Mbale is a Ugandan district situated on the volcanic slopes of Mount Elgon. The high altitude, fertile volcanic soil, and a humid tropical climate provide ideal conditions for the cultivation of Arabica coffee, which can fetch higher export prices compared to Robusta varieties due to its higher quality [20, 21]. Smallholder farming communities predominantly populate the region in scattered homesteads, managing mixed agroforestry systems on plots typically smaller than 1 ha [22, 23]. For these farming families, selling green coffee is often their primary source of cash income. However, with limited resources to buffer against fluctuations in harvests, many smallholder farmers across Uganda and in the Mount Elgon area are in an economically precarious situation [19].

In a first step we carried out a bibliographic review to identify all occupational hazards in agriculture and specifically in coffee [2–4, 11, 18, 24–34] followed by a rapid assessment in the field, resulting in a list of 80 possible hazards items.

The hazard questionnaire was then pre-tested with three local experts (Figure 1). Modifications based on that local feedback included the exclusion of hazards deemed not applicable to the local context, such as those associated with the use of large tools and machinery (e.g., gas bottles, vibrating tools) which are not commonly used in the region, as well as hazards unrelated to the type of work, such as “shift work” and “long commuting distances”. Several hazards were merged to improve clarity, for example, on weather perception (e.g., cold and rainy weather, sunshine and heat). In addition, local examples were added for specific hazards to enhance understanding and illustrate the hazard, e.g., “with knife, panga machete, hand-hoe, nails” for the physical hazard “accident with a sharp tool”. This resulted in a final list comprising 59 hazard items, (H_overall = 59) grouped into the six categories “biological” (h_bio = 10), “chemical”, (h_chem = 8), “ergonomic” (h_ergo = 5), “physical” (h_phys = 8), “environmental” (h_env = 8), and psychosocial (h_psycho = 20) (Figure 1).

The tool was applied as part of a farmer cohort study in the Mbale region (FarmCoUganda). Following the compilation of the occupational hazard list, we conducted a one-day in-person workshop with local stakeholders using a modified two-stage Delphi method. The goal was to prioritize and refine the long list of occupational hazards into a locally grounded set of the most relevant risks for coffee farmers in the Mbale district.

To develop a concise and relevant list of occupational hazards for use in a future cohort study, we applied a structured approach to prioritization. The goal was to reduce the number of hazards to a manageable set (no more than 25) while retaining those most relevant to the local farming context. This reduction was necessary to minimize the burden on study participants, many of whom have limited experience with surveys and low education level, and to focus the assessment on hazards that are both significant and recognizable.

To determine which hazards to retain, we applied four predefined criteria. A hazard was shortlisted if it met at least two of the following (Figure 1):

Mean risk score in Round 1 exceeded 12.

For Round 2A, average rankings within each category were calculated across the six discussion groups. In Round 2B, hazards were selected based on group consensus, and the number of groups selecting each hazard, along with the average rank across those groups, was recorded. Hazards not selected in Round 2B were not assigned a ranking.

This combined analytic and prioritization process ensured that the final list of hazards reflected both individual assessments and group-level consensus, balancing statistical rigor with practical relevance for smallholder coffee farmers in the region.

All Round 1 questionnaire responses were collected digitally using Open Data Kit (ODK) and analyzed in RStudio (version 2024.09.0). Descriptive statistics (counts, means, and standard deviations) were used to summarize participant characteristics. For each hazard, a risk score was calculated per participant by multiplying the likelihood and severity ratings. Mean and standard deviation values for likelihood, severity, and risk were then computed for each hazard and across all participants.

Out of 34 stakeholders contacted, 31 participated in the workshop (91%). The group included agricultural extension officers, public and occupational health professionals, representatives from the coffee sector and NGOs, as well as smallholder farmer leaders. The mean age was 44 years, and 42% were female (Table 1).

The likelihood of occupational hazards, given as means on a Likert scale from 1 (rare) to 5 (almost certain) and standard deviations, ranged from 2.4 ± 1.2 (“exposure to forest fire”/“attack by a wild animal”) to 4.3 ± 0.98 (“prolonged bending”). Severity scores, also given as means on a Likert scale from 1 (no harm) to 5 (potentially fatal) with standard deviations, varied from 2.4 ± 1.1 (“contact with a disinfectant”) to 4.3 ± 0.9 (“contact with exposed electrical connections”).

