When measuring health the need to go beyond bio-medical indicators (such as blood pressure and cholesterol levels) has long been recognised and a raft of health questionnaires have been developed over the past 30 years to help assess the functional health status of individuals, groups and populations. Population norms have been developed for a number of these questionnaires to facilitate comparisons between groups or populations. These are benchmark scores for the general population and permit analysis of a survey sample by demographic variables such as gender and age group [25].

Four distinct health related instruments were used in this study, three of these have been widely utilised in health research: the SF-36, EuroQol EQ-5D and the Visual Analogue Scale (VAS) [25–27]. The fourth, the Short Depression Happiness Scale (SDHS), is a more recently developed instrument which has not been widely used in other studies to date. A brief description of each of these instruments is given below.

The SF-36 is an eight-scaled multidimensional health instrument that measures different attributes of an individual’s health status: physical functioning (PF), role-physical (RP), bodily pain (BP), general health (GH), vitality (VT), social functioning (SF), role-emotional (RE), and mental health (MH) [25,28]. There are two summary scales: the physical component summary scale (PCS) and the mental component summary scale (MCS): Both of which are aggregated scores for the eight scales. The SF-36 has been translated for use in over 50 countries and its results have been reported in over 4,000 publications. It has been widely accepted as a valid instrument for measuring the health related quality of life (HRQoL) for samples from the general population and those with specific health conditions [26].

U.S. national norms exist for the eight health scales and the two component scores. Scores are transformed and normalised to facilitate comparison of individual or group aggregate scores with published national norms [25,29]. Unfortunately, norms do not exist for Uganda and as a substantial proportion of the UK horticultural workforce is multi-national the 1998 U.S. national norms were used as the comparator for this instrument (http://www.SF-36.org/). This is considered an acceptable practice in multinational studies that use carefully adapted and translated HQL questionnaires [30].

The EQ-5D instrument is a generic public domain HRQoL measure in which a respondent’s health status is assessed along five dimensions (mobility, self-care, usual activities, pain/discomfort and anxiety/depression) with three separate levels of severity for each [31,32]. The U.K. EQ-5D index tariff allows scores to be compared to the U.K. population norms [27,33].

The Visual Analogue Scale (VAS) is a conceptually simple health instrument comprising a vertical line with equally spaced gradations from 0 – 100 much like a thermometer. Respondents indicate their present health status by drawing a line on the scale with the understanding that zero represents their worst possible health status and 100 their best. Population norms for the U.K. exist for this instrument [27].

The Short Depression Happiness Scale (SDHS) is a public domain instrument which allows measurements of depression and happiness across sample populations [34]. It has previously been used in the study by Cross et al. [3] and was included in this study as it has the potential to provide information that may be missed by the other general health instruments. No population norms exist for this instrument, although a score of 9 or below is considered potentially indicative of mild clinical depression [34].

Differences between groups were analysed using non-parametric Mann-Whitney U test, Kruskal-Wallis and student t-tests. Where appropriate, associations between mean scale scores were explored using Spearman’s rank correlations. Differences between groups and population norms were investigated using student t-tests. The Ugandan health scores were compared with the U.K. farm worker health scores published in Cross et al. [3]. Where appropriate both Ugandan and U.K. SF-36 scores were compared with the U.S. population norms.

Multiple regression analysis was used to investigate the relationship between self-reported health status and twelve potentially relevant variables (house type, malaria within the past three months, distance travelled to work, number of children per respondent, whether the respondent smoked or had smoked in the past, level of education, annual income, bicycle ownership, radio ownership, mobile phone ownership, job status and number of tasks performed each day).

Candidate variables were entered into a backward stepwise elimination model to explore variation within PCS and MCS scores. Multicollinearity can be problematic when including a large number of variables in the analysis as parameter variance and the r² value can tend to increase leading to an increased probability of committing a type II error [35]. Consequently, multicollinearity was tested by setting the tolerance value at less than 0.2 and the Variance Inflation Factor (VIF) considerably less than 5.

