Prevalence of Bovine Tuberculosis in India: A systematic review and meta‐analysis

Abstract Bovine tuberculosis (bTB) is a chronic disease of cattle that impacts productivity and represents a major public health threat. Despite the considerable economic costs and zoonotic risk consequences associated with the disease, accurate estimates of bTB prevalence are lacking in many countries, including India, where national control programmes are not yet implemented and the disease is considered endemic. To address this critical knowledge gap, we performed a systematic review of the literature and a meta‐analysis to estimate bTB prevalence in cattle in India and provide a foundation for the future formulation of rational disease control strategies and the accurate assessment of economic and health impact risks. The literature search was performed in accordance with PRISMA guidelines and identified 285 cross‐sectional studies on bTB in cattle in India across four electronic databases and handpicked publications. Of these, 44 articles were included, contributing a total of 82,419 cows and buffaloes across 18 states and one union territory in India. Based on a random‐effects (RE) meta‐regression model, the analysis revealed a pooled prevalence estimate of 7.3% (95% CI: 5.6, 9.5), indicating that there may be an estimated 21.8 million (95% CI: 16.6, 28.4) infected cattle in India—a population greater than the total number of dairy cows in the United States. The analyses further suggest that production system, species, breed, study location, diagnostic technique, sample size and study period are likely moderators of bTB prevalence in India and need to be considered when developing future disease surveillance and control programmes. Taken together with the projected increase in intensification of dairy production and the subsequent increase in the likelihood of zoonotic transmission, the results of our study suggest that attempts to eliminate tuberculosis from humans will require simultaneous consideration of bTB control in cattle population in countries such as India.


| INTRODUCTION
Bovine tuberculosis (bTB) is a chronic granulomatous inflammatory disease that is predominantly caused by Mycobacterium bovis. While primarily affecting bovines, the pathogen has a broad host range that includes humans. It has been estimated that M. bovis causes~10% of the total human TB cases in developing countries and subsequently poses a significant threat to global health (Olea-Popelka et al., 2014) (Etchechoury et al., 2010) ("OIE, Bovine Tuberculosis: General Disease Information sheets,"). Prior to mandatory pasteurization of milk in many countries, M. bovis accounted for~25% of all TB cases in children (Roswurm & Ranney, 1973). In addition to being a threat to public health, bTB is also a major economic concern, costing an estimated USD 3 billion worldwide annually due to losses from reduced cattle productivity, culling and movement and trade restrictions (Waters, Palmer, Buddle, & Vordermeier, 2012).
Bovine TB is well controlled in most developed countries where national control programmes have been implemented, although complete eradication and maintenance of bTB-free status are challenging given the potential of spillover from wildlife reservoir hosts. Such control programmes for bTB were successfully adopted over a century ago in many developed countries by applying test and cull strategies, resulting in enormous benefits to human health and more than 10-fold return on investment in animal productivity (Olmstead & Rhode, 2004). In contrast, bTB remains endemic in developing countries like India that lack disease control programmes because of the associated economic costs and social barriers to test and cull strategies. This current level of endemicity is likely to increase in the coming years due to a confluence of factors including the growing intensification of dairy and cattle farming and increased emphasis on improving animal productivity in these countries.
In conjunction with possessing the largest population of cattle in the world (nearly 300 million cows and buffaloes) (Basic Animal Husbandry and Fisheries Statistics, Government of India 2017), India's lack of a control programme poses a potential threat for bTB infection and transmission worldwide. In the absence of a national surveillance programme, accurate prevalence data are lacking and, to our knowledge, there has thus far not been a comprehensive review of the existing literature to determine an estimate of the overall prevalence of bTB in the country. Such an estimate will prove crucial in future efforts to accurately assess risk and inform policy for the development of effective control strategies. In this systematic review and meta-analysis, we sought to address this critical gap and determine the overall prevalence of bTB in the cattle of India. This systematic review conforms to Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA) guidelines (Liberati et al., 2009).

| Literature search strategy
A systematic search for published articles reporting prevalence data for bTB in cows and buffaloes in India was conducted on 11th September 2017. The four databases used in our search (CAB Direct, Web of Science, Web of Science Biological Abstracts and PubMed) were selected in order to comprehensively capture articles published in both international and local journals and minimize journal biases.
After examining common MeSH terms for pre-identified and relevant publications, the following search terms were used across all four databases: (("mycobacterium bovis" OR tuberculosis) AND (cows OR cattle OR bovine) AND (epidemiolog* OR prevalen* OR inciden* OR surve*) AND (India)). No restrictions were placed on the date of publication. The citation software program EndNote X8 (Clarivate Analytics, Philadelphia, PA) was used to organize and remove duplicate articles between the databases. Additional articles were also identified manually from the reference lists of articles generated in the database search.

