Formerly with NCI OEEB.

An algorithm developed to estimate pesticide exposure intensity for use in epidemiologic analyses was revised based on data from two exposure monitoring studies. In the first study, we estimated relative exposure intensity based on the results of measurements taken during the application of the herbicide 2,4-dichlorophenoxyacetic acid (2,4-D) (n = 88) and the insecticide chlorpyrifos (n = 17). Modifications to the algorithm weighting factors were based on geometric means (GM) of post-application urine concentrations for applicators grouped by application method and use of chemically-resistant (CR) gloves. Measurement data from a second study were also used to evaluate relative exposure levels associated with airblast as compared to hand spray application methods. Algorithm modifications included an increase in the exposure reduction factor for use of CR gloves from 40% to 60%, an increase in the application method weight for boom spray relative to in-furrow and for air blast relative to hand spray, and a decrease in the weight for mixing relative to the new weights assigned for application methods. The weighting factors for the revised algorithm now incorporate exposure measurements taken on Agricultural Health Study (AHS) participants for the application methods and personal protective equipment (PPE) commonly reported by study participants.

The risk of adverse health effects associated with long-term exposure to pesticides is difficult to assess in epidemiologic studies due to various limitations that have been summarized in the literature [

The AHS algorithm has four variables that were combined as follows:

Exposure Intensity Score = ([MIX] + [APPLY] + [REPAIR]) × [PPE])

where [MIX] represents exposure from mixing and loading operations prior to application, [APPLY] represents exposure from applying pesticides, [REPAIR] represents exposure from contact with contaminated surfaces during the repair of pesticide application equipment, and [PPE] represents an exposure reduction factor to account for use of PPE.

The reliability of the version 1 algorithm intensity scores for correctly rank ordering various application scenarios has been evaluated based on four field monitoring studies; (1) a study among Canadian farmers [

In the AHS/PES, we selected 2,4-D and chlorpyrifos because 2,4-D is one of the most important agricultural and residential herbicides and chlorpyrifos is one of the most important agricultural insecticides. In addition, the pharmacokinetics of these chemicals are relatively well understood. Both chemicals are widely used by AHS cohort members. Similarly, the AHS/OFES measured captan, the second most frequently used fungicide in the AHS. These studies included some of the most frequently used application methods in the cohort.

Measurement results from the AHS field studies were used to examine relative differences in urinary biomarker concentrations associated with the algorithm exposure variables. These comparisons enabled us to modify the algorithm weights using AHS-derived field study data while still relying on information from the literature and PHED for algorithm weights, particularly where AHS-specific field data was lacking. Decisions on changing any algorithm weights were based on the field study data in combination with the body of information from the literature and PHED. In addition, we re-scaled the algorithm scores and assigned weights for application methods reported by cohort members in follow-up questionnaires but not in the enrollment questionnaire. These enhanced algorithm weights provide the basis for updated exposure intensity scores currently used in AHS epidemiological analyses.

The methodology and measurement results for the AHS/PES have been previously described in detail [

Arithmetic means, geometric means (GM) and geometric standard deviations (GSD) of post-application urine concentrations for AHS/PES applicators were calculated for application method and use of chemical-resistant or other waterproof gloves (referred to as CR gloves). We used a two-way analysis of variance procedure among study participants (GLM Procedure, SAS version 9.1, Cary, NC, USA) to evaluate whether CR-glove use or application method significantly affected the urine concentrations of the measured analyte, when controlling for the other factor. Urine concentrations were log-transformed to account for right skewed data.

We calculated the ratios of the GM’s to evaluate the relative exposure intensity for (1) for boom spray compared to an in-furrow/granular application method and (2) the reduction in post-application urine concentrations attributable to glove use. Spearman correlation coefficients were calculated between version 2

To provide a secondary method to evaluate the revised weighting factors, we fitted a nonlinear regression model to assess the joint influence of the algorithm variables on post-application urine concentrations (Y) in μg/L:

Y = {α_{0} + α_{1} Mix + α_{2} Method + α_{3} Repair} × {1 − (β_{1} Gloves) – (β_{2} PPE other)} (1)

where α_{0} represented the urinary concentration at the referent level of all factors, where α_{1}, α_{2} and α_{3} parameters represented the increase in Y for mixing (1 = yes, 0 = no), use of hand spray (method = 1) or boom spray (method = 0) for 2,4-D, or boom spray (method = 1) or in-furrow (method = 0) for chlorpyrifos, and repairing equipment (1 = yes, 0 = no), respectively, and where β_{1} and β_{2} parameters represented the reduction factors for use of CR gloves (1 = yes, 0 = no) and/or other PPE (1 = yes, 0 = no), respectively. We then compared the predicted values from the model to the algorithm scores. Because the regression coefficients were pesticide specific and based on relatively limited data in many of the exposure scenarios, we did not directly use the parameter estimates as weights, but rather to jointly assess the relative influences of the variables.

To evaluate the extent to which algorithm scores could be used to categorize applicators into exposure groups, we divided the 2,4-D applicators into three groups by algorithm score (<50, 50–100, >100), computed summary statistics, and conducted a nonparamteric test for trends based on rankings using the Stata nptrend command, an extension of the Wilcoxon rank-sum test. Due to a smaller number of applications and limited range of scores, the chlorpyrifos data were divided into two groups using a cut-point of 50.

CR glove use was associated with a significant difference in urinary 2,4-D GM levels overall, when controlling for application method (p < 0.0001). Among 2,4-D applicators who wore CR gloves, GMs of the post-application urine concentrations were 75% and 72% lower for boom (14 µg/L

Among chlorpyrifos applicators, the GMs of 3,5,6-trichloro-2-pyridinol (TCPy) post-application urine concentrations were 50% and 56% lower with CR glove use for in-furrow (granular formulation) and boom spray (liquid formulation) application, respectively, (GM = 6 µg/L and GM = 14 μg/L) compared with no glove use (12 µg/L and 32 μg/L). While CR glove use was associated with lower GM TCPy levels, the results were not statistically significant (p = 0.084) when we controlled for application method.

Based on a reduction of 72% to 75% among the 2,4-D applicators, and of 50% to 56% among the chlorpyrifos applicators, the reduction factor for use of CR gloves was increased from 40% in the version 1 algorithm to 60% in version 2.

Among 2,4-D applicators, the GMs for hand spray applicators were 1.6 times and 1.5 times higher than for boom spray applicators who did (23 μg/L

For chlorpyrifos applicators, the GMs for boom spray applicators were 2.3 and 2.7 times higher than for in-furrow applicators for those who did (14 μg/L

Based on the ratio of the GM’s by application method, we decided to increase the weighting factor for boom spray, thereby reducing the relative difference with hand spray from version 1 (

Post-application urine concentrations (µg/L) grouped by application method and CR glove use for 2,4-D ^{1} (N = 88) and chlorpyrifos ^{2} (N = 17) applications.

Application Method | CR Glove Use | N | AM | GM | GSD | CR Glove Use ^{3} | Application Method ^{3} |
---|---|---|---|---|---|---|---|

Boom Spray | Yes | 32 | 27 | 14 | 3.1 | P < 0.0001 | P = 0.092 |

No | 14 | 91 | 55 | 3.0 | |||

Hand Spray | Yes | 21 | 48 | 23 | 3.3 | ||

No | 21 | 200 | 81 | 4.9 | |||

In-furrow (granular) | Yes | 7 | 8 | 6 | 1.8 | P = 0.084 | P = 0.014 |

No | 6 | 14 | 12 | 1.8 | |||

Boom Spray(liquid) | Yes | 2 | 14 | 14 | 1.3 | ||

No | 2 | 47 | 32 | 3.6 |

^{1} 2,4-D measured as a urinary biomarker for 2,4-D.

^{2} TCPy measured as a urinary biomarker for chlorpyrifos.

^{3} P values from two-way analysis of variance using (independent variables: glove use and application method).

Abbreviations: AM = arithmetic mean; CR = chemically-resistant; GM = geometric mean; GSD = geometric standard deviation; N = number of application days monitored.

In the version 1 algorithm, hand spray and air blast had the same weight (

The version 2 algorithm retained the same four variables as version 1 because these variables were

In the version 1 algorithm, intensity scores ranged from 0.1 to 20, with scores that included decimal values. To use only integers with a minimum value of 1, the version 2 algorithm weights were re-scaled by a factor of 10, so version 2 intensity scores range from 1 to 220. Rescaling was done primarily for convenience and had no effect of the relative ranking by algorithm score.

