Article Text

Download PDFPDF

Determinants of agricultural injury: a novel application of population health theory
  1. William Pickett1,2,
  2. Louise M Hagel3,
  3. Andrew G Day4,
  4. Lesley Day5,
  5. Xiaoqun Sun4,
  6. Robert J Brison1,2,
  7. Barbara L Marlenga6,
  8. Matthew King1,
  9. Trever Crowe7,
  10. Punam Pahwa3,8,
  11. Niels Koehncke3,
  12. James Dosman3
  1. 1Department of Community Health and Epidemiology, Queen's University, Kingston, Canada
  2. 2Department of Emergency Medicine, Queen's University, Kingston, Canada
  3. 3Canadian Centre for Health and Safety in Agriculture, University of Saskatchewan, Saskatoon, Canada
  4. 4Clinical Research Centre, Kingston General Hospital, Kingston, Canada
  5. 5Accident Research Centre, Monash University, Melbourne, Australia
  6. 6National Children's Center for Rural and Agricultural Health and Safety, Marshfield Clinic Research Foundation, Marshfield, USA
  7. 7Department of Agricultural and Bioresource Engineering, University of Saskatchewan, Saskatoon, Canada
  8. 8Department of Community Health and Epidemiology, University of Saskatchewan, Saskatoon, Canada
  1. Correspondence to Dr William Pickett, Emergency Medicine Research, Queen's University, Angada 3, Kingston General Hospital, 76 Stuart St, Kingston, ON K7L2V7, Canada; will.pickett{at}queensu.ca

Abstract

Objectives (1) To apply novel population health theory to the modelling of injury experiences in one particular research context. (2) To enhance understanding of the conditions and practices that lead to farm injury.

Design Prospective, cohort study conducted over 2 years (2007–09).

Setting 50 rural municipalities in the Province of Saskatchewan, Canada.

Subjects 5038 participants from 2169 Saskatchewan farms, contributing 10 092 person-years of follow-up.

Main measures Individual exposure: self-reported times involved in farm work. Contextual factors: scaled measures describe socioeconomic, physical, and cultural farm environments. Outcome: time to first self-reported farm injury.

Results 450 farm injuries were reported for 370 individuals on 338 farms over 2 years of follow-up. Times involved in farm work were strongly and consistently related to time to first injury event, with strong monotonic increases in risk observed between none, part-time, and full-time work hour categories. Relationships between farm work hours and time to first injury were not modified by the contextual factors. Respondents reporting high versus low levels of physical farm hazards at baseline experienced increased risks for farm injury on follow-up (HR 1.54; 95% CI 1.16 to 1.47).

Conclusions Based on study findings, firm conclusions cannot be drawn about the application of population health theory to the study of farm injury aetiology. Injury prevention efforts should continue to focus on: (1) sound occupational health and safety practices associated with long work hours; (2) physical risks and hazards on farms.

  • Agriculture
  • cohort study
  • epidemiology
  • farming
  • population health
  • environment
  • farm
  • models
  • public health
  • socio economic status
View Full Text

Statistics from Altmetric.com

Introduction

Historically, studies of injuries and their determinants have most often been conducted in the absence of an implicit theoretical base.1 A promising advance in the field of injury control research is therefore the direct application of theory, stated a priori, to guide the development of aetiological modelling.

Population health thinking or the ‘new public health’ provides one possible theory for the development of aetiological models in injury research.2 3 Proponents of such theory state that determinants of events such as injury operate at a minimum of two levels: (1) contextual (eg, socioeconomic, physical, or cultural environments); and (2) individual (eg, personal health; individual work practices). The theory itself is based on past empirical findings that suggest that contextual and individual determinants may interact multiplicatively to produce varying levels of risk for injury.2 3 Study of these interactions may lead to advances in understanding of potential causes of injury.

