Introduction

Non-communicable diseases (NCDs), such as cardiovascular diseases (CVDs), diabetes and cancers are the leading causes of death and disability worldwide, accounting for more than half of the global disease burden.[1] Almost 75% of NCD-related deaths occur in low- and middle-income countries (LMIC),[2] often among working-age adults as young as 40 years.[3] Although disease patterns vary across world regions, CVDs remain the leading causes of death throughout.[4] Most CVDs develop from atherosclerosis (the hardening and narrowing of major blood vessels).[5] As such, atherosclerotic CVDs are largely preventable by addressing risk factors including unhealthy diets, physical inactivity, harmful use of alcohol, tobacco use, hypertension, overweight and obesity, diabetes and dyslipidaemia.[5–7] Urbanisation is demanding still more non-agricultural labour, and LMICs have undergone unprecedented labour-force transitions out of agriculture and into the industry and service sectors (Table 1).[8, 9] Labour-force transitions out of agriculture have been particularly steep in middle-income countries such as China, however, large low-income countries, such as India, are quickly catching up.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Percent employment in agriculture, services and industry of total employment in low-and middle-income countries.

https://doi.org/10.1371/journal.pone.0230744.t001

Type of employment is an important social determinant of health.[11] Types of employment contribute significantly to shaping the conditions of daily life that strongly associate with major immediate CVD risk factors such as hypertension, obesity, diabetes and dyslipidaemia.[11–13] For example, the type, amount and stability of labour and income influence people’s ability to acquire diverse and nutritious foods (through access to own-produce or purchase), assets and health-services.[13, 14] The type and duration of labour influence physical activity, nutritional needs, time available for food preparation and levels of exposure to biological and chemical hazards.[13, 15, 16] The environment in which people live and work influence the availability, access and affordability of commodities that may have beneficial or harmful effects on cardiovascular health, such as fruits and vegetables, highly processed energy-dense foods, tobacco and alcohol.[13, 16–18]

Agriculture has the potential to benefit nutrition and cardiovascular health, through the increased production and availability of nutritious foods and higher levels of physical activity that is associated with agricultural labour.[14, 15] Engaging in agriculture is also associated with cardiovascular health risks such as prolonged exposure to disease vectors, food borne diseases and toxic pesticides.[13] Most systematic reviews on the links between agriculture, nutrition and cardiovascular health synthesise evidence from studies that introduce, improve or intensify agriculture. In light of the expected continued labour-force transitions away from agriculture in LMICs, this paper aims to address two additional questions (i) is engaging in agriculture compared to types of non-agricultural employment (e.g. services and manufacturing) associated with lower levels of CVD incidence, prevalence and risk factors? (ii) Is the process of transitioning out of agriculture and into types of non-agricultural employment associated with higher levels of CVD incidence, prevalence and risk factors? Our initial systematic search returned only one eligible study pertaining to question (ii) and we therefore set out systematically to review the published evidence of the associations of engaging in agriculture compared to (any) types of non-agricultural employment with CVD incidence, prevalence or CVD risk factors in LMICs. We hypothesised that people who engage in agriculture have lower levels of CVDs and associated risk factors (not considering chemical, ambient and noise pollutants) than individuals who engage in other types of labour, particularly types of sedentary work. A review of the evidence might identify types of higher-risk employment, and with that, provide guidance to categorising employment in future longitudinal studies of employment transitions out of agriculture.

Methods

This systematic review asked the following questions (i) is engaging in agriculture compared to types of non-agricultural employment (e.g. services and manufacturing) associated with lower levels of CVD incidence, prevalence and risk factors?; and (ii) is the process of transitioning out of agriculture and into types of non-agricultural employment associated with higher levels of CVD incidence, prevalence and risk factors? The review is reported according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) [19] (see S1 Checklist) and the Guidelines for Meta-Analyses and Systematic Reviews of Observational Studies.[20] The protocol was published in advance (ID=CRD42015025488).[21] Eligibility criteria were defined in relation to PICOS (participants, interventions, comparisons, outcomes, and study design) as recommended by PRISMA.

