Non-Hodgkin lymphoma risk and organophosphate and carbamate insecticide use in the north American pooled project.

Organophosphates and carbamates have been among the most commonly used insecticides, with both agricultural and residential uses. Previous studies have suggested associations of non-Hodgkin lymphoma (NHL) with some of these chemicals; however, many studies have been limited in their ability to evaluate associations with lymphoma subtypes. We evaluated the use of eleven organophosphate and two carbamate insecticides in association with NHL in the North American Pooled Project, which includes data from case-control studies in the United States and Canada (1690 cases/5131 controls). We used unconditional logistic regression adjusting for potential confounders, including use of other pesticides, to estimate odds ratios (OR) and 95% confidence intervals (CI) for associations between these chemicals and NHL overall, and NHL subtypes, i.e., follicular (FL), diffuse large B-cell (DLBCL), small lymphocytic lymphoma (SLL) and others. Ever use of malathion was associated with increased risk of NHL overall (OR = 1.43; 95% CI: 1.14-1.81) compared with never users. Categories using tertiles of duration (<4 yrs., 4-12 yrs., and >12 yrs) also showed a significant exposure-response for increasing years of use of malathion and risk of NHL (OR<4vsUnex = 1.33 (0.88, 2.03), OR4-12vsUnex = 1.42 (1.02, 1.96), OR>12vsUnex = 1.55 (1.05, 2.28, p-trend < 0.01)). In addition, malathion use was statistically significantly associated with FL (OR = 1.58; 95% CI: 1.11-2.27) and DLBCL (OR = 1.61; 95% CI: 1.16-2.22) while there were no apparent associations with SLL or other subtypes, the p-value for heterogeneity across subtypes, however, was not significant. These results support previous studies suggesting an association between insecticide use and NHL overall, and provide new information on associations with NHL subtypes. Published by Elsevier Ltd.

Introduction

Organophosphates (OPs) and carbamates are classes of insecticides frequently used in agriculture and for public health and residential purposes. Over the last decade in the United States and Canada, chlorpyrifos, acephate and malathion were among the most commonly used OPs and carbaryl was one of the most widely used carbamate insecticides for conventional use in homes and gardens.(EPA, 2017; Health Canada, 2015) Pesticides, including some OPs and carbamates, have been linked to several adverse human health outcomes, including cancer.(Blair et al., 2015; Koutros et al., 2013) The International Agency for Research on Cancer (IARC) lists the OPs malathion and diazinon as probable human carcinogens (Group 2A) and dichlorvos as possibly carcinogenic to humans (Group 2B).(IARC, 2015) In earlier assessments, IARC classified carbaryl(IARC, 1987a) and the OP trichlorfon(IARC, 1987b) as not classifiable as to their carcinogenicity in humans (Group 3). Few other OP or carbamate insecticides have been assessed for carcinogenicity.

Population-based case-control studies from the U.S.(Cantor et al., 1992; Waddell et al., 2001; Zheng et al., 2001) and Canada(McDuffie et al., 2001) have demonstrated varied positive associations between non-Hodgkin lymphoma (NHL) and reported use of the OPs malathion, diazinon, coumaphos, chlorpyrifos, and fonofos, as well as the carbamate insecticides carbofuran and carbaryl. In addition, data from a large prospective cohort study of pesticide applicators found positive associations between NHL and the OPs terbufos and diazinon.(Alavanja et al., 2014; Bonner et al., 2010) Not all studies, however, have supported these positive associations.(Bonner et al., 2007; Mills et al., 2005) Associations between individual insecticides and NHL have varied in magnitude and statistical significance and many studies have been limited in their ability to evaluate associations with subtypes for this heterogeneous disease. Thus, more data are needed to explore these associations.

Here, we pooled data from four large population-based case-control studies from the U.S. and Canada to allow a more detailed evaluation of possible relationships between the use of several OP and carbamate insecticides and risk of NHL among 1690 cases and 5131 controls.

