Bayesian Estimation of True Prevalence and Associated Risk Factors for Leptospira spp. Among Slaughterhouse Workers and Slaughtered Cattle in the Bahr El Ghazal Region of South Sudan

ABSTRACT

Introduction

Leptospirosis is a major but under‐reported zoonotic disease, and epidemiological data from South Sudan remain limited. This study estimated the true prevalence of Leptospira spp. exposure and identified associated risk factors among slaughterhouse workers and slaughtered cattle in Western Bahr El Ghazal.

Methods

A cross‐sectional study was conducted in major slaughterhouses. Serum samples from workers and cattle were tested using the microscopic agglutination test (MAT). Bayesian hierarchical models were used to adjust for diagnostic test imperfections and to estimate true prevalence. Structured questionnaires captured occupational and animal‐level risk factors for analysis within the Bayesian framework.

Results

The estimated true prevalence was 10% in slaughterhouse workers and 85% in slaughtered cattle, indicating a high zoonotic exposure risk. Among workers, flaying, inconsistent use of protective equipment, and handling higher numbers of carcasses per day were significantly associated with seropositivity. In slaughtered cattle, exposure varied by breed, age, and sex. The model further indicated a 78% probability that a randomly selected slaughterhouse was affected and a 65% probability that infection levels among workers remained below 5%.

Conclusions

This study provides the first Bayesian‐based estimates of leptospiral exposure in slaughterhouse settings in Western Bahr El Ghazal. The findings underscore the need to improve occupational safety, strengthen surveillance, and apply One Health approaches to reduce zoonotic transmission. Despite limitations, including lack of environmental data, the Bayesian framework proved effective for generating robust prevalence estimates in a resource‐limited setting. Expanded geographic coverage and incorporation of environmental assessments are recommended to inform targeted leptospirosis control strategies.

Impacts

• Leptospirosis is a globally emerging zoonotic disease causing significant public health risks and economic losses, particularly in livestock‐dependent communities.

• Bayesian hierarchical modelling addressed the lack of a gold‐standard test, enhancing diagnostic accuracy for leptospirosis.

• The model informs resource allocation and prioritises interventions, supporting decision‐making in public health and veterinary sectors.

1. Introduction

Bayesian inference provides a flexible and robust framework for tackling complex diagnostic challenges, particularly by integrating prior knowledge with observed data (Gardner Gardner ²⁰⁰⁴). Accurate diagnostic testing is critical for detecting disease presence and assessing exposure in livestock populations (Moons, Biesheuvel, and Grobbee Moons et al. ²⁰⁰⁴). In prevalence surveys, the true prevalence must often be inferred from the apparent prevalence by adjusting for diagnostic sensitivity and specificity (Messam et al. 2008). Variability in these parameters can significantly alter prevalence estimates, highlighting the need for precise diagnostic performance metrics. In the case of leptospirosis, diagnostic tools demonstrate varying performance levels, complicating direct comparisons. For instance, the true prevalence of Leptospira spp. in slaughtered cattle is not directly observable but must be inferred through adjustments based on diagnostic test characteristics (Messam et al. 2008).

Bayesian methods have several desirable attributes, including the use of prior distributions to integrate existing knowledge, which enhances estimation accuracy, especially in small sample sizes or complex scenarios (Finch 2024). By incorporating expert opinions or historical data, Bayesian inference quantifies uncertainty more effectively, yielding robust conclusions (Bayman, Oleson, and Dexter Bayman et al. ²⁰²⁴). Unlike traditional null hypothesis significance testing (NHST) frequentist approaches, which rely exclusively on data from the current study, Bayesian methods provide a probabilistic framework for assessing both the null and alternative hypotheses by incorporating observed data and prior knowledge to estimate the likelihood of each. In contrast, frequentist approaches rely on hypothesis testing to determine whether to reject the null hypothesis, without quantifying the probability that either hypothesis is true (Bayman et al. 2024; Colling and Szűcs ²⁰²¹; Fornacon‐Wood et al. ²⁰²²). While Bayesian methods offer advantages in settings with uncertainty, small sample sizes, or complex dependencies, frequentist approaches remain robust and valid in well‐powered studies with minimal uncertainty (Smid et al. 2020).

Prior distributions are probability distributions on model parameters that specify beliefs about the relative plausibility of parameter values before seeing the data (Wagenmakers et al. 2018). These prior beliefs can then be updated with the data to obtain the posterior beliefs about the model parameters, following Bayes' rule (Jeffreys 1961).