All scales of the SF-36, EQ-5D, VAS and the SDHS were highly significantly correlated in both the U.K. and Uganda (p < 0.0001) although inter-scale correlations for the Ugandan sample were stronger than those found in the U.K. study which gives some degree of confidence concerning the translations. As a consequence of the strength of the correlations between the EQ-5D, VAS and SDHS only the SF-36 scores are reported in the results.

More than 95% of the U.K. sample population were aged between 18 and 34 and consequently only scores for this age group are reported here. The 18–34 population scores were significantly higher than the US population norm for vitality (VT) only and significantly lower for role-physical (RP), bodily pain (BP), general health (GH) social-functioning (SF), mental health (MH) and the physical component summary score (PCS) (Table 1).

The overall population scores were significantly higher than the U.S. population norm for physical functioning (PF) and the physical component summary score (PCS) and significantly lower for role-physical (RP), role-emotional (RE), mental health (MH) and the mental component summary score (MCS).

When the scores for those workers who had suffered malaria in the three months preceding the survey were removed from the sample the overall population scores were higher than the U.S. population norm for physical functioning (PF), bodily pain (BP), general health (GH), vitality (VT), social functioning (SF) and the physical component summary score (PCS). They remained significantly lower for role-physical (RP), role-emotional (RE), and the mental component summary score (MCS) (Table 2).

Ugandan males scored significantly higher than females for all SF-36 scales (Kruskall Wallis, df = 1, p ≤ 0.001) although the role-emotional scale (RE) significance value was less (df = 1, p = 0.002) (Figure 1). Males scored significantly higher than the U.S. population norms for physical functioning, bodily pain (BP), general health (GH), vitality (VT) and the physical component summary scale (PCS) and significantly lower for role-physical (RP), role-emotional (RE), mental health (MH) and the mental component summary scale (MCS) (Table 2). Ugandan female farm worker scale scores were significantly higher than the U.S. norms for physical functioning (PF) and significantly lower for all other SF-36 scales.

When the Ugandan scale scores were controlled for by age the 18 – 34 age group was significantly higher than the corresponding U.S. norms for vitality (VT) and significantly lower for role-physical (RP), role-emotional (RE), and the physical component summary score (PCS). When the scores for those workers who had suffered malaria in the three months preceding the survey were removed, scores were significantly higher than the U.S. population norms for bodily pain (BP), general health (GH), vitality (VT) and remained significantly lower for role-physical (RP) and role-emotional (RE) (Table 2).

As 96.5% of the U.K. sample was aged 18 – 34 the following comparisons between scores for the U.K. and Uganda refer solely to this age group. Ugandan farm worker 18–34 scores were significantly higher than the corresponding U.K. scores for physical functioning (PF), bodily pain (BP), general health (GH), social functioning (SF) and mental health (MH) and significantly lower for role-emotional (RE).

When the scores for those workers who had suffered malaria in the three months preceding the survey were removed Ugandan farm workers scored significantly higher than U.K. farm workers for physical functioning (PF), bodily pain (BP), general health (GH), social functioning (SF), mental health (MH), the physical component summary scale (PCS) and the mental component summary scale (MCS) scores of Ugandan farmers were not lower than the U.K. workers on any scale (Table 3).

In an attempt to better understand the relative contribution of different socio-demographic and occupational factors to U.K. and Ugandan health, the PCS and MCS scores were utilised as dependent variables in a multiple linear regression model. Independent variables entered into the first model for the U.K. were farm, farm size, farming method, number of tasks per day, wages, age, gender, nationality, marital status and children. A significant model emerged for the PCS (F_4,421 = 7.64, p ≤ 0.001, adjusted r² = 0.059) with the significant variables being tasks (β = 0.153, p = 0.001), marital (β = −0.17, p = 0.003), children (β = −0.127, p=0.027) and farm (β= −0.179, p ≤ 0.001). A significant model also emerged for MCS (F_4,421 = 9.799, p ≤ 0.001 adjusted r² = 0.076). Significant variables were farming method (whether the farm worker worked on an organic or conventional farm β = 0.134, p = 0.011) children (β = −0.133, p = 0.005) farm (β = −0.186, p ≤ 0.001) and farm size (farm size was measured by the number of seasonal employees β = −0.228, p ≤ 0.001).

