Study population and case ascertainment
The AHS is an ongoing prospective cohort that includes 52,934 licensed private pesticide applicators in Iowa and North Carolina and 4916 licensed commercial applicators in Iowa. The cohort has been described in detail previously [17]. Briefly, the cohort is composed of individuals (82% of the target population enrolled) seeking licenses for pesticides which the U.S. Environmental Protection Agency (EPA) designated restricted use. In the AHS, pesticide exposure information has been collected in 3 phases of questionnaires--Phase 1 (1993–1997), Phase 2 (1999–2003), and Phase 3 (2005–2010),—with each phase including self-administered questionnaires or a computer assisted telephone interview (CATI) covering demographic, lifestyle, and occupational characteristics. In Phase 1, a total of 57,310 applicators completed the enrollment questionnaire and 25,291 returned the ‘Take-Home Applicator’ questionnaire. Of those initially enrolled, 36,341 applicators completed the Phase 2 ‘Pesticide Use Module Applicator’ questionnaire, and 24,170 applicators returned the Phase 3 questionnaire (Fig. 1). Full text of all questionnaires is available at https://aghealth.nih.gov/collaboration/questionnaires.html.
We regularly linked cohort members to state mortality registries and the National Death Index to determine vital status. We also regularly linked participants to cancer registries in Iowa (cases reported through 2015) and North Carolina (cases reported through 2014) to determine incident cancers. Cancers were classified according to the International Classification of Diseases for Oncology, third revision (ICD-O-3) [18]. Aggressive prostate cancer was defined as having one or more of the following tumor characteristics: distant stage, poorly differentiated grade, Gleason score ≥ 7, or fatal prostate cancer (i.e. underlying cause of death was prostate cancer).
The study protocol, including implied informed consent for completion of questionnaires, was approved by all relevant institutional review boards.
Exposure assessment
We previously published an analysis of 50 pesticides first reported on the Phase 1 enrollment questionnaire [12]. For the current analysis, we focused on those pesticides that were first reported at take-home (Phase 1 take-home questionnaire) or follow-up (Phase 2 and Phase 3 questionnaires), but not included in prior analyses (n = 39 pesticides). We only evaluated pesticides with 15 or more exposed aggressive PCa cases. Questions regarding pesticide use differed by questionnaire (Fig. 1). The take-home questionnaire provided a checklist of specific pesticide names/active ingredients and applicators marked those that they had ever used. The follow-up questionnaires included open-ended questions and applicators provided the name of the pesticide(s) used. In the Phase 2 questionnaire, applicators reported on use for the last year that they farmed. In the Phase 3 questionnaire, applicators were asked about use since their last questionnaire or interview (which could have been at Phase 1 enrollment, Phase 1 take-home, or Phase 2). The Phase 3 questionnaire also asked applicators to report the lifetime duration (in years) of use for each of the reported pesticides. Since the take-home and follow-up questionnaires were structured differently, this resulted in two different analytic groups: ever/never use to pesticides first reported at take-home (20 pesticides) and ever/never use to pesticides first reported at follow-up (19 pesticides). For pesticides first reported at follow-up (19 pesticides), we also evaluated lifetime duration of use (years), which was ascertained on the Phase 3 questionnaire.
Because many of these pesticides have never been evaluated in the AHS, we also assessed the plausibility of users’ duration information (excluding arsenical pesticides). We began by calculating the maximum number of years a pesticide could have been used. This was calculated by subtracting the year a pesticide/active ingredient was first registered in the U.S. from the year the individual completed the phase 3 questionnaire (2005–2010; the year participants responded to questions about duration of use for the given pesticides in the AHS). We used four different sources of information to determine registration year, including the chemical specific Registration Eligibility Decision (RED) published by the U.S. EPA, the EPA’s document on “Chemicals Registered for the First Time as Pesticidal Active Ingredients Under FIFRA” [19], different versions of the Farm Chemicals Handbooks [20] covering the timespan of the AHS, and the EXTOXNET: the Extension Toxicology Network [21], which is a Pesticide Information Project of Cooperative Extension Offices of Cornell University, Michigan State University, Oregon State University, and the University of California at Davis. We also evaluated whether any pesticide had been canceled prior to 2010 using the EPA’s chemical specific REDs, U.S. Federal Register notices, as well as the Sittig’s Handbook of Pesticides and Agricultural Chemicals [22]. We found that fluazifop-butyl and disulfoton had been canceled in the latter halves of 2004 [23] and 2009 [24], respectively. To assess plausibility of duration responses by AHS applicators, we compared the maximum number of years a pesticide could have been available for purchase to self-reported life years of use. We evaluated the frequency and proportion of applicators using a given pesticide as well as the number and proportion of users with reported years of use that were plausible.
