0	44.4	42.3	39.6	36.7
Dairy Cattle	6.6	5.1	6.1	8.2	7.7	5.7
Hogs	42.2	29.0	38.8	43.7	47.5	52.1
Poultry/Eggs	2.5	1.9	2.1	2.6	2.8	3.3
Sheep	4.6	3.7	4.8	5.3	4.5	4.8
Any animal type	65.5	54.3	65.3	68.4	69.4	70.3
Livestock on Farmb (%)
No Livestock	21.4	25.6	21.7	20.3	20.0	19.4
<100	14.5	17.0	16.2	14.9	13.6	11.0
100 to 1000	36.1	30.3	36.7	38.6	37.4	37.5
>1000	16.3	8.2	13.4	16.3	19.8	23.9
Missing	11.7	18.9	12.0	9.8	9.3	8.2
Poultry on Farmb (%)
No Poultry	80.3	74.3	80.2	82.1	82.3	82.8
<1000	6.6	5.8	6.4	7.1	7.0	6.8
>1000	0.5	0.2	0.4	0.5	0.5	1.1
Missing	12.5	19.7	13.0	10.4	10.3	9.3
Farm Activity (%)
Handle stored grain	82.6	73.5	81.6	85.2	86.8	86.1
Handle stored hay	60.6	55.2	62.0	63.7	62.9	59.3
Grind animal feed	55.5	46.5	56.0	59.9	59.3	55.9
Work in swine areas	36.0	24.0	32.3	36.8	41.6	45.4
Work in poultry areas	1.9	1.5	1.6	1.5	1.9	3.0
Load or unload silage	26.6	22.1	25.2	29.6	28.1	28.1
Any animal-related activity	88.4	79.1	87.3	90.6	92.4	92.6

^aQuintiles of AUs from AFOs within 5km: Q1: 0-960 AUs, Q2: 961-3132 AUs, Q3:3133-6562 AUs, Q4: 6564-12842 AUs, Q5:12842+ AUs

^bAssessed as the maximum number of livestock or poultry on the farm in the last year

We generally observed consistency in multivariable associations between total AU and IDW AU metrics (Spearman rho=0.96-0.98) and therefore present IDW associations here. The risk of lymphohematopoietic cancer was not elevated among applicators across categories of IDW AUs within 2km (Table 2). However, positive associations across IDW AUs within 5km were evident and highest in the fourth quintile (AUs HR_Q4=1.61, 95%CI=1.21-2.14). No trends were evident (p-trend=0.57).

Table 2.

Hazard ratios (HR) and 95% confidence intervals (CI) for associations^a between lymphohematopoietic cancers overall and total and IDW animal units (AUs) within 2 and 5km among applicators in the Agricultural Health Study (N=32,635)

	AUs Categoryb	Cases	Total AUsc HR (95%CI)	p-trend	Cases	IDW AUsd HR (95%CI)	p-trend
Total AUs within 2km	No Exp.	256	1.00 (Ref)		256	1.00 (Ref)
	Q1	70	1.07 (0.82-1.39)		75	1.14 (0.88-1.48)
	Q2	67	1.07 (0.82-1.40)		60	0.93 (0.70-1.27)
	Q3	58	0.89 (0.67-1.18)		58	0.90 (0.67-1.19)
	Q4	68	1.08 (0.82-1.41)	p=0.85	70	1.15 (0.88-1.50)	p=0.51
Total AUs within 5km	Q1	78	1.00 (Ref)		78	1.00 (Ref)
	Q2	109	1.39 (1.04-1.86)		113	1.41 (1.06-1.88)
	Q3	112	1.50 (1.12-2.01)		109	1.44 (1.07-1.92)
	Q4	120	1.58 (1.19-2.11)		124	1.61 (1.21-2.14)
	Q5	100	1.31 (0.97-1.77)	p=0.37	95	1.27 (0.94-1.72)	p=0.57

^aCox-regression models were adjusted by age, BMI, education, cigarette smoking status, alcohol use, applicator type, family history of cancer. Associations did not change when adjusted by occupational exposure to pesticides previously linked to LH cancer risk ^26–28

^bTotal and IDW AUs within 2 or 5km were summed and categorized into quartiles or quintiles. For 2km exposures, the referent group was composed of participants with no total or IDW AUs within 2km