| Study inclusion criteria
The inclusion/exclusion criteria for data extraction are detailed in Table 1. Included studies reported the prevalence of bTB in cows and/or buffaloes in India based on commonly accepted methods for the diagnosis of bTB. More specifically, studies whose main objectives were not to determine bTB prevalence but required a preliminary prevalence study for determining initial disease status were included as long as data were reported and animals were not preselected for bTB symptoms. Prevalence studies that examined the effects of bTB control strategies were excluded in order to avoid the introduction of potential sampling bias, as the primary aims of these studies were to compare the effectiveness of control strategies. For instance, Dhanda et al. have reported an increase in prevalence in herds at Puri, Orissa, from 9.1% in 1937 to 84.7% in 1942 (Dhanda & Lall, 1959). The cattle populations that were tested were part of farms that did not practice any bTB control strategies. We believe that inclusion of studies conducted on pre-selected herds as

| Statistical analysis
All quantitative analyses were performed in RStudio (version 1.0.143) ("R core team, R: A language and environment for statistical computing.," R core team 2016) where the "meta" package was used to estimate models (Schwarzer, 2007) (Viechtbauer, 2010).
Codes used for the statistical analyses are publicly available at https://doi.org/10.18113/d37s9x. The prevalence estimates from individual studies were logit-transformed, and the pooled prevalence was estimated using meta-analytic models. Cochran's Q statistic (Cochran, 1954) was computed to test for heterogeneity, and Higgin's statistic (Higgins, Thompson, Deeks, & Altman, 2003) (I 2 > 50% represents at least moderate heterogeneity) helped describe the variability in the pooled prevalence estimate due to heterogeneity between studies.
A univariate screen was used to select a parsimonious set of moderator factors for multivariate analysis. Diagnostic test type was excluded from this selection procedure and forced into the final model in order to adjust for the well-known variability in sensitivity and specificity of diagnostic tests for bTB (Farnham, Norby, Goldsmith, & Wells, 2012). Univariable meta-regression models were estimated using both the random-effects (RE) and fixed-effects (FE) models for each potential moderator variable. Analysis of variance The fit of the resulting multivariable meta-regression models and evidence of publication bias was assessed through funnel plots, Egger's asymmetry test (Egger, Davey Smith, Schneider, & Minder, 1997) and Begg's rank correlation (Begg & Mazumdar, 1944) test.
To visualize the prevalence of bTB in the different states of India, we generated a map utilizing an open-source library called D3.js (Data-Drive Documents) (Bostock, Ogievetsky, & Heer, 2011).
This allowed us to plot data positions via the centroids of given shapefile locations represented in the map and control graphical elements based on their values (Cleveland & McGill, 1984). We utilized a continuous log scale for circle size to represent bTB prevalence and a power function for circle lightness to represent the confidence in the prevalence estimates of each state, ratifying values to visual variables on a linear scale (Bertin, 1983).

| Characteristics of included studies
From the 285 publications screened, 44 articles were included in the systematic review ( Figure 1). In the instance that a publication reported prevalence data for multiple states, years, cattle breeds, species or production systems, they were considered as separate strata level data. A total of 106 strata level data were extracted from the 44 articles for meta-analysis. For example, as seen in https://doi.org/10.18113/ d37s9x, the study by Iyer (1944) has been extracted into three strata level data, the strata being the three locations in which the study was performed. The same was done for other studies that included data on different production system, breed, species, etc. These studies included in the quantitative analyses spanned from 1942 to 2016 and provided bTB prevalence data for a total sample size of 82,419 of which 29,037 were buffaloes and 53,382 were cows ( and 4 mm. We did not use any cut-off values on the number of animals for classification of the various production systems. Most included publications explicitly mentioned the type of production system that was used in their studies. In the instance that a study did not specify the production system, we did not include that study under any production system strata. To examine any effect of time on the prevalence of bTB, the study periods were separated into four time intervals : 1941-1960; 1961-1980; 1981-2000; and, 2001-2016.

| Meta-analysis
To assess for potential publication bias, a funnel plot was constructed of the logit prevalence against standard error ( Figure 2).
The lack of symmetry in the funnel plot illustrates potential publication bias towards smaller studies with lower prevalence. Egger's asymmetry test was significant and showed presence of bias (p- This evidence of publication bias suggests that RE model will be more appropriate for these data. To explore this further, we estimated both random-and fixed-effects models and constructed funnel plots to compare their fit and look for evidence of systematic bias (Supporting information Figures S1a,b)

| Univariable meta-regression
Due to the presence of statistical heterogeneity, we conducted univariable meta-regression in order to determine the effect of studylevel covariates on the estimates of cumulative prevalence. The moderators considered for the analyses were study period, study location, sample size, production system, species, cattle breed and diagnostic technique used.
As seen in Table 3, the proportion of each predictor variable's effect on heterogeneity (R 2 ) ranged from 0% to 16.5% in the RE model. Further, under the RE model, the highest value of R 2 was observed for study location while, diagnostic technique, and sample size exhibited no effect on heterogeneity (R 2 = 0%).