Analysis

Where outcome units differed between studies, we converted results to a common unit using biomedical research conversion tables.[27] Some studies described one risk factor with multiple estimates, e.g. mean body mass index (BMI) and prevalence of overweight and obesity. The type of information reported by most studies were included in the analysis, resulting in the exclusion of continuous data. Characteristics of studies that were excluded from the analysis as a result of these restrictions are available in S2 Table in the online information.

We used formulas suggested by Sterne (Eq (1))[28] to calculate PRs with 95% CIs and produce forest plots that graphically describe patterns of outcomes by livelihood or occupation groups across studies. Adjusted PRs (95% CIs) could not be calculated because we did not have access to raw datasets. Narrative analyses of adjusted estimates were presented separately. It was not appropriate to perform formal meta-analyses or generate funnel plots because of the substantial heterogeneity of measurement methods, categorisation of exposures and outcomes, study settings and populations of the included studies. All analyses were performed in Stata 14.

(1)

Results

Our search yielded 3159 records, including 166 studies identified through other sources, such as manually searching bibliographies and contacting authors (Fig 1). After removing duplicates (n = 995) we screened 2164 titles and abstracts and reviewed 189 full-text publications against inclusion/exclusion criteria. We included 13 publications that reported data from 12 unique studies. Only one eligible study provided appropriately adjusted estimates (i.e. not including mediators in non-mediation analysis). Adjusted results on hypertension from this study is presented separately. A summary of identified studies on primary outcomes, atherosclerotic CVDs (n = 2), which did not meet inclusion/exclusion criteria for the main analysis, is further provided.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Flow diagram of the review process.

CVD–cardiovascular disease, HIC–high-income country, n–number of studies.

https://doi.org/10.1371/journal.pone.0230744.g001

Study characteristics

Characteristics of included studies are provided in Table 2 (see S3 Table for further details). We included one longitudinal[29] and 12 cross-sectional studies. The longitudinal study followed up participants 10 years from baseline. The cross-sectional studies comprised eight analyses of primary data[30–36]; two secondary analyses of existing cross-sectional data[37, 38]; and three cross-sectional analyses of longitudinal data,[39–41] of which two used data from different time points of one study.[40, 41] Study settings and characteristics differed substantially between studies. Most studies were from Asia (India[31, 33, 35–39] and China[29, 40, 41]); one was from Latin America (Guatemala[32]) and two were from Sub-Saharan Africa (Ghana[30] and Nigeria[34]). Eight studies analysed data from rural populations and five studies additionally included urban participants (including migrants). Sample sizes ranged from 195 to 77,220 participants. Twelve studies reported the age range of participants; eight included younger adults aged from 15 years,[38, 42] 18 years,[30, 36, 39] and 20 years.[34, 35, 41] The maximum age of participants varied from 49 to 99 years. One study did not report the age range, however, described participants as adults.[32]

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Characteristics of included studies from five low- and middle-income countries (n = 13).

https://doi.org/10.1371/journal.pone.0230744.t002

Secondary outcomes

Hypertension.