This study utilises Bayesian inference to estimate the true prevalence of pathogenic Leptospira spp. among slaughterhouse workers and slaughtered cattle in the Bahr El Ghazal region of South Sudan. It also reanalyzes existing data using Bayesian methods to reassess previous findings (Onafruo, Dreyfus, et al. ²⁰²⁴; Onafruo, Klein, et al. ²⁰²⁴) and to confirm risk factors associated with leptospiral positivity, aiming to build upon earlier conclusions through a different inferential perspective. In the context of South Sudan, where basic diagnostic infrastructure is severely limited, Bayesian methods offer a valuable alternative for addressing diagnostic challenges. This approach is particularly critical for zoonotic diseases like leptospirosis, which has been shown to have a high prevalence in both slaughtered cattle and slaughterhouse workers (Onafruo, Dreyfus, et al. ²⁰²⁴; Onafruo, Klein, et al. ²⁰²⁴). By incorporating prior knowledge from global and local epidemiological studies alongside current data, Bayesian methods can improve diagnostic accuracy and inform resource allocation in settings where conventional diagnostic tools are inaccessible or unreliable.

2. Materials and Methods

2.1. Target Population, Sampling, Testing and Data Collection

The data for this methodological study were derived from our cross‐sectional research conducted on slaughterhouse workers and slaughtered cattle in the Bahr El Ghazal region of South Sudan. Samples were collected between January and February 2023. Human blood sampling procedures, techniques, and testing followed the methods previously described (Onafruo, Dreyfus, et al. ²⁰²⁴).

In brief, approximately 5 mL of venous blood was collected from consenting slaughterhouse workers, processed to obtain serum, and stored at −20°C. Samples were transported under cold chain conditions to Makerere University's Central Diagnostic Laboratory in Uganda, ensuring integrity throughout collection, storage, and transport. Blood and urine samples were collected from slaughtered cattle at Lokoloko municipal slaughterhouse in Western Bahr El‐Ghazal State. Approximately 10 mL of blood was aseptically collected into gel‐activated tubes during bleeding, and 20 mL of urine was extracted from the urinary bladder post‐flaying using a sterile syringe and needle. Samples were labelled with unique animal IDs for traceability and immediately stored on ice to preserve integrity.

The slaughtered cattle were sourced by traders from the main livestock auction kraal on the Eastern Bank of Jur River, a central hub for livestock from across the Bahr El Ghazal region, including Northern Bahr El‐Ghazal, Western Bahr El‐Ghazal and Warrap States. Animals are initially gathered at the auction kraal before being purchased by traders and transported to different slaughterhouses. Lokoloko slaughterhouse was selected for its high slaughter volume, processing an estimated 47–50 cattle daily, or approximately 1500 monthly, making it the largest facility in the region. From an average of 50 cattle slaughtered daily, up to 13 animals were sampled per day over 1 month using a systematic random sampling method. Sampling commenced with one randomly selected animal from each of the four slaughtering lines, after which every fourth animal was systematically selected.

During slaughter, cattle were restrained on designated slaughter lines, enabling systematic sample collection. Restraining was solely carried out by slaughterhouse workers, during which animal welfare protocols were observed. This process was integral to ensuring ethical compliance during sampling. Individual animals were sampled, representing diverse farms within the Bahr El Ghazal region, including the Mbororo breed, which originates from West African countries. Although the breed may not inherently carry distinct infection risk, their extensive migratory patterns, often involving contact with wildlife, shared pastures, and contaminated water sources, significantly elevate their exposure to infectious diseases and contribute to epidemiological risk.

As the sampling focused on individual animals rather than herds, clustering effects were not considered in the study design (Onafruo, Klein, et al. ²⁰²⁴).

Bayesian statistics revolves around a core principle: knowledge and uncertainty about variables are represented using probability distributions. These probabilities can be updated, summarised, and analysed using the rules of probability. This approach sets Bayesian methods apart from other frameworks, like frequentist statistics, which focus on sampling distributions to handle uncertainty. What makes Bayesian statistics powerful is its ability to provide a clear, consistent, and intuitive way to interpret what is known, based on the assumptions made and the data available (Lee 2011).

Bayesian models incorporate data at different levels, allowing for the analysis of both individual and group‐level variations (Veenman, Stefan, and Haaf Veenman et al. ²⁰²⁴). Despite their flexibility and richness of interpretation, Bayesian approaches can be computationally demanding, and their results are sensitive to the choice of prior distributions, which must be carefully justified (Robert 2011).

