Factors associated with work accidents in a rural area in Minas Gerais, Brazil Fatores associados a acidentes de trabalho em zona rural de Minas Gerais

| Background: Noise and pesticide exposure is frequent among Brazilian agricultural workers, however, few studies analyzed its association with work accidents. Objective: To investigate the association between self-reported noise and pesticide exposure and self-reported work accidents among rural workers. Methods: Cross-sectional study performed in Conceição das Pedras, southern Minas Gerais, Brazil. We administered a questionnaire to 170 rural workers from April through October 2014. We subjected the data to unconditional multiple logistic regression analysis by means of the stepwise method. Results: The prevalence of work accidents was 11.8% (95%CI 7.3–17.6). Independent variables associated with occurrence of work accidents were: use of the fungicide Priori® (OR 11.8; p=0.007), use of hand-held power tools (OR 8.28; p=0.010), noise-induced hearing loss (OR 38.60; p=0.022) and length in the job (OR 1.04; p=0.034). Conclusion: We did not find association between work accidents and self-reported combined noise and pesticide exposure. The results, however, evidenced factors significantly related to the occurrence of work accidents, which identification might serve to ground preventive actions. We recommend performing additional studies likely to overcome detected limitations to establish the role of combined exposures in work accidents.

Factors associated with work accidents in a rural area in Minas Gerais, Brazil

INTRODUCTION
Work accidents are relevant in rural environments as a result of major changes which transformed the lives of workers 1 .While in the past accidents related to rural work mainly consisted of falls, injuries caused by manual tools, and poisoning by poisonous plants and animals, intensive use of pesticides and machinery significantly increased the hazards to which rural workers are exposed 2 .Therefore, risk factors for work accidents currently include: falls from heights, manual handling of heavy loads, driving tractors, noise, vibration, pesticides, cuts and wounds 3 .
Using pesticides in agriculture has considerable socioeconomic impacts as a function of their effects on the various social actors involved.Rural workers are the most affected in this regard, since pesticides are associated with several health outcomes, including hearing problems.For instance, several studies found that hearing loss is an early sign of poisoning [4][5][6] .In addition, using pesticides has also relationship with poorer quality of life 7 .
Agricultural machinery, including tractors, choppers, chainsaws and silage machines, among others, are potentially noisy.Available evidence indicates that workers exposed to an equivalent continuous sound level equal to or over 80 dB(A) along an 8-hour period (LAeq,8h) are at higher risk of accidents, because this condition might cause temporary hearing loss 8 , impaired communication, attention and concentration, stress and fatigue.All these factors are determinants of work accidents, therefore, there seems to be causal relationship between occupational noise and work accidents 9 .
Few studies addressed the relationship of noise exposure associated with machinery and tools used by workers with occurrence of accidents 10 .Therefore, the present study is highly relevant, given the need to elucidate better the relationship between health and work and thus contribute to the prevention, and consequent reduction of the number and severity of work accidents.
Combined exposures have been considered in several research agendas to establish possible interactions or synergism between hazards.Synergic and additive effects have been investigated through analysis of exposure to occupational noise and other stressors 11 .Some studies found that workers exposed to occupational noise and pesticides were at higher risk for hearing loss compared to workers exposed to noise alone 12,13 .
Given the aforementioned considerations, in the present study we sought to elucidate whether combined exposure to noise and pesticides has significant influence on the occurrence of accidents.Our main objective was to analyze the association between self-reported exposure to noise and pesticides and self-reported accidents among rural workers.

