The International Labour Organization has identified occupational health and safety as a priority for the agricultural industry, noting that together with mining and construction, it is one of the three most hazardous occupations worldwide.1 Further, fatality rates for the agricultural industry have remained consistently high, while those for other industries have decreased over the past decade.1 In Australia, the work related fatality rate for agriculture is three to four times higher than the all-workforce average, and it ranks third in terms of the numbers of occupational deaths.2 Similarly, the compensable injury rate for agriculture is almost double the all-industry average, with only manufacturing and transport having higher rates.3 Farm injury predominately affects men who comprise 95% of farm work related fatalities,2 make 82% of workers’ compensation claims for the industry, and experience 66–75% of farm injuries requiring hospital admission.2 4^–6 Common mechanisms of fatal injury are being hit by moving objects, vehicle accidents, rollovers, falls from a height and electrocutions.2 Animals, farm machinery and tools, motor vehicles and motor bikes, and falls are leading causes of non-fatal farm injury.5 6 Similar patterns have been reported in the USA, Canada, UK and New Zealand.7^–10

Until relatively recently, agricultural safety research has focused on identifying rates and patterns of farm work related injury. Previous analytical studies, published prior to our study, identified a range of risk factors in univariate analyses, including personal and health characteristics (farm owner, increasing age, male gender, increased experience, increasing hours worked per week, marital status, education, cardiovascular medications, stomach remedies) and farm characteristics (presence of non-resident workers, farm size, low value of sales, number of tractors, number of cattle, number of top-loading silos).11^–17 Some of these characteristics appear to be associated with exposure and property size. A small number of independent risk factors were identified, including hours worked per week, increased experience, the presence of non-resident workers on the property, and the use of either stomach remedies or cardiovascular medications. One study restricted to machinery related injury identified male gender, being married or having ever been married, auger operation and field cropping as independent risk factors.17 Milking and trimming or treating hooves are activities associated with an increased risk of dairy cattle related injury.18 Overall, these studies were generally limited by low power, the exclusion of non-farm resident workers, and the absence of exposure data for the period immediately prior to the injury.

Our aim, therefore, was to assess risk factors for farm work related injury among Australian male farmers and farm workers. The main risk factors of interest included personal characteristics (education, experience, safety training, job status, nature of involvement in farming, previous injury, medical conditions and medication use) and farm characteristics (commodity, size, income, injury hazards, farm safety features and systems, and injury history).

METHODS

Study design

We undertook a case–control study of male farmers and farm workers in Victoria, an Australian state producing a range of agricultural commodities.19 The study was approved by the Monash University Standing Committee for Research involving Humans (Melbourne, Australia), the Victorian Institute of Forensic Medicine Ethics Committee (Melbourne, Australia) and the 17 Human Ethics Committees responsible for the recruiting hospitals (Melbourne, and Victorian regional centres, Australia).

Definitions

The study region comprised the catchment areas (defined by zip code) of the major regional Victorian hospitals. The study base consisted of adult males, age 16 years and older, working in agricultural production on study region farms. Farm work included paid or unpaid work related to the agricultural livelihood of the person, their employer, a relative or a friend. Study base members, hereafter referred to as farmers, therefore included farm owners or operators, hired farm workers and unpaid family members. Hobby farmers were included if income was derived from their agricultural activity. Agriculture was defined as crop and livestock related activities classified under sub-divisions 1 (agriculture) and 2 (services to agriculture) of the Australian and New Zealand Standard Industrial Classification,20 and excluded forestry and fishing.

Cases were members of the study base who sustained a serious farm work related unintentional injury, from 21 January 2003 to 22 February 2006. Serious injuries were operationally defined as those with an Abbreviated Injury Scale (AIS) score of 2+, in addition to those with AIS 1 that resulted in hospital admission or treatment for infection. The AIS is a threat to life scale that provides a numerical measure of damage due to injury, and ranges from 1 for minor injury, to 6 for an un-survivable injury.21 Farmers with acute poisoning scoring 2 or higher on the Poisoning Severity Score (PSS) were also included. The PSS is a numerical scale based on symptoms and accurately assesses the clinical severity and likelihood of deterioration based on initial symptoms.22 It ranges from 0 for no symptoms to 4 for a fatal poisoning.