Nine studies reported sufficient data for calculating PRs (95% CIs) of hypertension in agricultural and non-agricultural groups. Seven studies were set in rural areas and two included rural and urban residents. The prevalence of hypertension varied considerably between compared groups and studies, from 0.3% among farmers in one study[40] to 48.5% among retired sedentary workers in another.[35] However, different cut-offs were used (range: 130/85mmHg, ≥ 160/95mmHg), which hampered direct comparisons of results (Fig 2). In three publications (from two studies), hypertension was less prevalent among farmers than migrants[40, 41] and urban workers,[32, 40, 41] with the exception of migrant women in one study.[40] The pattern was less clear in the studies that compared more specific occupation groups that combined men and women, who were predominantly living in rural areas. Farmers had a lower prevalence of hypertension than factory workers[34] and retired sedentary workers[35] in two studies. In two other studies, farmers had a higher prevalence of hypertension than manual labourers[35] and those with other employment than farming and trade.[30] Four studies did not observe differences in the prevalence of hypertension between agricultural and a range of non-agricultural groups comprising: non-farming rural workers,[32] housewives/housekeepers,[31, 35] daily wage earners,[33] service workers,[33] nomads,[35] sedentary workers,[35] monks,[35] the unemployed[33, 35] and a group of ‘others’ (who were not unemployed, business, service or daily wage workers).[33] Three comparisons of hypertension between agricultural workers and trade or business workers were contradictory: the prevalence was lower among agricultural workers than business workers in one study that analysed women and men separately[39] and similar between groups of agricultural and business/trade workers in analyses that combined women and men.[30, 33]

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Prevalence ratios and 95% confidence intervals of hypertension by employment status (n = 9).

Agri–agriculture, CI–confidence interval, n–sample size, PR–prevalence ratio, R–rural. Prevalence ratios were derived from comparing each non-agricultural group (coded 1) to the agricultural group (coded 0).

https://doi.org/10.1371/journal.pone.0230744.g002

Hypertension remained considerably less common among farmers than migrants (urban people were excluded) after adjusting the odds ratios (ORs) for age in one study[40] (blood pressure ≥ 140-159/90-94 mmHg: OR^men 1.38 [95% CI 1.19, 1.59], OR^women 1.03 [95% CI 1.14, 2.93]; Blood pressure ≥ 160/95: OR^men 1.96 [95% CI 1.52, 2.52], OR^women 1.83 [95% CI 1.14, 2.93]). The adjusted OR was a slight attenuation among men and a slight increase among women compared to the crude ORs.

Overweight and obesity.

Five studies reported enough information to calculate PRs (95% CIs) for overweight and/or obesity. Studies predominantly combined overweight and obesity using international cut-offs (BMI ≥ 25 kg/m²) (Fig 3). Data from all five studies suggested that overweight and obesity (analysed separately and combined) were less prevalent among farmers than migrant,[41] urban,[32, 41] business,[39] non-manual,[38] sedentary[35] and retired sedentary workers[35]; nomads,[35] housewives,[35] monks,[35] and the unemployed.[35, 38] Overweight (BMI 25–29.9 kg/m²) and obesity (BMI ≥ 30 kg/m²) were less prevalent among women engaging in agriculture than manual labour in one study,[38] whereas overweight and obesity combined did not differ between agricultural and manual workers in another study combining men and women.[35] Overweight and obesity (BMI ≥ 25 kg/m²), but not obesity separately (BMI ≥ 30 kg/m²), was less prevalent among agricultural than non-agricultural rural men in one study.[32] However, large confidence intervals of the association suggested low precision of the estimates.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Prevalence ratios and 95% confidence intervals of overweight and obesity by employment status (n = 5).

Agri–agriculture, BMI–body mass index, CI–confidence interval, n–sample size, PR–prevalence ratio; R–rural. Prevalence ratios were derived from comparing each non-agricultural group (coded 1) to the agricultural group (coded 0).

https://doi.org/10.1371/journal.pone.0230744.g003

Underweight.

Four studies presented sufficient data to calculate PRs (95% CIs) for underweight (BMI < 18.5kg/m²) (Fig 4). Three studies were set in rural areas and one included rural and urban workers. All studies suggested that farmers had higher prevalence of underweight than homemakers,[36] manual and non-manual women[38]; business,[39] government and non-government[36] women and men; and ‘other’ (than non-agricultural) workers in analyses combining genders.[37] The prevalence of underweight was higher among farming than among self-employed men but not women.[36] Underweight was more prevalent among farmers than among unemployed women,[38] but similar to students, retired and unemployed people (the latter two were combined in gender-specific analyses).[36] There were no differences in the prevalence of underweight between farmers and other (non-specific) non-agricultural workers.[37]

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. Prevalence ratios and 95% confidence intervals of underweight by employment status (n = 4).

