Determinants of respirable quartz exposure in farming

ABSTRACT

The objectives of this article are to quantify personal respirable quartz exposure on sandy, sandy loam, and clay soil farms and to identify exposure determinants. The methods applied included observing and examining the variables soil type, commodity farmed, activity, process, quartz % in respirable dust, and weather variables. Multiple linear regression was used to identify determinants of respirable quartz concentration and logistic regression was applied to identify determinants of respirable quartz concentration > 50 µg.m−3 (a commonly used reference value of over-exposure). The highest quartz concentration was 626 µg.m−3 and 30%, 22%, and 31% of measurements were > 50 µg.m−3 for sandy, sandy loam, and clay soil farms, respectively. In general, the commodities livestock farming and cereal grains as well as the activity cereal planter operator, decreased humidity on the day of measurement, the mechanical processes, and quartz % in respirable dust (in a confounding way) were associated with higher respirable quartz concentrations (p ≤ 0.10) as well as season (p = 0.14). Variables associated with quartz levels above 50 µg.m−3 were cereal planter operator, increased quartz % in respirable dust, decreased humidity on day of measurement, and increased respirable dust concentration. Cereal planter operator (Multivariate Odds Ratio (OR) 4.56, 95% CI: 1.79–8.89) and levels of quartz % > 10 µg.m−3 (Multivariate OR 6.01, 95% CI: 3.52–9.71 if quartz % > 10 but <= 20 µg.m−3, and Multivariate OR 5.32, 95% CI: 2.56–8.34 if quartz % > 20 µg.m−3) were clear determinants of quartz over-exposure. It can therefore be concluded that over-exposure to quartz in farming is possible. Joint influences of more farming characteristics and weather variables should be included, together with soil type in future farming exposure assessments.

METHODS

Sampling strategy

As was reported by Swanepoel et al., 298 personal respirable dust and quartz measurements were collected on a sandy, a sandy loam, and a clay soil farm located in the Free State and North-West provinces of South Africa. Access to the farms was granted as the researcher knew the farmers. All farm workers present on the days of measurement agreed to participate. There were no repeated measurements of the same worker doing the same job. Farming methods and products farmed on the three farms were typical of the regions in which the farms were located, and have been described previously by Swanepoel et al. The major tasks undertaken on the three farms over the annual farming cycle were identified in conjunction with the farmer. Crops and livestock produced on the sandy and sandy loam farms included maize, wheat, sunflower, water melon, pumpkin, potatoes, cattle, and sheep. The only product farmed on the clay soil farm was maize; therefore, on this farm respirable dust and quartz exposure were only measured during maize planting and maize harvesting operations. Even though farm activities can produce large dust clouds, no respiratory protective equipment (RPE) was observed to be used. The farms under study were small to medium sized farms and all tractor drivers measured during the study used “open-cabbed” tractors. As expected during farming, shift durations varied substantially depending on activity and often exceeded 12 hr, but personal breathing zone measurements were collected over a period of approximately 8 hr (mean = 460 min, range = 360–520 min). The same task was done over the longer shifts so the 8-hr assessment was likely to be representative of exposure during the shift on that day.
Walkthrough inspection

A Southern African Institute of Occupational Hygiene (SAIOH) registered occupational hygienist recorded detailed information on tasks performed, duration of tasks, commodities grown, and the process, i.e., manual vs. mechanical. Farming methods and products farmed on the three farms were typical of the regions in which the farms were located, and have been described previously.

Respirable dust and quartz measurement

The respirable dust and quartz measurement methodology has been reported in detail previously, but briefly, to measure respirable dust, Higgins-Dewell cyclones with 25 mm polyvinyl chloride filters were used according to the standard Health and Safety Executive (HSE) Methods for the Determination of Hazardous Substances (MDHS) 14/3 method, and quartz concentration was determined by X-ray diffraction according to the HSE MDHS 101 method. The limit of detection for quartz reported by the analytical laboratory was 22 µg of quartz. Respirable quartz concentrations are expressed as an 8-hr time-weighted average (TWA) in micrograms per cubic meter (µg.m−3).