3. Bayesian Model

3.1. Bayesian Approach and Statistical Analysis in Human

A Bayesian hierarchical model was used to estimate the true prevalence of leptospiral exposure among slaughterhouse workers. The model was informed by a systematic literature review, and prior distributions were derived from published estimates of prevalence and diagnostic test performance. This approach was applied to evaluate the apparent prevalence of Leptospira spp. in human serum using MAT. The observed number of MAT seropositive ($y_i$) out of the $n_i$ workers tested in the $i^{\mathit{th}}$ slaughterhouse and is assumed to follow a binomial distribution:

$$y_i\sim \text{Binomial}\left({\mathrm{ap}}_i,n_i\right)$$ (1)

Here, ${\mathit{ap}}_i$ is the apparent prevalence of Leptospira spp. i the corresponding slaughterhouse i. Subsequently, the apparent prevalence ${\mathrm{ap}}_i$ is modelled as a function of the true prevalence ${\mathrm{tp}}_i$ (Rogan and Gladen 1978):

$${\mathrm{ap}}_i={\mathrm{tp}}_i\mathrm{Se}+\left(1-{\mathit{tp}}_i\right)\left(1-\mathrm{Sp}\right)$$ (2)

Where Se, Sp is the sensitivity and specificity for MAT detection, respectively. Prior information on Se and Sp was incorporated in the form of Beta distributions:

$$\mathrm{Se}~\text{Beta}\left({\mathrm{a}}_{\mathrm{Se}},{\mathrm{b}}_{\mathrm{Se}}\right),\mathrm{Sp}~\text{Beta}\left({\mathrm{a}}_{\mathrm{Sp}},{\mathrm{b}}_{\mathrm{Sp}}\right),$$ (3)

The true prevalences ${\mathrm{tp}}_i$ are then conceptualised as exchangeable and parameterized as follows:

$${\mathrm{tp}}_i\mid \mu ,\psi ~\text{Beta}\left(\mathit{\mu \psi },\psi \left(1-\mathrm{\mu }\right)\right)$$ (4)

Here, $\mu$ denotes the mean observed leptospiral true prevalence, and $\psi$ represents a parameter describing the variability of the leptospiral prevalence across slaughterhouses. The global mean for the true prevalence $\mu$ is assumed to follow a Beta distribution:

$$\mu ~\text{Beta}\left(a,b\right)$$ (5)

The variability parameter $\psi$ follows a Gamma distribution:

$$\psi ~\text{Gamma}\left({\alpha }_{\gamma },{\beta }_{\gamma }\right)$$ (6)

This modelling framework permits the prevalence of leptospiral exposure to vary across slaughterhouses, with the degree of variability influenced by the $\psi$ parameter. Larger values of $\psi$ indicate more homogeneous prevalences. The utilisation of the Beta distribution to model prevalences as $\text{Beta}\left(\mathit{\mu \psi },\psi \left(1-\mathrm{\mu }\right)\right),$ where $\mu$ = $\frac{a}{a+b}$ and $\psi$ = a + b, provided a versatile approach to analysing diverse prevalence estimates due to the Beta distribution's inherent adaptability (Gelman et al. 2014).

Moreover, at each iteration of the MCMC process, we assess whether the observed slaughterhouse‐specific prevalence ${\mathit{ap}}_i$ surpasses a predetermined critical threshold ($a$). For each slaughterhouse $i$, we define an indicator variable $I_i$ as:

$$I_i=\left\{\begin{array}{c}1\text{if}{\mathrm{ap}}_i>a\\ 0\text{otherwise}\end{array}\right.$$ (7)

The Bayesian probability that the observed leptospiral prevalence for slaughterhouse $i$ exceeds $a$% is obtained by averaging across all Markov chain Monte Carlo (MCMC) iterations.

Lastly, the model incorporates replicated data generation, allowing for the prediction and model fit assessment:

$$y_{{\mathit{rep}}_i}~\text{Binomial}\left({\mathit{ap}}_{i,}{n}_i\right)$$ (8)

The input provided by the experts was used to generate priors for the Se, Sp, and true prevalence of leptospiral seropositivity, as well as the variability between slaughterhouses $\psi$. These expert estimates were used to inform the median and the 5th (or 95th) percentiles for Se, Sp, π, and $\psi$, which was then used to establish prior distributions for these parameters. This process employs the PriorGen package (Kostoulas 2021), offering a systematic approach to generating prior distributions based on expert elicitation in disease prevalence studies, translating expert beliefs into informative priors using Beta and Gamma distributions.

3.2. Bayesian Approach and Statistical Analysis in Slaughtered Cattle

This model was constructed using the apparent prevalence of Leptospira spp. in serum and urine of slaughtered cattle tested using MAT and qPCR. The observed number of slaughtered cattle that tested positive by either MAT or qPCR test is denoted as $y_{\mathrm{MAT}\mid \mathrm{PCR}}$ out of a total of $n$ slaughtered cattle tested. This count is assumed to follow a binomial distribution $y_{\mathrm{MAT}\mid \mathrm{PCR}}\sim \text{Binomial}\left(\mathit{ap},n\right).$

The outcomes of the MAT and qPCR tests were assumed to be independent, given the true infection status of slaughtered cattle, as they detect distinct biological markers: MAT identifies antibodies indicating past exposure, while qPCR detects leptospiral DNA, reflecting active infection.