Aetiological studies based on population health theory could point to the strongest underlying contextual modifiers of injury risk, and hence provide evidence-based information for the prioritisation of preventive efforts. To illustrate, in occupational settings, if socioeconomic contextual moderators are identified, emphasis would be placed on proven management practices and business models that encourage ongoing investments in safety. If cultural contextual moderators are identified, this would suggest that the injury problem is an inherently sociological issue, and that interventions must aim at changing work culture and beliefs. If physical contextual moderators emerged as most important, the emphasis would strictly be on environmental modification and hazard reduction.

Our research group has a particular interest in farm injuries and their potential causes. To that end, we developed a large cohort study in Saskatchewan, Canada.4 5 Data from this cohort provided the opportunity to perform novel longitudinal analyses that might: (1) demonstrate the value of applying population health theory to the modelling of injury experiences in one particular research context; and (2) potentially enhance understanding of the conditions and practices that lead to agricultural injury. Such research can in turn provide a potential evidence base for public health policy, knowledge translation, and preventive interventions.

Saskatchewan Farm Injury Cohort

The Saskatchewan Farm Injury Cohort (SFIC) involves an initial cohort of 5492 residents on 2390 Saskatchewan farms who participated in a baseline survey in 2007, and who are being followed longitudinally in subsequent years. The rationale and methodology for the SFIC are outlined in detail elsewhere.4 Major aims are: (1) to examine associations between individual farm work exposures and the occurrence of agricultural injury; and (2) to assess the importance of contextual factors (socioeconomic, physical, cultural) as potential moderators of associations between these individual exposures and the occurrence of injury. The primary hypothesis of the SFIC study is prevention oriented. Associations between individual farm exposures and farm injury were expected to become attenuated when contextual environments were safe versus hazardous. The current analysis is based on individual and contextual factors assessed at baseline, and injury occurrence during the subsequent two-year follow-up period.

Methods

Population health theory

Figure 1 outlines the framework via which population health theory was applied in our study. Work-related farm injuries were the primary study outcomes. The amounts of time that participants were engaged in farm work were the primary individual exposures. Characteristics of the farm work environment were the primary contextual moderators that potentially could influence individual exposure–outcome relationships. Contextual moderators considered to be important a priori were socioeconomic work conditions on the farm,6 hazards associated with physical farm work environments,7 and hazards associated with culturally-based farm work practices.8 Potential confounders considered a priori included demographic characteristics (age and sex, relationship to farm owner), health behaviours (alcohol consumption), and co-morbid conditions. Other farm-level confounders (eg, farm size) were identified during the analysis phase only.

Figure 1

Theoretical framework used to develop aetiological models within the Saskatchewan Farm Injury Cohort, adapted from population health theory.2 3 *Demographic: age, sex, relationship to farm owner, farm acreage, commodity produced; Behaviours: alcohol consumption index; Co-morbidity: index of common health conditions.

Study population and procedures

Study procedures were approved by the Biomedical Research Ethics Board at the University of Saskatchewan. Farms from within a stratified random sample of 50 rural Saskatchewan municipalities were selected for study. Stratification was based on: (1) soil type and hence type of agriculture (three strata); and (2) membership or not in an agricultural health and safety network (two strata).9 Study participation at the municipal level was requested during in-person meetings with each rural municipal council. In the case that councils refused participation (n=3), neighbouring rural municipalities from the same strata, defined by agricultural region and network participation, were approached and subsequently enrolled.

Baseline mail surveys were conducted from February to May 2007. After receiving updated lists of farms from individual councils, all listed farms (mean per rural municipality: n=163) in the selected municipalities were contacted to participate. Farms that were inactive were excluded. Baseline data collection followed the Dillman total design method.10 A knowledgeable adult on each farm was asked to complete a written questionnaire that provided information about: (1) the farm operation; (2) the health and work experiences of resident family members; (3) farm health and safety practices; and (4) contact details. The cohort was designed to be large, heterogeneous in terms of work-related exposures, but not necessarily representative.4

Following the baseline interview, all farms were re-contacted by mail on four occasions, in 6-month intervals over a 2-year period (October 2007, April 2008, October 2008, April 2009). On each occasion respondents were asked to report and describe the occurrence of farm injury events, defined below, using a one-page form. For each reported injury event, supplemental questions were used to describe the person involved as well as the timing, mechanism, and circumstances surrounding the injury, and its medical treatment. The follow-up forms were compiled to develop a longitudinal record of injury experiences for each farm.