AHS Pesticide Exposure Algorithm Weighting Factors. Algorithm Intensity Score = (MIX + APPLY + REPAIR) × PPE.

Did Not Mix | 0 | 0 |

Mix <50% of the time | 3 | 20 |

Mix >50% of the time | 9 | 50 |

No | 0 | 0 |

Yes | 2 | 20 |

Air blast | 9 | 150 |

Hand Spray | 9 | 90 |

Mist Blower Or Fogger | 9 | 90 |

Fog Or Mist Animals | 9 | 90 |

Greenhouse Sprayer | 9 | 90 |

Pour Fumigant From Bucket | 9 | 90 |

Powder Duster | 9 | 90 |

Backpack Sprayer | 8 | 80 |

Dust Animals | 7 | 70 |

Pour On Animals | 7 | 70 |

Garden Hose | None | 50 |

Hand Held Squeeze Or Squirt Bottle | None | 50 |

Watering Can/Sprinkling Can | None | 50 |

Soil Injected Or Drilled | 4 | 40 |

Spray Over Rows | 4 | 40 |

Boom On Tractor | 3 | 40 |

Broadcast Application | 3 | 40 |

Personally Applied To Seed | 2 | 40 |

Banded/Directed Spray (liquid) | 2 | 30 |

Banded Application (granular) | 2 | 20 |

Gas Canister | 2 | 20 |

Hang Pest Strips In Barn | 2 | 20 |

In-Furrow | 2 | 20 |

Incorporated | 2 | 20 |

Inject Animals | 2 | 20 |

Seed Treatment | 1 | 20 |

Hand Spreader Or Push Spreader | None | 20 |

Planter Box | None | 20 |

Aerial | 1 | 10 |

Chemical Resistant or Rubber Gloves | 40% | 60% |

Cartridge Respirator, Tyvek Coveralls | 30% for use of 1 or more | 10% each with max of 30% |

Face Shield, Goggles, Boots, Apron, Other | 20% for use of 1 or more | |

Fabric/leather gloves | 20% | none |

^{1} None indicates methods for which a version 1 weighting factor was not assigned

In the version 2 algorithm, the protection factor for glove use was increased from 40% to 60%. The increase was based on comparison of the GM urine concentrations for CR glove use relative to no CR glove use that ranged from 50% to 75% (

The enrollment questionnaire asked about use of “chemically" resistant gloves (for example, neoprene or nitrile gloves), and because we could not distinguish between different types of CR gloves based on the enrollment questionnaire, we assigned the same reduction for rubber, waterproof or disposable latex gloves as for CR gloves. The version 1 algorithm included a 20% reduction use of fabric/leather gloves. Data from our monitoring study AHS/PES study, however, did not support treating fabric/leather gloves as protective, and therefore, the version 2 algorithm does not assign any reduction in exposure for their use.

We increased the weight for boom spray application from 3 (on version 1 scale) to 40 (on version 2 scale) while retaining the banded/in-furrow application method weight at 2 (20 on the version 2 scale) to reflect the approximately 2-fold exposure difference observed in the chlorpyrifos data. Based on the detection frequency difference of THPI in the AHS/OFES, we increased the air blast application weight to 150 which was now 67% higher than the hand spray weight of 90. This change ensured that airblast would be the application method with the highest exposure potential under all exposure scenarios. Because post-enrollment AHS questionnaires expanded the number of application methods, we accommodated these additional methods in the version 2 algorithm by assigning weights based on similarities to previously assigned methods (

In version 1, the weight for mixing equaled the weight for hand spray (previously the highest application method weight). In version 2, we assigned a relatively smaller weight of 50 for mixing (

Because only five 2,4-D applicators did not personally mix or load on the morning prior to monitoring, the amount of data available to assess exposure that occurs during mixing compared with the rest of the application process was limited. The GM of the post-application urine concentrations for applicators who mixed on the morning of urine collection was ~50% higher than those who did not mix, which is somewhat lower than previously reported in the literature [

