Data were collected by way of triangulation through farmer interviews, field observations during the farmer interviews, and a survey of key informants. However, farmer interviews were considered the primary source of data for this study. The farmer interviews and field observations were conducted in four districts of central Uganda (Mpigi, Masaka, Luwero, and Wakiso) and one district in eastern Uganda (Kamuli) (Figure 1). These districts were selected based on data collected on clinical and pathologic lesions associated with ASF from pigs sampled from abattoirs around Kampala, Uganda. Districts with the highest number of pigs with splenic hemorrhages which is the most characteristic lesion of acute ASF (20) were selected. Convenience and purposive sampling were used to identify 19 to 20 respondents per district and sampling until saturation was used to identify trends (21, 22). The pig farmers who participated in the study were identified by the district veterinary officers (DVOs) and/or government animal health workers in each district with the intent of providing diversity in farmer gender, farm management styles, and farm size. The key informants were identified by the research team based at Makerere University and by the DVOs.

Questionnaires were developed for the farmers and key informants, and a field observation checklist was developed as well. The farmer questionnaire and the key informant’s questionnaire were semi-structured and captured information on the farmer’s production practices, biosecurity practices, knowledge on ASF, previous experience with ASF, as well as impact of ASF on the farm, including mortality, and herd characteristics. Regarding previous ASF incursions and mortality, a simplified description of the clinical signs of ASF was provided to the respondent before the questions on ASF occurrence and the associated mortalities were asked. The questions on biosecurity were aligned with the good practices for pig biosecurity in developing and transition countries as described by the Food and Agriculture Organization (FAO) (23) and where applicable, as described by the University of Ghent Biocheck on-farm biosecurity assessment system (24).

The farmer questionnaire was first developed by the research team, and reviewed by two epidemiologists at Cornell University, and 11 experts on African swine fever in Uganda to ensure the questions were complete and response option adequate. Next, seven pig farmers in the central region, but not in districts enrolled in the study, pre-tested the farmer questionnaire and provided feedback on clarity. The questionnaire was modified based on the feedback from the pre-testing. The pre-tested farmer questionnaire was sent to the Makerere University Institute of Languages for translation from English into Luganda and Lusoga, which are the common native languages spoken in the study districts. The translated questionnaires were examined for validity of the translations and later back translated into English by a second individual who had not seen the original questionnaire to check for the accuracy in the translations. All the versions of the questionnaire (English, Luganda, and Lusoga) were built into KoBo Toolbox (Kobo Organization, Cambridge, MA, United States), and later downloaded into the KoBo collect app (Kobo Organization, Cambridge, MA, United States) for administration using computer tablets. The captured responses in the tablets were then uploaded to the KoBo online repository. The farmer questionnaire in English is provided as Supplementary File 1. The Luganda and Lusoga versions of this questionnaire will be made available upon request from the corresponding author.

The key informants’ questionnaire and the field observation checklist were built based on the farmer questionnaire to enable meaningful triangulation. The key informants’ questionnaire was also sent to a team of ASF experts in Uganda for pre-testing and modified to reflect the feedback from the pre-testing (Supplementary File 2). The observation checklist (Supplementary File 3) was modified after administration in the first district surveyed (Supplementary File 4). The key informants’ questionnaire was administered via Qualtrics (Qualtrics, Provo, UT, United States) for a period of 6 weeks (August 8 through September 22, 2022), and a reminder was sent to non-respondents every 2 weeks. The field observation checklist was also built into KoBo Toolbox and the data was collected using the KoBo Collect app in the data collection tablets.

The farmer interviews and field observations were conducted in the months of June and July 2022. The farmers were interviewed by two field research assistants from Makerere University in Uganda, and the field biosecurity observations were recorded by two field research assistants from Cornell University. All the field research assistants were trained prior to the onset of field data collection to ensure uniformity in the farmer interviews and field observations. Following the interviewer training, the farmer interviewers practiced administering the questionnaire. The interviews and the field observations were overseen by the primary author (JE). A biosecurity protocol was followed by the field data collection team during the farm visits. Each member of the field research team wore disposable boot covers between farms and applied hand sanitizers following each farm visit to ensure that the team did not disseminate pathogens between farms. The farmer interviews were conducted at the respondents’ homesteads/farms in a language that the respondent preferred. Responses captured in Luganda and Lusoga were later back translated into English. All the respondents consented to the interviews and field observation and each respondent was offered 20,000 Ugandan shillings (approximately USD 6 at the time) to reimburse for their time.

