Farm-level risk factors associated with bovine tuberculosis in the dairy sector in Eritrea

The aim of our study was to determine the association of selected potential risk factors with the presence of bovine tuberculosis (BTB) in dairy herds in Eritrea. A case – control study was conducted in the three major milk-producing regions of the country by strati ﬁ ed random sampling of 61 case and 65 control herds combined with completion of a standardized pretested questionnaire pertaining 36 relevant risk factors (variables). The variables were divided into two clusters, based on potential association with either “ introduction ” or “ establishment ” of BTB on the farms to elucidate association with inci-dent or prevalent cases separately. Subsequent to univariable analysis of the 36 risk factors at herd level, 14 of these were offered to multivariable logistic regression models. Farms with higher numbers of cows, and those with concrete ﬂ oors, were 3.6, and 7.5 times more at risk for presence of BTB, respectively, compared with their references. These ﬁ ndings will be useful as entry points for future informed decision-making towards BTB control and eradication programme in the country.

a 1:1 ratio. A herd was considered a "case herd" when it included at least one SICTT reactor (positive) animal during the 2011 BTB prevalence survey in the dairy sector in Eritrea (Ghebremariam et al., 2016).
In the previous prevalence study (Ghebremariam et al., 2016), all herds (3,149) and all individual animals (15,354) above 6 weeks of age, from the three selected regions (Maekel, Debub and Anseba), were tested using the SICTT according to OIE. A herd was considered a "control herd" when none of the animals reacted to the test. Herds with (n = 545) and without SICTT (n = 2,604) reactors were identified in each of the three major milk-producing regions (Maekel, Debub and Anseba; Figure 1). Relative to the number of reactor herds in Maekel and Debub, the largest proportions of herds, n = 56 and n = 65, respectively, were selected from these two regions and only few (n = 5) from Anseba. Prior to the selection, all the herds in the three regions were stratified according to their herd sizes, categories being 9-20, 21-30, 31-40 and >40 animals. Only farms with a herd size of ≥9 head of cattle were selected. Numbers of herds, representative for the number of herds in each stratum, were randomly drawn using a computer-generated random table (Epi Calc 2000 version 0.1), resulting in inclusion of 61 case and 65 control herds.
To obtain information on the potential risk factors associated with the reactivity to SICTT in a standardized fashion, a pre-tested questionnaire was used. It included 36 relevant variables that captured the farmers' and farms' characteristics. The variables were divided into two clusters, based on potential association with either "introduction" (entry to the farms from an external source) or "establishment" (persistence on the farm) of BTB. The selection process was based on epidemiological importance of the factors as described by Griffin, Martin, Thorburn, Eves, and Hammond (1996). All 36 factors selected for the study and their categories are shown in Table S1.

| Data analysis
Data were first entered in Excel and then exported to SPSS IBM version 20 software and analysed. Associations between the farm-level risk factors (variables) and herds with reactor(s) were tested in two stages: first, all the variables (n = 36; Table 1) were screened using Fisher's exact test in univariable analyses and those variables with p-value <.25 (n = 14) ( Table 2) were tested for collinearity. The variable "reproduction method" could not provide any interpretable differences between case and control herds (Table 1). Due to the lack of artificial insemination (AI)-only reproduction methods in control herds, Firth's correction was applied to allow for estimation of the odds ratio (OR) ( Table 2), but the variable ("reproduction method") was excluded from further analysis as it did not give interpretable result in the multivariable analysis. Hence, 13 variables remained to be offered to either or both of the multiple logistic regression models, for "introduction" and "establishment" of BTB (presence of SICTT reactors) on the farms (Table 3). During the multivariable logistic regression model building, using manual (Enter) and automated (Backward) procedures for the possible risk factors screened, respectively, potential confounders were identified by the changes in the coefficients (b) (i.e., if inclusion of a variable to the baseline model altered the coefficient of the model by >10%, one of the variables, depending on its importance, was retained/dropped), and p-values <.05, to decide for definite inclusion or exclusion of the variables in the models.
Two variables, "age of respondents" and "major activity," representing farmers' characteristics for "establishment" of BTB in the farms were highly related to case-control status. However, they were not used in the analysis as they overshadowed the potential influence of more biological factors. Besides, two variables ("number of bulls" and "availability of bulls") that showed high correlation with "number of cows" and "herd size" were excluded from the models, as the latter two were considered more plausible risk factors for "introduction" and "establishment" of BTB, respectively. The herd size, number of cows and number of heifers were arbitrarily categorized as shown in Table 2. The model for introduction of BTB included four variables ("number of cows," "species of animals sharing water point," "sharing of water point" and "same source of bull purchase") and that for establishment of BTB six variables ("herd size," "number of heifers," "housing type," "housing system," "building status" and "feeding system"). Finally, two variables ("number of cows" and "building status" from both "introduction" and "establishment") retained in the final model (Model 3; Table 3) as potential risk factors for the presence of BTB reactors in the farms.

