A causal approach to understanding management factors, reproductive underperformance and reproductive infections in dairy cattle

The objective of this study was to investigate causal associations between cattle farms’ management practices and reproductive disorders (abortion, stillbirth, retention of placenta, metritis). Besides, direct causal associations between farms’ management and reproductive infections (Chlamydia abortus, Coxiella burnetii, and Neospora caninum), reproductive disorders and infections were also investigated in this study. As a secondary objective, constraints that affect the production in cattle farms were examined. The study was carried out in the north-western (Una-Sana), western (Canton 10) and central part (Central Bosnia Canton) of Bosnia and Herzegovina. A total of 201 farms were selected for participation. A semi-structured questionnaire-based interview was conducted among farmers/managers from January 1st to August 31st, 2015. The 40 questions were divided into three groups: socio-demographic, management, and information on reproductive performances in cattle. Supplementary questions were asked about the perceived primary constraints of the production. A multivariable mixed-effects logistic regression was used to screen management factors for potential statistical influence. All investigated outcomes were associated with farms’ management. The final multivariable models were merged into a Structural Equation Model (SEM). The causal model was then specified graphically. The SEM model showed that herds that experienced abortions (OR=4.3) and stillbirth (OR=6.7) were associated with N. caninum seropositivity. Also, herds that experienced retention of placenta were strongly associated with the occurrence of metritis (OR=10.1). C. abortus and C. burnetii herd seropositivities were mainly associated with environmental factors and contact with potential intermediate hosts. Our study demonstrated that management practices on dairy farms in Bosnia and Herzegovina contributed to the occurrence of severe reproductive outcomes and reproductive infections. N. caninum seems to be an infectious agent that substantially contributed to the reproductive underperformance. Further we demonstrate the need for using causal models in understanding complex relationships.


4
Introduction 50 Agriculture is one of the most vital sectors of economy for the countries in Western Balkan. 51 Most of them have substantial share in the total state gross value added (GVA) ranging from 52 7.6% in Bosnia and Herzegovina (BH) to 20.1% in Albania. [1]. The situation in the agricultural 53 sector is gradually improving in most of Western Balkan countries, but there are still some 54 critical steps in reforming the agricultural policy and its harmonising with the EU standards. August 2016) represents an opportunity for further intensification of this production. This 64 opportunity makes milk one of the principal export commodities of the country. To cope with such as prolonged calving intervals, repeat breeder cows and losses due to early embryonic 74 death require more systematic recording. These reproductive problems are usually calculated 75 from the recorded reproductive events including calving date, insemination dates, and 76 pregnancy outcome [4]. Fertility measures calculated from reproductive events can be divided 77 into two categories as fertility scores and interval traits [5]. 78 Losses due to infectious diseases are typically reflected in impaired (re)productivity and lack 79 of market access. In a previous paper, a relatively high occurrence of exposure to infectious hectares and a high level of land fragmentation. Furthermore, there is a limited number of 95 animals per farm (1-10) with approximately 75% of the farms having one or two cows [14]. 96 Poor housing conditions followed by poor management linked to the inadequate education of 97 the farmers may also contribute substantially to low productivity and losses in animal 6 98 production. Management may also directly influence the occurrence of infectious agents. In this 99 case, management will have an indirect effect on reproduction through increasing exposure to 100 infectious agents. To be able to suggest sound interventions for improving reproductive 101 performance, there is a need to evaluate which factors are dominant and their causal 102 relationships with reproduction performance. As an improved production will also typically 103 imply larger herds, new factors may also become more critical as, e.g., the dynamics of 104 infections, that depend on herd size.

105
Although improvement of the cattle sector in BH partially depends on the views and motivation 106 of the farmers/ owners, the scientific data from farms often do not include the perspective of 107 the farmer/ owner. Importantly, their perceived main constraints on production will 108 complement the data collected through epidemiological studies.

110
The primary objective of the present study was to describe the importance of management 111 practices and to identify the causal interaction between management practices and selected 112 infectious diseases (C. abortus, C. burnetii, and N. caninum) on reproductive underperformance 113 and specific reproduction disorders. The secondary objective was to identify perceived 114 constraints that affect production in the cattle farms in the study area.

116
Study area and study design 117 The study was carried out in the north-western (Una-Sana Canton), western (Canton 10) and 118 central (Central Bosnia Canton) parts of BH, which covers a 23.9% of the total BH territory.