For each biosecurity question, the risk was determined using a method previously described (25) and was as follows. First an overall value was assigned to the biosecurity measure the question asked about. Then, a weight was assigned to each of the response options that the farmer could choose for each question. A pig farmer who practiced good biosecurity scored all or almost all the points allocated to that specific question while one who did not, scored zero points. For binary response options, the farmer got all the points if the specific biosecurity aspect was practiced. However, responses measured on a Likert scale were weighted on a scale of 0 to 1, in increments of 0.25 as previously described (25). The farmer’s final scores for each question were obtained from the product of the selected answer weight and the biosecurity question value. For the Likert scale responses, a maximum score of one and a minimum score zero was deemed untenable for some biosecurity aspects depending on how the question was structured. For example, some questions had response options structured as “no, rarely, sometimes, and almost all of the time,” but with no “all of the time” option. Once all questions were scored, the sum of the scores was calculated and converted into a percentage. A biosecurity score of >90% was categorized as excellent, >80–90% as good, >70–80% as fair, and ≤ 70% as poor (25).

The values of different biosecurity measures were based on the Biocheck biosecurity scoring algorithm used at the University of Ghent and previously described (24). However, some biosecurity questions used in this study’s questionnaire either did not properly align or were not included in Ghent Biocheck scoring algorithm. As a result, the values provided by the Biocheck tool had to be validated for Ugandan pig farmers in the assessed districts. A total of 15 veterinarians evaluated the biosecurity value scheme and provided their expert opinion on the values or suggested alternative values for each biosecurity aspect in the questionnaire. Three members of the research team (JE, EM, and KH) were among the 15 veterinarians who evaluated the algorithm. The other 12 were practicing veterinarians in Uganda and included two veterinarians from each of the five districts surveyed and two academic veterinarians at Makerere University. The practicing veterinarians were selected based on their experience working with farmers in the study districts, and their knowledge of pig production and biosecurity in the selected districts. The 15 veterinarians were requested to either agree or disagree with the biosecurity measure value in the questionnaire as assigned by the Biocheck survey, and to assign a new value if they disagreed. They were also asked to assign a biosecurity measure value to aspects without any assigned values in the Biocheck tool. They provided an alternative value with a maximum score greater than zero and less than or equal to 10. The responses from all 15 veterinarians for each biosecurity aspect were summarized in Excel version 2303 Build 16.0.16227.20202 (Microsoft, Redmond, WA, United States) and the modal value for each biosecurity aspect was used in the final biosecurity scoring as it represented the most popular expert opinion. A summary of the modal values used to score farmer’s biosecurity is provided in Supplementary Table 1.

The data from the farmer interviews and field observations were downloaded from KoBo into Excel and collated. The key informant survey data was downloaded from Qualtrics into Excel as well. The data were analyzed using commercial statistical software (SAS, version 9.4, SAS Institute Inc., Cary, NC, United States). Data on common diseases affecting the farmers’ pigs was captured as free text and was analyzed qualitatively into two categories, farmers who mentioned ASF and/or at least one clinical sign of ASF, and those who did not. Categorical data were then summarized using frequencies and proportions, and confidence intervals for the proportions were calculated using the Agresti-Coull method (26). The Pearson chi-squared test was used to explore the association between district, years involved in pig farming, pig husbandry type, pig breed type, herd size, and source of farm labor with self-reporting of ASF in the past 12 months. Respondents who were not sure if their pigs had ASF in the past 12 months were excluded from the analysis. If the chi-squared test assumptions were violated, the Fisher’s exact test was used. Multiple comparisons of proportions following a significant chi-squared or Fisher’s exact test was conducted using a post hoc Tukey-style multiple comparisons test of proportions using the COMPPROP SAS macro (27).

Biosecurity scores were tested for normality using a visual assessment of a histogram and the Shapiro–Wilk test and they showed the data were normally distributed (p = 0.729). The mean, standard deviation, and range of the biosecurity scores were calculated. The one-way analysis of variance (one-way ANOVA) test was used to compare mean biosecurity scores between different groups of farmers by pig production type (farmers that use confinement in corrals versus all other husbandry systems), pig type (farmers who raised local breed pigs, mixed breed types, exotic breed pigs, and all other types), and between districts (Mpigi, Masaka, Luwero, Kamuli, and Wakiso). Levene’s test was used to check for the assumptions of homogeneity of variances. If the assumption of homogeneity of variances was not met, Welch’s ANOVA was used. Multiple pairwise comparisons were done if the ANOVA was significant and used a Tukey’s studentized range (HSD) test to control for type 1 experiment-wise error rate. The level of significance for all statistical tests was 0.05.