| Univariable analysis
A summary of the results of the univariable analysis is shown in Table 2.
The variables "age of respondent" and "major activity" were selected to demonstrate the role of farmers' characteristics in relation to SICTT reactivity between the case and control farms. "Age of respondents" coded in three categories (25-45 years, 46-65 years and >65 years) was significantly associated (p = .036) with the presence of SICTT reactor F I G U R E 1 Study area map: Maekel (1), Debub (2) and Anseba (3), the three major milk-producing regions of Eritrea. Adopted from Ghebremariam et al. (2016) T A B L E 1 Names, descriptions and coding of the 36 variables and their categories included in the study as potential risk factors for "introduction" (Intro) and "establishment" (Est) of BTB in the 61 case and 65 control farms in Eritrea  One of the potential risk factors, "number of cows" on the farms, screened in the univariable analysis, was significantly (p = .001) associated with the presence of SICTT reactors. Farms with larger (>10) numbers of cows were more at risk (OR = 4.0) compared to those with smaller numbers of cows. The mean number of cows in the dairy herds included in the study was 10.2 (range: 0-120), whereas the mean numbers of cows in the cases and controls were 13.52 (range: 0-120) and 7.09 (range: 2-20), respectively.
"Herd size" was also significantly (p = .002) associated with the presence of SICTT reactors and was coded in two categories (Table 3). The risk increased with herd sizes >26 animals (OR = 5.0).
The number of heifers was significantly associated (p = .018) with the presence of SICTT reactors in dairy farms and herds with larger (>8) numbers of heifers had a higher risk (OR = 4.1) than herds with fewer heifers. The mean number of heifers in case farms was 6.8 (range: 0-40) as compared to 4.1 (range: 0-20) in controls.
Finally, the "availability of bulls" (categories: yes/no) and "number of bulls" were associated with the presence of SICTT reactors The use of stanchions on the farm was identified as a risk factor (OR = 1.8) and found to be associated (p = .135) with being a case herd.
The variable "housing system" was grouped into two categories: farms that kept their animals indoor permanently, compared with those that did not do so. Farms that kept their cattle entirely indoor were at higher risk (OR = 1.8) of having SICTT reactor(s) as compared with farms that allowed animals outdoors as well.
The "building status" was grouped into two categories: dairy farms with concrete floors and without concrete floors. This variable was significantly (p = .033) associated with the presence of SICTT reactor animal(s) in the farms, and those that housed their animals on concrete floors were more (OR = 8.4) at risk.
The variable "feeding system" was categorized into two groups (zero-grazing and grazing) and was significantly associated (p = .061) with the presence of SICTT reactor(s) on the farms. Farms using zero-grazing were at higher (OR = 4.1) risk than the others. Feeding of green feeds (Est) Provision of green feed to the animals either by grazing, by cut and carry or purchase.