Independent variables entered into the first model for Uganda were house type, malaria within the past three months, distance traveled to work, number of children per respondent, whether the respondent smoked or had smoked in the past, level of education, annual income, bicycle ownership, radio ownership, mobile phone ownership, job status and number of tasks performed each day at work. A significant model emerged for the PCS (F_5,504 = 18.86, p ≤ 0.001, adjusted r² = 0.149), with the significant variables being education (β = −0.132, p = 0.002), annual income (β = 0.23, p ≤ 0.001), malaria (β = −0.119, p = 0.004), number of tasks (β = 0.179, p ≤ 0.001) and house type (β = 0.091, p ≤ 0.033). A significant model also emerged for the MCS (F_5,504 = 10.633, p ≤ 0.001, adjusted r² = 0.086). Significant variables were smoking (β = 0.088, p ≤ 0.039), annual income (β = 0.084, p ≤ 0.05), malaria (β = −0.204, p ≤ 0.001), travel (β = −0.14, p = 0.001) and house type (β = 0.09, p = 0.037).

The mean self-reported annual income per capita was $US 398, with males earning more than twice that of females (males $US 553, females $US 248). Ninety one percent of the sample population earned less than $US 1000 per annum. The incomes of these workers were aggregated into five income category groups to explore possible relationships between health scores and income categories. The mean PCS and MCS scores differed significantly between annual income classes (PCS df 3 p ≤ 0.001; MCS df 3 p ≤ 0.001) (Figure 2). As income increased so did the mean score for the health scales. Annual income in Uganda differed significantly with respect to the level of educational attainment (n = 437 p = 0.02). Mean annual income for those who attended primary school was $US 347 compared to $US 455 for those who attended secondary school.

The type of house occupied by workers appeared to be a function of their annual income. There were significant differences in house type dependent upon income levels (Mann Whitney n = 522, p ≤ 0.001). Those who described their house as mud and wattle (n = 204, mean annual income $US 275) had an annual income almost half that of those who owned a brick house (n = 319, mean annual income $US 475, Kruskal Wallis df = 1, p ≤ 0.001).

Bicycle ownership had a significant impact on both mental and physical scales (Mann-Whitney U PCS n = 565, p = 0.003; MCS n = 565, p = 0.015). The mean cost of a bicycle was $US 44.25. The positive effect on health scores was more marked if the respondent owned a radio (Mann-Whitney U, PCS n = 561, p ≤ 0.001; MCS n = 561, p ≤ 0.001). The mean cost of a radio was $US 22.13.

All SF-36 scale and component summary scores (with the exception of physical functioning (PF)) were significantly higher for U.K. farm workers at induction for the 2007 cohort than those recorded for workers mid-season in 2006 (Table 4). The sample was drawn from the four largest farms used in the survey of 2006. There were no significant differences in the gender or age composition for each cohort.

There is now an established relationship between income level and health status [12]. As income for a population increases so too does their health status, although at an ever diminishing marginal rate. The relationship implies that the greatest improvement in population health for a unit increase in income would be expected to accrue to the lowest income workers. The Ugandan farm worker data appears to suggest that the relationship between income and health is still positive and linear whereby a unit increase in income corresponds to an equivalent increase in good health status (Figure 2). It is interesting to note that the U.K. health scores were largely independent of income. This suggests that attempts to improve worker health through the lever of increased income are likely to be limited. Previous analysis of health drivers in the U.K. horticultural workforce identified the monotonous nature of the work and possibly workers’ material living conditions and homesickness as being primarily responsible for poor health [3]. By contrast Ugandan workers’ health appears to be much more sensitive to income.