Here, apMAT, PCR, is the apparent prevalence of Leptospira spp. based on either MAT or PCR. Subsequently, the evident prevalence is modelled as a function of the true prevalence Tp (Rogan and Gladen 1978).

$$\mathrm{ap}=\mathrm{tpS}{\mathrm{e}}_{\mathrm{par}}+\left(1-\mathit{tp}\right)\left(1-\mathrm{S}{\mathrm{p}}_{\mathrm{par}}\right)$$

with the Se, Sp of the parallel interpretation being:

$$\mathrm{S}{\mathrm{e}}_{\mathrm{par}}=1-\left(1-\mathrm{S}{\mathrm{e}}_{\mathrm{MAT}}\right)·\left(1-\mathrm{S}{\mathrm{e}}_{\mathrm{PCR}}\right)$$

$$\mathrm{S}{\mathrm{p}}_{\mathrm{par}}=\mathrm{S}{\mathrm{p}}_{\mathrm{MAT}}·\mathrm{S}{\mathrm{p}}_{\mathrm{PCR}}$$

yMATy_{MAT}yMAT and yPCRy_{PCR}yPCR, respectively, out of the total number of cattle tested (n). The data were assumed to follow a binomial distribution. Estimation of the true prevalence was based on the sensitivity (Se) and specificity (Sp) of both diagnostic tests, and the model was defined by the following equations:

$$\left[\mathrm{y}\_\left\{\mathrm{MAT}\right\}\left(\mathrm{ap}\_\left\{\mathrm{MAT}\right\},\mathrm{n}\_\right),\mathrm{y}\_\left\{\mathrm{PCR}\right\}\left(\mathrm{ap}\_\left\{\mathrm{PCR}\right\},\mathrm{n}\right)\right]$$ (9)

$$\begin{array}{cc}\text{apMAT}=\hfill & \text{tpSeMAT}+\left(1-\mathit{tp}\right)\left(1-\text{SpMAT}\right)\text{apPCR}\hfill \\ \hfill & =\text{tpSePCR}+\left(1-\mathit{tp}\right)\left(1-\text{SpPCR}\right)\hfill \end{array}$$ (10)

Here, $\mathrm{S}{\mathrm{e}}_{\mathrm{MAT},\mathrm{PCR}},\mathrm{S}{\mathrm{p}}_{\mathrm{MAT},\mathrm{PCR}}$ are the Se and Sp for MAT and PCR, respectively, which are assumed to follow a Beta distribution:

$$\mathrm{S}{\mathrm{e}}_{\mathrm{MAT}}\sim \text{Beta}\left(a_{\mathrm{S}{\mathrm{e}}_{\mathrm{MAT}}},b_{\mathrm{S}{\mathrm{e}}_{\mathrm{MAT}}}\right),\mathrm{S}{\mathrm{p}}_{\mathrm{MAT}}\sim \text{Beta}\left(a_{\mathrm{S}{\mathrm{p}}_{\mathrm{MAT}}},b_{\mathrm{S}{\mathrm{p}}_{\mathrm{MAT}}}\right)$$

$$\mathrm{S}{\mathrm{e}}_{\mathrm{PCR}}\sim \text{Beta}\left(a_{\mathrm{S}{\mathrm{e}}_{\mathrm{PCR}}},b_{\mathrm{S}{\mathrm{e}}_{\mathrm{PCR}}}\right),\mathrm{S}{\mathrm{p}}_{\mathrm{PCR}}\sim \text{Beta}\left(a_{\mathrm{S}{\mathrm{p}}_{\mathrm{PCR}}},b_{\mathrm{S}{\mathrm{p}}_{\mathrm{PCR}}}\right)$$ (11)

The true prevalence Tp is then parameterized using a Beta distribution:

$$\mathrm{tp}\sim \text{Beta}\left(a_{\mathrm{mu}},b_{\mathrm{mu}}\right)$$ (12)

The mean expected value for the Se of the diagnostic process was expected to be 0.60, and we were 95% certain that it was lower than 0.80. This corresponds to a Beta (8.396, 5.597).

The mean expected value for the Sp. of the diagnostic process was expected to be 0.98, and we were 95% certain that it was higher than 0.95. This corresponds to a Beta (80.892, 1.651).

A priori, we thought that the probability of a species being free from infection was 0.80, and we were 95% certain that it was lower than 0.90. This corresponds to a Beta (27.454, 6.864).