Agri–agriculture, BMI–body mass index, CI–confidence interval, Govt–government, n–sample size, PR–prevalence ratio, stud–student. Prevalence ratios were derived from comparing each non-agricultural group (coded 1) to the agricultural group (coded 0).

https://doi.org/10.1371/journal.pone.0230744.g004

Tobacco.

Five studies presented sufficient data to calculate PRs (95% CIs) for current smoking (n = 2), ever smoked (n = 1), unspecified period of smoking (n = 1) and chewing, snuffing and smoking tobacco (n = 1) (Fig 5). Two studies were set in rural areas and three were set in rural and urban areas. In one longitudinal study,[29] the prevalence of smoking decreased over 10 years among men who transitioned out of agriculture and into to unspecified occupations. The prevalence of smoking did not change among men who remained in agriculture, factory or office work over the 10 years.[29] Smoking prevalence did not differ between the four groups at the 10-year follow-up (Fig 5). Farmers were more likely to smoke than urban workers in three publications (from two studies).[32, 40, 41] Two publications used data from different time-points of the same study and found that migrant men were more likely to smoke than farming men in 1991[40] and less likely to smoke than farming men in 2010[41]. Although PRs for migrant versus farming women were similar to those of men in the two studies, they had wide 95% CIs that included the null-value of one. Two studies did not support differences between agricultural workers and non-agricultural rural workers[32] or business workers.[39]

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 5. Prevalence ratios and 95% confidence intervals of tobacco use by employment status (n = 5).

Agri–agriculture, CI–confidence interval, n–sample size, PR–prevalence ratio, R–rural. Prevalence ratios were derived from comparing each non-agricultural group (coded 1) to the agricultural group (coded 0).

https://doi.org/10.1371/journal.pone.0230744.g005

Discussion

We aimed systematically to review the published evidence on the association of engaging in agriculture compared to other types of non-agricultural employment with atherosclerotic CVDs and over 50 associated modifiable risk factors. Following the protocol, we found two studies on the primary outcomes, atherosclerotic CVDs and 13 studies on four secondary outcomes, hypertension, overweight and obesity, underweight and tobacco use. Included studies were predominantly from India (n = 7) and China (n = 3). Heterogeneity in study settings, populations under study, measurement methods and categorisation of exposures and outcomes prevented formal meta-analyses with pooled results and generation of funnel plots.

Older agricultural workers had a lower CVD mortality rate than retired individuals in a cohort of rural residents from Vietnam, whereas the prevalence of CHD did not differ with employment in a study from rural India. Five out of nine studies suggested that people who engaged in agriculture had a lower prevalence of hypertension than migrant, urban, factory and retired sedentary workers. The evidence suggested no difference or contradictory results on the prevalence of hypertension for a number of other occupations with no clear pattern. One study reported that migrants had a higher likelihood of hypertension than farmers after adjusting for age and gender. Most evidence suggested that people who engaged in agriculture were less likely to be overweight and obese and more likely to be underweight than most non-agricultural workers they were compared to. Urban men and women appeared less likely to smoke than farmers in two studies, whereas results were contradictory for migrants. The prevalence of smoking declined among men transitioning out of agriculture and remained unchanged among men who did not change their occupation during 10 years. However, there were no differences between groups at the 10-year follow-up. The associations of type of labour could not be separated from those of location of residence for any of the outcomes because of the way different types of employment were sampled, for example, rural agricultural workers and urban government workers.

