A Task-Specific Algorithm to Estimate Occupational (1→3)-β-D-glucan Exposure for Farmers in the Biomarkers of Exposure and Effect in Agriculture Study

Abstract

Objectives

Farmers may be exposed to glucans (a cell component of molds) through a variety of tasks. The magnitude of exposure depends on each farmer’s activities and their duration. We developed a task-specific algorithm to estimate glucan exposure that combines measurements of (1→3)-β-D-glucan with questionnaire responses from farmers in the Biomarkers of Exposure and Effect in Agriculture (BEEA) study.

Methods

To develop the algorithm, we first derived task-based geometric means (GMs) of glucan exposure for farming tasks using inhalable personal air sampling data from a prior air monitoring study in a subset of 32 BEEA farmers. Next, these task-specific GMs were multiplied by subject-reported activity frequencies for three time windows (the past 30 days, past 7 days, and past 1 day) to obtain subject-, task-, and time window-specific glucan scores. These were summed together to obtain a total glucan score for each subject and time window. We examined the within- and between-task correlation in glucan scores for different time frames. Additionally, we assessed the algorithm for the ‘past 1 day’ time window using full-shift concentrations from the 32 farmers who participated in air monitoring the day prior to an interview using multilevel statistical models to compare the measured glucan concentration with algorithm glucan scores.

Results

We focused on the five highest exposed tasks: poultry confinement (300 ng/m³), swine confinement (300 ng/m³), clean grain bins (200 ng/m³), grind feed (100 ng/m³), and stored seed or grain (50 ng/m³); the remaining tasks were <50 ng/m³ and had similar concentrations to each other. Overall, 67% of the participants reported at least one of these tasks. The most prevalent task was stored seed or grain (64%). The highest median glucan scores were observed for poultry confinement and swine confinement; these tasks were reported by 2% and 8% of the participants, respectively. The correlation between scores for the same task but different time windows was high for swine confinement and poultry confinement, but low for clean grain bins. Task-specific scores had low correlation with other tasks. Prior day glucan concentration was associated with the total glucan ‘past 1 day’ score and with swine confinement and clean grain bin task scores.

Conclusions

This study provides insight into the variability and key sources of glucan exposure in a US farming population. It also provides a framework for better glucan exposure assessment in epidemiologic studies and is a crucial starting point for evaluating health risks associated with glucans in future epidemiologic evaluations of this population.

Introduction

(1→3)-β-D-glucans, commonly referred to as glucans, are glucose polymer components of the cell walls of fungi, bacteria, and plants (Fogelmark et al., 1994; Douwes et al., 2003). Studies have suggested that exposure to airborne glucans plays a role in bioaerosol-induced inflammatory responses and resulting respiratory symptoms (Douwes et al., 2003). Glucan exposure has been associated with an increase in the severity of symptoms of nose and throat irritation, potentially due to its immune stimulatory properties and activation of neutrophils and stimulation of macrophages and eosinophils (Rylander, 1999; Douwes, 2005; Tischer et al., 2011).

High glucan exposure has been observed in several types of industries, including metal plants, wastewater treatment facilities, waste composting plants, greenhouses, and poultry facilities, and in several occupations, including farmers, cotton growers, veterinarians, and workers handling animal feed and grain (Douwes et al., 2003; Wouters et al., 2006; Adhikari et al., 2011; Cyprowski et al., 2011; Sykes et al., 2011; Samadi et al., 2013; Madsen et al., 2014; Viegas et al., 2017). The current work on glucan exposures in farming populations is limited. Excluding one study (Sauvé et al., 2020), the few studies evaluating glucan exposure in agricultural settings have been conducted outside of the USA (Lawniczek-Walczyk et al., 2013; Viegas et al., 2017) or focused on worst-case exposures (Singh et al., 2011).

In this paper, we developed a task-specific algorithm to assign quantitative estimates of glucan exposure for farmers in the Biomarkers of Exposure and Effect in Agriculture (BEEA) study as a crucial first step in characterizing their glucan exposure for future risk analyses. We first calculated task-specific geometric mean (GM) concentrations of glucan exposure from a prior air monitoring study conducted within a subset of BEEA participants (Sauvé et al., 2020). We then multiplied the task-specific GMs by the duration each participant reported doing those tasks for three different time windows (the past 30 days, past 7 days, and past 1 day) based on participants’ questionnaire responses to obtain subject-, task-, and time window-specific glucan intensity estimates. The task-specific intensity estimates were then summed together to get a total glucan estimate for each participant. This approach aims to capture between-subject exposure variability and assign exposure with greater consistency, transparency, and reproducibility (Friesen et al., 2015; Sauvé and Friesen, 2019). Although this approach has not been applied to glucans, studies of other exposures have successfully used similar methods (Sauvé and Friesen, 2019).