| Multivariable meta-regression
All moderators from the univariable meta-regression were subjected to multivariable meta-regression (

| Effect of moderators on prevalence of bTB
Prevalence estimates using both the RE and FE models are reported in

| Geographical distribution of included studies in India
Study reports from included publications encompassed 18 states and one union territory in India. No reports were found for Arunachal Lakshadweep, comprising a total of 11 states and six union territories. It can be observed from the map that the prevalence of bTB varied highly between states (Figure 4) (Table 7).

| DISCUSSION
After screening of 285 publications, we extracted data from 44 cross-sectional studies published in peer-reviewed journals that report the prevalence of bTB in India and conducted meta-analysis.
The pooled prevalence estimate (RE model) for all of India was found to be 7.3% (95% CI: 5.6, 9.5). Despite being a disease of antiquity with significant animal and public health costs that have been controlled in most developed countries over a half-century ago, bTB has a high and widespread prevalence in India as no national control strategies have been implemented in the country (Figure 4). These data suggest that India, as the world's largest producer of milk (~156 MMT), accounting for~18.5% of the world's total milk production and the world's largest red meat exporter (~1.9 MMT), has an urgent and as yet unmet need for control of bTB for both economic and public health reasons (DADF, 2015).
Following the White Revolution (Bellur, Singh, Chaganti, & Chaganti, 1990) (Nair, 1985), a rural development programme in India that resulted in making India the largest producer of milk and milk F I G U R E 3 Forest plot visualizing the varying bTB prevalence reported for each included publication in the meta-analysis. Weightage given to each included publication by both RE and FE models have been shown for rigorous comparison. "Total" refers to the number of animals in each publication, while "Events" refers to the number of bTB-positive animals. "Proportion" reports the bTB prevalence for each publication Overall, the ordering of prevalence estimates determined using the FE model for different production systems follows the same trend as in the RE model (i.e., prevalence in Gaushala > Organized farms > Rural farms) (Table 6). However, given the observed heterogeneity in the studies, it is difficult to assess the validity of the FE model, and hence, further study is necessary to clarify the exact influence that each production system has on bTB prevalence before definitive conclusions can be made. We note that accurate estimates of prevalence rates for each production system are particularly important in the Indian context where the magnitude of animals housed in Gaushalas and the increasing population of cattle being reared under intensive conditions have the potential to considerably impact overall prevalence and influence assessment of bTB transmission rates and targeted interventions.
The high prevalence observed in buffaloes using the FE model is most likely driven by a single study that sampled 21,592 buffaloes (of a total buffalo sample size of 29,037 included in this meta-analysis) and recorded a prevalence of 17.4% (Dhanda & Lall, 1959). As per the Government of India's Department of Animal, Dairy and Fisheries (DADF) 2016-2017 Annual report, the share of milk contribution from buffaloes is 49% and that of cows is 48% (DADF, 2016(DADF, -2017. Assuming a conservative 10% loss in milk productivity due to bTB (Thoen, 2008) and the overall estimated bTB prevalence rates based on the RE model, the annual costs to farmers only from loss in milk production in cows and buffaloes in India are estimated to range from 375 to 544 million USD (Supporting information Table S1). We note that the need for intensification of dairy production to meet increased milk demand and national priorities for nutritional improvement and rural development is likely to significantly increase bTB disease prevalence as the disease is known to more easily spread amongst intensively reared cattle. With the inevitable increase in bTB prevalence, this already large economic cost will only continue to grow if no intervention measures are implemented.
Published studies on the influence of breed on genetic susceptibility to bTB showed that the native breed of cattle is more resistant to the disease than exotic breed   1925) (Liston & Soparkar, 1917) (Sharma, Vanamayya, & Parihar, 1985), affirming a generally held and commonly disseminated dogma. In contrast, our results note no significant differences in bTB prevalence between cow breeds in either the RE or the FE models (Table 3). However, given the heterogeneity observed in the studies, rigorous investigations of the true differences in susceptibility amongst different cattle breeds to bTB will be essential for evidence-based formulation of a rational approach to control this disease in India.
India has a cattle population of nearly 300 million, and we We note that M. bovis may not necessarily be the only causative agent of bTB in all reactor animals as isolation of M. tuberculosis from cattle samples has also been reported (Srivastava et al., 2008) (Sweetline Anne, Ronald, Kumar, Kannan, & Thangavelu, 2017).
The multivariable logistic regression model accounted for~31% of the heterogeneity between studies, suggesting that additional factors not part of our model are also contributors to bTB prevalence.
These factors may include animal age, sex and herd size that were not represented with enough frequency in the papers included in the systematic review to be subject to robust and rigorous metaanalysis. Hence, future studies should strive to understand how these factors contribute to overall bTB prevalence.