Yes/No
Source of green feed(Est) Where does the green feed come from: own farm or purchase from other farms?
Not applicable, home-grown, purchased, both Feeding system (Intro) The system of feeding that allows grazing of animals or that does not allow grazing but rather depends on cut and carry (bringing the green fodder to the barn).
Zero-grazing, grazing Concentrate feeding (Est) The provision of the animals with concentrate (formulated) feed that fulfils the requirement of dairy cattle for milk production Farms using both "AI and natural breeding" had (OR = 0.08) a protective effect (Table 2).

| Multivariable analysis
From the three potential risk factors related to "introduction" of BTB on the farm that were offered to the first multivariable model (Model 1,  in the farms. Farms with larger numbers of cows (>10) were more at risk (OR = 4.0; 95% CI: 1.7-9.7) than farms with ten cows or less.
Likewise, from the six variables that were offered to the second model (Model 2, Table 3) as possible risk factors for "establishment" of BTB in the farms, two variables, namely "herd size" and "building status," were retained in the model. Larger herd sizes (>26) were associated with a higher risk (OR = 4.8; 95% CI: 1.6-14.4) compared with smaller herd sizes (9-26).
Herds kept on concrete floors were more at risk (OR = 8.6; 95% CI: 1.0-75.7) to have SICTT reactors when compared with those kept on other types of floors (Table 3). This variable ("building status") had a borderline significance (p = .052), but was retained in the model given its biological importance as a risk factor. In the final multivariable model (Model 3; Table 2), only two variables "number of cows" from the "introduction model" and "building status" from the "establishment model" were retained.