Determinants of exposure

Possible determinants of interest were soil type, season, commodity, activity, farming process, quartz percentage in the respirable dust, respirable dust concentrations, and five measures of weather conditions (Tables 1 and 2). Soil grain size analysis showed that three soil types were correctly classified as sandy, sandy loam, and clay. Only two seasons were involved, namely winter (dry) and summer (rainy). The six farming commodities were pooled as follows for the purpose of data analysis: cereal grains (maize, wheat, sunflower), ground crops (watermelon, pumpkin and potato), and animal livestock (cattle and sheep). Fourteen farming activities were measured and pooled as follows: cereal planter operators (maize and wheat planter operators), cereal tractor operators (maize and wheat tractor drivers, maize harvester operators, and maize harvesting transport tractor operators), cereal soil preparation (maize and sunflower planting soil preparation, maize harvesting soil preparation), and other (water melon and pumpkin harvesting and picking, maize silo workers, and maize sorting and packaging, and cattle and sheep handling). Two farming processes were identified, i.e., manual and mechanical. Five weather variables were obtained from the South African Weather Service. Instruments were situated within 3 km of the farms under study and weather variables included humidity, air pressure, wind speed, temperature, and rainfall. Weather data were obtained for 8 consecutive days prior to quartz measurements as well as on the day of airborne quartz measurement.

Quality assurance

A quality assurance plan was in place; a qualified occupational hygienist registered with the Southern African Institute for Occupational Hygiene (SAIOH) monitored all aspects of data collection, from possible determinant identification through to all gravimetric and analytical analysis procedures. All of the gravimetric and analytical laboratories were experienced South African National Accreditation System (SANAS) accredited laboratories. The results were expressed in µg.m−3 and the quartz limit of detection reported by the analytical laboratory was 22 µg. Two field blanks were taken on each field-sampling day and were included in the analyses to assess any contamination.

Weather data were collected with calibrated instruments as prescribed by the South African National Standards (SANS) organization and results were compiled by a qualified climatologist working for the South African Weather Service. Soil mineralogy was done by an experienced scientist in the School of Geochemistry, University of the Witwatersrand.

Data analyses

Determinants of exposure

Quartz concentrations below the limit of detection (LOD) were estimated using a multiple imputation method. A full conditional-model method was used to impute data, which imply that it is assumed that each farm has its own distribution. Means as well as variances differ for the three farms. Various imputation methods exist in the literature. All data and regression analyses were done using the TIBCO Spotfire S-plus (version 8.1) software package.

Determinants of exposure included both categorical and continuous variables. The following variables were measured on a categorical scale: soil type (sandy soil, sandy loam soil, and clay soil), season (summer and winter), commodity (cereal grains, ground crops, and livestock), activity (cereal planter operator, cereal tractor operator, cereal soil preparation, and other), and process (manual and mechanical). The following variables were measured on a continuous scale: personal respirable dust exposure, percentage quartz in respirable dust, and weather data (humidity, air pressure, temperature, wind strength, rain fall), measured in various time windows.

All analyses for Tables 1-3 were done, using 10 sets of imputed data, due to unobservable observations that originated from the inability of instruments to observe measurements below the detection limit of 22 µg in the original dataset. The methods suggested by Rubin et al. for the analysis of imputed data were used throughout. However, analyses involving exposures > 50 µg.m−3 (Table 4) were performed on the original data, not on imputed data.

The following information is important to clarify the statistical methods used in the analysis of the imputed data.

- Kruskal-Wallis tests were performed to determine significant differences in quartz exposures between more than two categories of the variables. Where differences were found, subsequent Wilcoxon tests were done to determine the direction of relationships that exist between the subgroups of the variables. The mean value for each weather variable was determined for three time windows: the average over eight days before measurement, the day before measurement, and on the day of measurement. These variables were evaluated against respirable quartz concentrations. Pearson's correlation coefficient was calculated to determine the linear correlation between respirable quartz exposure and all the continuous variables.

- Multiple linear regression was used to identify possible determinants of respirable quartz concentrations and to explain the variation in these concentrations. Respirable quartz concentration data was best described by a log normal distribution and all potential determinants were tested against the logarithm of the quartz concentrations, due to the skewness of the distribution. To acquire initial insight into possible relationship-structures in the data set, the covariance matrix and pair-wise scatter plots of all the variables of interest were considered, in order to identify possible covariates.

- Multi-collinearity was avoided by carefully choosing dummy variables to represent the categorical variables and by omitting correlated weather variables, while minimizing loss regarding fitting criteria. Inclusion criteria for obtaining a suitable model involved three steps: (i) one-by-one stepwise sequential addition of continuous variables to the model using sum of square patterns, Cp – values, R2, and adjusted R2 values to identify variables for inclusion; (ii) sequential addition of dummy variables representing the categorical variables (allocated to prevent multi-collinearity) using the same processes and criteria explained above; and (iii) identifying interaction terms via stepwise procedures and including it in hierarchical models. The resulting regression model was tested for violation of the model assumptions, by using standard diagnostic steps, keeping in mind in a realistic way, that the data set was on the small side for the number of parameters to be estimated.