Measures

A panel consisting of the primary research team and knowledgeable farm producers developed the baseline study questionnaire. Where possible, items that had been used in other research contexts were selected or adapted for use.11–13 The questionnaire and recruitment strategy were tested in pilot studies and refined on multiple occasions.4 5

Outcomes

The primary outcome measures were times to first farm injury and then specific types of injury, reported on follow-up. Farm injuries were defined as those: (1) that happened on a farm during a work activity or during the course of farm work in non-farm locations; and (2) that resulted in treatment of the injury by a doctor or nurse, or the victim missing at least four hours from work or usual activities due to the injury. Three specific outcomes were times to: (1) any reported farm injury; (2) farm injuries involving machinery; and (3) farm injuries that required hospitalisation or any medical treatment.

Individual exposures

The individual measure used was exposure to farm work, measured in terms of the duration (amount of time) that people were exposed during each of four seasons. These measures were averaged over the full year and then subdivided into categories. Full-time (≥30 h/wk) and part time work hours (1 to <30 h/wk) were defined as per Canadian occupational standards.14

Contextual exposures

The socioeconomic farm work environment was measured using Likert-like scales that describe the frequency of worry on the farm (five categories: never through daily) attributable to cash flow shortages and debt, each reported seasonally. These measures were developed with the input of farm operators who recommended not asking direct questions about money, farm assets, or debt in order to maximise response rates. Reports of cash flow shortages and debt were each highly collinear across seasons. Reports from one season (spring) were therefore used to represent general socioeconomic conditions on the farm. If this measure was not reported for the spring, reports from other seasons (summer, then fall, then winter) were substituted in order to maximise sample sizes available for analysis. Exploratory bivariate analyses were conducted to identify appropriate cut-points. An additive scale was then created and then used to categorise respondents into two categories (low vs high), split at the median.

The physical farm work environment was inferred from injury-related physical hazards within the farm work environment, each measured at the farm level. A list of illustrative hazards was derived from the findings of the Canadian Agricultural Injury Surveillance Program.15 16 From these lists, we developed items describing: (1) the presence of safety features on major classes of injury-producing machinery in Saskatchewan (tractors, combines, and augers); and (2) safety features surrounding other hazardous physical features of farm work environments (buildings, water bodies). An additive measure that characterised the general degree of physical hazard on the farm was developed from these items, and following exploratory bivariate analyses respondents were again divided into two categories (low vs high), split at the median.

The cultural farm work environment was measured using items that described normative farm safety practices that reflected the work culture of the farm and pertained to three vulnerable farm groups (young children, young workers, and older farm workers). These were measured at the farm level. Examples of such practices assessed using Likert-like response options from never to daily are listed below (for a complete list of items see Pickett et al4): Young children: (1) how often are young children present in the farm worksite? (2) how often do young children ride in a cabbed tractor or cabbed combine with an adult operator? Young workers and older farm workers: (1) how often do these workers operate tractors without rollover protection structures? (2) how often do these workers operate farm equipment that is more than 20 years of age? Valid responses to each of these questions were only provided when the farm had at least one person in any of the vulnerable groups. Responses were averaged for each farm and the resultant additive score was used to categorise farms into three levels: (1) no valid responses (indicating that the farm had no members in the three vulnerable groups); (2) low exposure to known cultural hazards; and (3) high exposure to known hazards. Following exploratory analyses, low and high exposure levels were derived by splitting the additive score at the median for those with valid responses.