Controls were randomly selected members of the study base who were age matched to a case within a ±5-year age range, and who worked on a farm at the time of the case injury.

Case recruitment

Fatally injured cases were recruited through the Principal Registrar at the Victorian Coroner’s Court, to whom all fatal work related incidents in the state are reported. A written invitation to participate was dispatched 6–8 weeks after the date of death to the case next of kin, who returned the consent form to Monash University if participating, and was subsequently interviewed.

Non-fatally injured cases were recruited through the emergency departments of the 14 participating regional hospitals, and five metropolitan hospitals. The metropolitan hospitals included those with major trauma centres and those with specialist services such as microsurgery and ophthalmology, to which these 14 hospitals would refer the most serious cases. The 19 study hospitals treat 74% of those with farm work related injuries occurring in the study region and admitted to hospitals in Victoria (unpublished data from the Victorian Admitted Episodes Database). Male patients at least 16 years of age who sustained an unintentional injury on a farm were approached by hospital staff seeking approval to give patient contact details to Monash University for the purposes of contacting them regarding the study, as required by the Health Records Act 2001 (Victoria). Following this approval, the remainder of the screening (confirming work related injury and determining injury severity) and recruitment process was conducted by study staff (fig 1). Patients declining to provide contact details were asked one question regarding farm experience in an attempt to assess the potential for response bias (60% responded).

• Download figure
• Open in new tab
• Download powerpoint

Figure 1

Case recruitment and non-fatally injured cases (Farm Injury Risk among Men Study, Australia, 2003–2006).

A hospital records audit conducted after the first 5 months of recruitment revealed that up to 50% of potentially eligible cases were not being approached during their treatment in the regional hospital emergency departments. Therefore, an additional recruitment process was developed and received ethics committee approval. Hospital staff generated regular (eg, monthly) lists of eligible cases using the injury surveillance system to identify patients potentially meeting the study criteria. Patients who had not been approached were sent a letter from the hospital providing the opportunity to opt-out of telephone contact by study staff. The hospital then provided contact details for those who did not opt out, and study staff completed the screening and recruitment process.23 A total of 31 cases were recruited by this process who would otherwise not have been included, resulting in a decrease in the case sample size by 12%.

Control recruitment

We employed risk set sampling to ensure controls were at risk of injury at approximately the same time as the cases. When a case was enrolled, two matching controls were recruited by telephone using contact numbers from the study areas selected randomly from the electronic community telephone listings. Proxy interviews with the next of kin were conducted for controls matched to a fatally injured case.24 Eligible controls that did not age-match a case were placed in a pool for possible matching with cases occurring in the future. If an age match was possible from the pool, controls were selected in order of decreasing time in the pool. Potential controls remained in the pool for a maximum of 1 month after initial contact. If the pool could not provide a control, then recruitment reverted to the random telephone number approach. Among the 504 controls, 116 (23%) were recruited from the pool.

Data collection and management

Demographic, farm and exposure data were collected by interview using a structured questionnaire (www.monash.edu.au/muarc/projects/resrural.html). For cases, the interview occurred as soon as possible after the injury. If possible, the farm manager was interviewed for the farm characteristics section if the case was unable to answer these questions. The questionnaire collected data on: (1) farm characteristics, (2) personal characteristics and (3) injury incident (where applicable). Risk factors were selected for study on the basis of (1) an a priori hypothesis of a role in farm injury, (2) being part of prevention programs despite limited aetiological evidence and (3) having been identified in previous studies. Farm characteristics included details on major commodities grown, farm size, income, farm work-force, injury history on the property, presence of hazards, presence of protective equipment, and safety policies and practices. Personal characteristics included details on the individual’s employment in the agricultural sector, relationship to farm owner, age, education and specific agricultural training, farm injury history, use of glasses/hearing aids, health conditions, medications, pesticide exposure, hours of sleep, alcohol consumption and excessive daytime sleepiness (score >10 Epworth Sleepiness Scale25). The injury incident details included the event itself and details of any machinery involved. The referent time period for controls was the day last worked on the farm.