^cQuartiles of total AUs from AFOs within 2km were: Q1:1-899 AUs, Q2:900-1658 AUs, Q3:1660-3119 AUs, Q4:3120+ AUs; Quintiles of AUs from AFOs within 5km were: Q1: 0-960 AUs, Q2: 961-3132 AUs, Q3:3133-6562 AUs, Q4: 6564-12842 AUs, Q5:12842+ AUs

^dInverse distance-weighted (IDW) AU metric was calculated as the sum of the AUs of each AFO within 5km weighted by the continuous distance between the AFO and residence

In analyses by cancer site, applicators’ risk of NHL was elevated across IDW AU quintiles within 5km (Table 3); however, risks did not increase monotonically (p-trend=0.52). Risks for applicators were also elevated in a few quintiles for several NHL subtypes, though there were no monotonic trends with increasing AUs. Associations with IDW AUs were evident for B-cell NHL across quintiles, with the highest risks observed for Q4 (HR_Q4=1.88; 95%CI=1.35-2.64). An increased risk of follicular lymphoma was evident in Q2 and Q4 of IDW AUs (HR_Q2=2.97; 95%CI=1.08-8.20; HR_Q4=3.16; 95%CI=1.14-8.74). There were no clear differences in associations between total or IDW exposure metrics.

Table 3.

Hazard ratios (HR) and 95% confidence intervals (CI) for associations^a between lymphohematopoietic cancers and total and inverse distance-weighted (IDW) animal units (AUs) within 5km among applicators in the Agricultural Health Study (N=32,635)

Cancer Site	AUs Categoryb	Cases	Total AUsc HR (95%CI)	p-trend	Cases	IDW AUsd HR (95%CI)	p-trend
Hodgkin lymphoma
	T1	5	1.00 (Ref)		6	1.00 (Ref)
	T2	9	1.72 (0.57-5.15)		9	1.44 (0.51-4.08)
	T3	9	1.68 (0.56-5.06)	0.48	8	1.23 (0.42-3.60)	0.86
All Leukemias
	Q1	12	1.00 (Ref)		16	1.00 (Ref)
	Q2	19	1.54 (0.75-3.18)		14	0.84 (0.41-1.72)
	Q3	11	0.98 (0.43-2.22)		13	0.85 (0.41-1.77)
	Q4	15	1.30 (0.61-2.79)		17	1.09 (0.55-2.16)
	Q5	15	1.27 (0.59-2.72)	0.81	12	0.79 (0.37-1.67)	0.73
Acute myeloid leukemia
	T1	13	1.00 (Ref)		13	1.00 (Ref)
	T2	16	1.28 (0.61-2.67)		16	1.27 (0.61-2.65)
	T3	18	1.43 (0.70-2.94)	0.36	18	1.46 (0.71-3.01)	0.33
Non-Hodgkin lymphoma (NHL)
	Q1	64	1.00 (Ref)		60	1.00 (Ref)
	Q2	83	1.30 (0.93-1.80)		93	1.52 (1.09-2.10)
	Q3	97	1.59 (1.16-2.19)		91	1.57 (1.13-2.17)
	Q4	99	1.59 (1.16-2.19)		103	1.74 (1.26-2.40)
	Q5	81	1.31 (0.94-1.82)	0.36	77	1.36 (0.97-1.91)	0.52
NHL B cell
	Q1	58	1.00 (Ref)		53	1.00 (Ref)
	Q2	76	1.31 (0.93-1.84)		88	1.63 (1.16-2.29)
	Q3	87	1.59 (1.14-2.21)		82	1.61 (1.14-2.27)
	Q4	97	1.73 (1.25-2.40)		98	1.88 (1.35-2.64)
	Q5	76	1.36 (0.96-1.92)	0.22	73	1.46 (1.02-2.09)	0.35
Chronic lymphocytic leukemia, small lymphocytic lymphoma
	Q1	16	1.00 (Ref)		17	1.00 (Ref)
	Q2	24	1.52 (0.80-2.85)		24	1.39 (0.75-2.59)
	Q3	25	1.69 (0.90-3.18)		27	1.67 (0.91-3.07)
	Q4	30	1.94 (1.06-3.58)		27	1.63 (0.88-2.99)
	Q5	24	1.58 (0.84-2.99)	0.31	24	1.53 (0.82-2.86)	0.39
Diffuse large B cell lymphoma
	Q1	13	1.00 (Ref)		12	1.00 (Ref)
	Q2	12	0.90 (0.41-1.97)		16	1.27 (0.60-2.69)
	Q3	16	1.25 (0.60-2.62)		11	0.92 (0.40-2.09)
	Q4	19	1.46 (0.72-2.97)		24	1.97 (0.98-3.97)
	Q5	17	1.31 (0.63-2.71)	0.31	14	1.19 (0.55-2.59)	0.58
Follicular lymphoma
	Q1	5	1.00 (Ref)		5	1.00 (Ref)
	Q2	13	2.62 (0.93-7.37)		15	2.97 (1.08-8.20)
	Q3	11	2.36 (0.82-6.83)		10	2.13 (0.72-6.25)
	Q4	18	3.83 (1.42-10.4)		15	3.16 (1.14-8.74)
	Q5	10	2.17 (0.74-6.41)	0.41	12	2.70 (0.94-7.74)	0.29
Multiple myeloma
	Q1	15	1.00 (Ref)		15	1.00 (Ref)
	Q2	13	0.85 (0.40-1.79)		13	0.83 (0.40-1.76)
	Q3	16	1.10 (0.54-2.23)		18	1.22 (0.61-2.42)
	Q4	18	1.22 (0.61-2.43)		18	1.20 (0.60-2.38)
	Q5	18	1.18 (0.59-2.36)	0.43	16	1.08 (0.53-2.21)	0.69