Repairing equipment increased exposure for 2,4-D applicators (GM = 34 µg/L, n = 26 who repaired

Spearman correlation coefficients between version 2 algorithm score and measurements of 2,4-D in post-application urine were greater than the Spearman correlation between version 1 algorithm scores and measurements of 2,4-D in post-application urine but not for chlorpyrifos (

Spearman correlation coefficients between Version 1 algorithm scores and measurements of post-application urine 2,4-D and chlorpyrifos and modeled post-application urine concentrations for 2,4-D (N = 88) and chlorpyrifos (N = 17) and Version 2 algorithm scores with post-application urine concentrations and modeled post-application urine concentrations for 2,4-D and chlorpyrifos.

Algorithm | ||
---|---|---|

Version 1 | Version 2 | |

2,4-D | ||

Version 1 | 1 | |

Version 2 | 0.95 | 1 |

Post-apply urine conc. | 0.42 | 0.48 |

Predicted post-apply urine concentration ^{1} | 0.96 | 0.97 |

Chlorpyrifos ^{2} | ||

Version 1 | 1 | |

Version 2 | 0.97 | 1 |

Post-apply urine conc. | 0.53 | 0.52 |

Predicted post-apply urine concentration | 0.52 | 0.59 |

^{1} Modeled value from a non-linear regression mode l.

^{2} TCPy measured as a urinary biomarker for chlorpyrifos.

We fitted a nonlinear model based on the algorithm formula (1) to compare the updated weights with parameter estimates from a joint analysis of all component variables simultaneously. Coefficients were in the expected direction and the application method and CR-glove PPE terms were significant (see

Nonlinear regression of post-application urine concentration on algorithm.

Y = [{α_{0}} + {α_{1}} × mix + {α_{2}} × method + {α_{3}} × repair] × [1 − {β_{1}} × gloves − {β_{2}} × ppe_other].

2,4-D (n = 88) | R-Squared = Regression | 0.36 |
---|---|---|

^{1} | ||

Intercept α_{0} | 27 | 0.76 |

Mix α_{1}, | 58 | 0.53 |

Method α_{2} | 123 | 0.02 |

Repair α_{3} | 32 | 0.59 |

Gloves β_{1} | 0.75 | <0.001 |

PPE other β_{2} | 0.26 | 0.26 |

Chlorpyrifos (n = 17) | R-Squared = Regression | 0.77 |

^{1} | ||

Intercept α_{0} | 8 | 0.22 |

Mix α_{1}, | Na ^{2} | Na ^{2} |

Method α_{2} | 33 | 0.006 |

Repair α_{3} | 15 | 0.89 |

Gloves β_{1} | 0.51 | 0.014 |

PPE other β_{2} | 0.21 | 0.59 |

^{1} α_{0} represented the urinary concentration at the referent level of all factors, where α_{1}, α_{2} and α_{3} parameters represented the increase in Y for mixing (1 = yes, 0 = no), use of hand spray (method = 1) or boom spray (method = 0) for 2,4-D, or boom spray (method = 1) or in-furrow (method = 0) for chlorpyrifos, and repairing equipment (1 = yes, 0 = no), respectively, and where β_{1} and β_{2} parameters represented the reduction factors for use of CR gloves (1 = yes, 0 = no) and/or other PPE (1 = yes, 0 = no), respectively.

^{2} na: all participants mixed chlorpyrifos and the regression omitted the variable.

When grouped by approximate tertile of the algorithm scores, we found a statistically significant trend (p ≤ 0.01) in the post-application 2,4-D GM concentrations (

Arithmetic means, geometric means and geometric standard deviation of post-application urine concentrations by Version 2 algorithm score category.

Category | Range | N | AM | GM | GSD |
---|---|---|---|---|---|

<50 | 12–48 | 40 | 30 | 15 | 3.2 |

50–100 | 59–90 | 24 | 78 | 39 | 3.6 |

>100 | 110–160 | 24 | 178 | 69 | 4.7 |

All | 88 | 84 | 30 | 4.2 | |

p-trend | <0.01 | ||||

^{1} | |||||

Category | Range | N | AM | GM | GSD |

<50 | 24–36 | 9 | 10 | 8 | 2.1 |

≥50 | 70–110 | 8 | 22 | 16 | 2.1 |

All | 17 | 11 | 10.6 | 2.3 | |

p-trend | 0.03 |

^{1} TCPy measured as a urinary biomarker for chlorpyrifos.