Statistical analysis

Continuous variables were summarised by mean and standard deviation (SD), while categorical variables were summarised using percentages. We used sampling weights to account for the non-constant probability of selection of controls due to people being clustered in households contacted by telephone. Odds ratios and 95% confidence intervals were obtained from logistic regression. Adjustment was made in logistic regression for the variables used in matching: age and season. Season was defined by categorising the date of the case’s injury according to the usual 3-month definitions of spring, summer, autumn and winter. Controls were recruited in the same season as the case’s injury date. Additionally, adjustment was made for the key exposure of average weekly hours worked.

RESULTS

Recruitment

Among the next of kin of 25 fatally injured farmers invited to participate, four (16%) responded and were interviewed.

A total of 971 non-fatally injured farmers were approached (table 1). A substantial proportion of these did not consent to be contacted by the study staff (32%) or were not contactable (13%). The eligibility of this group could therefore not be determined as study staff did not have access to medical records to determine injury severity. Slightly less than half (45%) of those whose eligibility could be determined, did not meet the inclusion criteria, predominately due to an AIS of 1. Among the non-fatally injured farmers who were eligible, 84% participated and were interviewed. Recruitment in emergency departments resulted in lower proportions declining to be contacted (28% compared with 43% for those approached by letter) and higher proportions of eligible patients agreeing to participate and being interviewed (90% compared with 56%).

Almost 28 000 households were contacted for control recruitment, and the eligibility of 81% of these could be determined (table 1). Eligibility could not be determined due to no answer (12%), the call being terminated by the respondent (7%) or language difficulties (0.6%). The majority of those who were not eligible were not farms or farm worker households (99%), with the remainder not suitable age matches. Among the 548 households in which an age match could be identified, 92% participated.

Case characteristics

The average age of cases was 49 (SD 14) years, and ages ranged from 17 to 88 years. Other case characteristics are shown in tables 2–4. Just over two thirds of cases were farm owners/managers (table 2). Cases most commonly reported dairy (23%), beef (20%) and horticulture/fruit (16%) as the main commodity produced (table 4). The majority (67%) worked on properties smaller than 500 hectares (table 4).

The most common external causes of injury were machinery (65 cases, 26%), falls (47, 19%), transport (45, 18%), animal related causes (44, 17%) and being struck by an object (28, 11%). The four fatally injured cases comprised 2% of all cases in the study, and 82% of cases (n = 206) were admitted to hospital. Excluding fatalities, injury severity ranged from a score of 1, representing minor injuries, to a score of 5, representing critical injuries. Most cases had a maximum AIS of 2 (65%) and 3 (23%), with the majority at each level requiring admission to hospital (77% and 96%, respectively).

Risk factors for injury

Personal characteristics associated with an increased odds of injury included farm work hours, being an employee or contractor (compared with owner/manager) and not having attended farm training courses (table 2). There was weak evidence that seasonal involvement in agriculture was associated with increased odds. Given that many employees and contractors are engaged on a seasonal basis, and that employees may be less likely to attend farm training courses, job status was further adjusted for these factors, with marginal effect (OR 1.64, 95% CI 1.05 to 2.55).

Several health related characteristics were associated with reduced odds of injury: history of a chronic medical condition, back pain in the previous 12 months and being unwell on the previous day (table 3). An unexpected finding was that excessive daytime sleepiness was also associated with reduced odds of injury. There was little evidence that the following health related characteristics were associated with increased odds of injury: previous injury or previous multiple injuries in the last 3 years, regular use of stomach remedies or heart/circulatory medications, having stopped analgesics or arthritis medications in the last 12 months, medical conditions requiring medication, pesticide/herbicide use, having a second job, hours of sleep in the previous 24 h, drinking alcohol, poor eyesight or wearing a hearing aid. There was some evidence that having had a hospital stay for an injury in the last 3 years increased the odds of injury, based on the unadjusted odds ratio. However, this was not the case for the adjusted odds ratio.