- Logistic regression analyses were performed on the original, censored and uncensored data, to identify determinants of over-exposure to respirable quartz. Over-exposure is defined as respirable quartz > 50 µg.m−3 in this case. The value > 50 µg.m−3 was selected because it is a widely used reference concentration. This study provided additional insight into identifying possible determinants of quartz exposure above this commonly used reference concentration. Furthermore, besides identifying determinants of over-exposure in this section, it should be mentioned again that the logistic regression analysis of this section is independent of the uncertainty of multiple imputations since all values below the detection limit of 22 µg were also <= 50 µg.m−3. The logistic regression analysis was therefore applied to the original multivariate data set, and the response variable was defined as respirable quartz values greater than 50 µg.m−3.

- Potential determinants were carefully chosen to prevent multi-collinearity. Humidity on the day of measurement was the only weather variable that satisfied the inclusion criteria, although wind-direction and wind-strength might have proved to be meaningful determinants if larger samples were available. The categorical regressor variables soil type (sandy, sandy loam and clay soil), season (winter, summer), commodity (cereal grains, ground crops, livestock), activity (cereal planter operator, cereal tractor operator, cereal soil preparation and others), and process (manual or mechanical) were considered for univariate and multivariate logistic modeling. For the logistic regression analysis continuous variables (quartz %, humidity on the day of measurement, and respirable dust exposure) were transformed to discrete classes. Each variable was divided into three subgroups. Quartz % was divided into groups: 0–10%, > 10–20%, and > 20%; similarly, humidity on the day of measurement was factored into three groups: <= 25 g.m−3, > 25 – <= 45 g.m−3, and > 45 g.m−3, while respirable dust exposure was grouped as follows: <= 0.2 mg.m−3, > 0.2 – <= 0.5 mg.m−3 and > 0.5 mg.m−3. Odds ratios (ORs) and 95% confidence intervals (CIs) for the ORs were calculated to compare the association between group differences in quartz exposures while adjusting for the above mentioned variables. For comparative modeling analyses, one of the categories of the possible determinant, often the lowest level of the determinant, was used as the base level and conditioning was done on this baseline category, denoted by 1 in column 3 of Table 4.

Ethics

Confidentiality regarding farms, farm owners, and farm workers was maintained throughout the study and written informed consent was obtained from all study participants. This study was approved by the University of the Witwatersrand Human Research Ethics Committee (clearance number M070252).
RESULTS

Over-exposure to quartz was found on all three farms and for most tasks. 30%, 22%, and 31% of measurements were > 50 µg.m−3 and 14.5%, 9.1%, and 4.8% of measurements were > 100 µg.m−3 (another commonly used OEL) for the sandy, sandy loam, and clay soil farms, respectively. The highest concentration was 626 µg.m−3. The number of measurements below the LOD of 22 µg on the sandy, sandy loam and clay soil farms in the original set of observations, were 48 (35%), 21 (27%), and 53 (64%), respectively.

Determinants of quartz exposure on three South African farms

Associations determined and combined from ten imputed data sets, between categorical variables and respirable quartz, are shown in Table 1. Respirable quartz concentrations on the sandy, sandy loam, and clay soil farm were very similar according to the median and GM-values displayed in Table 1. Note that, due to the skewness of the data, and the influence of some very large outliers, the AM-values may be not reliable in this case.

Quartz concentrations for commodities differed (Kruskal-Wallis test, p = 0.09). The livestock category of the commodities-variable produced higher exposure than ground crops and cereal grain farming. The median of livestock was 13.2 µg.m−3 and 12.9 µg.m−3 higher than that of ground crops and cereal grain, respectively (Wilcoxon tests, p < 0.015 in both cases).

Quartz concentration varied over the categories of the variable activity (Kruskal–Wallis test; p = 0.01) with cereal planter operators' exposures exceeding those of cereal tractor operators, cereal soil preparation, and Other activity (Wilcoxon tests; all p-values were < 0.01). From the location measures in Table 1 (medians for example), quartz exposures for cereal planter operator exceeded exposures for cereal tractor operator by at least 9.6 µg.m−3. No significant differences were found between cereal tractor operators and cereal soil preparation (Wilcoxon tests; p = 0.16), nor between cereal tractor operators and Other activity (Wilcoxon tests; p = 0.93) or between soil preparation and Other activity (Wilcoxon tests; 0.29).

As expected, respirable quartz concentrations were significantly higher during mechanical than manual processes (Wilcoxon test; p = 0.04), as the GM column clearly shows. Surprisingly, quartz exposure values were significantly higher during summer (rainy season) than winter (Wilcoxon test; p < 0.01).

Table 2 shows the correlation (r), between respirable quartz concentration and weather data during three time windows. Increased humidity on the day of measurement was slightly associated with a decrease in respirable quartz concentrations (r = −0.09; p = 0.08). Three other interesting relationships not involving weather data were determined. Respirable dust concentration was correlated with respirable quartz concentration (r = 0.36; p < 0.01). Also, quartz % in respirable dust was not strongly correlated with respirable quartz (r = 0.10; p = 0.08), and quartz % in respirable dust was negatively correlated with respirable dust exposure (r = –0.39; p < 0.01). These somewhat unexpected results will be discussed briefly in the second last paragraph of the “Discussion” section below.