Methods

Study population

The current pooled analysis is comprised of three population-based case-control studies conducted by the U.S. National Cancer Institute in Kansas, Iowa/Minnesota, and Nebraska in the 1980s(Cantor et al., 1992; Hoar et al., 1986; Zahm et al., 1990) and the Cross Canada Study of Pesticides and Health (CCSPH) which was conducted in Quebec, Ontario, Manitoba, Saskatchewan, Alberta, and British Columbia between 1991 and 1994.(McDuffie et al., 2001) Cases included incident NHL diagnoses who were at least 19 years of age. NHL histology codes from each study were reexamined for the pooled effort and categorized to the International Classification of Diseases for Oncology First Edition (ICDeOe1) to classify NHL overall and the following subtypes: follicular lymphoma (FL), diffuse large B-cell lymphoma (DLBCL), small lymphocytic lymphoma (SLL), and “other”. The “other” sub-type included all cases whose histologies were unknown or not FL, DLBCL, or SLL. Pathology reviews in the original studies were conducted to standardize and validate NHL diagnoses. The current analysis includes 1690 NHL cases and 5131 controls. Controls for the studies were selected from the general population using different methods (depending on the study), including random digit dialing, voter lists, health insurance records, Medicare listings for those 65 years or older, and from state mortality files for deceased cases. Cases and controls were frequency matched on location (state/province) and age (± 2 or 5 years); in some states, the matching was done on additional variables, including sex and race, as well as vital status and year of death for deceased cases (Iowa, Minnesota, Nebraska, Kansas). Demographic data, agricultural exposures (including pesticide use), and other covariates related to NHL risk factors, such as lifestyle, medical and occupational history, were obtained via postal or interviewer-administered questionnaires (either in person or by telephone) from study participants or proxy respondents (if subjects were deceased (U.S. studies) or required assistance due to illness or disability (CCSPH)).

Investigators of individual studies received human subjects approval prior to collection of data. The pooling project was approved by the Health Sciences Research Ethics Board, University of Toronto (#25166) and an exemption was obtained from the Office of Human Subjects Research, U.S. National Institutes of Health (#11351).

Pesticide exposure

Information on pesticide use was harmonized across the four studies, including data on self-reported use of 11 OPs and two carbamate insecticides. For the current analysis, we included OP and carbamate insecticides with at least ten exposed cases of NHL overall. This resulted in the inclusion of the following specific insecticides for analysis (in order from highest to lowest prevalence of use): malathion (OP), diazinon (OP), carbofuran (carbamate), carbaryl (carbamate), fonofos (OP), phorate (OP), dimethoate (OP), terbufos (OP), coumaphos (OP), and dichlorvos (OP). Additional information on numbers of days per year of use and number of total years of use for specific pesticide active ingredients was also collected. Information was not collected on days per year (Iowa, Minnesota, and Kansas) or duration (Kansas) for individual pesticides (only for herbicides and insecticides as groups). For studies with duration of use information for specific pesticides, we imputed values for those with missing duration information using a stratified sampling scheme. Using individuals with duration information, we first cross-classified data by categories of duration, age, sex, respondent type, and study site. For subjects missing duration information, we created five imputed values by sampling with replacement a value for duration of exposure conditional on the levels for the other factors. We carried out five separate analyses and combined OR results using the SAS MIANALYZE procedure (SAS, version 9.4 (Cary, North Carolina)).