Abbreviations: AM = Arithmetic Mean, GM = geometric mean, GSD = Geometric Standard Deviation.

Developing estimates of pesticide exposure intensity for large-scale cohort studies is a challenging, but critical task for exposure–response analysis. The use of simple exposure metrics, such as duration, fails to account for large differences in cumulative exposure that can occur because of the amount and concentration of active ingredients in the pesticide products applied, mixing and application methods, equipment size and design, PPE use, individual work practices and personal hygiene [

Although we fitted a model to compare the updated algorithm weights with parameter estimates from a joint analysis of all component variables simultaneously, we did not use the coefficients from the model directly to change algorithm weight because coefficients were pesticide specific, based on relatively limited data and encompassed relatively few exposure scenarios. Nonetheless, coefficients were in the expected direction and the application method and PPE terms were significant, supporting the usefulness of the exposure algorithm.

Previous evaluations of the AHS algorithm (version 1) in both non-AHS and AHS applicators demonstrated its usefulness [

Information about several commonly used application methods was obtained using the enrollment questionnaire. Additional application methods used by members of the cohort have been identified in subsequent follow-up data collections. Robust exposure measurement data were not available for assigning algorithm score weights for these methods, so scores previously developed for similar methods were assigned. The uncertainty in these assignments is a limitation of the updated algorithm.

Because liquid chlorpyrifos was always applied by spraying and granular chlorpyrifos was always applied using banded or in-furrow methods in the AHS/PES study, we could not distinguish between application method or formulation type. Both dermal measurements and urine concentrations were higher for liquid spray applications than for in-furrow granular applications. Formulation type was not included in the algorithm because it was not collected in the enrollment questionnaire.

While exposure levels varied by chemical, we lacked sufficient measurement data on determinants of exposure for multiple pesticides under different application scenarios to develop pesticide-specific weights, and therefore algorithm weights apply to all pesticides. In addition, differences in absorption, metabolism and excretion rates for different pesticides and tissue-specific effects did not allow algorithm intensity scores to estimate internal doses directly. Nonetheless, it was clear from the results that the algorithm scores, on average, provided an indicator of exposure intensity for applicators using the most commonly reported application methods in the AHS cohort. Epidemiologic analyses of the AHS cohort have used the algorithm score (version 1) extensively as a measure of exposure intensity (

Both version 1 and 2 of the algorithm are based on an extensive review of the world’s literature and the use of the Pesticide Handlers Exposure Database (PHED) which included many different chemicals (6). With the addition of revised algorithm weights derived from the two field studies within the AHS we were able to adjust the weights to account for local variations in farming practices and conditions. We judge version 2 to be superior to version 1 but the correlations between version 1 and version 2 are high r = 0.95 for 2,4-D and r = 0.97 for chlorpyrifos. This demonstrates that local conditions and characteristics can have some influence on algorithm weights, although the degree of influence is not substantial. The revised algorithm (version 2) will be used in future AHS epidemiologic analyses.

Revised weighting factors in a pesticide exposure intensity algorithm were developed for use in epidemiologic analyses for the Agricultural Health Study by using exposure monitoring data from two monitoring substudies in combination with the world’s exposure literature and PHED.

This work has been supported in part by the Intramural Research Program of the NIH: National Cancer Institute (Z01-CP010119-12) and National Institute of Environmental Health Sciences (Z01-ES049030-1) and by the Centers for Disease Control and Prevention, National Institute for Occupational Safety and Health and the United State Environmental Protection Agency. The United States Environmental Protection Agency also funded, in part, the research described here under Contracts 68-D99-011 and 68-D99-012, and through Interagency Agreement DW-75-93912801-0. We thank the participants of the Agricultural Health Study for their valuable contributions to this research.

The authors declare no conflict of interest.

Mention of trade names or commercial products does not constitute endorsement or recommendation for use. This manuscript has been subjected to U.S. Environmental Protection Agency review and approved for publication. The findings and conclusions in this report are those of the authors and do not necessarily represent the views of the National Institute for Occupational Safety and Health.