DISCUSSI ON
To the best of our knowledge, this is the first study in Eritrea that has systematically identified farm-level risk factors for the presence of SICTT reactor(s) in the three major milk-producing regions. Both univariable (Table 2) and multivariable logistic regression (Table 3) analyses for potential risk factors related to presence of SICTT reactor(s) on the farms have been performed.
Overall, the major risks seemed to be associated either with numbers of cattle on the farms ("number of cows," "herd size," "number of heifers," "number of bulls" and "availability of bulls"), or with whether or not the animals were kept outdoors for parts of the day, for example, for grazing and/or access to and sharing of water.
"Major activity" of the farmers and "age of respondents" were the two farmers' characteristics found significant risk factors for "establishment" of BTB in the farms as identified by the univariable analysis. The majority of farmers, above 45 years old, were full-time dairy farmers. Both "age of respondents" and "major activity" were independently associated with the SICTT reactivity in the herds.
The lower risk associated with farms owned by the younger age groups may be explained by younger age groups engaging more actively with and seeking advice from animal health experts and veterinary professionals if their dairy cattle showed reduced productivity or ill health. Surprisingly, farms owned by full-time dairy farmers were three times more at risk as compared to part-time dairy farmers.
"Number of cows" was identified as a risk factor for the "introduction" of BTB to the case farms (Model 1, Table 2) and was retained in the final Model 3 (Table 3)  Availability and number of bulls were significant risk factors in the univariable analyses for "introduction" of BTB in the case farms which might have been attributed to the discontinuation of artificial insemination and thus the use of bulls for reproduction by natural mating in the study area. Sharing of bulls is a common practice and those having more bulls are inclined to share them more frequently.
The observed higher SICTT reactivity in those herds having bulls could be due to frequent and direct close contact between the bulls and animals in several herds (Skuce, Allen, McDowell, & Branch, 2011). Although farms not owning bulls also use bulls from other farms to breed their animals, they experience a shorter contact time and hence a lower risk of transmission from potentially infected bulls.
"Sharing of water points" and "species of animals sharing water point" were associated with a decreased risk of BTB "introduction" in the farms. The water points used in our study areas were surface T A B L E 3 Final logistic regression models of the risk factors associated with herds having SICTT reactor(s) in the study areas within the dairy sector of Eritrea, as they are grouped into factors for "introduction" (Model 1) and "establishment" waters (rivers and dams) that were indicated not to be significant risk factors for BTB transmission in cattle (Griffin et al., 1993), and in buffaloes under free ranging condition (Michel, De Klerk, van Pittius, Warren, & Van Helden, 2007). During the dry season, in our study area, farmers used their own mobile watering troughs to water their animals from water holes manually dug out from river beds. In addition, "sharing of water points" entails walking the animals to water points that may result in minimizing, within herd, animal-toanimal contacts, due to lower density of cattle on pasture, besides providing them with an opportunity of, beneficial, physical exercise "en route".
"Herd size" was identified as one of the major risk factors for the "establishment" of BTB in the case farms (Model 2). Farmers try to increase herd sizes for increased efficiency gains, but increase in herd size may lead to overcrowding causing enhanced cattle-tocattle transmission of M. bovis. This finding is in agreement with those of similar studies that indicated the association of herd size with the prevalence of BTB (More & Good, 2015;Pavlik, Matlova, Dvorska, Shitaye, & Parmova, 2005;Proaño-Perez et al., 2009;Wright et al., 2015), including that of Omer et al. (2001) who identified herd size as one of the major risk factors for SICTT reactivity in the Maekel region, one of our current study areas in Eritrea. As shown by the univariable analyses, farms with larger numbers (>8) of heifers were four times more at risk to have SICTT reactors than farms with fewer heifers (Menzies & Neill, 2000;Neill, Pollock, Bryson, & Hanna, 1994) in relation to "establishment" of BTB in the farms. Increased animal-to-animal contact, normal behaviour during oestrous, might explain this finding (Hurnik, King, & Robertson, 1975).
The other major risk factor for the "establishment" of BTB in our study was "building status" (floor types) (Model 2), which was the second most important risk factor retained in the final Model 3 (Table 2). Those farms using concrete floors were about eight times more at risk when compared to those farms without concrete floors.
Most of the dairy farms in the urban and peri-urban areas in Eritrea house their animals on concrete floors. Due to lack of maintenance such floors are indicated as culprits for causing digital dermatitis which was shown as one of the major constraints in the dairy farms in and around the capital by Nsahlail and Moges (2007). Eventually, due to the tendency of affected animals to lie down for prolonged times; their feed intake time may be compromised, causing emaciation and enhanced susceptibility to disease, including BTB (Cook & Nordlund, 2009;De Vries et al., 2015;Nordlund, Cook, & Oetzel, 2004).
"Feeding system" was one of the risk factors for "establishment" of BTB in the farms identified by the univariable analysis, where farms with a zero-grazing system were four times more at risk than farms that do not make use of this practice. Similarly, farms that kept their cattle permanently indoor encountered double the risk for "establishment" of BTB in the farms compared with those letting animals go outdoors. The plausible reasons for this, as already indicated by other studies (Ameni et al., 2006;Ayele, Neill, Zinsstag, Weiss, & Pavlik, 2004;Griffin et al., 1993Griffin et al., , 1996Skuce, Allen, & McDowell, 2012;Skuce et al., 2011), may be that indoor housing enhances closer contact between cattle, and more likely transmission of respiratory disease (Skuce et al., 2011). For infection to occur in cattle via inhalation, it requires as little as one bacillus as compared to ingestion that requires quite a large number of M. bovis (10 7 ) (Humblet et al., 2009). On the other hand, being outdoors, the animals literally reduce the contact among themselves and thus with the pathogenic agent (M. bovis). Besides, allowing animals to graze is directly connected with the animals' health and welfare. Relatively intense solar radiation of animals (e.g., hair, skin) (Kazda, 2010) and their excrements might play a bactericidal role and thus reduce mycobacterial burden (Fine, Bolin, Gardiner, & Kaneene, 2011).
"Housing type" was also one of the significant risk factors for In our attempt to assess a relatively large number of risk factors, it appeared impossible to rule out all potential biases, unobserved confounders and lack of independence between some of the factors. For that reason, we retained only those factors that were most strongly associated with the risk of having SICTT reactor(s) in a case farm. Although, currently, SICTT is the most widely applied screening test for BTB in live cattle (De la Rua-Domenech et al., 2006, OIE, 2009, it is imperfect but performs better than other ante mortem tests currently available for the detection of BTB.
Although the advantages of grazing on pasture are apparent in relation to hoof health, exercise and the access for more space that might minimize animal-to-animal contact, switching from indoor housing to pasture is not a realistic option for many farms in the urban areas because of unavailability of land for pasture. In such herds, a regular test and control system related to animal movements might help to control BTB in the future.
Dairy farms that allowed their animals outdoors, that kept smaller numbers of cattle (fewer cows and smaller total herd size) or that did not have concrete floors were less at risk to be BTB positive as | 111