Dataset Information

Using machine learning improves predictions of herd-level bovine tuberculosis breakdowns in Great Britain.

ABSTRACT: In the United Kingdom, despite decades of control efforts, bovine tuberculosis (bTB) has not been controlled and currently costs ~?£100 m annually. Critical in the failure of control efforts has been the lack of a sufficiently sensitive diagnostic test. Here we use machine learning (ML) to predict herd-level bTB breakdowns in Great Britain (GB) with the aim of improving herd-level diagnostic sensitivity. The results of routinely-collected herd-level tests were correlated with risk factor data. Four ML methods were independently trained with data from 2012-2014 including ~?4700 positive herd-level test results annually. The best model's performance was compared to the observed sensitivity and specificity of the herd-level test calculated on the 2015 data resulting in an increased herd-level sensitivity from 61.3 to 67.6% (95% confidence interval (CI): 66.4-68.8%) and herd-level specificity from 90.5 to 92.3% (95% CI: 91.6-93.1%). This approach can improve predictive capability for herd-level bTB and support disease control.

SUBMITTER: Stanski K

PROVIDER: S-EPMC7838174 | biostudies-literature | 2021 Jan

REPOSITORIES: biostudies-literature

ACCESS DATA

Publications

Using machine learning improves predictions of herd-level bovine tuberculosis breakdowns in Great Britain.

Stański K K Lycett S S Porphyre T T Bronsvoort B M de C BMC

Scientific reports 20210126 1

In the United Kingdom, despite decades of control efforts, bovine tuberculosis (bTB) has not been controlled and currently costs ~ £100 m annually. Critical in the failure of control efforts has been the lack of a sufficiently sensitive diagnostic test. Here we use machine learning (ML) to predict herd-level bTB breakdowns in Great Britain (GB) with the aim of improving herd-level diagnostic sensitivity. The results of routinely-collected herd-level tests were correlated with risk factor data. F ...[more]

PMID: 33500436

Similar Datasets

Project description:The number of cattle herds placed under movement restrictions in Great Britain (GB) due to the suspected presence of bovine tuberculosis (bTB) has progressively increased over the past 25 years despite an intensive and costly test-and-slaughter control program. Around 38% of herds that clear movement restrictions experience a recurrent incident (breakdown) within 24 months, suggesting that infection may be persisting within herds. Reactivity to tuberculin, the basis of diagnostic testing, is dependent on the time from infection. Thus, testing efficiency varies between outbreaks, depending on weight of transmission and cannot be directly estimated. In this paper, we use Approximate Bayesian Computation (ABC) to parameterize two within-herd transmission models within a rigorous inferential framework. Previous within-herd models of bTB have relied on ad-hoc methods of parameterization and used a single model structure (SORI) where animals are assumed to become detectable by testing before they become infectious. We study such a conventional within-herd model of bTB and an alternative model, motivated by recent animal challenge studies, where there is no period of epidemiological latency before animals become infectious (SOR). Under both models we estimate that cattle-to-cattle transmission rates are non-linearly density dependent. The basic reproductive ratio for our conventional within-herd model, estimated for scenarios with no statutory controls, increases from 1.5 (0.26-4.9; 95% CI) in a herd of 30 cattle up to 4.9 (0.99-14.0) in a herd of 400. Under this model we estimate that 50% (33-67) of recurrent breakdowns in Britain can be attributed to infection missed by tuberculin testing. However this figure falls to 24% (11-42) of recurrent breakdowns under our alternative model. Under both models the estimated extrinsic force of infection increases with the burden of missed infection. Hence, improved herd-level testing is unlikely to reduce recurrence unless this extrinsic infectious pressure is simultaneously addressed.

