The Northern Ireland group held a meeting at 4pm on Wednesday the 23rd of October, 2013 in the David Bates Building in Queen's University Belfast. The speaker was Dr Gabrielle Kelly of the School of Mathematical Sciences, University College Dublin, Ireland.
Dr Kelly set out the epidemiological and statistical problems attending the evaluation of the 'Badgers cause bovine TB' hypothesis. These are by no means trivial and include the observations that: (a) bovine TB exists in the absence of infected badgers, (b) culling in some studies reduced the the incidence of bovine TB by 30% (leaving 70% of the incidence unaccounted for from this source) and (c) the there exists the possibility of cattle infecting badgers. She also noted that Irish badgers were genetically different from badgers in the UK. The problems in relation to statistical analysis relate largely to the quality and resolution of the data, given a pattern of relatively rare disease incidence.
Her talk centred on the analysis and results from the Irish Four Areas Project [G Kelly (2013), Statistical Communications in Infectious Diseases: 5(1), 1-16]. The four areas were counties in the Republic of Ireland: Cork, Donegal, Kilkenny and Monaghan. These geographically distant counties were studied for 5 years (divided into 2 periods; early, late). The study aims were to establish: (i) if there is spatial association between TB in cattle and badgers (ie, cross infection), (ii) if the spatial association changed over time and (iii) to present relevant statistical methodology.
Gabrielle used linear geostatistical models with spatial random effects having various covariance structures (geometrical isotropic and anisotopic) to fit the joint (stacked) badger and cattle data after first allowing for the effects of different covariates on the two species. The notion of spatial association between the species rests on setting various blocks of the covariance matrix equal to zero in order to test the null hypothesis of no association. Such tests are on the boundary of the parameter space and consequently the resulting likelihood ratio tests are not asymptotically distributed as chi-squared. Special adjustments are therefore required.
Overall, the results showed that there was only evidence of significant cross-infection in county Kilkenny (stage 1) and in county Monaghan (stage 2) therefore in only two, out of the eight county by stage, spatial-temporal units studied. Dr Kelly explained that there were many limitations to the study, which might have contributed to this finding, but, nevertheless, the audience was left with the impression that the evidence for cross-infectivity from this carefully conducted study was not compelling.
Dr Kelly's talk was received with general acclaim and there followed an interesting question and answer session.