Statistical analysis

Unconditional logistic regression was used to calculate ORs and 95% CIs for associations between ever use of individual OP and carbamate insecticides and the risk of NHL overall and for histological subtypes (FL, DLBCL, SLL, and other). Categories for days per year and years of use of each insecticide were created based on the distribution among controls and were grouped as either above or below the median or according to tertiles. Based on the observed relationship between these factors and risk of NHL, all models were adjusted for age, sex, state/province, and family history of lymphatic or hematopoietic cancer in a first-degree relative. Proxy respondent status was also evaluated as a potential covariate and effect modifier of the observed pesticide relationships (using a likelihood ratio test). The relationship between individual pesticides reported by participants (ever/never) was evaluated using the Phi coefficient. Additional modeling was done to further adjust for ever/never pesticide co-exposures based on this relationship (those phi coefficient ≥ 0.35). We also explored adjustment of all models with other pesticides that have been previously linked to NHL (2,4-D, dicamba, glyphosate, atrazine, lindane, chlor-dane, and DDT) in the individual studies comprising the pooled analysis or other studies. Finally, we mutually adjusted models for pesticides with significant associations observed in the current pooled analysis. A sensitivity analysis was conducted to compare risk estimates for duration of use with and without imputed duration. As an alternate approach, we conducted a weighted quantile sum logistic regression analysis using duration of OP and carbamate insecticide use to analyze the pesticides as a mixture.(Carrico et al., 2015) This was done using the same covariates as adjustment variables and with the imputed datasets. Since the prevalence of ever use for a number of pesticides was very low (e.g., for Trichlorfon even the 99th percentile for duration was zero), we used normalized continuous durations (Z-scores) as measures of exposure. As a sensitivity analysis, we also fit a model using pesticide exposure percentiles. These mixture analyses were done using the gWQS package (R package that is available in the R-CRANS Library to perform logistic regression with weighted quantile sums).

We computed tests for linear trend using the Wald test, treating the median value for each category among control subjects as continuous. P-values for heterogeneity across NHL subtypes were conducted using polytomous logistic regression using a Wald test. All statistical analyses were performed using SAS software version 9.1.3 (SAS Institute, Inc., Cary, NC).

Results

The mean age of NHL cases (62.7 years) was comparable to that of controls (61.7 years) (Table 1). Most participants were male (cases: 89.1%, controls: 86.2%) and about a third of case and control interviews were provided by proxy respondents (cases: 31.5%, controls: 33%). A larger proportion of NHL cases had a first-degree relative with lymphatic or hematopoietic cancer (8.2%) than controls (3.9%). Cases were also slightly more likely to have ever been diagnosed with a medical condition (32.3%) than controls (27.1%). Around two-thirds of cases and controls had reported to have ever lived on a farm or ranch (65.2%, v. 63.9%) and < 10% had reported to have ever used personal protective equipment (6.2% v. 6.0%). Proxy respondent status did not materially impact observed point estimates (by > 10%) and was not retained in the final models, nor did we observe significant effect modification by proxy status.

Table 1

Demographic characteristics of lymphoma cases and controls in the NAPP.

Variable	Cases N (%)	Controls N (%)
	1690	5131
Histological sub-type
Follicular Lymphoma (EL)	468 (27.69)
Diffuse Large B Cell Lymphoma (DLBCL)	647 (38.28)
Small Lymphocytic Lymphoma (SLL)	171 (10.12)
Follicular Lymphoma (FL)	404 (23.91)
State/Province
U.S.
Iowa	292 (17.28)	604 (11.77)
Minnesota	329 (19.47)	642 (12.51)
Nebraska	368 (21.78)	1449 (28.24)
Kansas	188 (11.12)	930 (18.13)
Canada
Quebec	117 (6.92)	291 (5.67)
Ontario	142 (8.40)	585 (11.40)
Manitoba	34 (2.01)	113 (2.20)
Saskatchewan	29 (1.72)	91 (1.77)
Alberta	65 (3.85)	196 (3.82)
British Columbia	126 (7.46)	230 (4.48)
Age (years)
≥19 to ≤29	26 (1.54)	277 (5.40)
≥30 to ≤39	97 (5.74)	445 (8.67)
≥40 to ≤49	159 (9.41)	514 (10.02)
≥50 to ≤59	288 (17.04)	726 (14.15)
≥60 to ≤69	564 (33.37)	1264 (24.63)
≥70 to ≤79	402 (23.79)	1189 (23.17)
≥80 to ≤89	137 (8.11)	610 (11.89)
≥90	17 (1.01)	106 (2.07)
Sex
Male	1506 (89.11)	4424 (86.22)
Female	184 (10.89)	707 (13.78)
Respondent type
Self	1140 (67.46)	3372 (65.72)
Proxy	533 (31.54)	1692 (32.98)
Unknown/mi s s ing	17 (1.01)	67 (1.31)
Lymphohematopoietic cancer in a first-degree relative
No	1493 (88.34)	4790 (93.35)
Yes	139 (8.22)	202 (3.94)
Unknown/missing	58 (3.43)	139 (2.71)