^aCox-regression models were adjusted by age, BMI, education, cigarette smoking status, alcohol use, applicator type, family history of cancer. Associations did not change when adjusted by occupational exposure to pesticides previously linked to these cancers (Alavanja 2014)

^bTotal and IDW AUs within 5km were summed and categorized into tertiles or quintiles

^cTertiles of total AUs from AFOs within 5km were: T1:0-2242 AUs, T2:2243-8115 AUs, T3:8116+ AUs; Quintiles of AUs from AFOs within 5km were: Q1: 0-960 AUs, Q2: 961-3132 AUs, Q3:3133-6562 AUs, Q4: 6564-12842 AUs, Q5:12842+ AUs

^dInverse distance-weighted (IDW) AU metric was calculated as the sum of the AUs of each AFO within 5km weighted by the continuous distance between the AFO and residence

In analyses of specific animal types (Table 4), we observed a significant increasing trend in total lymphohematopoietic cancers and NHL risk among applicators across quartiles of cattle AUs compared to those with no cattle AUs within 5km (p-trend=0.02). For NHL, elevated risks were observed in the highest category of IDW AU exposure to dairy cattle (HR_≥median=1.45; 95%CI=1.02-2.07; p-trend=0.04) and in the second and third quartile for beef cattle (HR_Q2=1.39; 95%CI=1.05-1.85; HR_Q3=1.39; 95%CI=1.05-1.84; p-trend=0.12). In analyses by NHL subtype, these findings were consistent among B-cell NHL cases and an elevated risk of MM was evident among those with greater beef cattle exposure (HR_≥median=2.17; 95%CI=1.26-3.73; p-trend<0.01; Supplemental Table 1). No increased risks were observed across quartiles of other animal types. Leukemia risk was associated with dairy cattle (HR_≥median=2.12, 95%CI=0.98-4.56), and for AML specifically HR_≥median=2.41, 95%CI=1.01-5.75; p-trend=0.04; Supplemental Table 1).

Table 4.

Hazard ratios (HR) and 95% confidence intervals (CI) for associations^a between lymphohematopoeietic cancers overall, leukemia, or non-Hodgkin lymphoma and type-specific total and inverse distance-weighted animal units (AUs) within 5km among applicators in the Agricultural Health Study (N=32,635)