Analytic population
All respondents completed the enrollment questionnaire; however, responses about pesticide use from the take-home and follow-up questionnaires were structured differently resulting in two different analytic groups (Fig. 1). The first analytic group included participants that responded to the take-home questionnaire. Among the 25,291 total participants who completed the take-home questionnaire, we excluded 4368 individuals (2375 commercial applicators [did not receive the Phase 3 follow-up questionnaire], 556 women, 80 who had moved out of state, 620 with prevalent cancer or those diagnosed before completing the questionnaire, and 737 who were diagnosed with non-aggressive PCa during follow-up), leaving 20,923 private applicators (20,040 non-cases/883 aggressive PCa cases) for analysis. This analytic set was used to evaluate: (1) ever/never use as indicated on the take-home questionnaire, and (2) ever/never use considering take-home questionnaire use plus information reported at one or both follow-up questionnaire(s) (Phase 2 and/or 3). The second analytic group included participants that responded to both follow-up questionnaires (Phase 2 and Phase 3). Among the 21,142 participants that completed both follow-up questionnaires (Phase 2 and Phase 3), we excluded 2209 individuals (640 women, 70 who had moved out of state, 963 individuals with prevalent cancer or those diagnosed before completing the Phase 2 or Phase 3 questionnaire, and 536 who were diagnosed with non-aggressive PCa during follow-up), leaving 18,933 individuals (18,199 non-cases/734 aggressive PCa cases) for analysis.
Statistical analyses
We used Cox proportional hazards models, with age as the time scale, to calculate hazard ratios (HRs) and 95% confidence intervals (CIs) for the association between individual pesticide use and aggressive PCa risk. We censored follow-up at the time of aggressive prostate cancer diagnosis, death, movement out of the state, or at December 31, 2014 for North Carolina and 2015 for Iowa, whichever came first. Supplemental Tables 1 and 2 report associations for overall PCa risk. We adjusted all models for year of birth; state (North Carolina, Iowa); family history of prostate cancer in first-degree relatives; race; and cigarette smoking status (never, former, current, missing) based on information on the Phase 1 enrollment questionnaire. Other covariates, including body mass index (BMI: underweight, normal, overweight, obese), fruit servings (< 1/day, ≥ 1/day), and leisure-time physical activity in the winter (none, > 0–2 h/week, ≥ 3 h/week), did not materially impact observed point estimates (≥ 10%) and, therefore, were not retained in models. We evaluated correlations between ever/never exposure to individual pesticides using the Pearson correlation coefficient and conducted additional adjustment for ever/never use of the three pesticides mostly highly correlated with the pesticide of interest to account for co-exposures (rho range = 0.001–0.2). We also explored adjustment of all models with pesticides previously linked to aggressive PCa in the AHS (malathion, fonofos, terbufos, and aldrin) [12] as well as with those significantly associated with PCa in the current analysis (Supplemental Table 3). However, adjustment for co-exposures did not materially impact observed point estimates and were not retained in models. We conducted exposure-response analyses for those pesticides with valid reported duration data from the Phase 3 questionnaire. We created categories for years of use based on the distribution of years reported, split at the median, among all cases. Those with reported years of use that were implausible were coded as having missing exposure in duration analyses. To compute tests for linear trend, the Wald test was used, treating the median value for each category as continuous. We evaluated potential effect modification by family history of PCa. Likelihood ratio tests were used to assess differences between strata.
Using multivariate logistic regression, we also explored possible predictors of prostate specific antigen (PSA) screening including demographic characteristics and individual pesticide use to assess whether screening could explain any observed relationships between pesticides and prostate cancer risk. Participants provided PSA screening (yes/no) information on two follow-up questionnaires (N = 28,880 men).
All analyses were conducted using SAS, 9.4 (SAS Institute, Cary, North Carolina) and use AHS data release P1REL201701, P2REL201701, and P3REL201701.