Reproducibility of African giant pouched rats detecting Mycobacterium tuberculosis.

BACKGROUND: African pouched rats sniffing sputum samples provided by local clinics have significantly increased tuberculosis case findings in Tanzania and Mozambique. The objective of this study was to determine the reproducibility of rat results. METHODS: Over an 18-month period 11,869 samples were examined by the rats. Intra-rater reliability was assessed through Yule's Q. Inter-rater reliability was assessed with Krippendorff's alpha. RESULTS: Intra-rater reliability was high, with a mean Yule's Q of 0.9. Inter-rater agreement was fair, with Krippendorf's alpha ranging from 0.15 to 0.45. Both Intra- and Inter-rater reliability was independent of the sex of the animals, but they were positively correlated with age. Both intra- and inter-rater agreement was lowest for samples designated as smear-negative by the clinics. CONCLUSION: Overall, the reproducibility of tuberculosis detection rat results was fair and diagnostic results were therefore independent of the rats used.

Tuberculosis detection by pouched rats: Opportunities for reinforcement under low-prevalence conditions.

Giant African pouched rats (Cricetomys ansorgei) have been employed successfully in two operational tuberculosis-detection projects in which they sniff sputum samples from symptomatic individuals who have visited tuberculosis clinics. The prevalence of pulmonary tuberculosis in this population is high, approximately 20% in the regions where the rats have been used. If the rats are to be used to screen individuals from lower-prevalence populations, their performance under such conditions must first be evaluated. In this study, the prevalence of tuberculosis-positive samples presented to eight pouched rats was reduced to approximately 5%, and the percentage of known-positive samples included as opportunities for reinforcement was varied in sequence from 10 to 8, 6, 4, 2, 4, and 2. Liquid food reinforcers were delivered for identification responses to known-positive samples and at no other time. The rats' accuracy was clinically and statistically significantly lower at 2% than at the other values. These results indicate that the rats can perform well in low-prevalence scenarios but, if they are used under the conditions of the present study, at least 4% of the samples presented to them must be opportunities for reinforcement.