There was a significantly increased risk of NHL among ever users of malathion (OR = 1.63, 95% CI = 1.33, 2.00), diazinon (OR = 1.69, 95% CI = 1.27, 2.24), fonofos (OR = 1.75, 95% CI = 1.21, 2.55), and carbaryl (OR = 1.62, 95% CI = 1.20, 2.18) in comparison with never users of that pesticide (Table 2). After mutual adjustment for these four chemicals, the risk of NHL was attenuated for each insecticide, with only the risk of NHL among ever users of malathion remaining statistically significant (OR = 1.43, 95% CI = 1.14, 1.81). This increased risk was not impacted by additional adjustment for other pesticides correlated with malathion or with any pesticide previously linked to NHL in these studies (data not shown). Elevated OR for ever use of dimethoate, coumaphos, trichlorfon, chlorpyrifos, and carbofuran were also further attenuated after adjustment for other insecticides (data not shown). Exposure-response analyses evaluating years of use of malathion showed that the OR was highest among those who reported using malathion ≥6 years compared to never users (n = 103 cases) (OR = 1.57; 95% CI = 1.18, 2.10, p-value for trend < 0.01, Table 3). Expanded categories using tertiles of duration of malathion use (< 4 yrs., 4–12 yrs., and > 12 yrs) also showed a significant exposure-response for increasing years of use of malathion and risk of NHL (OR < 4vsUnex = 1.33 (0.88, 2.03), OR4–12vsUnex = 1.42 (1.02, 1.96), OR > 12vsUnex = 1.55 (1.05, 2.28, p-trend < 0.01)). These results were similar when considering data without imputation (data not shown) and across studies (no heterogeneity by study, p = 0.93). There were no additionally significant trends between risk and duration of use (years) for any other insecticide after mutual adjustment (Table 3); results for increasing years of dimethoate and coumpahos approached unity after adjusting for other insecticides (not shown). Results evaluating days per year of use were limited (data available for only for 2 of the 4 studies). These analyses also show a significant association between malathion use and NHL (Supplemental Table 2).

Table 2

Associations between ever use of organophosphate and carbamate insecticides and lymphoma in the NAPP[a].

Pesticide—Ever use	Cases	Controls	OR (95% CI)	OR (95% CI)[b]
Organophosphate (OP)
Malathion
Never used.	1518	4839	1	1
Ever used	172	292	1.63 (1.33, 2.00)	1.43 (1.14, 1.81)
Diazinon
Never used	1605	4994	1	1
Ever used	85	137	1.69 (1.27, 2.24)	1.25 (0.90, 1.74)
Fonofos
Never used	1640	5053	1	1
Ever used	50	78	1.75 (1.21, 2.55)	1.23 (0.81, 1.87)
Phorate
Never used	1644	5022	1
Ever used	46	109	1.07 (0.75, 1.54)
Dimethoate
Never used	1665	5056	1
Ever used	35	75	1.34 (0.89, 2.03)
Tefbufos
Never used	1657	5046	1
Ever used	33	85	0.94 (0.62, 1.44)
Dichlorvos
Never used	1665	5078	1
Ever used	25	53	1.04 (0.64, 1.71)
Coumaphos
Never used	1665	5091	1
Ever used	25	40	1.56 (0.93, 2.62)
Famphur
Never used	1670	5078	1
Ever used	20	53	1.02 (0.60, 1.73)
Trichlorfon
Never used	1680	5115	1
Ever used	10	16	1.80 (0.80, 4.05)
Chlorpyrifos
Never used	1680	5110	1
Ever used	10	21	1.77 (0.82, 3.81)
Carbamate
Carbofuran
Never used	1612	4969	1
Ever used	78	162	1.29 (0.97, 1.71)
Carbaryl
Never used	1615	5004	1	1
Ever used	75	127	1.62 (1.20, 2. 18)	1.17 (0.84, 1.64)

Adjusted for age, study, gender, family history of lymphohematopoietic cancer.

Mutually adjusted for: malathion, diazinon, fonofos, carbaryl; results not shown for non-significant pesticides (all further attenuated towards unity).