Overall, a total of 99 pig farmers were interviewed in the five districts and results are summarized in Table 1. Nineteen respondents were interviewed in Mpigi, and 20 in each of the districts of Masaka, Luwero, Kamuli and Wakiso. Overall, more males (56.6%, 56) were interviewed when compared to females (43.4%, 43), 39.8% (28) of farmers had spent five or less years raising pigs, and 63.6% (63) of farm labor was provided by family members.

Among farm characteristics, 80.5% (80/99) of farmers across all districts confined their pigs in corrals. Mpigi had the greatest diversity in production systems with 63.2% (12/19) confining pigs, 21% (4/19) confining and tethering pigs, 10.5% (2/19) using confinement and free-range, and 5.3% (1/19) using tethering and free range. Farmers raised a variety of pig types as well. The three major pig types included exotic or European breed pigs, local Ugandan breed of pigs, and mixed breeds of the two. Farmers reported raising one type, two different types or all three types. In all districts, 33.3% (33/99) of the farmers raised mixed breed pigs and 30% (30/99) raised exotic pigs (Table 1). Mpigi farmers differed, the majority of them raised local breed pigs (31.9%; 6/19). There was diversity in the size of farm, but 39.4% (30.99) of farmers raised 4 to 11 pigs, followed by 19.2% (19/99) that raised 12 to 20, 16.2% (16/99) that raised 21 to 30, and 15.2% (15/99) that raised >50. Less than 5% raised 1 to 3, 31 to 40, and 41 to 50 pigs. Masaka and Mpigi had more farmers raising >50 pigs, 25% (5/20) and 20% (4/20), respectively.

When asked about the common diseases affecting their pigs, 82.8% of the 99 farmers interviewed mentioned ASF and/or at least one clinical sign of ASF, 12.1% (12) did not mention ASF and/or at least one clinical sign of ASF, 4% (4) reported they had not experienced disease on their farms, and 1% (1) did not know common diseases on their farm. There were 84 (84.9%) farmers that stated they had knowledge about a disease of pigs called ASF. When asked about ASF occurrence, 42.7% (41/96) reported having ASF sick pigs in the past 12 months, and 3% (3/99) were unsure if their pigs had ASF in the past 12 months.

There were 41 farmers who reported having pigs with ASF. Among them, 85.4% (29) said that some or all the pigs with ASF died. Four out of 19 (21.1%) farmers in Mpigi reported having pigs with ASF in the past 12 months, 45% (9/20) in Masaka, 50% (10/20) in Luwero, 30% (6/20) in Kamuli, and 60% (12/20) in Wakiso. Significant omnibus associations were found between reporting ASF on farms in the past 12 months and farming experience (p = 0.0083) as well as herd size (p < 0.0001), but no post-hoc pairwise significant differences were found.

The summary of field observations had to be different for Mpigi compared to all the other districts, because it was necessary to modify the observational checklist while in Mpigi, the first district to be surveyed. The observations were able to support that on the 19 farms visited in Mpigi, it was observed that human food leftovers were in feeding troughs in 31.6% (6) of the farms and 50% (3/6) had meat scraps. Livestock (57.9%; 11), dogs and cats (26.3%; 5), as well as flies and rodents (89.5%; 17) had some contact with the pigs, as did some visitors that came by (15.8%; 3). It was observed that the majority of pig pens appeared clean (89.5%; 17). It appeared that 73.7% (14) spread pig manure to nearby crop fields or gardens, and no farms had a hand washing facility near the area where pigs were kept. No wild pigs or Ornithodoros species of tick were seen either. A detailed breakdown of the biosecurity aspects observed by the research team in Mpigi District is provided in Supplementary Table 5.