Confounders

Age, sex and relationship to farm owners were included as confounders because they have been associated with injury outcomes, exposures to occupational situations, and yet were not in the causal pathway that links exposures to injury outcomes.17 Comorbid conditions were included as there is a strong correlation between health status and the ability to work, as well as between health status and the occurrence of injury, especially among older people.18 Health behaviours (alcohol consumption patterns) were included as these are known risk factors for injury (and hence potential confounders) in farm and other contexts.19 20 Characteristics of farms (eg, acreage, commodities produced) were also considered as potential confounders on an exploratory basis, but only included in models if they met conventional statistical criteria for confounding (as above).

Analysis

Individual farm work exposures, contextual factors, potential confounders, and injury outcomes were described. Participants were evaluable for this analysis if they had a baseline assessment and came from a farm with known measurements of the contextual factors and at least one follow-up assessment. Cox proportional hazards models were used to evaluate whether relationships between the individual exposure (hours of farm work) and time to first injury were modified by the contextual factors (socioeconomic status, physical farm work environment, cultural farm work environment). This analysis was organised according to the underlying population health theory (figure 1).

To increase the effective sample size of the multivariable analysis, hours of farm work for any season were imputed from other seasons when missing using the expectation–maximisation algorithm method.21 This allowed us to estimate the exposure category for the 96% of subjects who reported hours worked in at least one season. Robust SEs were estimated in order to account for the complex and clustered nature of our sampling frame (these SEs were of the same magnitude to those estimated via multiple techniques, including frailty models). The proportional hazards assumption was tested using the method of Lin et al.22 All continuous predictors were categorised due to their non-linear, and often non-monotonic, relationship with the log hazard rate of time to injury.

Regression analyses were conducted for three time to first injury outcomes: (1) any farm injury; (2) farm injuries caused by machinery; (3) farm injuries that resulted in hospitalisation or medical treatment. Interactions were considered by putting a product term into the Cox proportional hazards models; all analyses considered the list of confounders determined a priori, as well as those identified via exploratory analyses involving backwards elimination methods. The statistical significance of any interactions was assessed via the likelihood ratio test. We also examined the direct effects of the three contextual factors on risks for injury. All estimates of effect are presented as HRs and associated 95% CIs.

A priori, the study was 90% powered (α=0.05; 2-sided) to identify large shifts in proportions of farm people experiencing injuries (eg, 10% to 15%) between highest versus lowest quartiles of exposure, if the full study sample (5038 people on 2169 farms) was employed.4 Study power obviously diminished for analyses based on fewer observations per group, or from smaller effect sizes, but was >80% for tests of interaction that involved such shifts in proportion of 10% or more.

Results

Descriptive findings

During cohort assembly, a response rate of 33% was achieved at the farm level. In total, 5038 people from 2169 farms were evaluable for the longitudinal analysis. These people contributed 10 092 person-years of follow-up, and a total of 450 farm injuries were reported from 370 individuals on 338 farms during the follow-up period. There was on average 2.3 respondents and 4.9 person-years of follow-up per farm. The cohort included a wide range of age groups, with a clear male predominance (table 1). The farms involved were primarily family farm operations. Major commodities produced were grain and beef products, consistent with provincial agricultural practices.13

Table 1

Characteristics of farm people and farm operations participating in the Saskatchewan Farm Injury Cohort: longitudinal dataset

Testing of population health theory (interactive effects)

Table 2 shows findings from the primary analysis designed to test the population health theory surrounding determinants of farm injury. The unadjusted analysis was based on 4767 (95%), 4426 (88%), and 5036 (99.9%) for the socioeconomic, physical, and cultural farm work contextual measures, respectively. After adjustment for all covariates in the model (see footnotes to table), these reduce to 4317 (85%), 4024 (84%), and 4512 (90%), respectively. Due to small cell sizes in groups that reported no farm work, we selected those with: (1) full-time work exposure; and (2) the highest levels of contextual risk, as the referent group in each analysis. HRs for time to first injury were highest among those with full-time work exposures, and lowest among those reporting no work for all three injury outcomes. Relationships between farm work exposures and the occurrence of injury were not modified by any of the contextual factors, as indicated by the tests for interaction and visual inspection of the HRs.