Table 3

Associations between duration of organophosphate and carbamate insecticide use and lymphoma in the NAPP.

Pesticide	Cases	Controls	OR[a] (95% CI)	OR[b] (95% CI)
Organophosphate (OP)
Malathion
Unexposed	1352	3903	1	1
< 6 years	65	128	1.40 (1.01, 1.92)	1.25 (0.90, 1.75)
≥ 6 years	103	152	1.79 (1.38, 2.32)	1.57 (1.18, 2.10)
Ptrend			< 0.0001	< 0.01
Diazinon
Unexposed	1437	4048	1	1
< 8 years	49	66	2.03 (1.36, 3.02)	1.51 (0.97, 2.34)
≥ 8 years	34	69	1.30 (0.84, 2.01)	0.97 (0.61, 1.55)
Ptrend			0.08	0.96
Fonofos
Unexposed	1471	4107	1	1
< 6 years	26	37	1.80 (1.02, 3.16)	1.25 (0.69, 2.28)
≥ 6 years	23	39	1.64 (0.90, 3.00)	1.20 (0.63, 2.27)
Ptrend			0.03	0.47
Phorate
Unexposed	1474	4078	1
< 6 years	22	52	0.98 (0.58, 1.67)
≥ 6 years	24	53	1.21 (0.72, 2.02)
Ptrend			0.45
Dimethoate
Unexposed	1485	4113	1
< 8 years	17	35	1.36 (0.75, 2.47)
≥ 8 years	18	35	1.46 (0.81, 2.65)
Ptrend			0.07
Terbufos
Unexposed	1487	4098	1
< 4 years	12	37	0.77 (0.40, 1.51)
≥ 4 years	21	48	1.06 (0.62, 1.81)
Ptrend			0.94
Dichlorvos
Unexposed	1496	4131	1
< 10 years	10	18	1.18 (0.53, 2.60)
≥ 10 years	14	34	0.88 (0.46, 1.69)
Ptrend			0.80
Coumaphos
Unexposed	1496	4148	1
< 4 years	7	13	1.11 (0.40, 3.11)
≥ 4 years	17	22	1.89 (0.99, 3.62)
Ptrend			0.05
Carbamate
Carbofuran
Unexposed	1449	4040	1
< 8 years	44	71	1.50 (1.01, 2.22)
≥ 8 years	27	72	0.91 (0.57, 1.45)
Ptrend			0.83
Carbaryl
Unexposed	1447	4064	1	1
< 6 years	38	63	1.50 (0.97, 2.30)	1.11 (0.70, 1.74)
≥ 6 years	35	56	1.75(1.13, 2.70)	1.24 (0.78, 1.99)
Ptrend			< 0.01	0.35

Adjusted for age, study, gender, family history of lymphohematopoietic cancer.

Mutually adjusted for: malathion, diazinon, fonofos, carbaryl; results not shown for non-significant pesticides (all further attenuated towards unity).

When we evaluated the association between a mixture of all OP and carbamate insecticides and NHL risk, we found a statistically significant association between pesticide duration and NHL risk. For a unit change in the weighted normalized (Z-score) across pesticides, there was a 38% (95% CI: 18% to 61%, p-value < 0.0001) increased risk of NHL. In the weighted normalized sum, malathion, carbaryl, fonofos, diazinon, and coumaphos duration play the biggest role with all of their associated weights being above 0.10 (0.31, 0.19,0.17,0.13, and 0.10, respectively). The results were very similar when using the percentiles rather than standardized exposure durations (42% increased risk per weighted percentile unit, p-value < 0.0001, with the same five pesticides playing the largest role in the weighted sum).