Calculated risk values, integrating likelihood and severity, spanned from 6.7 ± 4.7 (“tick bites”) to 16.5 ± 4.6 (“contact with a pesticide” from the chemical category) (Figure 2, Supplementary Material 1). The hazards presented in the upper part of Figure 2 were shortlisted as they met the cutoff criteria for Round 1, i.e., had a calculated risk value exceeding 12, and ranked among the top three per category based on risk values. The second criterion added hazards in the psychosocial and environmental category. With 13.0 ± 4.1, the “mosquito bites” showed the lowest standard deviation of all hazards, while “high legal compliance requirements” showed the highest standard deviation (13.1 ± 6.9).

During Round 2A the environmental and psychosocial categories were subject to extensive debate, leading to considerable rearrangement of hazard priorities compared to the individual ranking in Round 1 (Figure 3). As a result, and contrary to the ranking in Round 1 (criterion 2), the hazards “working in heavy rains” and “working in a cold environment” from the environmental category were prioritized in Round 2A. These additional hazards had also already met criterion 1 (risk >12 in Round 1). The same applies to “unexpected events”, “failure of coffee crop” and “limited access to farm inputs” in the psychosocial category (Figure 3).

In contrast, there was broad agreement in the other categories (biological, chemical, ergonomic, physical), and the top three hazards remained largely unchanged. “Repetitive movements” in the ergonomic category fulfilled criterion 3, but not shortlisted as it did not fulfil the other criteria (Figure 4).

Results from Round 2B (criteria 4) confirmed the prioritization established in Round 1 and Round 2A. Applying the initial rule that at least two cutoff criteria had to be met for a hazard to be shortlisted, a total of 23 hazards were selected (Figure 4).

Stakeholders prioritized hazards across six categories, selecting three biological, three chemical, three ergonomic, and three physical hazards, as well as five environmental and six psychosocial hazards. The hazard “contact with a pesticide”, e.g., during mixing, loading, transport, storage, application, or disposal, was consistently ranked as the top hazard across all assessment rounds (Figures 2, 4, Supplementary Material 1). The related hazard, “re-entry into sprayed areas,” was also ranked highly. This is consistent with broader concerns about pesticide exposure in LMICs: in Uganda’s tropical climate, coffee growing is affected by numerous biotic constraints, including insect pests, diseases, weeds, and nematodes [40, 41]. Farmers therefore rely on synthetic pesticides as their primary, and sometimes only, means of crop pest and disease control. However, recommended safety precautions are often not followed [6, 7, 42–47]. In addition, some of the pesticide products used may contain active ingredients that are internationally banned, although evidence for this in Uganda is mixed, compared with more consistent findings reported from other developing countries [45, 47–49].

The extent of pesticide use among Ugandan smallholder farmers, however, varies considerably across studies and regions. Fuhrimann et al., for example, reported that 72% of smallholder farmers in a study in Wakiso district in Central Uganda used pesticide products that posed substantial health risks [50]. By contrast, Kagezi et al. concluded that pesticide use was generally limited in Mount Elgon region, with only 19% of coffee farmers using pesticides [49]. The contrasting findings and the high likelihood score assigned by stakeholders in our workshop (4.3 ± 0.8, corresponding to “likely to almost certain”) suggests either substantial regional variation in pesticide use or a possible overestimation of regional pesticide spraying for Arabica coffee by workshop participants possibly influenced by their experience with other crops and regions. Given the mixed farming systems common in the region, it also remains unclear to what extent pesticide use on non-coffee crops may have shaped the expert assessment.

Information on the specific pesticide active ingredients used in Ugandan smallholder systems remains limited and fragmented. Available studies indicate the use of the herbicide glyphosate, the fungicide mancozeb, and the insecticides cypermethrin [45, 51, 52] and fenitrothion [53]. More recently, Uganda’s Ministry of Agriculture, Animal Industry and Fisheries (MAAIF) has severely restricted the use of the insecticides imidacloprid and thiamethoxam on coffee, reflecting safety concerns [54]. Glyphosate and cypermethrin, in particular, have been associated with measurable internal exposure among farmers in urinary biomarker studies, and have been linked to adverse health outcomes, including sleep disturbances [51, 52]. Most pesticide active ingredients used by Ugandan farmers fall into WHO hazard class II, indicating moderate toxicity [45].

“Exposure to landslides” (15.4 ± 6.3) stood out in Round 1 due to its particularly high severity score (4.2 ± 0.9) (Figure 2). This is consistent with the Mount Elgon region being one of the most landslide-prone areas in Africa, where landslides contribute to fatalities, casualties, land loss, soil degradation, damage to property and infrastructure, and reduced agricultural productivity [55–57].