Animal Type	AUs Categoryb	Cases	Total AUsc HR (95%CI)	p-trend	Cases	IDW AUsdHR (95%CI)	p-trend
All LH Cancers
Swine	No Exp.	86	1.00 (Ref)		86	1.00 (Ref)
	Q1	107	1.11 (0.83-1.48)		105	1.09 (0.82-1.45)
	Q2	100	1.07 (0.80-1.43)		115	1.22 (0.92-1.61)
	Q3	120	1.23 (0.92-1.63)		113	1.16 (0.87-1.55)
	Q4	106	1.09 (0.80-1.47)	0.69	100	1.05 (0.78-1.43)	0.89
All Cattle	No Exp.	197	1.00 (Ref)		197	1.00 (Ref)
	Q1	77	1.16 (0.89-1.51)		67	1.00 (0.76-1.32)
	Q2	63	0.97 (0.73-1.29)		87	1.35 (1.05-1.74)
	Q3	97	1.50 (1.17-1.92)		80	1.23 (0.95-1.61)
	Q4	85	1.39 (1.05-1.82)	0.02	88	1.47 (1.13-1.93)	<0.01
Dairy Cattle	No Exp.	444	1.00 (Ref)		444	1.00 (Ref)
	< median	25	0.75 (0.50-1.12)		27	1.17 (0.94-1.45)
	≥ median	50	1.55 (1.13-2.12)	<0.01	48	1.34 (1.08-1.66)	0.01
Beef Cattle	No Exp.	214	1.00 (Ref)		214	1.00 (Ref)
	Q1	75	1.22 (0.93-1.59)		62	0.98 (0.74-1.30)
	Q2	69	1.10 (0.84-1.44)		83	1.36 (1.05-1.75)
	Q3	81	1.33 (1.02-1.72)		83	1.32 (1.02-1.70)
	Q4	80	1.25 (0.94-1.67)	0.17	77	1.26 (0.94-1.68)	0.16
All Poultry	No Exp.	447	1.00 (Ref)		447	1.00 (Ref)
	< median	35	1.05 (0.74-1.48)		37	1.12 (0.80-1.58)
	≥ median	37	1.13 (0.80-1.60)	0.74	35	1.06 (0.74-1.50)	0.80
Chickens	No Exp.	472	1.00 (Ref)		472	1.00 (Ref)
	< median	25	1.09 (0.55-2.14)		27	1.08 (0.56-2.08)
	≥ median	22	0.67 (0.35-1.31)	0.26	20	0.64 (0.32-1.25)	0.20
Sheep, lambs, goats, horses	No Exp.	488	1.00 (Ref)		488	1.00 (Ref)
	< median	13	0.61 (0.35-1.06)		12	0.56 (0.31-1.01)
	≥ median	18	0.88 (0.54-1.44)	0.64	19	0.92 (0.57-1.48)	0.74
All Leukemia
Swine	No Exp.	12	1.00 (Ref)		12	1.00 (Ref)
	< median	29	1.08 (0.54-2.13)		34	1.24 (0.63-2.41)
	≥ median	31	1.10 (0.54-2.22)	0.86	26	0.92 (0.45-1.90)	0.40
All Cattle	No Exp.	28	1.00 (Ref)		28	1.00 (Ref)
	< median	19	0.98 (0.55-1.77)		21	1.08 (0.61-1.91)
	≥ median	25	1.28 (0.72-2.29)	0.34	23	1.21 (0.67-2.18)	0.54
Dairy Cattle	No Exp.	59	1.00 (Ref)		59	1.00 (Ref)
	< median	4	0.92 (0.33-2.56)		4	0.89 (0.32-2.48)
	≥ median	9	1.98 (0.91-4.31)	0.08	9	2.12 (0.98-4.56)	0.05
Beef Cattle	No Exp.	31	1.00 (Ref)		31	1.00 (Ref)
	< median	19	1.01 (0.57-1.80)		20	1.06 (0.60-1.87)
	≥ median	22	1.07 (0.59-1.95)	0.82	21	1.06 (0.58-1.95)	0.88
All Poultry	No Exp.	60	1.00 (Ref)		60	1.00 (Ref)
	< median	7	1.54 (0.69-3.42)		8	1.75 (0.83-3.72)
	≥ median	5	1.11 (0.43-2.84)	0.58	4	0.94 (0.34-2.66)	0.32
Chickens	No Exp.	64	1.00 (Ref)		64	1.00 (Ref)
	< median	5	1.01 (0.21-4.79)		5	0.67 (0.16-2.80)
	≥ median	3	0.62 (0.10-3.79)	0.61	3	1.29 (0.13-12.56)	0.89
Sheep, lambs, goats, horses	No Exp.	65	1.00 (Ref)		65	1.00 (Ref)
	< median	3	1.02 (0.31-3.36)		3	1.09 (0.33-3.57)
	≥ median	4	1.54 (0.52-4.51)	0.43	4	1.58 (0.54-4.63)	0.40
Non-Hodgkin lymphoma
Swine	No Exp.	71	1.00 (Ref)		71	1.00 (Ref)
	Q1	85	1.07 (0.78-1.47)		83	1.04 (0.76-1.44)
	Q2	82	1.07 (0.77-1.47)		92	1.19 (0.87-1.63)
	Q3	102	1.27 (0.93-1.74)		97	1.22 (0.89-1.67)
	Q4	84	1.07 (0.76-1.50)	0.67	81	1.06 (0.76-1.49)	0.88
All Cattle	No Exp.	159	1.00 (Ref)		159	1.00 (Ref)
	Q1	62	1.17 (0.87-1.57)		55	1.03 (0.76-1.40)
	Q2	52	1.00 (0.73-1.37)		71	1.37 (1.04-1.82)
	Q3	82	1.58 (1.20-2.07)		67	1.28 (0.96-1.72)
	Q4	69	1.43 (1.06-1.94)	0.02	72	1.52 (1.13-2.05)	<0.01
Dairy Cattle	No Exp.	365	1.00 (Ref)		365	1.00 (Ref)
	< median	20	0.73 (0.46-1.15)		21	0.75 (0.48-1.17)
	≥ median	39	1.50 (1.05-2.13)	0.03	38	1.45 (1.02-2.07)	0.04
Beef Cattle	No Exp.	172	1.00 (Ref)		172	1.00 (Ref)
	Q1	63	1.29 (0.96-1.72)		51	1.01 (0.74-1.38)
	Q2	56	1.11 (0.82-1.51)		68	1.39 (1.05-1.85)
	Q3	68	1.39 (1.05-1.85)		70	1.39 (1.05-1.84)
	Q4	65	1.30 (0.95-1.79)	0.15	63	1.32 (0.96-1.82)	0.12
All Poultry	No Exp.	370	1.00 (Ref)		370	1.00 (Ref)
	< median	26	0.94 (0.63-1.41)		27	0.99 (0.67-1.47)
	≥ median	28	1.04 (0.70-1.54)	0.72	27	0.97 (0.65-1.45)	0.95
Chickens	No Exp.	390	1.00 (Ref)		390	1.00 (Ref)
	< median	18	1.00 (0.46-2.18)		20	1.11 (0.52-2.37)
	≥ median	16	0.62 (0.29-1.32)	0.22	14	0.53 (0.25-1.13)	0.11
Sheep, lambs, goats, horses	No Exp.	400	1.00 (Ref)		400	1.00 (Ref)
	< median	10	0.58 (0.31-1.09)		9	0.52 (0.27-1.02)
	≥ median	14	0.85 (0.49-1.48)	0.58	15	0.89 (0.52-1.52)	0.67