Table 4 shows the association between ever use of OP and carbamate insecticides and lymphoma subtypes. There were positive associations between ever use of malathion and both FL (OR = 1.80, 95% CI = 1.31, 2.46) and DLBCL (OR = 1.77, 95% CI = 1.33, 2.35) subtypes compared to never users; these associations held after mutual adjustment for other pesticides (FL: OR = 1.58, 95% CI = 1.11, 2.27; DLBCL: OR = 1.61, 95% CI = 1.16, 2.22); increases in risk were also evident with increasing duration of exposure for these subtypes (Supplemental Table 3). Ever use of diazinon was associated with FL (OR = 1.75, 95% CI = 1.14, 2.71), DLBCL (OR = 1.74, 95% CI = 1.17, 2.60), and SLL (OR = 2.24, 95% CI = 1.17, 4.30) subtypes, however these associations were attenuated and became non-significant after adjustment for malathion, fonofos, and carbaryl. A similar attenuation in risk was observed between ever use of fonofos and FL and DLBCL subtypes. A significant increased risk was observed between ever use of coumaphos and SLL (OR = 3.97, 95% CI = 1.60, 9.84), however this was based on only six exposed cases. Ever use of the carbamate insecticide carbaryl was significantly associated with FL (OR = 1.66, 95% CI = 1.05, 2.64) and SLL (OR = 2.59, 95% CI = 1.39, 4.85) NHL subtypes, with only the association between ever use of carbaryl and SLL remaining significant after mutual adjustment for other pesticides (OR = 2.12, 95% CI = 1.01, 4.42). A significant increased risk was also observed for ever use of carbofuran and DLBCL. This, however, was no longer significant after adjusting for the top five correlated pesticides, (OR = 1.43, 95% CI = 0.86, 2.40). None of the tests for heterogeneity across NHL subtypes for each pesticide were statistically significant (all p-value for heterogeneity > 0.05).

Table 4

Associations between ever use of organophosphate and carbamate insecticides and lymphoma subtypes in the NAPP[a].

Pesticide	FL (n = 468)			DLBCL (n = 647)			SLL (n = 171)			Other (n = 400)
	Cases	OR (95% CI)	OR (95% CI)[b]	Cases	OR (95% CI)	OR (95% CI)[b]	Cases	OR (95% CI)	OR (95% CI)[b]	Cases	OR (95% CI)	OR (95% CI)[b]
Organophosphate (OP)
Malathion
Never used.	413	1	1	579	1	1	154	1	1	368	1	1
Ever used	55	1.80 (1.31, 2.46)	1.58 (1.11, 2.27)	68	1.77 (1.33, 2.35)	1.61 (1.16, 2.22)	17	1.50 (0.89, 2.54)	1.04 (0.57, 1.91)	32	1.27 (0.86, 1.87)	1.18 (0.77, 1.83)
Diazinon
Never used	441	1	1	615	1	1	160	1	1	385	1	1
Ever used	27	1.75 (1.14, 2.71)	1.20 (0.72, 2.02)	32	1.74 (1.17, 2.60)	1.26 (0.79, 2.03)	11	2.24 (1.17, 4.30)	1.67 (0.77, 3.61)	15	1.30 (0.75, 2.25)	1.10 (0.59, 2.07)
Fonofos
Never used	450	1	1	629	1	1	166	1	1	391	1	1
Ever used	18	1.95 (1.14, 3.34)	1.35 (0.73, 2.48)	18	1.77 (1.04, 3.02)	1.22 (0.67, 2.22)	5	1.87 (0.73, 4.81)	1.02 (0.35, 2.95)	9	1.39 (0.68, 2.84)	1.18 (0.54, 2.61)
Phorate
Never used	456	1		631	1		163	1		390	1
Ever used	12	0.83 (0.45, 1.54)		16	1.09 (0.63, 1.88)		8	1.91 (0.89, 4.10)		10	1.05 (0.54, 2.05)
Dimethoate
Never used	457	1		634	1		169	1		391	1
Ever used	11	1.62 (0.85, 3.12)		13	1.26 (0.69, 2.31)		2	¶		9	1.42 (0.70, 2.89)
Terbufos
Never used	458	1		637	1		166	1		392	1
Ever used	10	0.83 (0.43, 1.64)		10	0.85 (0.43, 1.67)		5	1.46 (0.57, 3.75)		8	1.02 (0.48, 2.15)
Dichlorvos
Never used	457	1		644	1		169	1		391	1
Ever used	11	1.31 (0.67, 2.57)		3	¶		2	¶		9	1.75 (0.84, 3.64)
Coumaphos
Never used	462	1		638	1		165	1		396	1
Ever used	6	1.09 (0.45, 2.63)		9	1.63 (0.78, 3.42)		6	3.97 (1.60, 9.84)		4	¶
Carbamate
Carbofuran
Never used	444	1		615	1		164	1		385	1
Ever used	24	1.25 (0.80, 1.97)		32	1.50 (1.00, 2.23)	1.43 (0.86, 2.40) ±	7	1.10 (0.50, 2.42)		15	1.08 (0.63, 1.88)
Carbaryl
Never used	445	1	1	621	1	1	159	1	1	386	1	1
Ever used	23	1.66(1.05, 2.64)	1.12 (0.66, 1.90)	26	1.52 (0.99, 2.35)	1.03 (0.63, 1.68)	12	2.59 (1.39, 4.85)	2.12 (1.01, 4.42)	14	1.29 (0.73, 2.27)	1.10 (0.58, 2.06)