METHODS
The present cross-sectional study was performed from August through November 2014.The study population comprised 978 agricultural workers in Conceição da Pedras, southern Minas Gerais, Brazil, as per data obtained from the Brazilian Institute of Geography and Statistics 14 .
Participants were randomly selected based on the municipal Family Health Program (Programa Saúde da Família-PSF) records.This program is charged of all health actions involving the entire local population, therefore, all 978 workers were enrolled.As is the case of all PSF units, the covered area is divided into seven micro-areas, each assigned to a health community agent; the covered families are attributed an ID number.Based on these ID numbers, we randomly selected, using software Excel, a sample proportional to the number of households in each micro-area.Interviews were conducted exclusively with rural workers.For households with more than one rural worker, we interviewed those with longest time in the job.
We only included active rural workers 18 years old or older with at least 5 years in the job.Workers on sick leave, retired or off work due to work accidents having occurred more than 12 months earlier were excluded.
Based on a previous cross-sectional study 14 , which found a prevalence of 11% of work accidents among rural workers in Pelotas, Rio Grande do Sul, Brazil, margin of error of 5% and 95% confidence interval, the calculated sample size was 130 participants.To this we further added 10% to compensate for eventual refusals or losses, and 20% to control for confounding variables, to a total of 172 participants.This calculation was made with software Excel.Despite differences in the characteristics of rural Minas Gerais and Rio Grande do Sul with probable impact on the physical activity of workers, and thus also on the profile of work accidents, we chose to use Fehlberg et al. 's 15 study to calculate the sample size.
Data collection was performed through a questionnaire, which included: sociodemographic data -age, years of formal schooling, years in the job, weekly working hours, place of residence (rural or urban); employment relationship (landowner or self-employed); occupational exposures during daily activities -tractors, agricultural machinery, hand-held power tools, handling animals, use of agricultural trailers and pesticides (recorded per commercial name; exposure was categorized as: very frequent, usual, sporadic or rare; participants were considered exposed when they selected any of the two former options); behavioral habits -smoking (yes, no, formerly) and use of alcohol (categorized as was also done for daily activities).
As victims of work accidents we considered all the participants who responded affirmatively to the following question: "Have you had a work accident in the past 12 months?" Investigation of noise exposure and some relevant hearing complaints was intended to establish the usual intensity of noise in the workplace; answers options were: "weak or none," "moderate," "strong," and "very strong." We also inquired the participants about the intensity of noise on the day they suffered an accident or the last day at work; answers options were: "weak or none," "moderate," "strong," and "very strong." In addition, we investigated whether the participants had or not some hearing problem, whether noise in the workplace was or not uncomfortable, and whether they had already been or not diagnosed with noise-induced hearing loss (NIHL).
We did not include items on personal protective equipment (PPE) in the questionnaire, given that the literature 16,17 has already shown that it is seldom, inadequately or partially used by rural workers.In addition, the protection provided by PPE against noise and pesticides is rather limited 18,19 .
The questionnaire was administered by one of the investigators.The data were analyzed using software EpiInfo 5.3.1 20 .A pilot study was performed with 21 rural workers not included in the study sample to test and adjust the questionnaire to avoid redundancies and improve its comprehensibility.

DATA ANALYSIS
We first performed descriptive analysis, then investigated factors possibly associated with the analyzed outcome.We had resource to unconditional multiple logistic regression analysis to investigate the association of the dependent (work accident) with the independent variables, which was tested as per the following hypothesis: General model hypothesis test (likelihood ratio test): H0: ß1 = ß2 = ... ßk = 0 Ha: at least one ß≠0 Univariate analysis was performed to select the independent variables to be included in the multiple regression model.The significance level for variable selection was 20% on the Wald test: Wald test (applied to each ß): H0: ßj = 0 RC(Xj)=1 Ha: ßj≠0 RC(Xj)≠1 To identify the model with the best goodness of fit we had resource to the forward stepwise method, i.e. we included the variables in decreasing order of significance and excluded the non-significant ones, with concomitant analysis of the variation in odds ratio (OR), 95% confidence interval (CI) and significance.For variables to remain in the final model the significance level was set to 5% on maximum likelihood estimation 21 .The final model structure is expressed by Equation 1.The odds of occurrence of work accidents as per the fitted model were calculated using Equation 2. The ß coefficients calculated for the independent variables (X i ) represent the incremental change of a dependent variable function per unit of change in the independent variables.

RESULTS
Only two of the 172 questionnaires were not returned, corresponding to 1.2% of losses/refusals.