Farm characteristics associated with an increased odds of injury included absence of roll-over protective structures on all/almost all tractors, absence of personal protective equipment for chemical use, and a low average annual income of AUD$5000 or less (table 4). The absence of anyone on the property having done safety training in the last 12 months was associated with increased odds of injury in the univariate analysis (OR 1.60, 95% CI 1.06 to 2.41), but this effect did not remain after adjustment. Factors associated with decreased odds of injury included the absence of power take-off guards from all/almost all operational tractors, and serious injury on the farm in the previous 3 years.

DISCUSSION

In this age-matched case–control study, we identified several risk factors for work related injury among male farmers and farm workers. Some of our findings are new. Few other studies, in which both owners and employees have been included, have specifically examined job status. Two studies in the United States found no evidence for an association between being the owner and injury,27 28 while one Canadian study found that owner-operators had significantly higher injury rates than others on the farm, including employees.11 However, respondents in this retrospective survey were the farm owners, creating the potential for information bias. Our finding of increased odds for injury among employees and contractors does not seem to be entirely explained by the seasonal nature of work by employees, or the likelihood of a lower level of farm training among employees. As experience was not associated with injury odds in our study, this too is an unlikely explanation. It is possible that exposure to specific farm work tasks may be different, thereby influencing injury risk.

We found evidence to suggest that farm training, but not safety training alone, may be protective for farm injury. Farm training courses attended by the controls were varied and included general agricultural and horticultural certificates, diplomas and apprenticeships, and business management courses, in addition to courses on specific issues such as chemical handling, animal husbandry, pasture management, machinery and equipment operation, and wool classing. These courses may include a safety component, and it is possible this contextual safety training is more readily applied than that received in stand-alone safety courses. Others have also reported no evidence for any protective effect of farm safety training.27 29

Little previous research on farm safety features and systems has been reported. We examined a range of relevant variables and found few associations. Working on farms where most tractors were not protected with roll-over structures, and where personal protective equipment for chemical use was not available was associated with increased odds of injury. These factors are most likely markers for farms with the least developed safety systems. An unexpected finding was decreased odds associated with the absence of guarding for power take-off shafts.

A number of our results confirm those of others. Our finding that previous farm work related injury for an individual is associated with subsequent injury is similar to that of others.7 Previous non-farm related injury has also been associated with increased odds for subsequent farm work injury among older farmers.30 In contrast, we found that working on a farm where there had been a previous serious injury was associated with decreased odds of subsequent injury among individuals working on that farm. It is possible that farms with an injury history have better developed farm safety features and systems as a result of this previous injury experience. Post hoc analysis of our data found no evidence to support this explanation.

A number of studies examining farm income have found no evidence for an association with injury,14 27^–29 with the exception of one restricted to farm machinery injury which found higher rates of injury on farms with low farm sales.15 Our result showing increased odds for low farm income should be treated with some caution due to the high proportion of missing data, particularly for the cases (table 3).

Other findings in our study that are consistent with the literature include no evidence for an association with experience, drinking alcohol and having a second job.16 27 28 We found no evidence for an effect of average sleep hours on injury odds, consistent with results from a study of part-time farmers31 but inconsistent with studies examining sleep and any injury (not just farm injury) in a rural population32 and among adolescents living on farms.33

We found no evidence of an association of injury with educational levels, despite consistent evidence from North American studies that higher educational levels are associated with increased farm injury risk, including in age and gender adjusted models.7 14 27 29 Similarly, we did not replicate a previously reported age adjusted association between wearing a hearing aid and injury,27 although our study may have been underpowered for this risk factor. Nor did we confirm previous findings in relation to using stomach remedies or heart/circulatory medications and injury.14 However, given dis-similarities between health systems in different countries including access to medical care, pharmaceuticals and insurance structures, it is difficult to compare our medication results with those from studies in other countries.