Results suppressed where case count was low (n < 5).

Adjusted for the top five correlated pesticides.

Adjusted for age, study, gender, family history of lymphohematopoietic cancer; All tests for heterogeneity across NHL subtypes resulted in p > 0.05.

Mutually adjusted for: malathion, diazinon, fonofos, carbaryl; results not shown for non-significant pesticides (all further attenuated towards unity).

Discussion

Pooling of data from three case-control studies from the United States and one from Canada provided one of the largest efforts to evaluate possible associations between commonly used insecticides and NHL. Our results show that increasing duration of exposure to OP and carbamate insecticides significantly increased the risk of NHL. Analysis of these pesticides as a mixture or by individual use suggests that the OP insecticide malathion is particularly associated with risk of NHL overall and that use may be associated with FL and DLBCL subtypes. Few other positive individual associations were observed for other OP and carbamate insecticides and risk of NHL or its subtypes, after controlling for other pesticides.

Malathion is currently classified by IARC as probably carcinogenic to humans (Group 2A) in 2015 (IARC, 2015) and the by U.S. Environmental Protection Agency as having “suggestive evidence of carcinogenicity” in a draft review in 2016.(EPA, 2016) In both instances, it was noted that human data were limited, with the most consistent evidence being for NHL and for prostate cancer. Data from the individual case-control studies in the U.S.(Cantor et al., 1992; Waddell et al., 2001) and Canada,(McDuffie et al., 2001) included in the current pooled analysis, contributed to these earlier evaluations of a malathion use and NHL risk. Our pooled analyses of the data from these studies confirm prior associations with ever use of malathion and NHL risk in 1690 cases after adjustment for use of other pesticides. In addition, we show a significant exposure-response relationship with years of malathion use. Data from the prospective Agricultural Health Study (AHS) cohort (n = 495 cases with malathion use information, n = 179 exposed cases with quantitative data) indicate no association between ever use of malathion and NHL (RRever = 0.9, 95% CI = 0.8, 1.1) or for lifetime-days of use in the highest exposure category (RRhigh = 0.9, 95% CI = 0.6, 1.3).(Alavanja et al., 2014) Data from two other large cohort studies, which used crop exposure matrices to assess malathion use, also showed no association between use and NHL risk. (Leon et al., 2019) Differences in results between the current study and these cohorts could be due to different times between exposure and disease development (latent period), different uses of products containing malathion over time (including use on animals), different referent populations (general population versus occupationally exposed), or study design (cohort versus case-control).