Factors associated with work accidents in a rural area in Minas Gerais, Brazil
The final sample comprised 170 participants, most of whom were male.
As Table 1 shows, 29.4% of the participants were aged 32 to 43 years old, and their mean number of years of formal schooling was 5.81.About 70% of the participants had 20 or more years of work in rural jobs, and 78.3% worked 40 to 50 hours/week.Twenty participants had already been involved in work accidents in the past year, corresponding to a prevalence of 11.8% (95%CI 7.3-17.6).
About 61.8% of the sample reported to be exposed to noise caused by machinery (tractors, choppers, chainsaws, silage machines).
The pesticides reported by the participants are described in Chart 1, together with their corresponding chemical class and probable date of introduction in Brazil.
According to Table 2 self-reported exposure to some pesticides was significantly associated with occurrence of work accidents on univariate analysis.However, as Table 3 shows, Priori ® was the single pesticide that remained in the multiple regression model.The odds for the participants who used Priori™ to be involved in a work accident were 1.18 times higher compared to all the others.Combined use of six or more of the 12 listed pesticides did not exhibit significant association.
Table 2 describes the number of participants rated exposed, i.e., those who answered affirmatively the items corresponding to the investigated independent variables, and the corresponding prevalence rates.This table further includes the results of univariate analysis expressed as OR and p value.Table 3 shows that independent variables with significant association with the analyzed outcome were "use of handheld power tools," "NIHL," "use of Priori ® " and "years in the job." None of the other sociodemographic, labor, occupational and behavioral variables or use of pesticides exhibited statistically significant association on multiple analysis.
We calculated the odds for occurrence of work accidents under several circumstances using Equation 2; the results are depicted in Figure 1.

DISCUSSION
In the present study we did not find significant association between work accidents and self-reported combined exposure to noise and pesticides.However, we could establish that the participants had long working hours, were exposed to noise from agricultural machinery and tools, and used pesticides.
About 55% of the participants had attended 3 to 6 years of formal schooling, which agrees with the results of a the study performed by Marques and Silva 23 , in which 58.4% of rural workers reported to have attended 4 years of schooling.About 78.2% of the participants worked 40 to 49 hours/ week.In the study by Fehlberg at al. 24 , 65.2% of rural workers worked more than 48 hours/week.Variable "years in the job" was associated with occurrence of work accidents.This finding agrees with the results of Sprince et al. 's study 25 , which evidenced significant association between work accidents and shorter experience in agricultural jobs.Workers with longer experience probably feel more confident and thus become less careful, which might account for the fact that variable "years in the job" was associated with occurrence of work accidents.
Pesticide Priori ® is a strobilurin; this class of relatively new fungicides, introduced in 2000, is being increasingly used, but data are still scarce 26 .Despite the small number of studies on the effects of strobilurins on human health, some authors reported lower counts of white blood cells as one of the main findings 27 .Only 4.1% of the participants reported to use Priori ® , which nonetheless might be associated with higher odds of work accidents as a function of its wide use in tomato plantations.This crop demands intensive farming, involving frequent use of pesticides, and thus also continuous and excessive exposure, eventually leading to acute and chronic poisoning and making workers increasingly susceptible to accidents.Combined use of six or more of the 12 pesticides listed in the questionnaire was not significantly associated with the analyzed outcome.The relationship between use of pesticides and work accidents deserves more thorough investigation, with particular emphasis on the duration of use.
The odds for occurrence of work accidents were 38.6 times higher for the participants who reported to have been previously diagnosed with NIHL compared to those without this condition.This finding corroborates the results of the study by Sprince et al. 25 , who found that the odds of accidents among rural workers who wore hearing aids -which presupposes previous hearing loss -were 4.37 times higher compared to non-wearers, i.e., allegedly with normal hearing.
Use of hand-held power tools was associated with 8.28 times higher odds of accidents compared to non-use.This finding agrees with those in Sprince et al.'s study 25 , which evidenced significant association between machinery-related injuries and weekly hours spent in farm work.The association between potentially noisy hand-held power tools and work accidents might also be attributed to factors other than noise.For instance, choppers include blades, with consequent risk of cuts, in addition to propelling things while in operation.
According to the odds estimates depicted in Figure 1, the chances of work accidents for workers with 30 years in the job, previous NIHL and using hand-held power tools and pesticide Priori ® are 99%.In turn, the chances are just 1% for workers with 10 years in the job, without NIHL, not using hand-held power tools and not exposed to Priori ® .The plotted curves do not overlap, but represent different ranges of odds.
Curves A, F and G correspond to the highest odds, for which previous diagnosis of NIHL was determinant.This finding corroborates the results of the study performed by Girard et al. 28 , who found that the risk of accidents increased proportionally to the degree of hearing loss, and was significantly higher in noisy environments.
None of the variables related to noise exposure was significantly associated with the analyzed outcome.This result disagrees from those reported by Dias et al., 19 according to which self-reported moderate or strong environmental noise behaved as risk factor for work accidents.Similarly, our findings also disagree from Cordeiro et al. 29 , who reported association between work accidents and exposure (always/ Factors associated with work accidents in a rural area in Minas Gerais, Brazil  sometimes) to strong noise at work.Lack of association in our study might have been due to the fact that the information on these variables was self-reported, thus liable to information bias.
In the present study, we considered work accidents in the past 12 months as a way to control for memory bias, since the participants could have forgotten some minor accident.In turn, information bias cannot be ruled out, because the data had not external valdation, e.g. in medical or hospital records.However, self-reported and morbidity information is used in many countries to investigate the state of health of different populations 30 .
Information bias might occur in epidemiological studies, especially when the explanatory variables originate in participants' self-reports.In this regard, we cannot rule out the possibility of some participants having been unaware of presbycusis, among other disorders.The fact we did not subject the participants to quantitative assessments for diagnosis of hearing loss or noise exposure is a limitation of the present study.An additional limitation derives from the wide confidence interval found for the results relative to explanatory variables "hand-held power tools," "NIHL," "use of Priori ® " and "years in the job" in the multiple model, even though all of them exhibited statistically significant association with the analyzed outcome.Confidence intervals depend on the sample size, and the narrower they are, the higher the likelihood that sample-based estimates represent the true effect size for the analyzed population.
One further source of bias to be considered is the healthy worker, or survivor, effect, which might have occurred in the present study because we included only active workers, with exclusion of those who had retired or changed their occupation.