To the best of our knowledge, this is the first systematic review to assess associations of engaging in agriculture compared to types of non-agricultural employment with CVD incidence, prevalence and risk factors. A main contribution of this review was the calculation and comparison of PRs and 95% CIs across studies with available data, including studies that did not perform statistical tests. We attempted to reduce overestimation of associations from studies with high outcome prevalence by calculating PRs (95% CIs), as ORs are prone to overestimate the strength of associations in this context.[45] There are also a number of limitations to this review. A main limitation was the lack of eligible studies, particularly on the primary outcomes, atherosclerotic CVDs. The lack of studies on primary outcomes was likely due to a lack of this type of data from LMIC, e.g. from national surveys and disease surveillance.[3] It is also possible that we missed some relevant studies. The lack of studies on secondary outcomes were more surprising and is likely rooted in a disconnect between sectors and sciences concerned with employment/labour and chronic diseases that are inherently interlinked.[11, 13, 46] A number of studies and risk factors were excluded from the review as a result of restricting analyses to risk factors that were reported in four or more studies and for which clear descriptions of measurement methods and outcome categorisations were provided. Our list of risk factors was comprehensive, but not exhaustive and we may in turn have neglected some important relationships of agriculture and CVD risk factors, e.g. mental health.

Substantial limitations of the included studies made it difficult to draw conclusions. The absence of definitions and descriptions of the measurement methods and categorisations of exposure and comparators complicated the interpretation of results. It may be challenging to measure livelihoods and types of employment in LMIC settings where the informal sector dominates and livelihood diversification and seasonal migration are common.[47] It is possible that the ‘current status’ in cross-sectional studies reflect the work done on the day or season of the survey, at least for some employment groups, e.g. daily wage earners. In turn, some misclassification of results is expected to dilute our results (towards the null). In all studies, participants may have belonged to their current exposure or comparator group for varying lengths of time. Survivor bias arising from ill-health or death from the reviewed risk factors or related CVD events in the time of employment prior to study start may explain some of this review’s inconsistent and contradictory results.[48] To varying degree, healthy worker bias may have diluted and possibly reversed associations, particularly in studies of older adults or other high-risk groups.[48, 49] However, studies of younger adults in LMICs may also be at risk of this kind of bias, as premature deaths and disability from chronic diseases become more common.[3] Some employment groups, e.g. highly physical work, may attract healthier individuals. For example, it is common for young adults in LMICs to migrate from rural to urban areas for work in e.g. construction (men) or housekeeping (women),[47] leaving older, potentially less healthy, adults behind in agriculture.[50] None of the included studies adequately addressed differences between included and excluded participants. Large differences in time between included studies (1991 to 2015) further pose challenges to directly comparing results within and between studies, as well as inferring relevance of the observed associations in the present contexts of LMICs.

We could not determine the extent to which potential confounders might have accounted for observed associations (or absence of associations) because of the lack of appropriately adjusted analyses. An important example is socio-economic status, which particularly may mask results of studies that included or compared rural and urban workers. For example, wealthier urban residents may enjoy health benefits from availability of and access to goods and services, such as food, water, sanitation and health-services. In contrast, the health of the urban poor may be worse than that of the rural poor,[18] for example as a result of unsafe living conditions and high living costs.[47] This may explain some of the inconsistent findings, e.g. among manual workers, business/trade and the unemployed, in studies sourcing participants from different settings. We appreciate that hypothesis-generating studies, that analysed multiple exposure-outcome relationships may have limited their number of analyses to reduce the risk of producing ‘statistically significant’ results by chance. Few studies reported that they based sample sizes on power calculation and wide confidence intervals in several studies suggested that the statistics power might not be high enough to detect existing differences. Missing data may have additionally favoured or diluted associations in several studies. Finally, there were some indications of selective reporting of results from most studies. To some extent, this could be explained by the large scope of several exploratory studies, which may not allow for reporting all analyses. It is also possible that some studies gave preference to ‘statistically significant’ results in order to improve chances of publication.[26]