Table 2

Empirical testing of population health theory; adjusted HRs for time to first farm injury associated with combinations of individual farm work exposure levels and contextual factors

Direct contextual effects

We next examined the direct effects of the contextual factors on risks for injury in a further series of Cox regression analyses. There was no evidence of an important influence of the socioeconomic and cultural work exposures on the risks for injury (table 3). Those with a score that indicated a higher level of physical farm work risks did report a 1.4- to 1.6-fold increase in the relative hazard for each of the three types of medically treated injury, after adjustment for other potential confounders. Each of these hazards (by CIs) represented significant increases in injury risk.

Table 3

Adjusted HRs for time to first injury associated with exposures to contextual factors (socioeconomic, physical, and cultural farm risks)

Discussion

We performed a cohort analysis of farmers and their families on the risks for injury associated with individual and contextual aspects of farm work and the farm work environment. In doing so, we examined some new conceptual thinking surrounding injury and its determinants. As expected, this analysis demonstrated that the amount of time that people are engaged in farm work is a major risk factor for farm injury. Contrary to theoretical expectations, while some contextual factors were associated with injury risk, these factors as measured did not consistently modify relationships between individual farm work exposures and the occurrence of injury.

Existing analytic studies that have studied potential causes of farm injury include a limited number of prospective cohort studies,23–25 along with case–control and cross-sectional studies.6 26–29 These have examined personal factors associated with increased risk for injury, including musculoskeletal disorders,23 stress, depression, and other psychological problems,6 and psychotropic and other medication use.24 28 Operational factors that have been studied include hours of farm work exposure,24 26 29 farm size and mechanisation,23 29 commodities produced,24 and farm management practices.23 Findings from these studies informed our study methods, including the selection of our measures as well as confounders considered during our modelling process. Noted limitations of existing analytic studies include the absence of documented farm work exposure information, small sample sizes, and homogeneity of farm types under study. Our cohort was large and heterogeneous, with detailed estimation of individual exposures to farm work as well as contextual factors on the farm. We also based our study on a new theoretical framework. Because of this, we view our study approach as novel.

The population health model and its depiction in figure 1 describe one conceptual approach to the application of theory to the modelling of injury. Unlike some visual representations of this theory,2 it is straightforward to infer from figure 1 the way in which our analytic strategy was conceived (see table 2). This approach can be adapted to almost any interactive relationship involving individual exposures and contextual factors affecting health, including mortality and morbidity outcomes. It provides a helpful roadmap for the planning of similar analyses for determinants of health outcomes.

However, our primary study idea was not supported in that we failed to identify consistent modification of the association between individual work hours and the occurrence of injury by socioeconomic, physical, and cultural contextual risk factors. Our major hope from this effort was that it would point to the strongest underlying contextual modifiers of injury risk, and hence provide evidence for the prioritisation of preventive efforts. Our analysis failed to provide clear direction as to which of these three contextual modifiers were strongest. This finding may be due to insufficiencies in the theory. Other explanations include a lack of statistical power for tests for interaction, misclassification in exposure assessment due to insensitive measures or the collapsing of exposure categories, and possible bias attributable to sampling and response.

While physical farm hazards did not modify the exposure–outcome relationships, we did identify direct increases in risk for injury associated with physical farm hazards. This is consistent with the well accepted principle that optimisation of physical environments is an efficacious strategy for the prevention of injury. Physical features of the farm work environment play an important role in the aetiology of injury. Similar influences of socioeconomic and cultural environments on time to injury could not be detected.

Our major finding was that risks for farm injury appear to be primarily driven by the amounts of time that people are spending performing farm work, as well as the contextual nature of their physical work environments. Findings with respect to the testing of population health theory may also demonstrate a need for the application of more homogeneous injury outcomes in such analyses (eg, tractor rollovers and runovers, specific machinery entanglements, blunt animal trauma, falls from specific structural hazards). Causal pathways implied by this theory are inherently complex, especially for farm occupational environments that involve a multiplicity of hazards and risk circumstances. It is possible that our use of a heterogeneous series of injury outcomes may have masked the existence of important relationships.

