Can more information be extracted from bovine TB skin test outcomes to inform animal risk management? A retrospective observational animal-level study

Abstract

Continual tailoring of control programmes of endemic pathogens during long-term eradication campaigns requires detailed analysis of surveillance data to inform evidence-based policy. Bovine tuberculosis is a disease where long-term control and eradication programs are in train in several countries. The primary diagnostic tool, the intradermal tuberculin test, used to identify infected animals can be interpreted using different criteria and cut-offs, facilitating flexibility in its use as a basis to inform interventions.We investigated the comparative risk of animals failing a single intradermal comparative tuberculin test (SICTT) based on their previous tuberculin test result following a higher risk test-type (reactor retest of an infected herd). The study was a retrospective cohort design, and the primary exposure was the test status following a reactor retest classified as mutually exclusive categories based on bovine and avian tuberculin reactions: standard interpretation inconclusive (Sdi), severe interpretation inconclusive (Svi), single intradermal test (SIT) reactors (SITr), SIT inconclusive (SITi), avian tuberculin reactors (Ar), and test negative animals. Random effects multivariable logistic regression was used to investigate future risk. Cross-validation and downscaling was used to explore model performance. Alternate models with differing outcome test types were also explored.The models were trained on 844,207 observations from June 2018 to June 2021. Sdi, Svi, SITr, SITi and Ar were associated with the following odds ratios 12.242 (95 %CIs: 5.236–28.625; p < 0.001), 4.101 (95 %CIs: 3.423–4.913; p < 0.001), 2.503 (95 %CIs: 1.878–3.338; p < 0.001), 1.741 (95 %CIs: 1.195–2.538; p = 0.004) and 1.065 (95 %CIs: 0.833–1.361; p = 0.616) for failing the next test, respectively. High model performance was achieved with inclusion of random effects for both training and test evaluation datasets (AUC: 0.94; Balanced accuracy: 0.84), but fixed-effects only predictions exhibited moderate performance (AUC: 0.70; Balanced accuracy: 0.69). This reflects that 55 % of the risk of test failure relates to between herd heterogeneity based on intra-class correlation, while controlling for fixed effects. Other factors that were associated with increasing risk included age (older cohorts were at greater risk than the youngest cohort), breakdown history of the herd (greater number of breakdowns prior to the study period), and the time between exposure test and outcome test.These results provide further evidence to inform risk-based management policies for TB, including the removal of Sdi animals in higher risk situations, supplementary testing of cattle based on tuberculin responses and the provision of risk management advice to herd owners. The results characterise the future animal-level risk posed by Svis and suggest this risk may require policy led interventions.