We note that the findings of this study must be considered in conjunction with the limitations inherent in systematic reviews and metaanalyses. For example, studies were limited to those included in the four databases used and no single study sample can provide a perfect representation of the cattle in a state or across the country. Further, our review consisted of only published studies written in English and did not capture any unpublished data, subjecting our analysis to F I G U R E 4 Geographical distribution and pooled prevalence estimates (RE model) of bTB in the different states of India. For the confidence intervals, a power scale was used to map colour lightness and represented as y = mx k + b, where k = 0.5. Prevalence was mapped to a logscale where data were uniformly corrected with allow for visual properties. Note that although scales were altered, the original data set is provided in Table 7 to afford accurate measures [Colour figure can be viewed at wileyonlinelibrary.com] publication bias (Figure 4). This potential publication bias is supported by the significant result of Egger's regression test, the test with greater statistical power (Hayashino, Noguchi, & Fukui, 2005).
In addition, our findings are limited by the variation in experimental design and methodology of each included article. Variation in the reporting details of each article also contributes to variations in study quality. It is important to note that the pooled prevalence estimate of 7.3% (95% CI: 5.6, 9.5) was derived from studies that, despite efforts during the screening process to exclude studies with biased sampling methods, may not have been entirely random surveys and needs to be refined with proper cross-sectional national level surveys using internationally recognized and well-standardized methods. In addition, acceptance of each study's reported number of positive animals as truly positive is also a limiting but unavoidable reality of lacking access to the original data and an inability to know the sensitivity, specificity and other performance characteristics of the tests used. In addition, while most studies' objectives were to determine bTB prevalence in the study population, studies differed in their examination or reporting of specific moderators of interest.
Taken together, the meta-analysis highlights a critical and hitherto unmet need for the development of a national surveillance programme and the implementation of an effective strategy for control of bTB in India-a need that will only continue to grow in conjunction with India's increasing cattle population and demands on milk production and an inability to cull potentially diseased cows. Given the likely inability of implementing a test and cull programme at any scale due to social and economic considerations, the need for a vaccine that can reduce the burden of infection and transmission is critical. In this context, we note that recent reports suggest that the century-old BCG vaccine may have considerable utility in this regard (Ameni, Vordermeier, Aseffa, Young, & Hewinson, 2010)  , but requires further study to evaluate its ability in reducing onward transmission. However, if effective, there is also an unmet need for a validated and accepted fit-forpurpose DIVA (Differentiating Infected from Vaccinated Animals) diagnostic test for the detection of bTB-infected cattle that can be used in conjunction with a vaccine programme.
Mycobacterium bovis has also been isolated from milk samples of tuberculous cattle (Aswathanarayana, Rao, Krishnappa, Ramanatha, & Raghavan, 1998) (Veerasami et al., 2012). Given the fact that over 70% of the milk in India is sold unpasteurized (FAO/OIE/WHO, 1993), this raises concerns regarding the potential for zoonotic transmission of bTB and continued spread of human tuberculosis (India has the world's largest burden of human TB) (Thoen, LoBue, & de Kantor, 2006). In May 2014, the World Health Assembly adopted a new strategy to attain an ambitious goal of ending the global TB | 1637 epidemic by 2035: the End TB strategy (Uplekar et al., 2015). Given the prevalence of bTB and the potential for zoonotic transmission, particularly to children and others who consume unpasteurized or unprocessed milk from infected cows, there is a critical need for a national bTB control programme in India and other developing countries as attempts to eradicate the disease from humans without eradicating it from cattle are likely to prove futile. Importantly, implementation of a national control programme would not only enable accurate temporal trends and estimates of bTB prevalence, risk and economic costs, but would equally importantly improve the health and productivity of cattle in India.

| CONCLUSION
Overall, the results of our systematic review and meta-analysis conducted on 44 publications indicate high and widespread bTB prevalence in India of 7.3% (95% CI: 5.6, 9.5). Further study is necessary to obtain more robust state-by-state prevalence estimates and explore other moderators of risk (including herd size, animal sex, and age, amongst others) that are likely to impact development and implementation of a rational and effective bTB control strategy.
Taken together with the expected dairy farm intensification, growing demands for increased milk production and the zoonotic nature of M. bovis, the results of our current studies highlight the importance of developing and implementing a national bTB control programme that will need to include a national surveillance plan using (a) wellstandardized method(s) and evidence-based intervention(s) that are likely to work in India and other developing country settings.