119
Based on a detectable herd-level prevalence of 5% and 95% level of confidence, a minimum 120 sample size of 60 farms (total n=180) was selected in each region to achieve the detection of 121 the agent. In the period from January 1 st to August 31 st , 2015, a total of 201 cattle farms were 122 visited, blood samples were collected, and a questionnaire-based interview was conducted. Four 123 farms were excluded due to insufficient farm records, leaving 197 farms in the final dataset for 124 statistical analyses (Fig 1). A complete farm sampling frame was unavailable, and farm 125 selection was made using the available list of farms at the municipal level, within each region.

126
In each farm, all female animals older than 12 months were selected for sampling. The breed N. caninum has not been implemented in BH, and hence, seropositive animals were considered 139 to be naturally infected. Details of this serological survey are given in a former companion 140 publication [6]. In the current study, we used the herd classification (presence or absence) for 141 these agents based on prior publication.

142
Reproductive outcomes (questionnaire) 143 The questionnaire was developed to collect information about management factors and 144 reproductive underperformance in selected farms. Subsequently, the questionnaire was pre-145 tested in a small number of farms (n=5). After the pre-testing, the questionnaire was The primary reproduction outcomes defined in the present study were abortion, stillbirth, 155 retention of placenta and metritis. Abortion was defined as any termination of the pregnancy 156 after its confirmation by transrectal palpation or no return to oestrus. Transrectal palpation was 157 done between 40 and 70 days after artificial insemination or natural breeding. In case of natural 158 breeding, the date of such event was recorded by the farmer/owner. Stillbirth was recorded as 159 a part of perinatal mortality. Perinatal mortality is defined as a death of the calf before, during 160 or within 48h of calving at full term [16]. Retained placenta was considered to be any retention 161 at 12 to 24h postpartum, while metritis was regarded as any condition with abnormal vaginal discharge, which was documented and treated by the veterinarian. We were not able to extract 163 figures on the frequency of reproductive outcomes in each farm and had to rely upon 164 information about farms' and veterinarians written records in the period of five years.

165
The infectious outcomes defined in the current study were C. abortus, C. burnetii and N. 166 caninum serostatus in the selected farms, based on results of our previous study [6]. Also, 167 seropositivity to the listed reproductive infections was considered as an exposure factor for the  Table 6.

174
The questionnaire data were entered into a Microsoft Excel ® spreadsheet before inspected and 175 merged with data from the previously reported serological survey (Softic et al., 2018). After 176 cleaning data and initial analyses in Microsoft Excel ® , using Pivot tables, data were transferred 177 to Stata (Stata/SE 15 for Windows, StataCorp, College Station, TX, USA) for further statistical 178 analyses.

179
Descriptive analyses and tabulation were done for each canton, and potential differences 180 between cantons were assessed using a chi-square test for categorical variables and the Kruskal-

181
Wallis test for continuous variables. After descriptive analyses, management-related factors 182 were screened for their influence on abortion, stillbirths, retained placenta and metritis, as well 183 as, infectious agents. As all reproduction and infection outcomes were coded as present/ absent 184 (seropositive/ seronegative herd), a mixed-effects univariable logistic regression was used to 185 screen management factors for potential statistical influence. Herd size effect on the occurrence of severe reproductive outcomes was expected, e.g. larger herds were more likely to report an 187 abortion, stillbirth, metritis or retention of the placenta. Hence, an adjustment was made for 188 herd size by including its natural logarithm (ln (herd size)) as an offset in the logistic model for 189 the disease outcomes. This was not done for the infection outcomes, as this was based upon a 190 more systematic sampling in the herds. Factors with a p-value less than or equal to 0.20 from 191 univariable analyses were used as candidate variables for subsequent multivariable modelling.

192
The multivariable mixed-effects logistic models were established using the melogit procedure 193 in Stata with canton as a random effect. Models were built using a backward selection procedure

194
[15] with a set p-value less than or equal to 0.05 (5%) of the Likelihood Ratio test as exclusion analysis, utilising all options built into Stata's multilevel mixed-effects models. The initial 208 model was constructed using the graphical sembuilder interface and then refined using the Stata 209 command-language syntax. The model was built using the same strategy as the logistic model, 210 but with a causal model specified. Additionally, the causal model was then specified graphically 211 using the online software for Directed Acyclical Graphs (DAGitty) (Fig 2) [20]. As SEM allows 212 an assessment of direct and indirect effects, we were able to reduce the standard logistic models 213 further into more realistic causal models.

214
Assumed farm production constraints were analysed graphically using pivot functions in 215 Microsoft Excel ® and visualised through a radar graph in Microsoft Excel ® . Graphs were split 216 according to herd size groups (quartiles), the variable found to determine the owners' 217 approaches to production constraints.