It is common for LMIC governments, e.g. in Indian and China, to facilitate labour-force shifts out of agriculture and into industry and service sectors to promote economic growth during development.[51, 52] Concurrently, national age-standardised prevalence of death and disability from CVDs have risen and the Employment Conditions Knowledge Network expresses concern that LMICs will be unable to provide the growing urban labour-force with fair employment opportunities.[11, 53] Progress reports on China’s recent urbanisation have shown unexpected challenges with integrating agricultural workers into new types of employment and urban settings. Chinese migrant workers, for example, suffered social exclusion and were denied equal rights to fair employment, housing and health services, which is associated with infectious and chronic diseases.[51, 54] As previously discussed, our search only returned one study that examined the association of employment transitions out of agriculture with cardiovascular health in LMICs. The unexpected developments in China warrant a deeper understanding of how employment transitions are associated with CVDs and risk factors to ensure current economic growth strategies do not add to already expected rises in CVD burdens with development in LMICs.

Evidence on associations of types of employment, and particularly employment transitions, with chronic diseases and related risk factors is fragmented. Where the health sciences typically assess an association of an intervention with an outcome within a single type of employment (e.g. agriculture [55, 56]) the employment sector tends to focus on shorter-term issues relating to health and safety regulation-outcomes (e.g. ergonomics, injuries or exposure to noise or hazardous agents).[11] The data collected across sectors vary widely, for example in relation to sources and collection methods, and are not easily combined for epidemiological studies that can assess health risks or benefits that are associated with types of employment. Prospective longitudinal studies should take advantage of the current momentum for interdisciplinary initiatives across the 2030 agenda for sustainable development, such as ‘decent work and economic development’, ‘sustainable cities and communities’ and ‘health and well-being for all’ in LMICs.[57] New study models and data collection approaches are likely to be required to address these challenging needs. Studies should also address the methodological shortcomings identified in this review by complying with appropriate reporting guidelines,[58] namely ensuring appropriate selection of participants that accommodates conditional comparability of outcomes across exposure and comparator groups; use of appropriate measurement methods for collection of all data; and appropriate reporting of study protocol, methods and results (including the reporting of missing data and ‘non-significant’ findings).

Conclusion and implications

There was some, however limited, evidence of negative associations of engaging in (rural) agriculture compared to types of (usually urban) employment with prevalence of hypertension, overweight and obesity, and positive associations with prevalence of underweight and smoking. There were no clear patterns indicating which types of employment were associated with higher or lower prevalence of outcomes. High quality evidence is lacking on how engaging in agriculture compared to types of non-agricultural employment may act on CVDs and risk factors in LMICs. Rigorous studies that address the methodological shortcomings identified by this review are needed. They should cultivate a consensus of how best to measure and categorise employment when investigating chronic disease outcomes as well as appropriately control for potential confounders. We further call for new models of interdisciplinary longitudinal studies, for example across health sciences, livelihoods, demography and policy, that assess the association of livelihood transitions and migration with core health outcomes in LMICs. These sorts of studies should seek to inform evidence-based chronic disease prevention measures in national economic growth strategies to safeguard health during development in line with the sustainable development goals.

Supporting information

Acknowledgments

We would like to thank Ms Katharine Ker (Faculty of Epidemiology Faculty of Epidemiology and Population Health, London School of Hygiene and Tropical Medicine (LSHTM), United Kingdom (UK)) for guidance on the search strategy, data extraction form and risk of bias assessment strategy; Dr MinHae Park (Faculty of Epidemiology Faculty of Epidemiology and Population Health, LSHTM, UK) for contributing to the data extraction form and risk of bias assessment strategy; Ms Jane Falconer, User Support Services Librarian (Library and Archives Service, LSHTM, UK) for developing the search on ‘low- and middle-income countries’; Dr Clara Calvert (Faculty of Epidemiology and Population Health, LSHTM, UK) for advising on analysis strategy and manuscript preparation; and MD Karin Østergaard (Department of Internal Medicine, Hospital of South West Jutland, Denmark) for manuscript revisions.