We found, unexpectedly, that several health related characteristics (history of a chronic medical condition, back pain in the previous 12 months, and being unwell the previous day) were associated with decreased odds of farm work related injury. It is possible that farmers and farm workers with these conditions restrict the range of farm work tasks undertaken, incidentally restricting their exposure to more hazardous activities. An aspect of our study that requires exploration was an apparent protective effect for excessive daytime sleepiness. This is in contrast with Sprince et al who used the same instrument and definition and found no association between sleepiness and farm injury in a case–control study that was comprised primarily of men.27

Study design and methodological limitations need to be considered when interpreting these findings. Two important principles are adequate capture of all cases and that should controls become cases, then they would be captured by the case recruitment processes. The definition of the study region as the catchment areas of the major regional Victorian hospitals, recruitment of cases through these hospitals, and recruitment of controls from the same geographical areas ensured that cases and controls were recruited from the same underlying source population. However, it is possible that some potentially eligible cases with an AIS 2 injury could have sought treatment at minor regional hospitals or through their general practitioner. Higher level severity injuries would almost certainly be treated at the recruiting hospitals (or the participating metropolitan hospitals). Detection bias could occur if the reasons for not presenting to a study hospital were associated with any of the risk factors of interest. This would generally be true in our study, with the same exception for some AIS 2 injuries as noted above for cases.

We employed risk set sampling to ensure that the group from whom we recruited our controls were at risk of becoming cases at approximately the same time that the case was injured. Injuries are acute events and, if not fatal or permanently disabling, a recovered previously-injured person can continue to be exposed to the risk of being injured again. Therefore, it is quite appropriate that controls for our study may have been injured in the past. Finally, an inherent limitation of the case–control study design is the retrospective nature of collection of information on risk factors and our use of self-report for this. Some of the unexpected and unexplained findings, such as an apparent protective effect for absence of guarding of power take-off shafts, may have arisen as a consequence.

Although participation was high among eligible non-fatally injured cases who were contacted (84%), a substantial proportion of potentially eligible injured farmers (45%) did not consent to be contacted. The direction and extent of any potential response bias is difficult to assess as there is limited information available for those who declined to be contacted. Based on data for 60% of this group, it is likely that participating non-fatally injured cases were more experienced than those who did not participate, which could bias the result for inexperience towards the null. Given the relatively small number of fatal cases included, any response bias in this group is likely to have limited impact on the results. Recall bias is possible, given the potential influence of an acute injury event on recall of relevant information by the cases.34 The number of risk factors examined could increase the probability of a chance finding.

Strengths of the study include an early audit which identified that a large proportion of potentially eligible cases were not being approached,23 thereby allowing a modification in the recruitment process reducing the potential for case detection bias. Few other farm injury risk factor studies have included employees and contractors as well as owner/operators. Further, we sought individual level exposure data directly from the employees rather than via the owner. The high participation rate among the controls reduced the potential for response bias. Household telephone recruiting for the controls in our study removed potential selection bias that may have arisen with other recruitment strategies, for example the use of farmer organisation membership lists.

CONCLUSION

This study adds to the now growing literature on farm injury risk factors both by confirming some previously reported findings and identifying some new ones. We found little or no effect for many of the postulated risk factors, a theme which is emerging from other well designed and executed studies. This suggests that these factors may not be as important as first speculated. It is possible that more proximal and transient factors may have a greater significance in the genesis of some of these injuries than previously considered. Our study did identify some at risk subgroups, including employees and contractors, those who have already had a farm work related injury, and farms lacking roll-over protection on their tractors, for whom it might seem logical to recommend targeted interventions. However, the evidence base for effective interventions is limited.35 There is still much to be learned about farm injury aetiology and the translation of this knowledge to effective interventions.