219
The study participants were distributed across three cantons of BH. The data were collected in 220 a total of 19 municipalities and 94 local communities. There were a total of 66 (33.5%) farms   Housing and management 231 The distribution of farms by housing characteristics in three cantons in BH is shown in Table   232 1. In most of the visited farms (193/197, 98%), animals were tethered in the barns on wooden 233 or concrete flooring.

234
Insert Table 1 here 235 A quarter of all farms used a combination of these two types of flooring. A total area of the 236 barn's facilities ranged from 20 to 4000 square meters, depending on herd size. Consequently, 237 the laying area ranged from 1.5 to 5 m 2 per animal. The majority (85%) of the farms had 238 separate pens for calves, while a small number of the farms had separate calving pens.

239
Substantial differences in farm management between cantons were observed. They are reflected 240 in the farm size, number of animals per farm, and the breed composition. Also, there was a substantial difference in the presence of dogs and poultry on the farms in three cantons. On the 242 other hand, there was no difference in feeding management between cantons.

243
Insert Table 2 here 244 In Table 2, basic information about herd sizes and management strategies are presented. The 245 overall median number of herd size in the selected farms was six animals (range from 1 to 584).   Table 3. Depending on the season, roughage was based on grazing, feeding by cereals, industrial by-265 products and hay. Predominantly larger farms, i.e. farms with the greater number of animals 266 (37.6%) have also used corn and grass silage. Water supply was ad-libitum usually coming 267 from public or private sources (Table 4).

268
Insert Table 4 Table 5. Reproductive problems such as abortion, stillbirth, 276 metritis, and retained placenta were observed with substantial differences between cantons.

277
Less observable disorders such as anoestrus, silent oestrus, and repeat breeder cows were 278 recorded, but information on these was incomplete and are not included in the paper.

279
Insert Table 5 here 280 All potential factors listed in Tables 1-5 were tested in the univariable logistic regression.  Table 7 shows the multivariable models for candidate variables that were associated with 285 abortion, stillborn calves, retention of the placenta and metritis, as well as with the occurrence 286 of N. caninum, C. burnetii, and C. abortus seropositive herds. Table 7 here   288   Table 8 Table 8.   Independent of herd size, the lack of market access due to production underperformance, and 310 low purchasing prices of bovine products were the most common production constraints of all 311 cattle farms in three cantons in BH (Fig 3). Canton were mostly small farms. Consequently, the number of animals per farm, and the breed 320 composition were found significantly different. Also, the presence of poultry, companion 321 animals (dogs and cats), wild birds, and rodents on the farms in three cantons were found 322 significantly different. There were no differences between feeding practices between cantons.

323
In this study, the SEM was used as a confirmatory approach for assessing potentially 324 unobserved direct associations between outcomes and predictors. Based on the results from the 325 multivariable logistic regression, four different causal complexes were identified. In the 326 following, we discuss the findings based on the SEM model in Table 8    Herds that have contact with other herds on the common pastures had more seropositive to C. 413 burnetii compared to isolated herds (OR=2.8). Contact with infected herds, contaminated 414 equipment or biological material or other intermediate host seemed to be a possible explanation 415 for this finding. Additionally, partial grazing system could favour seropositivity to C. burnetii.

416
After re-housing in the winter period, seropositivity could be partially explained by exposure 417 to aerosols or contaminated material [37]. Recording of reproductive disorders is still at insufficient levels in BH. This was also reflected 441 in our study, partially explained by the lack of an organised national cattle database. Farmers 442 have limited skills in recognising a reproductive problem, and they seem to only on exceptional 443 occasions seek professional assistance for contributing to data in the poor recording system.

444
Reproductive outcomes such as abortion, stillbirth, retained placenta, and metritis are perceived 445 as relatively severe, and the farmers can recognise the disease. Additionally, these outcomes 446 affect (re)production, which also can motivate for consultancy. Subtler reproductive outcomes 447 such as anoestrus and silent oestrus require more data and were not included in this study.

448
Unfortunately, farms do typically not keep proper farm records, so we were not able to assess 449 the frequency of these outcomes. Furthermore, farmers could overestimate or downscale the 450 importance of specific factors investigated in this study. This was partially prevented by 451 analysis of perceived production constraints.

479
This study has demonstrated that management practices contributed to the occurrence of severe 480 reproductive outcomes and reproductive infections. N. caninum seems to be an infectious agent 481 that substantially contributed to the reproductive underperformance, particularly in inducing 482 abortions and perinatal mortality. Also, there was a causal relationship between retention of 483 placenta and metritis, which seems to be associated with poor managerial decisions. Our results illustrate the benefits of applying the SEM approach in elucidating potentially causal pathways 485 in analysing complex epidemiological data.

488
The authors express sincere gratitude to all study participants, veterinarians and contributors 489 who helped in the development of this study.