Core Concepts
Incorporating herd-level epidemiological risk factors into the interpretation of the standard skin test for bovine tuberculosis can substantially improve the sensitivity of disease detection without compromising specificity.
Abstract
The content describes the development and evaluation of a machine learning-based model that augments the standard skin test for bovine tuberculosis (bTB) in cattle. The key highlights are:
The model was trained on a comprehensive dataset of over 1.3 million bTB skin test records in Great Britain, incorporating various herd-level risk factors.
By considering the epidemiological context of each test, the model was able to improve the herd-level sensitivity of the skin test from 63.8% to 78.4%, while maintaining the same herd-level specificity of 89.5%.
The model identified several important risk factors, including herd location, time since last breakdown, herd size, badger abundance, and animal movements, that influence the likelihood of a herd having a confirmed bTB breakdown.
Simulation modeling showed that applying the machine learning-augmented skin test could lead to a reduction in the number of confirmed bTB breakdowns and individual reactor animals, with the magnitude of the effect varying between high-risk and edge areas.
The authors discuss the potential for implementing this approach in practice, including the need for further regulatory and policy considerations.
Stats
The model was trained on a dataset of over 1.3 million bTB skin test records in Great Britain, covering the period from January 2012 to September 2021.
The dataset included the following key variables:
Herd-level result of the skin test (clear or not clear)
Date of the test
Month of the year the test was conducted
Whether the severe interpretation was applied
Number of animals tested
Herd location
Results of the two previous skin tests in the same herd
Time since the last test in the same herd
Time since the herd last entered breakdown
Number of prior interferon-gamma tests conducted on the herd
Test type (routine, pre-movement, etc.)
Herd type (dairy, beef, etc.)
Numbers of cattle moved into/out of the herd in various time periods
APHA bTB herd risk score
Mean badger abundance at the holding location
Veterinary practice that conducted the test
Tuberculin batch number used for the test