Methods

Study population

The BEEA study was designed to facilitate investigations to better understand the biological mechanisms underlying observed associations between agricultural exposures and adverse health outcomes. BEEA is a molecular epidemiology subcohort within the Agricultural Health Study (AHS), a prospective cohort study that includes farmers and pesticide applicators from Iowa and North Carolina who were recruited into the study between 1993 and 1997 (Alavanja et al., 1996; Hofmann et al., 2015). The BEEA study recruited 1681 male AHS farmers, aged 50 years or older, who had completed questionnaires at AHS enrollment (1993–1997) and two follow-up AHS interviews (1999–2003 and 2005–2010).

Participants completed an in-home interview and provided biological samples at the time of BEEA study enrollment. The agricultural (non-pesticide exposure) section of the BEEA interview questionnaire included questions regarding crops grown, animals raised, time spent in animal confinement, and performance of specific farming tasks (Hofmann et al., 2015). The questionnaire was updated to expand the assessment of the agricultural activities (see Table 1), in particular, to ascertain more details on the duration and frequency of bioaerosol-related agricultural activities. Thus, 1235 participants completed the first version (Version 2010) and 446 participants completed the second more detailed version (Version 2014).

Available exposure data

A subset of 32 BEEA farmers in Iowa participated in a bioaerosol air sampling study the day before their BEEA interview/phlebotomist visit (Sauvé et al., 2020). Twenty-four farmers were visited two or more times and 204 task samples were collected (average 3 per day) with an average sampling duration of 34 min (range 4–105 min). On the day of air sampling, an industrial hygienist visited the farm to collect task-based and full-shift personal inhalable fraction air samples using Button aerosol samplers (SKC Inc, No. 225–360, Eighty Four, PA) containing pre-weighted 25 mm, binder-free glass fiber filters connected to air sampling pumps (BGI OMNI-400, MesaLabs, Butler, NJ or SKC model 224-PCXR4, SKC Inc, Eighty Four, PA) operated at a flow rate of 4 l/min. The industrial hygienist also recorded the agricultural activities conducted by the farmer. The samples were analyzed for (1→3)-β-D-glucans at the University of Iowa using a Limulas Amoebocyte Lysate (LAL) assay (Glucatell Reagent Kit, Associates of Cape Cod). Using these data, Sauvé et al., 2020 developed several models to identify determinants of glucan exposure. In addition, for this paper, we associated each task-based sample with one of the above-mentioned agricultural tasks.

Glucan exposure intensity estimates

We calculated task-specific glucan concentrations using two methods: (i) derived from previously developed models and (ii) calculated from the BEEA air sampling data. For most tasks, we calculated the predicted geometric mean (GM) from the glucan-specific model coefficients presented in Sauvé et al. (2020), which included an overall task model, a crop-specific model, and a livestock-specific model. For each BEEA task, we multiplied the coefficients contributing to that task together to calculate the GM. For example, for the task harvest grain, we multiplied the coefficients for the model intercept, work location = field, activity = harvest, and crop type = grain for a predicted glucan GM of 15 ng/m³. (See Supplementary Table S1, available at Annals of Occupational Hygiene online, for coefficients used for other tasks and the resulting predicted GMs.) We used the second approach, calculating the GM directly from air sampling data, for tasks that were subsets of tasks in Sauvé et al. (2020). For example, the task grind feed was included in the general feed work category in the model, so we used the raw sampling data for the subset grinding feed. We also used raw data where we grouped tasks a bit differently from the Sauvé et al. (2020) models. As a result, GMs for the tasks haul alfalfa, stored seed or grain, and clean grain bins were calculated directly from the BEEA air sampling data. Because of the lack of exposure data and low prevalence of cotton farming in our population (0.3% of study population), the tasks harvest cotton and haul cotton were excluded.

Glucan exposure scores

To develop the glucan algorithm, we first calculated task-specific GMs (in ng/m³) for each agricultural task. All farmers are exposed to glucans throughout their agricultural activities. With an aim to identify broad contrasts in exposure, we focused on the five highest exposed tasks (Table 1 and Supplementary Table S1, available at Annals of Work Exposures and Health online), all with glucan GMs of ≥50 ng/m³, and assigned task-intensity scores at one-significant digit because the measurements were based on a single, small study: poultry confinement (300 ng/m³), swine confinement (300 ng/m³), clean grain bins (200 ng/m³), grind feed (100 ng/m³), and stored seed or grain (50 ng/m³). We omitted the cleaning and feeding sub-tasks within the poultry and swine confinement questions because the sub-task GMs were similar to the main task GMs and their durations were already accounted for in the primary question on work in the confinement area.