Whilst self-reported health is commonly used to measure a sample population’s health status, it has also been frequently employed as an indicator of mortality rates [40,41]. In these studies, respondents were asked to rate their health as very good, good, fair, poor or very poor (the equivalent in the SF-36 is excellent, very good, good, fair and poor). Follow up studies suggested that those answering ‘poor’ or ‘very poor’ had a subsequent increased risk of mortality compared to those answering ‘good’ or ‘very good’. For example, in the Whitehall study Singh-Manoux et al. [41] found that 3.7% of middle-aged men and 7.1% of middle-aged women described their health as poor or very poor. Their subsequent mortality rate was 3.8 times higher over a ten year period than those who described their health as good or very good. In a further study of middle-aged British males [42], those reporting poor health had a mortality rate of 45 deaths per thousand compared with 5.5 deaths per thousand for those reporting excellent health. This equated to an eight fold increase in mortality per 1,000 per year.

In the present study, the proportion of U.K. and Ugandan farm workers aged 18 – 34 describing their health as poor or fair (the lowest two categories) was 14% and 18% respectively. If the relationship, as described by Singh-Manoux [41] and Wannamethee and Shaper [42], between an individual’s self-reported poor health and subsequent mortality rate holds, then these individuals may have an increased risk of mortality. The proportion of individuals in this category is far greater than that reported by Singh-Manoux [39] and Wannamethee and Shaper [42].

After controlling for the effects of malaria the Ugandan farm workers’ scores were significantly higher than those of the U.K. workforce and not significantly different to the U.S. population norm. Income and malaria were important explanatory variables in the multiple regression analysis. The causality appears to be bidirectional with income levels correlated with malarial infection rates and malarial incidence impacting income levels [43]. The importance of this relationship is borne out by a number of studies that suggest that there is an important financial cost incurred following malaria illness [44] which can impact upon a household’s ability to maintain living standards [21,45]. Thus, the potential of economic improvement at a local scale to positively influence the health status for farmers and farm workers in Uganda appears to be very large. This is in contrast to the expected health problems that accrue in the short-term to East European workers in the U.K.

There are a number of considerations that may need to be evaluated with regard to the ethical appropriateness of continuing to grow vegetables in the U.K., which extend beyond the ‘food miles’ debate [46]. For instance, health costs incurred as a consequence of working in the U.K. may ultimately have to be borne by the workers’ home country. In addition migrant workers may gain a financial benefit from working in the U.K. but the extent to which this compensates for a decline in health remains unexplored.

Consumers whose preferential purchasing of British produce is based upon notions of ethics may wish to consider the relevance of human health as a factor in their purchase decisions for two reasons. Firstly, when British produce is bought from a U.K. supermarket the perceived health status of a worker declines, at least in the short-term. Secondly, there are potential alternative production centres outside the U.K. meriting further consideration such as Uganda, where health may improve as a consequence of consumers purchasing. If such comparative poor health is due to the U.K. working conditions then the ethical permissibility of such a system needs to be tackled on a number of fronts. For instance, ‘local food’ proponents may wish to consider how increased local food production at a local scale may impact the health of workers in the local food chain. Furthermore, managers on horticultural farms will need to respond to the fact that worker health declines during the season. Attempts to redress the problem may require the adoption of new working practices that can counteract these effects, such as offering a diversity of tasks in the workplace, as well as improving the employees’ social environment.

It is against this backdrop that the topic of farm worker health is becoming increasingly prominent in food production debates particularly as movements such as Fairtrade and IFOAM re-evaluate concepts of agricultural social justice. The factoring in of the costs of poor health to the farmer and society as a whole needs to be considered by policy makers for both the short and long-term.

Long-term health costs may be difficult to detect, particularly for those workers who return to their home country and receive medical care at a later date. Costs may be incurred by the donor country and the extent to which this would be morally acceptable remains unexplored. At a European level the cost of palliative care in one country may be compensated for by the health benefits derived from increased vegetable consumption in another country. For instance, if a Polish worker experiences a decrease in health status of one unit for every ten thousand lettuces that he or she harvests, consumers’ health status may need to increase by an equivalent amount through the consumption of lettuce in order to balance the population health status.

Further research should assess the costs and benefits that accrue to migrant workers working in U.K. horticulture, and how these impact their health following their return to their homeland. In addition it would be useful to work with farm managers in order to alter work patterns such that worker health improves. Meanwhile in developing countries considerable work is needed to fully understand the relationship between income, participation in the export market and farmer health.