The mean prevalence of Leptospira spp. positivity was thought to be 0.16, and we are 99% confident that it was not more than 0.20. We were also confident that 90% of all species have a prevalence less than or equal to 0.40, and we are 95% certain that it does not exceed 0.50. These correspond to Beta (80.162, 420.848) and Gamma (6.147, 1.435). Beta distribution was used to model true prevalence, while Gamma prior is not on prevalence itself but on the precision parameter distribution governing slaughterhouse‐level prevalences.

Priors generation for the Se., Sp. (MAT, PCR), and true prevalence of leptospiral positivity employs the PriorGen package (Kostoulas 2021). Microscopic Agglutination Test (MAT), Se., and Sp. can vary depending on the serovar, geographic region, and laboratory practices. The Bayesian model accounted for these limitations by incorporating prior distributions for test sensitivity and specificity based on published estimates relevant to similar settings (Clarke and Jones 2015).

An initial univariable logistic regression analysis was conducted to examine the relationship between each covariate, including demographic characteristics, occupational and non‐occupational factors, and leptospiral positivity. Covariates with a p‐value less than 0.25 were included in a multivariable logistic regression model, using stepwise selection based on the Akaike Information Criterion (AIC) to identify the most parsimonious set of predictors. Risk factor analysis was further conducted using a random‐effects logistic regression (RL) model within a frequentist framework, alongside the Bayesian methods employed elsewhere in the study.

4. Prior Specification

The mean expected value for the Se of the diagnostic process was expected to be 0.90 and we were 95% certain that it was lower than 0.80. This corresponds to a Beta (27.787, 3.087). The mean expected value for the Sp of the diagnostic process was expected to be 0.98 and we were 95% certain that it was higher than 0.95. This corresponds to a Beta (80.892, 1.651). A priori we thought that the probability of a species being free from infection was 0.80 and we were 95% certain that it was lower than 0.90. This corresponds to a Beta (27.454, 6.864). The mean prevalence of infection was thought to be 0.10 and we are 99% confident that it was not more than 0.15. We were also confident that 90% of all species have a prevalence less or equal to 0.30 and we are 95% certain that it does not exceed 0.35. These correspond to a Beta (23.706, 213.35) and Gamma (8.495, 2.369).

5. Results

5.1. Human Results

Out of 250 human samples collected from 3 slaughterhouses, 71.2% (178/250) of the workers sampled were from Lokoloko slaughterhouse, 10.8% (27/250) from Zagalona slaughterhouse, and 18% (45/250) were from Eastern Bank. Eighty percent of the participants (200/250) were male.

Although the sample distribution was skewed, with 71.2% of samples from Lokoloko, the Bayesian hierarchical model mitigated this imbalance by using the hyperparameter $\psi$ in the Beta–Gamma prior to borrow strength across slaughterhouses, thereby stabilising estimates for under‐sampled sites like Zagalona.

The median probability of detecting at least one Leptospira spp. positive case in a randomly selected slaughterhouse was 80%, with 95% CI (70.0–90.0) Figure 1.

FIGURE 1.

Plot for the median probability of Leptospira spp. being present at a slaughterhouse (n = 3) in Western Bahr El Ghazal South Sudan, from January to February 2023.

While adjusting for the imperfect Se/Sp of the diagnostic process, the probability of Leptospira spp. being present in a randomly selected slaughterhouse was found at a median of 79% with 95% CI (64.0–91.0).

The Bayesian model also estimated the mean true prevalence of leptospiral seropositivity among slaughterhouse workers in the Bahr El Ghazal region to be 10% (Figure 2).

FIGURE 2.

Plot for the mean true prevalence of leptospiral seropositivity among slaughterhouse workers (n = 250) sampled from Western Bahr El Ghazal State of South Sudan, from January to February 2023.

When adjusted for an affected slaughterhouse, the mean expected true prevalence was 10% with 95% CI (7.0–14.0).

The likelihood of a randomly selected slaughterhouse being entirely free from infection was estimated at a mean of 22%. Additionally, the probability that a randomly selected slaughterhouse has a true prevalence below 5% was calculated and reported at a mean of 65%.

5.2. Cattle Results

Of the 402 sampled slaughtered cattle, the majority of the slaughtered cattle population (68.4%, 275/402) consisted of adult animals aged 2 years or older, while younger animals under 2 years made up 31.6% (127/402). Regarding sex distribution, 45.8% (218/402) of the slaughtered cattle were female. In terms of breed composition, 64.7% (260/402) belonged to the Nilotic local breed, whereas the Felata (Mbororo) breed accounted for 35.3% (142/402).

The Bayesian model estimated that the true prevalence of leptospiral seropositivity among slaughtered cattle in the Bahr El Ghazal region was 85% (342/402, 95% CI 80.0–90.0). Figure 3 illustrates the estimated true prevalence of leptospiral seropositivity.