Next, for participants who completed BEEA questionnaire Version 2014, we calculated task-specific glucan exposure scores by multiplying the task-specific GMs (ng/m³) with the task duration (in hours) from BEEA questionnaire responses for three time windows (the past 30 days, 7 days, and 1 day; in hours) for a quantitative glucan intensity score in ng/m³-h. The frequency responses were categorical, so category midpoints were used in the calculation. If the farmer indicated that they did not perform a particular activity, the activity duration was assigned 0 h. Each participant’s total glucan score for each time window was calculated by adding together the task-specific glucan scores.

For participants who completed Version 2010, frequency information was less detailed or unavailable for some tasks, requiring us to adapt the algorithm and to focus only on the past 30-day time window. For the tasks clean grain bins and grind feed, Version 2010 included a question on the number of times the farmer conducted that activity but did not include a question on the average duration per time. For the tasks poultry confinement and swine confinement, Version 2010 included only a yes/no question for these activities and no measure of frequency. For these four tasks, we derived median durations from the Version 2014 participant responses, with the median duration for the animal confinement tasks stratified by the number of animals. For the task stored seed or grain, no information was collected in Version 2010; thus, we applied group medians from the Version 2014 participants stratified by crop and animal type.

Statistical analysis

For the five tasks included in the glucan algorithm, we calculated the median and inter-quartile range (IQR) of the time spent doing each task in the past 30 days, 7 days, and 1 day based on the questionnaire Version 2014 to represent the contrast in exposure observed in our study population. We also calculated the median and IQR of the glucan exposure score for each task using Version 2014, Version 2010, and both versions combined, respectively. We evaluated the correlation between the scores for the three time windows (past 30 days, 7 days, and 1 day) for each task in the algorithm using the Spearman correlation statistic, restricting each analysis to those with a 30-day task score greater than 0 and to those who completed Version 2014. We evaluated the correlation between the task-specific and total glucan scores for the 30-day time window for those with a non-zero total glucan score, stratified by questionnaire version, using the Spearman correlation coefficient. These analyses were restricted to the first interview for those participants who completed more than one interview.

Algorithm assessment

To determine how well the calculated total glucan exposure score predicted actual glucan concentrations, we used the time-weighted average (TWA) full-shift concentration from the 32 farmers in the bioaerosol sampling study and the ‘past 1 day’ glucan scores calculated from their interview responses. There were 56 paired TWA concentration/questionnaire responses, including 24 for farmers who were visited twice. We used mixed-effects models with the total log-transformed glucan TWA concentration as the dependent variable and subject as a random effect. We evaluated three models. The task model evaluated the association between measured exposure and whether the farmer conducted the five glucan-specific tasks (each included in the same model as yes/no). The task score model evaluated the association with the task-specific glucan scores for the five tasks in the same model. The total score model evaluated the association with the total glucan score. We were unable to evaluate the poultry confinement task in these models due to only one participant reporting that task. All analyses were performed in STATA, Version 17.

Results

The prevalence of the five tasks in the glucan algorithm and their reported durations are shown in Table 2 for those who completed Version 2014. Overall, 54% of the participants reported at least one of these activities in the past 30 days. The most prevalent task was stored seed or grain (46% in past 30 days). Poultry confinement and swine confinement had the highest median glucan scores; however, these tasks were reported by only 1% and 8% of the participants who completed Version 2014, respectively.

Using the Version 2014 participant responses, we calculated and used the following group averages to estimate glucan task scores for Version 2010 participants. For clean grain bins and grind feed, we used the median duration per ‘time’ conducting that activity of 0.75 h from Version 2014, which was then multiplied by the subject-specific ‘number of times’ reported in Version 2010 to obtain the duration for those participants. For poultry confinement, we assigned Version 2014 group medians of 1200 (<100 poultry) and 6000 ng/m³-h (≥100 poultry). For swine confinement, we assigned Version 2014 group medians of 2700 (<500 hogs), 4200 (500 to <2500 hogs), 9000 (2500 to <7500 hogs), and 18000 (>7500 hogs) ng/m³-h. For stored seed or grain, we assigned Version 2014 group medians of 150 ng/m³-h for the participants who both raised cattle and farmed corn, soy, wheat, or similar crop types; 110 ng/m³-h for those who raised hogs but no cattle and farmed the aforementioned crops; and 30 ng/m³-hfor those without animals but who farmed the aforementioned crops.