^aModels included all non-overlapping animal-specific exposures concurrently and were adjusted by age, BMI, education, cigarette smoking status, alcohol use, applicator type, family history of cancer. Associations did not change when adjusted by occupational exposure to pesticides previously linked to LH cancer risk ^26–28

^bTotal and IDW AUs within 5km were summed and categorized by median or quartile and compared to the group with no exposure of that animal type

^cQuartiles for swine AUs: Q1:1-1980; Q2:1981-4858;Q3:4859-10168;Q4:10169+ ; Quartiles for all cattle AUs: Q1:1-375, Q2:376-949, Q3:950-2500, Q4:2501+; median split of dairy cattle AUs: 635 AUs; median split of beef cattle AUs: 891 AUs; median split of all poultry AUs: 1000AUs; median split of chicken AUs: 1000 AUs ; median split of sheep, lamb, goat, horse AUs: 23 AUs

^dInverse distance-weighted (IDW) AU metric was calculated as the sum of the type-specific AUs of each AFO within 5km weighted by the continuous distance between the AFO and residence

Lymphohematopoietic cancer risk associated with animal intensity near the home was higher among applicators reporting any animal-related work compared to the common referent group of those in Q1 without animal-related work (Figure 1). Tests of heterogeneity were not statistically significant for animal-related work (p_int=0.61), raising animals on farm (p_int=0.51) or working in confinements (p_int=0.66).