Strengths of this analysis warrant comment. First, our study represents a novel attempt to formally evaluate some current ideas about injury and its determinants on a population basis. Our approach to this aetiological analysis was well reasoned and based on contemporary theory. There are few like examples of this type of investigation available in the injury control literature. Second, our study was created from a large and well established cohort of farm operations with longitudinal follow-up over two years. Such cohorts are rare in the injury control literature. Third, recommendations based on this analysis provide practical direction to health and safety practices. The results are inconsistent with theories that suggest that health risks vary in direct or interactive manners with underlying social or cultural determinants. This finding could be real or not, due to aforementioned limitations. Studies in more general populations have found consistent gradients in health risks according to income and social class factors.30–32 Our findings do suggest that our preventive efforts should be focused on amelioration of physical risks and hazards. They also suggest that since time of work exposure is a critical factor, a modern occupational industrial approach to the manner in which that time is spent, including hours of work per rest period, and other standard safety practices for occupations that involve long work hours, may be important to consider in prevention efforts.33

Limitations of our analysis include some lack of statistical power to test the underlying theory, specifically for the cultural factor, and within vulnerable subgroups of the farm population and for more specific types of farm injury. Because our response rate was 33%, this may have resulted in some loss of heterogeneity of exposure and associated power to detect effects.4 Second, our reliance on self-reported measures to characterise both the individual work hours and contextual aspects of the farm work environment were practical, but subject to obvious misclassification errors that might attenuate associations. This may be particularly true for the socioeconomic and cultural farm work measures that, although based on standard and tested formats and multiple pilots,4 represented more indirect questions than were the physical farm hazard measures. There is a need for onsite validation of self-reported measurements of contextual aspects of the farm work environment. Finally, the findings may not be generalisable beyond the farm occupational environments in the grain and cattle growing areas of Saskatchewan.

Implications for prevention

This cohort study was developed to test some new conceptual thinking about the aetiology of injury. While our study methodology provides one model for the testing of population health theory, the study findings were inconclusive with respect to that theory. Injury prevention efforts might continue to focus on amelioration of physical risks and hazards on farms, and the effects of long work hours, in order to minimise injury risk.

What is already known on the subject

  • Studies of injury may benefit from the application of theory to guide the development of aetiological models.

  • Population health theory suggests that individual and contextual risk factors interact to produce varying risks for injury.

What this study adds

  • In this large cohort study set in Canadian agricultural population, possible determinants of injury using population health theory were examined.

  • Study findings were inconsistent with the patterns of injury hypothesised by that theory. Direct effects were observed between individual (higher levels of farm work hours) and also contextual (physical farm hazards) factors reported at baseline, and increased risks for farm injury observed on follow-up.

  • Injury prevention efforts on farms should therefore continue to focus on occupational health and safety practices associated with long work hours, and the amelioration of physical risks and hazards. Further study of the application of population health theory to the aetiological study of injury is warranted.

Acknowledgments

We thank Phyllis Snodgrass, Iris Rugg, Deborah Emerton, Murray Purcell, Debra Gronning, Louise Singer, Suzanne Dostaler, and Catherine Isaacs, as well the Saskatchewan Association of Rural Municipalities, the 50 participating Saskatchewan rural municipal councils, and the farm families from Central Saskatchewan who took the time to assist us with this research.

References

View Abstract

Footnotes

  • See Editorial, p 361

  • Linked article 028175.

  • Funding This study, initiated and conducted by the investigators, was supported financially, in part, by a research agreement with the Canadian Institutes of Health Research (Operating Grant: MOP-145294) and a pilot study grant from the Canadian Centre for Health and Safety in Agriculture (also funded by the Canadian Institutes of Health Research, Operating Grant: CDA-66151). LD is supported by a Senior Research Fellowship (ID 236880) from the National Health and Medical Research Council of Australia.

  • Competing interests None.

  • Ethics approval This study was conducted with the approval of the University of Saskatchewan.

  • Provenance and peer review Not commissioned; externally peer reviewed.

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.

Linked Articles