FIGURE 3.

Plot for the true prevalence of Leptospira spp. in slaughtered cattle (n = 402) sampled from Bahr El Ghazal region in South Sudan, from January to February 2023.

5.3. Risk Analysis Results for Humans and Slaughtered Cattle Leptospiral Positivity

Individuals not involved in flaying had 0.08 times the odds of leptospiral positivity (95% CI: 0.0–0.0, p = 0.001) compared to those who participated in flaying. Those who did not wear an apron had 0.1 times the odds of seropositivity (95% CI: 0.0–0.8, p = 0.03) compared to those who wore an apron. Additionally, each additional carcass handled per day was associated with a 2.3 times increase in odds of leptospiral positivity (95% CI: 1.5–3.8, p = 0.001).

For slaughtered cattle, the final model identified breed, sex, and age as significant risk factors for leptospiral exposure. The Falata breed exhibited 2.5 times higher odds of leptospiral positivity (95% CI: 1.3–4.6) compared to the Nilotic breed. Sex also played a role, with females showing 0.5 times lower odds of positivity (95% CI: 0.3–0.8) than males. Age was another influential factor, with older slaughtered cattle being twice as likely (95% CI: 1.1–3.5) to test positive compared to younger slaughtered cattle (Table 1). The associations reported were adjusted for potential confounding variables by including relevant covariates in the multivariable Bayesian logistic regression model. However, interactions between risk factors were not explicitly modelled in this initial analysis.

TABLE 1.

The final models for leptospiral seroprevalence and associated risk factors were analysed using Bayesian multivariable logistic regression of sampled slaughterhouse workers (n = 250), and slaughtered cattle (n = 402) in the Bahr El Ghazal Region, South Sudan, from January to February 2023.

Slaughterhouse workers						Slaughtered cattle
Covariate	Category	OR	95% CI		p	covariate	Category	OR	95% CI		p
			Lower	Upper					Lower	Upper
Flaying	Yes	Ref.			0.001	Breed	Nilotic	Ref.
	No	0.08	0.0	0.4			Falata	2.5	1.3	4.6	0.001
Wearing apron	Yes	Ref.			0.03	Sex	Male	Ref.
	No	0.1	0.0	0.8			Female	0.5	0.3	0.8	0.01
Number of carcasses handled per day		2.3	1.5	3.8	0.001	Age	Young	Ref.
							Old (≥ 2 years)	2	1.1	3.5	0.02

6. Discussion

The Bayesian findings reveal a 22% probability that a given slaughterhouse is entirely infection‐free, indicating a 78% likelihood of infection within such facilities. This infection‐free estimate likely reflects a combination of factors, such as geographic variation in exposure risk, differences in slaughterhouse hygiene and biosecurity practices, and variability in cattle sourcing. However, due to the limited availability of data on these contextual variables in the current study, we are unable to attribute this estimate to any specific factor. We, therefore, identify this as a key area for future research. This result aligns with the high leptospiral prevalence of 6.4% among slaughterhouse workers reported in our previous study (Onafruo, Dreyfus, et al. ²⁰²⁴) and corroborates prior research (Ghasemian et al. 2020; Gonçalves et al. ²⁰¹⁸).

Bayesian analysis indicates a 65% probability that the true prevalence of leptospiral infection in a given slaughterhouse is below 5%. However, there is also a 35% probability that some slaughterhouses exceed this threshold, reflecting variability in infection levels and emphasizing ongoing zoonotic and occupational health risks. The elevated probability of infection underscores the critical role that both environmental and occupational factors play in the transmission of Leptospira spp. The high prevalence observed in this study may be amplified by regional environmental conditions such as heavy rainfall and flooding (Meng et al. 2024), which are known to facilitate pathogen spread. Additionally, the occupational setting increases exposure risk, as reflected by the 81.8% seropositivity and 6% PCR‐confirmed urine shedding among slaughtered cattle (Onafruo, Klein, et al. ²⁰²⁴). However, the lack of systematically collected environmental data across the study sites limited our ability to assess these influences within the model. We acknowledge this as a key limitation and emphasize the need for future studies that integrate environmental and ecological data to improve understanding of exposure pathways and regional variations in risk.

The Bayesian modelling approach confirmed specific occupational risk factors for leptospiral seropositivity among slaughterhouse workers. Notably, those who did not engage in animal flaying had significantly lower odds of infection (OR = 0.08, 95% CI: 0.0–0.4), reinforcing evidence that direct contact with animal tissues during flaying increases exposure risk (Colavita and Paoletti 2007; Ghasemian et al. ²⁰²⁰). Additionally, workers not wearing an apron had lower exposure (OR = 0.1, 95% CI: 0.0–0.8) compared to those who did, a finding that contrasts with previous studies suggesting aprons are protective (Cook et al. 2017). Leptospires can be transmitted via different contact transmission, hence reducing the role of aprons as protective gears (Astuti et al. 2019; Cook et al. ²⁰¹⁷).