Figure 1.

Associations^a between lymphohematopoietic cancers overall and total animal units (AUs)^b within 5km stratified by animal-related work^c among applicators in the Agricultural Health Study (N=32,635)

^aCox-regression models were adjusted by age, BMI, education, cigarette smoking status, alcohol use, applicator type, family history of cancer. Associations did not change when adjusted by occupational exposure to pesticides previously linked to LH cancer risk ^26–28

^bTotal AUs within 5km were summed and categorized into quintiles. Quintiles of AUs from AFOs within 5km were: Q1: 0-960 AUs, Q2: 961-3132 AUs, Q3:3133-6562 AUs, Q4: 6564-12842 AUs, Q5:12842+ AUs

^cAnimal-related work defined as either raising animals on the farm or working in confinements. Raising animals on farm assessed on enrollment questionnaire as a response to whether cattle, swine, poultry, sheep, eggs, or other farm animals were a major source of income on the farm. Work in confinement assessed on the enrollment questionnaire as a positive response to working in either swine or poultry confinement areas at least once each year

Among the spouses, nearly two-thirds reported direct contact with animals in the last 12 months, most commonly with beef cattle (37.7%) and swine (39.3%) (Supplemental Table 2). We observed no increased risks of any specific lymphohematopoietic malignancies or subtypes in association with IDW AUs within 2 or 5km (Supplementary Tables 3–5). We also observed no interaction with self-reported direct animal contact (Supplementary Figure 1).

In sensitivity analyses, ever/never and lifetime use of pesticides previously associated with risk of lymphohematopoietic cancers in the AHS was weakly or not correlated with AU metrics within 5km (Supplemental Tables 6 and 7), and our main associations did not change with their adjustment (Supplemental Table 8). When we restricted exposures to permitted facilities, associations between total and IDW AUs within 5km and total lymphohematopoietic cancers were attenuated but still evident for applicators in higher AU categories. Results were also similar when restricting exposures to permitted facilities known to be operating prior to 2005, to cases diagnosed after 2005, or to white, male applicators (Supplemental Table 8).

Discussion

In this prospective investigation of lymphohematopoietic cancer risk and residential proximity to intensive animal agriculture, we observed increased risk among applicators with the highest estimated farm animal exposure within 5km of the home. We noted associations between total and IDW AUs with NHL risk and some subtypes. Cattle exposures were also associated with an increased risk of NHL and leukemia. Associations did not differ by whether the applicator did animal-related work, nor did risks change when accounting for occupational pesticide use or farming activities. We found no evidence of these associations among spouses of applicators, although small case numbers limited analyses of certain subtypes.

Our finding of increased risk of NHL in relation to proximal animal exposure is generally consistent with many, but not all, studies reporting higher risk of lymphohematopoietic malignancies in farmers¹⁹. However, only a few of these studies have examined risk in relation to intensive animal agriculture near the residence, and they relied on ecologic estimates of exposure, such as county-level animal inventories. A case-control study of death certificates in Wisconsin found highest NHL mortality among farmers living in counties with the highest levels of dairy sales or hog inventories²¹. Several studies have found positive associations between county-level hogs, cows and chickens and adult leukemias, including unspecified acute leukemia in Nebraska²⁰, and acute lymphocytic leukemia³⁴ and unspecified lymphatic leukemia³⁵ in Iowa. Additionally, one study found positive associations between deaths from multiple myeloma and county-level poultry inventories³⁶. Our study builds on previous research by using more refined metrics that reflect animal intensity proximal to the residence, which are likely to better capture individual-level exposures.