Project description:Bovine tuberculosis (bTB), caused by Mycobacterium bovis, remains a significant problem for livestock industries in many countries worldwide including Northern Ireland, where a test and slaughter regime has utilised the Single Intradermal Comparative Cervical Tuberculin (SICCT) test since 1959. We investigated the variation in post-mortem confirmation based on bTB visible lesion (VL) presence during herd breakdowns using two model suites. We investigated animal-level characteristics, while controlling for herd-level factors and clustering. We were interested in potential impacts of concurrent infection, and therefore we assessed whether animals with evidence of liver fluke infection (Fasciola hepatica; post-mortem inspection), M. avium reactors (animals with negative M. bovis-avium (b-a) tuberculin reactions) or Bovine Viral Diarrhoea Virus (BVDV; RT-PCR tested) were associated with bTB confirmation.The dataset included 6242 animals removed during the 14 month study period (2013-2015). bTB-VL presence was significantly increased in animals with greater b-a reaction size at the disclosing SICCT test (e.g. b-a?=?5-9 mm vs. b-a?= 0 mm, adjusted Odds ratio (aOR): 14.57; p?<?0.001). M. avium reactor animals (b-a?<?0) were also significantly more likely to disclose VL than non-reactor animals (b-a?= 0; aOR: 2.29; p?=?0.023). Animals had a greater probability of exhibiting lesions with the increasing number of herds it had resided within (movement; log-herds: aOR: 2.27-2.42; p?<?0.001), if it had an inconclusive penultimate test result (aOR: 2.84-3.89; p?<?0.001), and with increasing time between tests (log-time; aOR: 1.23; p?=?0.003). Animals were less likely to have VL if they were a dairy breed (aOR: 0.79; p?=?0.015) or in an older age-class (e.g. age-quartile 2 vs. 4; aOR: 0.65; p?<?0.001). Liver fluke or BVDV variables were not retained in either multivariable model as they were non-significantly associated with bTB-VL status (p?>?0.1).Our results suggest that neither co-infection of liver fluke nor BVDV had a significant effect on the presence of VLs in this high-risk cohort. M. avium tuberculin reactors had a significantly increased risk of disclosing with a bTB lesion, which could be related to the impact of co-infection with M. avium subsp. paratuberculosis (MAP) affecting the performance of the SICCT however further research in this area is required. Movements, test history, breed and age were important factors influencing confirmation in high-risk animals.

Project description:Bovine tuberculosis is endemic in cattle herds in Great Britain, with a substantial economic impact. A reservoir of Mycobacterium bovis within the Eurasian badger (Meles meles) population is thought to have hindered disease control. Cattle herd incidents, termed breakdowns, that are either 'prolonged' (lasting ? 240 days) or 'recurrent' (with another breakdown within a specified time period) may be important foci for onward spread of infection. They drain veterinary resources and can be demoralising for farmers. Randomised Badger Culling Trial (RBCT) data were re-analysed to examine the effects of two culling strategies on breakdown prolongation and recurrence, during and after culling, using a Bayesian hierarchical model. Separate effect estimates were obtained for the 'core' trial areas (where culling occurred) and the 'buffer' zones (up to 2 km outside of the core areas). For breakdowns that started during the culling period, 'reactive' (localised) culling was associated with marginally increased odds of prolongation, with an odds ratio (OR) of 1.7 (95% credible interval [CI] 1.1-2.4) within the core areas. This effect was not present after the culling ceased. There was no notable effect of 'proactive' culling on prolongation. In contrast, reactive culling had no effect on breakdown recurrence, though there was evidence of a reduced risk of recurrence in proactive core areas during the culling period (ORs and 95% CIs: 0.82 (0.64-1.0) and 0.69 (0.54-0.86) for 24- and 36-month recurrence respectively). Again these effects were not present after the culling ceased. There seemed to be no effect of culling on breakdown prolongation or recurrence in the buffer zones. These results suggest that the RBCT badger culling strategies are unlikely to reduce either the prolongation or recurrence of breakdowns in the long term, and that reactive strategies (such as employed during the RBCT) are, if anything, likely to impact detrimentally on breakdown persistence.

Dataset Information

Using machine learning improves predictions of herd-level bovine tuberculosis breakdowns in Great Britain.

Publications

Using machine learning improves predictions of herd-level bovine tuberculosis breakdowns in Great Britain.

Similar Datasets

OmicsDI is part of the ELIXIR infrastructure