Table 3 shows the multiple linear regression model results. Note again that, because of the skewness of the respirable quartz concentration distribution, regression coefficients were determined by using the logarithm of respirable quartz concentration as response. This complicates the interpretation of the estimated regression parameters. Also note that full hierarchical models were fitted but in Table 3 only some interesting interaction terms are displayed.

Furthermore, the following general information is important to clarify the statistical interpretation of the estimated regression parameters in the model. Regression coefficients were determined for regressor variables measured on a continuous scale as well as for categorical regressor variables with two, three or four categories. Interpretation of the estimated parameters is therefore complicated. Regression parameter estimates (denoted by beta in Table 3) are interpreted differently for the four cases, according to the following general rules:

(a) For a continuous variable, the logarithm-response, i.e., the logarithm of respirable quartz exposure, will increase by beta units per unit change of the regressor variable (determinant), or, for the back transformed response, the response, i.e. respirable quartz exposure, will change by per unit change of the regressor variable of concern.

(b) For categorical regressor variables coded by 1 or 0, i.e., for the two-category case, the beta-value indicates how much higher for positive values of beta (or lower for negative beta-values) the log-response function is for the category coded by 1 than for the category coded by 0.

(c) For three and four categories, one category will be chosen as the baseline category. A parameter estimate will be interpreted (if the response is in logarithm form) as the expected difference between the logarithms of the response of the regression variable category of interest and the baseline category. In Table 3, the baseline category is omitted and only the remaining categories are displayed. By comparing the categories displayed in Tables 1 and 3, the baseline category of interest can easily be identified.
	
From Table 3 the following deductions can be made: with regard to the variable “soil type”, and by basing comparative differences of effects on the clay soil farm, the three soil types (i.e., sandy, sandy loam, and clay) showed similar differences on the log-transformed quartz exposure variable and can be interpreted as having similar effects in the model. The p-values: p = 0.29 and 0.12 states, respectively, that the hypotheses beta = 0 cannot be rejected, in which case = 1, as was assumed for the baseline.

Similarly, at least two of the levels (cereal grain and livestock) of the determinants “commodity” were statistically significant to the model at 10% level (p = 0.10 and p = 0.09, respectively). Similarly, at least one of the levels (cereal planter operator) of the determinants “activity” was statistically significant to the model (p = 0.08).

Process was contributory (p < 0.01) with quartz concentrations for the manual process 0.38 µg.m−3 lower than that of the mechanical process for the fitted model. This trend is also verified by the median column in Table 1. Season seems not to be significant in this model (p = 0.14). It should be kept in mind that for this model R² = 0.70, so that a less conservative approach would consider Season with a p-value of 14%, for inclusion in the model.

Regarding continuous potential determinants, respirable dust exposure and humidity on the morning of measurement were found to be significant determinants of quartz exposure (all p-values were < 0.10), as well as some of the two-fold and three-fold interactions between quartz % in respirable dust, respirable dust exposure, and process (applicable p-values were < 0.01).

To summarize, the most significant determinants of exposure were commodity, activity, process, humidity on the morning of measurement, and the interactions of quartz percentage in respirable dust with respirable dust concentration and process. This regression model explained 70% of the variability in quartz concentration (multiple R-squared value = 0.70; adjusted R–squared value = 0.68). Assumptions of normality of residuals, homogeneity of variance and linearity were satisfied or dealt with during modeling.

Logistic regression results are shown in Table 4. In univariate analyses, soil type, season, commodity, and process were not significant determinants of respirable quartz exposure > 50 µg.m−3, but the activity cereal planter operator, increased quartz % in respirable dust, decreased humidity on the day of measurement and increased respirable dust concentration were all significantly associated with quartz concentrations > 50 µg.m−3. Odds Ratios (ORs) in the multivariate analysis are adjusted for these latter four variables. In this study soil type, season, commodity and process were not determinants of whether exposure exceeded 50 µg.m−3 (all CIs spanned 1 widely). But the activity cereal planter operator remained a strong determinant of quartz exposure relative to Other activity category, which is the baseline in this case (OR 4.56, 95% CI: 1.79–8.89). A trend of high values of ORs were found with increasing percentage of quartz. ORs for increasing humidity on the day of measurement, when adjusted for the other variables, were well above one (95% CIs did not span 1) as was the case for univariate analysis. Lastly, lower levels of respirable dust concentrations were associated with lower respirable quartz exposure (ORs close to zero and CIs below 1).