We observed associations between NHL and residential proximity to cattle that are supported by studies of occupational animal contact, which have generally shown increases in risk from poultry, swine, and cattle exposures^15,16,37,38. A large population-based case-control study in Canada found that occupational contact with beef cattle increased the risks of NHL (OR=1.8; 95% CI=1.1-2.9) and leukemia (OR=2.0; 95%CI=1.2-3.3) compared to no occupational contact, but there was no increased risk of MM (OR=1.1; 95%CI=0.3-3.5) and none of these cancers were associated with dairy cattle contact³⁸. In this same study, risk of these cancers combined was elevated for contact with swine (OR=1.7; 95% CI=0.7-3.8)³⁸. Similarly, a case-control study in Germany found occupational contact with cattle increased lymphoma risk (OR=1.3; 95%CI=1.0-1.7)³⁷. An ecologic evaluation of lymphohematopoietic cancer mortality in relation to occupational contact with animals by U.S. regions found the highest risks in the North Central region (an analysis that did not include Iowa), an area with large swine inventories¹⁵. Although Iowa is the leading producer of both swine and poultry in the U.S.³⁹, we observed inconsistent positive associations with swine AUs and total lymphohematopoietic cancers, NHL, and leukemias and no associations with poultry. Possible reasons for varying associations by animal type may include differences in facility construction, waste management strategies, and use of feed, pesticides, antimicrobials and hormones across different operation types which may lead to differing emissions into the surrounding environment^40,41. However, the variability of these factors, even among operations of the same animal type, make characterization of emission profiles from AFOs challenging⁴².

In contrast to our findings, a previous investigation in the AHS found increased risk of NHL among applicators who raised poultry (RR=1.6; 95%CI=1.0-2.4) and positive, though imprecise associations between raising swine and leukemia (RR=1.3; 95%CI=0.9-1.9) and MM (RR=1.7; 95%CI=0.96-3.0). Additionally, there was an increased risk of NHL for those who worked in hog confinements (RR=3.6; 95%CI=1.2-10.3)¹⁶. One explanation for our findings may be that occupational work with these animal types, often conducted inside confinements, confers different exposures of varying magnitude than does residing near these animals. In our data, AFO type was too highly correlated with animal type (i.e., swine and chickens were commonly being raised in confinements whereas cattle were not) to evaluate this directly. Additionally, most permitted AFOs (93%) were confinements, whereas more than half (56%) of the voluntarily-reported farms were open feedlots. Our findings did not change when we excluded voluntarily-reported farms from our analyses, likely because these farms have fewer animals on site.

Our associations with animal exposures were upheld with adjustment for occupational pesticide use, an important observation in this agricultural cohort with exposure to both animals and pesticides. Several studies have found positive associations with occupational animal contact after adjustment for pesticide use. In a multi-center population-based study in Canada, McDuffie et al found elevated risk of NHL among farmers with over 13 head of swine (OR=1.96; 95%CI=1.21-3.18) or who raised bison, elk, or ostriches (OR=3.26; 95%CI=1.20-8.89) after adjusting for occupational exposure to multiple broad categories of pesticides²². The aforementioned prior work in the AHS that found positive associations with occupational animal contact also persisted after adjustment for pesticide use¹⁶. Taken together, these studies and our findings suggest that exposures other than pesticides may be responsible for the observed increases in lymphohematopoietic risk among individuals with direct or indirect animal exposures.

Our proximity-based exposure metrics could reflect multiple environmental contaminants emanating from CAFOs that are plausibly associated with lymphohematopoietic cancer risk, including both air and water pollutants. Animal waste, when applied to land as a fertilizer, transported via flooding, or from overflow or leakage from facilities and dedicated waste lagoons, can lead to high concentrations of soluble nutrient breakdown products, such as nitrate, in surface and groundwater^43,44. Nitrate is a precursor to N-nitroso compounds, which are known carcinogens⁴⁵. One population-based case-control study in Nebraska found that participants with long-term consumption of community water with nitrate levels in the highest quartile (≥4mg/L) had significantly increased risk of NHL (OR=2.0; 95%CI=1.1-3.6)⁴⁶, though other studies in Iowa⁴⁷ and of men in Minnesota found no association⁴⁸. Air emissions from AFOs can include numerous organic particulates such as fecal matter, feed, pollen, bacteria, fungi and viruses, skin cells, and microbial byproducts, as well as gaseous emissions of ammonia, H₂S, malodorous organic compounds, and diesel exhaust^1,5,49,50. Some of these exposures, including infectious agents, are known or suspected lymphomagens⁵¹. Few studies have examined associations between ambient air pollutants and adult lymphohematopoietic cancer risk with mixed results^52,53 and have focused on traffic-related pollution, not the chemical constituencies or pollutants specific to AFOs. High levels of endotoxins have been reported on animal farms⁵⁴ and have been posited as a potential NHL risk factor⁵⁵. However, few studies have included objective measures of endotoxin exposure. Endotoxins in household dust were not associated with NHL in a multicenter population-based case-control study in the U.S. that included Iowa⁵⁵.