CONCLUSION
The aim of the present study was to investigate the association between combined self-reported noise and pesticide exposure and work accidents.Such relationship was not found, however, the results pointed out some relevant factors which contributed to the occurrence of work accidents among the analyzed population.For future studies on this subject we suggest exploring how pesticides are used through a more thorough and  Factors associated with work accidents in a rural area in Minas Gerais, Brazil consistent evaluation, with emphasis on the amounts and length of use.The same applies to noise, i.e., historical analyses of exposure.Hearing loss, use of the fungicide Priori ® , years in the job and use of hand-held power tools were associated with occurrence of work accidents.
The results of the present study might serve to ground health promotion and preventive strategies targeting the health of rural workers, centered on the risk factors evidenced here and the considerable use of pesticides in Brazil, in addition to noise exposure as part of the work process, inasmuch as it might lead to hearing loss.
The health of rural workers, i.e., the focus of the present study, is a complex matter, given the nature of the hazards to which they are exposed.The goal of overcoming the analyzed limitations should orient future studies on combined exposures to contribute to the elucidation of the relationship between health and work, and thus also to the prevention and consequent reduction of the number and severity of work accidents.

Figure 1 .
Figure 1.Estimated odds for occurrence of work accidents relative to 170 rural workers according to a multiple model that included variables diagnosis of noise-induced hearing loss, use of power tools, use of pesticide Priori ® and years in the job, Conceição das Pedras, Minas Gerais, Brazil, 2014 (n=170).

Table 1 .
Distribution of variables age, years of formal schooling, years in the job and working hours for the analyzed sample, Conceição das Pedras, Minas Gerais, Brazil, 2014 (n=170).

Table 2 .
Univariate analysis and distribution of absolute and relative frequencies (prevalence) of independent variables associated with work accidents among rural workers, Conceição das Pedras, Minas Gerais, Brazil, 2014 (n=170).