The limited protective effect of aprons may be attributed to factors such as improper usage (e.g., loose fitting or inadequate coverage) and material limitations that permit leptospiral penetration. Moreover, aprons alone may not provide adequate protection, as effective infection control typically involves multiple measures, including hand hygiene and respiratory safety. Environmental exposure to contaminated water or surfaces may further compromise apron effectiveness. Therefore, the use of additional protective gear such as gloves, boots, and face safety goggles may be necessary to reduce the risk of infection effectively (Andersen and Andersen 2019). Further investigation is warranted to evaluate apron usage practices, material properties, and the integration of complementary protective measures.

The risk of exposure also increased with the number of animals handled per day (OR = 2.3,95% CI:1.5–3.8), a result consistent with research highlighting that frequent exposure raises exposure risk (Cook et al. 2017; Gizamba and Mugisha ²⁰²³; Wainaina, Wasonga, and Cook Wainaina et al. ²⁰²⁴). We recommend implementing comprehensive protective strategies for slaughterhouse workers, including the use of personal protective equipment (PPE), strengthened hygiene practices, and improved environmental management (e.g., waste disposal and rodent control), supported by regular training to reduce leptospiral transmission risks.

Applying the Bayesian approach to slaughtered cattle data estimated a true prevalence of leptospiral positivity at 85%, representing the first estimate for slaughtered cattle in South Sudan using both MAT and qPCR test results. This estimate is slightly lower than the 92.2% reported in Thailand (Yatbantoong and Chaiyarat 2019), yet higher than the 78.7% in Brazilian dairy cattle (da Silva et al. ²⁰²²) and the 63.5% previously reported in South Sudan (Sebek et al. 1989), suggesting not only geographic variations but also an increasing trend in leptospiral infections in the area. Geographic variations may stem from a range of factors, including environmental conditions such as rainfall, flooding, and temperature, which influence leptospires' survival and transmission. Ecological aspects, including the density of reservoir hosts (e.g., rodents and livestock), also play a key role. Additionally, differences in public health infrastructure, diagnostic capacity, and disease surveillance systems may contribute to discrepancies in reported prevalence between countries like South Sudan, Thailand, and Brazil.

The study also highlighted breed, sex and age as significant factors associated with leptospiral positivity in slaughtered cattle. The Falata breed, known for its migratory grazing practices, was 2.5 times more likely to be positive than the Nilotic breed, likely due to broader environmental exposure (Onafruo, Klein, et al. ²⁰²⁴). The migratory grazing practices of the Falata breed involve transboundary movement across various African countries into Sudan and South Sudan, exposing them to diverse ecological zones such as wetlands and communal grazing areas. These environments, particularly during or after the rainy season, favor leptospiral survival and are often contaminated by infected animals. Frequent contact with other herds, wildlife, and shared water sources further increases the risk of exposure.

Additionally, female slaughtered cattle showed lower odds of infection than males, contrasting with prior South Sudan studies suggesting females as more susceptible (Onafruo, Klein, et al. ²⁰²⁴). This difference may be influenced by the predominantly male population among slaughtered cattle sampled (Onafruo, Klein, et al. ²⁰²⁴), indicating how demographic factors can shape observed patterns of infection. Age also emerged as a significant factor, with older slaughtered cattle being twice as likely to test positive, a finding consistent with previous research showing that longer environmental exposure increases leptospiral infection risk (Alinaitwe et al. 2020; Onafruo, Klein, et al. ²⁰²⁴). Although individual‐level sampling was employed to reduce herd‐level clustering, we acknowledge the possibility of residual clustering, particularly when cattle originate from the same supplier, transport group or geographic area. This potential non‐independence among sampled animals may have influenced the results.

These epidemiological insights, derived from Bayesian modelling frameworks, provide a valuable foundation for designing targeted surveillance and intervention strategies, ultimately enhancing the prevention and control of zoonotic diseases in both human and livestock populations.

Despite their advantages, Bayesian methods are not without limitations. One key challenge is the reliance on prior information, which, if poorly informed, can skew results and lead to biased conclusions (Mulder 2014; Veldhuis et al. ²⁰²³). However, this risk can be mitigated by using well‐grounded informative priors based on empirical data, which align the prior distribution with the actual scenario, thereby improving model accuracy (Merkle et al. 2023). Tools such as prior‐generation packages (Kostoulas 2021) can further support the development of appropriate priors.