For the rest of the four districts (Masaka, Luwero, Kamuli, and Wakiso), human food leftovers were observed in 11.4% (9/79) of the farms visited, but none were observed in Kamuli or Wakiso. Only one farm in Luwero (11.1%) of the nine farms with food leftovers had meat scraps. There was a fence on 23.8% of farms (19) and 89.5% (17/19) surrounded the entire property. Hand washing (31.6%; 6/19) and farm-specific clothing (5.3%; 1/19) were most observed in Masaka, and no farm-specific footwear was seen in any district. All districts had some observations of cats and dogs mixing with pigs (61.5%; 49/80); poultry mingling with pigs (78.8%; 63/80), and other livestock mingling with pigs (35%; 28/80). Thirty-six out of 80 (45%) farms visited spread pig manure to nearby crop fields or gardens, and this was observed on some farms from every district. Thirty-three out of 79 farms (41.8%) had an enclosure or area that could be dedicated for quarantine of new pigs. There were some visitors seen to have contact with pigs as well (3.8%; 3/80), but it was not commonly observed (80.5%; 70/80). A detailed description of the field observations in the districts of Masaka, Luwero, Mpigi, and Wakiso is provided in Supplementary Tables 6A,B.

Of the 54 invited key informants, 35.2% (19/54) took the survey and completed some or all the questions in the questionnaire. In total, 84.2% (16/19) were local government employees, one was an employee of the agriculture ministry, one worked in academia, and the other worked in the private sector. Of the 16 local government key informants, 81.3% (13) worked in Wakiso district, 12.5% (2) were from Masaka, and 6.2% (1) was from Luwero. There were no key informant respondents from the districts of Mpigi or Kamuli (Supplementary Table 7).

Key informants reported on similar biosecurity methods asked about in the farmer questionnaires. As for external biosecurity, 63.2% (12/19) of key informants reported that only 1 to 25% of the farmers they interacted with quarantined new animals before their introduction to other pigs on the farm. Only 11.1% (2/18) of key informants reported that it was very common for farmers to involve animal health workers in evaluating the pre-acquisition health status of pigs, 44.4% (8/18) reported that this practice was both common and not common. There was 68.4% (13/19) of key informants that reported that use of household waste as feed was very common among pig farmers and 33.3% (6/18) mentioned that it was very common for farmers to feed their pigs on restaurant waste. The most common response among key informants was that 1–25% of farmers boiled swill before feeding it to their pigs (38.9%;7/18), and the next most common was that no farmers boiled swill before feeding it to pigs (27.8%; 5/18). Most key informants also reported that no farmers or 1–25% of farmers used or had visitors wear farm specific clothing or shoes. The most commonly reported visitors were the animal health worker (72.2%; 13/18) and the pig buyer (50%; 9/18). Among key informants, 50% (9/18) said it was common for farmers to isolate and treat sick pigs, 70.6% (12/18) said it was common for farmers to consume pork from sick pigs, and 50% (9/18) said it was common for farmers to sell pork from sick pigs. Most key informants also stated it was very common for farmers to sell off sick pigs (50%; 9/18). The majority of the key informants (77.8%; 14/18) mentioned that just 1–25% of pig farmers provided footbaths with disinfectant at the entrance to pig housing. Supplementary Tables 8A,B provide a detailed description of the key informant’s answers regarding pig farmers’ biosecurity practices.

Table 5 provides a summary of the biosecurity mean scores of farmers. Overall, the biosecurity scores for the respondents were considered poor with 98/99 (99%) scoring <70%. The mean biosecurity score for all the respondents was 49% (standard deviation (SD) = 8.5%, range: 29.3–72.2%). There was a statistically significant difference in biosecurity scores between districts (p = 0.0481). The mean biosecurity score for Kamuli district of 52.6% (SD = 6.5%, range: 41.9–63.8%) was significantly higher than for Luwero (44.9%; SD = 6.2%, range: 33.6–57.3%) with a difference in means of 7.72% (95% confidence interval (95%CI): 0.46–14.99%).

The statistical differences found include the following. The mean scores among farmers that used confinement compared to other husbandry systems were significantly different (p = 0.014). Confinement had a score of 50.1% (SD = 8.5%: 29.3–72.2%) and all other husbandry systems had a score of 44.8 (SD = 7.2%, range: 29.7–60.1%) with a difference in means of 5.3% (95%CI: 1.1–9.5%). For pig breed type, there was a statistical difference in scores between farmers that raised only exotic breed and those that raised only local breeds (p = 0.004). Farmers who raised only exotic pigs had a significantly higher mean score (52.1%; SD = 8.6%, range: 37.4–72.2%) when compared to those who raised local breed pigs (43.6%; SD = 7.2% range: 29.7–60.1%) with a difference in means of 8.5% (95%CI: 1.5–15.5%). Similarly, farmers who raised mixed breed pigs had significantly higher biosecurity mean score (50.5%; SD = 7.5%, range: 37.9–64.2%) when compared to those who raised local pigs giving a difference in means of 7% (95%CI: 0.06–13.9%).