A major advantage of this pooled analysis is larger numbers to evaluate possible links between insecticide exposure and NHL subtypes. This is important given the etiologic heterogeneity observed in lymphoma.(Morton et al., 2014) Farming occupations have been linked to FL (Fritschi et al., 2005), DLBCL(Cerhan et al., 2014; Ferri et al., 2017) and multiple myeloma(Ferri et al., 2017; Perrotta et al., 2008) in some studies. Here, we observed an association between malathion and FL and DLBCL subtypes as well as an association between SLL and the carbamate insecticide carbaryl. Few studies have evaluated specific pesticide use by NHL subtypes. No increased risks were reported for carbaryl use and SLL/chronic B-cell lymphocytic lymphoma (CLL)/mantle-cell lymphoma (MCL) subtype in the AHS cohort.(Alavanja et al., 2014) Given the inconsistent risk associations between carbaryl and NHL, as well as the small number of exposed SLL cases (n = 12) in this pooled analysis, it is possible that either of these contrasting observations could be due to chance. The observed associations between malathion and risk were robust in the current analysis, however, these too are not consistent with results from other studies.(Alavanja et al., 2014; Leon et al., 2019) The AHS reports relative risks (RR) for malathion and FL of RRever vs. never = 1.3, 95% CI = 0.7, 2.4 (60 cases) and RRhigh use = 1.6, 95% CI = 0.6, 4.4 (20 cases).(Alavanja et al., 2014) These are similar in magnitude to those reported here in 468 cases, ORever vs. never = 1.58, 95% CI = 1.11, 2.27, suggesting that if an association truly exists, the magnitude of effect is modest. No statistically significant heterogeneity was indicated by subtype in our study. Thus, more powerful studies will be needed to further explore this link.

There is mechanistic data supporting the carcinogenic potential of malathion. Purported mechanisms of action include direct genotoxicity (of either malathion or its metabolite malaoxon, (S-(1,2-dicarboethoxyethyl)O,O-dimethyl phosphorothiolate))(Blasiak et al., 1999; Pluth et al., 1996), disruption of critical cellular pathways involved in cellular proliferation(IARC, 2015), and the induction of oxidative stress (Abdollahi et al., 2004) and inflammation.(Banerjee et al., 1998; Galloway and Handy, 2003) Aside from the known links between autoimmune and chronic inflammatory disorders and lymphoma(Ekstrom Smedby et al., 2008), none of the above noted pathways have been concretely linked to the development of lymphoma. Some studies have suggested that pesticide exposure is associated with common chromosomal alterations t(14;18)(q32;q21) occurring in FL and DLBCL.(Chiu et al., 2006; Schroeder et al., 2001) Future studies examining potential biological mechanisms directly linking malathion exposure and lymphoma will be valuable.

OP and carbamate insecticides have short half-lives and leave the body within 24–48 h after exposure, thus, exposure assessment is typically questionnaire-based.(Bouchard et al., 2003) Pooling of data from four case-control studies created one of the largest resources with detailed historical use of specific OP and carbamate insecticides. This allows for a powerful assessment of risk for NHL overall, for exposure-response analyses, and by NHL subtypes. Despite these advantages, we cannot rule out the possibility of recall bias. In addition to differential case recall, some interviews were obtained from proxy respondents which could also be a source of differential exposure misclassification. A previous methodologic effort to evaluate case-response bias in the Nebraska study (included in this pooled effort) showed no evidence of case-response bias for any pesticide or for the number of pesticides reported.(Blair and Zahm, 1993) Further, when we compared risks from proxy versus self-respondents, however, we found no significant differences in the reported risk estimates for the positive associations observed in this study. In addition, our results for malathion show consistent associations with NHL overall and for FL and DLBCL in exposure-response analyses which are less susceptible to modest errors in exposure misclassification.(Marshall et al., 1981) Since there were more missing data on days/year of use (only available in two of four studies) we chose not to impute for days/year. Results for malathion use, however, are still significantly elevated despite the decrease in power in these analyses (Supplemental Table 2). And finally, although the study is quite large, the numbers of exposed cases is small in some analyses by subtype of NHL. This could explain the lack of observed statistical heterogeneity by subtype. Further, classification for subtypes of NHL have evolved over time and thus, the current analysis which includes ‘other’ types which were unclassifiable, as well as those defining SLL, would not be comparable with current classifications.

In conclusion, data from this large, pooled analysis of four case-control studies suggest a positive association between increasing years of malathion use and risk of NHL. These data add to a growing body of evidence, much of which comes from the individual case-control studies included here, suggesting that malathion may be carcinogenic in humans. Although there are few other studies that can explore the observed findings, more data are needed for evaluation of subtype-specific findings. In addition, mechanistic data are still needed to provide a biologically plausible link between malathion use and NHL.