Using machine learning to improve bovine tuberculosis control in herd level outbreaks.

Bovine tuberculosis (bTB), a chronic disease of cattle, is caused by the Mycobacterium bovis infection. Despite having a serious social and economic impact in the United Kingdom and Ireland, there is no antemortem gold standard diagnostic test. Tuberculin skin tests (CICT) are commonly used as a control measure with the interferon gamma (IFN-γ) assay being applied in certain circumstances. This paper utilizes data gathered describing tuberculin regression in reactors (test positive cattle) following the CICT at 72 ± 4 h post injection in herds with large bTB outbreaks. The work then applies machine learning techniques (Decision Trees, Bagging Trees and Random Forests, alongside several balancing approaches) to predict which cattle were likely to be truly infected with tuberculosis, enabling identification of atypical breakdowns that require extra investigation and providing a mechanism for quality assurance of the existing CICT bTB surveillance scheme. The analysis showed that Random Forests (RF) trained using SMOTE balancing had the joint best performance and accuracy (0.90). The importance of the two components of the interferon gamma assay within the RF model also indicated that varying the assay threshold for large outbreaks would be beneficial. Furthermore, the combined use of the RF and IFN- γ models could lead to the improved detection of infection within breakdown herds, reducing the scale and duration of outbreaks. An additional use of these models would be for quality assuring the current bTB surveillance based on CICT and post mortem inspection. Quality control is well recognized as an essential component of a disease surveillance/eradication programme.Clinical Relevance- Bovine tuberculosis remains a disease that is hard to control on a national level. The use of the machine learning model could lead to significant improved detection of infection within breakdown herds, reducing the scale and duration of outbreaks. Advanced modelling, such as this, has the potential to strengthen the efficacy of disease surveillance and the eradication strategy and can meaningfully contribute to animal disease national control plans.

The CP-ABM approach for modelling COVID-19 infection dynamics and quantifying the effects of non-pharmaceutical interventions.

The motivation for this research is to develop an approach that reliably captures the disease dynamics of COVID-19 for an entire population in order to identify the key events driving change in the epidemic through accurate estimation of daily COVID-19 cases. This has been achieved through the new CP-ABM approach which uniquely incorporates Change Point detection into an Agent Based Model taking advantage of genetic algorithms for calibration and an efficient infection centric procedure for computational efficiency. The CP-ABM is applied to the Northern Ireland population where it successfully captures patterns in COVID-19 infection dynamics over both waves of the pandemic and quantifies the significant effects of non-pharmaceutical interventions (NPI) on a national level for lockdowns and mask wearing. To our knowledge, there is no other approach to date that has captured NPI effectiveness and infection spreading dynamics for both waves of the COVID-19 pandemic for an entire country population.

Expediting assessments of database performance for streams of respiratory parameters.

A new methodology is proposed to compare database performance for streams of patient respiratory data from patients in an intensive care unit. New metrics are proposed through which databases may be compared both for this and similar streaming applications in the domain of the Internet of Things. Studies are reported using simulated patient data for four freely available databases. The statistical technique of non-parametric bootstrapping is used to minimise the total running time of the tests. We report mean values and bias corrected and accelerated confidence intervals for each metric and use these to compare the databases. We find that, among the four databases tested, ScaleDB is an optimum database technology when handling between 200 and 800 patients in this application, while PostgreSQL performs best outside of this range. Comparing the non-parametric bootstrapping method to a complete set of tests shows that the two approaches give results differing by a few percent.