Diagnostic test imperfections also pose challenges, as inaccuracies in test sensitivity and specificity can distort prevalence estimates. Studies on brucellosis and paratuberculosis have demonstrated the impact of misclassification, necessitating adjustments for test performance (Meletis, Sakhaee, and Kostoulas Meletis et al. ²⁰²⁴; Al Naeem et al. ²⁰²⁴). In this methodological study, we applied a hierarchical Bayesian framework that jointly models diagnostic sensitivity and specificity as probabilistic parameters incorporated via informative Beta priors and uses the resulting posterior to infer the true prevalence of infection, thereby correcting for misclassification due to imperfect MAT (Gelman and Carpenter 2020).

Additionally, selecting an appropriate Bayesian model can be challenging, as outcomes may vary depending on factors such as the choice of prior distributions, the specification of the likelihood function, and the overall model structure (Flor et al. 2020; de Van Schoot et al. ²⁰¹⁴). These elements can significantly influence the results, making interpretation more complex, particularly when compared to frequentist approaches that rely on different inferential assumptions.

This challenge underscores the importance of robust model‐checking techniques and sensitivity analyses to identify the most appropriate model for the data, ultimately enhancing the clarity and reliability of interpretations (Merkle et al. 2023). By addressing these limitations, Bayesian methods demonstrate their potential as a powerful tool for advancing public health efforts, particularly in resource‐limited settings such as South Sudan, where reliable diagnostic infrastructure remains scarce. The scientific impact of this study extends beyond its regional context by illustrating how Bayesian approaches can enhance inference and interpretation in epidemiological research. Through the reanalysis of existing data within a probabilistic framework, this study offers a methodological contribution that can be applied to other contexts where data limitations challenge traditional analytical methods.

7. Conclusion

This study presents the first integrated Bayesian analysis of leptospiral positivity among slaughterhouse workers and slaughtered cattle in Bahr El Ghazal region, South Sudan. The findings reveal a substantial burden of Leptospira spp., with an estimated true prevalence of 10% among slaughterhouse workers and 85% among slaughtered cattle. These results highlight significant zoonotic transmission risks in occupational and environmental contexts, particularly in settings with limited disease surveillance and control infrastructure.

Key occupational risk factors including involvement in flaying, lack or improper use of protective gear, and the number of carcasses handled underscore the urgent need for improved workplace safety practices and regular training. In slaughtered cattle, breed, age and sex were significantly associated with infection risk, reflecting both biological and management‐related influences on disease dynamics.

The Bayesian framework enabled adjustment for diagnostic test imperfections and provided deeper insights into prevalence and risk patterns. However, the study also faced limitations, such as the lack of environmental exposure data, potential clustering among slaughtered cattle, and the absence of slaughterhouse‐level effects. Despite these constraints, the study lays a strong foundation for evidence‐based interventions and supports targeted surveillance, improved biosecurity measures, and integrated One Health strategies to mitigate zoonotic transmission risks.

Future research should integrate environmental data, evaluate the effectiveness of personal protective equipment, and expand sampling across broader geographic areas and production systems to enhance understanding of leptospiral epidemiology.

Importantly, the scientific impact of this study extends beyond its regional context by demonstrating how Bayesian approaches can strengthen inference and interpretation in epidemiological research. Through the reanalysis of existing data within a probabilistic framework, this study offers a methodological contribution applicable to other settings where data limitations challenge traditional analytical approaches.

Author Contributions

Conceptualization of the study: D.O.; C.K.; J.K.; A.J. Funding acquisition: C.K.; A.J. Design of the study: D.O.; C.K.; J.K.; J.E.; I.K. Fieldwork and laboratory analysis: D.O.; E.K.; E.S.; P.M.; A.J. Data analysis: D.O.; P.K. Supervision: C.K.; J.E.; J.K. Writing original draft preparation, writing – review and editing the final manuscript: D.O.; J.K.; C.K.; J.E.; A.J.; I.K.; P.K.; J.S.

Funding

The Norwegian Agency for Development Cooperation (NORAD) funded the study through the NORHED II Program (Grant number 68802; https://en.uit.no/project/onehealthcidimoh).

Ethics Statement

Procedures involving sample collection from human subjects were approved by the Ministry of Health, Research Ethics Review Board (MOH‐RERB), Juba, South Sudan (MOH/RERB 02/7/2022—MOH/RERB/A/02/6/02/2023). Ethical approval was obtained from the Ministry of Livestock and Fisheries, Juba‐South Sudan (RSS/MLF/DVS/39/023) to collect animal samples. All methods were carried out following relevant guidelines and regulations under the ethics approval and consent to participate.

Conflicts of Interest

The authors declare no conflicts of interest.

Data Availability Statement

The findings of this study can be replicated using the cross‐sectional study test results cited and presented within this article.