Table 3 presents the number of participants who reported performing each of the five glucan-related tasks, the median time spent performing the task, and the median glucan exposure score for the task for the last 30 days. Overall, 67% of participants were assigned a non-zero total glucan score. Poultry confinement and swine confinement had the highest median glucan scores; however, these tasks were reported by only 2% and 8% of the participants. Less variability was observed in Version 2010 because of the assignment of group averages when subject-specific frequency was unavailable. For example, no variability was observed for Version 2010 clean grain bins because those identified as exposed were assigned the same score. Figure 1 shows the median and IQR glucan exposure scores for each glucan-exposed activity and total glucan exposure score in the last 30 days in the combined versions.

Table 4 presents the within-task correlations comparing glucan exposure scores during different time windows. The correlations varied widely between time windows and across the tasks. The highest correlation was seen for the task spend time in poultry confinement, with the past 30 day and 7 day correlation of 0.87; the past 30 day and 1 day correlation of 0.94; and the past 7 day and 1 day correlation of 0.91. The lowest within-task correlation was seen for clean grain bins, with the past 30 day and 7 day correlation of 0.12; the past 30 day and 1 day correlation of 0.03; and the past 7 day and 1 day correlation of 0.49.

Table 5 shows the between-task correlations between exposure scores and tasks in the past 30 days. Total glucan score was moderately correlated with four of the five tasks; however, correlations between tasks were low. When we restricted analyses to respondents of Version 2014 (data not shown), correlation between tasks remained low and the correlations with total glucan score were lower for stored seed or grain (rho = 0.43), clean grain bins (0.37), and grind feed (0.29).

Statistical modeling of total glucan exposure

Table 6 presents the results comparing the measured TWAs to the estimated task and total glucan metrics. In the task model, clean grain bins and swine confinement were associated with higher glucan TWAs; the latter was at borderline significance. Similarly, in the task score model, these same two tasks were significantly associated with higher glucan TWAs; however, the magnitude of the parameter estimate was much higher for clean grain bins than for swine confinement. The total glucan score was also significantly associated with higher glucan TWAs. In all models, there was very little between-subject variability but very large within-subject variability. In a sensitivity analysis, the task score model was expanded to include all tasks (as yes/no), rather than just the four reported in Table 6, to see if we may have missed an important task; none of these additional tasks had a model parameter that differed meaningfully from the intercept (data not shown).

Discussion

Glucan exposure often occurs in farming environments but has been infrequently characterized. This study adds to the limited work on glucan exposure in farmers by identifying tasks associated with higher glucan concentrations and characterizing variability in the prevalence and duration of those tasks in a US farming population. By incorporating questionnaire responses on farming activities, we were able to capture important subject-specific variability in exposures.

The highest exposures occurred for those working in poultry and swine confinements, but these activities were conducted by less than 10% of our population. These are recognized as high-hazard tasks and are often performed by hired hands and contractors or in large confinement operations, and were thus less prevalent in our population of predominantly farm owners. High exposure levels were also observed for cleaning grain bins, reported by 18% of our population. The most prevalent of these five tasks was working with or around the stored seed or grain (64%), but this only modestly contributed to the total glucan exposure because of its typically short durations, and is thus unlikely to provide a meaningful glucan signal above agricultural activities not included in the algorithm.

We had limited glucan exposure data on which to base our task-specific GMs. All air sampling data used came from our prior BEEA air monitoring study (Sauvé et al, 2020). In general, our exposure levels are lower than those reported in the few other studies that have included glucan exposure assessment. Singh et al. (2011) focused on worst-case exposure scenarios for grain bin work (150 000 ng/m³), hog handling (311 ng/m³), and horse handling (37 ng/m³). Lawniczek-Walczyk et al. (2013) examined glucan concentrations in poultry handling and cleaning barns (6900–8400 ng/m³), but these were area samples and a smaller sample size fraction (PM₁₀) than the measurements in this paper. These studies align with our findings that spending time in animal confinement are contributors to glucan exposure. Glucan measurements collected using the inhibition enzyme immunoassay (EIA) are not discussed here for lack of comparability to the LAL method (Brooks et al., 2013; Madsen et al., 2014).

Working with moldy hay was not identified as a high exposure task in this study. The few samples assigned to working with moldy hay were associated with feeding silage to cattle, were generally outdoors, and had low concentrations. When wet, moldy hay yields fewer airborne mold spores. However, we also had limited field measurements exclusive to working with moldy hay in this study. Some measurements that may have involved working with moldy hay or straw were attributed to time spent cleaning swine and poultry confinement areas, if used as bedding in smaller operations.