Our buffer-based metrics are a simple surrogate for these environmental exposures and a commonly used approach to account for dispersion of contaminants surrounding a point source⁵⁶. However, residing within 2 and 5km of an AFO was common in the AHS population; the homogeneity in AU exposures might have limited our ability to discern associations with cancer risk. Our IDW-based approach should better capture the intensity of animal exposures as a function of the distance from the home to nearby AFOs, and subsequently the relative risk associated with AUs. Perhaps not surprising given their strong correlation, the positive associations and trends with total AUs were consistently observed with these more refined IDW metrics. Additionally, we note that associations with AUs were mostly evident within 5km rather than 2km. One reason for this result may be the required separation of animal housing and waste structures from residences by up to 3000 ft (0.91km) depending on the animal, type of structure (lagoons, confinement buildings), and total AUs⁵⁷.

While we observed positive associations among applicators, we did not see these same risks among their spouses. Small numbers of cases may have limited our analyses of certain subtypes and animal-specific associations in this group. However, only a few studies have examined lymphohematopoietic risk among female farmers^17,58, with one showing stronger associations with NHL for male farmers⁵⁸. We were not able to separate gender from participant status (applicator or spouse), as applicators were predominantly male (98.2%) and spouses were predominantly female (99.7%). Thus, it remains difficult to disentangle whether biological factors related to gender, or other unmeasured variables may be responsible for these differences. Another potential explanation for the differences we observed between applicators and spouses is if applicators engage in behaviors that might increase their exposure to AFO-related emissions, such as more time spent outdoors. These determinants of exposure would be unlikely to be captured in our exposure metrics.

Strengths of this study include the long follow-up period and large numbers of lymphohematopoietic cancer cases, allowing for separate evaluation of NHL and leukemia and their major subtypes. We were able to evaluate potential confounding by pesticide use and effect modification by direct animal contact in our analyses. Given existing evidence that occupational exposure to specific pesticide active ingredients is linked to risk of certain LH cancers, future analyses of possible joint effects of pesticides with proximal animal exposures are warranted. Whereas previous studies have examined associations with county-level animal inventories, this is the first to use geocoded AFO locations and permitted animal counts to investigate risks from living near intensive animal agriculture. The AFO database includes all animal farms large enough to be regulated by the IDNR, therefore our exposure metrics likely capture most high intensity animal exposures near the home³¹. It also includes some smaller, voluntarily-reported facilities. Although we could not determine how many additional smaller farms were not included, these farms raise fewer animals than those that require an MMP. Moreover, our associations remained after their exclusion in sensitivity analyses. Additionally, while positional errors in the AHS residential geocodes are small (median=39m)⁵⁹, uncertainty in AFO locations could have resulted in misclassification of exposure. Our large exposure buffers should partially mitigate this concern. Although the AFO database included operating records dating back to 1991, most were during the period after cohort enrollment (>99% had permit dates 1995 or later), and we had limited ability to evaluate the timing of operation in relation to cancer risk. We expect non-differential misclassification of exposures as a result of these limitations, which may have diminished our ability to discern associations. We found no evidence of confounding by pesticide use or farming characteristics assessed at enrollment, though we recognize these exposures may have changed over time. The lack of congruency in some covariate measures between applicators and spouses was another limitation. We excluded North Carolina participants, who comprise one-third of the cohort, because AFO data were not of comparable completeness to Iowa; they were primarily satellite-derived with limited information on animal counts and operation status. Efforts to harmonize multiple and varied data sources may allow inclusion of these participants in future analyses. Finally, as study participants were limited to pesticide applicators and their spouses in Iowa, the generalizability of our results to other populations living near intensive animal agriculture may be limited.

Conclusion

Results from this study suggest that residential proximity to intensive animal agriculture increases the risk of NHL among AHS cohort members in Iowa, even after consideration of occupational exposure to pesticides and contact with farm animals. Our findings of increased risk of NHL and leukemia with cattle exposures warrant further investigation, ideally in a study population with greater heterogeneity in exposure. Future studies that can temporally or spatially refine animal exposure metrics or identify etiologic mechanisms would provide support for our findings.