“Mosquito bites” was rated as the highest biological risk (13.0 ± 4.1), likely reflecting both farmers’ frequent outdoor exposure to mosquitoes, and the country’s high malaria burden. The relatively low standard deviation suggests strong agreement among workshop participants, possibly because malaria and its health consequences are widely recognized through public health campaigns like the Uganda National Malaria Control Program (UNMCP) [58].

Ergonomic hazards received the highest ratings among all hazard categories in Round 1 (13.4 ± 5.5 in Figure 3). This is not surprising, given that agricultural workers in LMICs frequently experience musculoskeletal disorders linked to physically demanding occupational tasks [59]. In coffee production, these risks are particularly prevalent during harvest, when workers perform repetitive movements in awkward postures and carry heavy bags of coffee berries, contributing to pain symptoms [59–63]. The literature also highlights wider consequences of these issues, including absenteeism, increased production costs, and long-term reductions in productivity [64, 65]. However, assistive equipment for coffee harvesting is currently limited. One promising solution is an ergonomic redesign of the collection bag [60, 66]. Mechanized harvesting alternatives, such as vibrating machines used for other crops like cherries and olives, could further reduce the ergonomic strain. However, these approaches face substantial technological and financial challenges. Key obstacles include the difficulty of distinguishing ripe from unripe fruit [67], the operation of heavy equipment on steep and uneven terrain without causing soil damage, dependence on fuel or electricity [68], and prohibitively high initial costs.

Psychosocial factors such as stress, fatigue, and long working hours may further contribute to pain symptoms alongside mechanical strain [59, 69]. The workshop discussions around health impacts in the psychosocial category (Figure 3), might suggest that there is a lack of awareness and understanding around causes and indirect health consequences of stress, and that these risks therefore remain insufficiently understood.

A strength of this study is the participation of 31 stakeholders in the workshop, which constitutes an adequate panel size [37, 39]. Another notable strength is the integration of gender perspectives in a context where coffee production is typically male dominated. Previous studies conducted with coffee farmers on the northern slopes of Mount Elgon have reported a share of female coffee farmers ranging from 10% to 47% [22, 49, 70]. In contrast, female stakeholders comprised 42% of participants in our workshop. However, among these female participants, direct experience with practical coffee cultivation was notably lower than that of their male counterparts (Table 1).

On the other hand, there may be several limitations in the assessment related to how the hazards and their consequences are interpreted by the stakeholders in the workshop. These limitations arise from differing conceptual understandings in four key areas: (1) single events versus repeated exposure, (2) permanent versus reversible health effects, (3) short-term versus long-term health impacts, and (4) averaging all possible health impacts on a Likert scale. We attempted to address the first point and improve the accuracy of likelihood ratings by carefully introducing the topic and providing illustrative examples for each hazard. However, severity assessments may still be influenced by how stakeholders perceive the reversibility and duration of effects and how they mentally average these impacts. The same applies to the overall risk concept applied in Round 2, which encompasses both likelihood and severity.

To further verify the assessment, we need to ask farmers directly about their exposures, for example, regarding pesticide use intensity and products, in the subsequent cohort study, both for Arabica coffee and other crops.

We developed TIERRA, a practical and context-sensitive tool designed for prioritizing occupational risks in agriculture. This tool was successfully applied in the context of smallholder coffee farming in Uganda (FarmCoUganda), demonstrating its relevance in a real-world, resource-constrained setting.

Our participatory approach offers several advantages for occupational risk assessment in agriculture. It fosters local stakeholder ownership, supports the co-creation of knowledge, and encourages dialogue for action. Importantly, it ensures that the identified priorities are based on the lived experiences of farmers and reflect specific local conditions. Additionally, TIERRA allows for the development of tailored farmer surveys that focus only on relevant hazards, thereby reducing the burden on respondents. This makes data collection among farmers more focused and feasible in terms of time and logistics. Ultimately, improving the identification of critical occupational risks lays the foundation for evidence-based interventions and training programs that align with local realities and contribute to safer, healthier, and more sustainable agricultural livelihoods.

Our approach provides a methodological blueprint that can be easily adapted to other agricultural contexts, including resource-constrained settings in LMICs. The structured prioritization criteria offer a consistent framework for cross-context comparison, supporting broader efforts toward cleaner and more sustainable agricultural production.