Our evaluation of the correlations within the same task across different time windows showed that the ‘past 1 day’ and ‘past 7 days’ were moderately to highly correlated with the ‘past 30 days’ for animal-related tasks that likely occur on a near-daily basis. In contrast, the recent time windows were only poor to moderately correlated with the ‘past 30 day’ time window for infrequent tasks like clean grain bins. This suggests that it is important to query participants about different time windows if exposures may be related to acute effects. We observed very low correlations between the five tasks, indicating that each of the five tasks was likely unique contributors to the total glucan exposure level experienced in our population.

We performed a limited assessment of the algorithm scores for the ‘past 1 day’ glucan scores. We observed significant associations with the prior day’s glucan time-weighted average exposure for swine confinement and clean grain bins, as well as total score. Although the units of each glucan task score are designed to be on the same scale, we identified some differences in the magnitude of the task-specific slopes. It is unclear whether this is a real difference. It may be related to the low prevalence of the activities or due to measurement error in the reported activity duration and measured TWA concentration (which represented an incomplete work day). Unfortunately, we were unable to assess poultry confinement due to low prevalence in these data. Indirect validation by examining the associations with the task-specific scores and the total scores in future epidemiologic studies will provide additional insight.

The strengths of our study include the use of study-specific measurements, subject-specific task durations, and the estimation of task-specific exposure levels that can be used individually or summed together in future analyses. We were also able to perform a limited assessment of the algorithm. Our study has several limitations. First, the glucan exposure data available were limited to a small study in Iowa and may not represent the experience of the North Carolina farmers in the BEEA study. For example, tobacco, cotton, and sweet potatoes are raised in North Carolina but not in Iowa. Second, the Glucatell kit used here may overestimate glucan concentrations due to sensitivity to plant glucans (Cherid et al., 2011); thus, not all of the measured glucans reported here necessarily reflect exposure to molds. However, our development of separate source-specific scores will allow for the individual sources, as well as their aggregate, to be examined in future epidemiologic analyses. Third, the limited data did not allow us to account for potentially important determinants. For instance, our estimates using the models in Sauvé et al., 2020 did not allow us to derive separate intensity estimates for swine and poultry. Similarly, Sauvé et al., 2020 found that glucan levels may be higher for harvesting soybeans versus other grains; however, the BEEA questionnaire did not distinguish between time spent harvesting soybean versus other grains or corn, and thus, we could not take this potential determinant into account. Finally, we focused on the higher-exposed tasks, which excluded most crop-related activities other than grain bin work. Higher duration activities with low glucan concentrations, such as harvesting or hauling grain, have the potential to contribute to the total glucan exposure; however, these tasks were not associated with the glucan TWAs (data not shown). However, the transparency of the algorithm allows for future modification when new data on glucan concentrations during farming activities become available. In subsequent epidemiologic analyses, we will be able to distinguish between BEEA participants who were farming but had a glucan total score of ‘0’ from those participants who were not farming in the past 12 months.

Conclusions

Our assessment of glucan exposure through combining glucan exposure measurements with task frequency of activities associated with glucan exposure provides insight into where variability in glucan exposure may occur within- and between-farmers. Although the developed algorithm is specific to the BEEA population, our approach provides a framework that can be applied to other studies relying on questionnaire data to characterize glucan exposure. In addition, it provides a framework that can be adapted upon the availability of additional studies and data. Algorithms such as this one identify broad contrasts in exposure that will provide insight into associations between exposure and disease in future epidemiologic analyses.

Acknowledgements

Amy Miller, Kate Torres, Sarah Woodruff, Marsha Dunn, and other staff at Westat, Inc. (Rockville, MD) contributed to the study coordination and data management. We thank the field research team in Iowa, including Charles Lynch, Debra Lande, Debra Podaril, and Jennifer Hamilton. We thank Nervana Metwali for preparing the air sampling equipment and conducting the glucan analyses and Emma Stapleton for assisting with the equipment preparation and conducting the field air monitoring. We also thank Anne Taylor at Information Management Services, Inc. (Calverton, MD) for her programming support. Finally, we gratefully acknowledge the participation of the Biomarkers of Exposure and Effect in Agriculture study participants that made this work possible.

Funding

This work was supported by the Intramural Research Program of the National Institutes of Health, National Cancer Institute, Division of Cancer Epidemiology and Genetics (Z01 CP 010119).

Data availability

The data underlying this article will be shared on reasonable request to the corresponding author.

Disclaimer

The authors declare no conflict of interest relating to the material presented in this article. Its contents, including any opinions and/or conclusions expressed, are solely those of the authors.

References