Search

Abstract Rabies causes 59,000 human deaths annually in over 150 countries. Mass dog vaccination (MDV) is key to controlling dog rabies, requiring 70% coverage in the susceptible dog population to eliminate rabies deaths. MDV campaigns must achieve geographical homogeneity of coverage. Although rabies elimination is feasible, operation challenges exist, especially in hard-to-reach areas. We conducted geospatial modelling to identify geographical factors affecting MDV success in terms of campaign completeness and vaccination coverage across 25 districts in south-eastern Tanzania. From October 2016 to January 2017, about 81,000 dogs were vaccinated in 1,379 (68%) villages within these districts. Multivariable regression analysis revealed that land cover, residence, poverty, and elevation were associated with campaign completeness. The odds of achieving completeness in croplands were 1.75 times higher compared to forests. Vaccination coverage was influenced by residence, area, poverty index, and elevation, with urban areas having lower odds of achieving coverage compared to rural areas. Coverage probabilities exceeding 70% were lower on islands, highlands, urban areas, and areas bordering protected areas. As the 2030 deadline for "zero dog-mediated human rabies deaths" approaches, operational and logistical challenges in MDV campaigns persist. Our data provide insights into MDV success and failure, guiding future control efforts to improve their effectiveness.

Abstract A lack of methods to identify individual animals can be a barrier to zoonoses control. We developed and field-tested facial recognition technology for a mobile phone application to identify dogs, which we used to assess vaccination coverage against rabies in rural Tanzania. Dogs were vaccinated, registered using the application, and microchipped. During subsequent household visits to validate vaccination, dogs were registered using the application and their vaccination status determined by operators using the application to classify dogs as vaccinated (matched) or unvaccinated (unmatched), with microchips validating classifications. From 534 classified dogs (251 vaccinated, 283 unvaccinated), the application specificity was 98.9% and sensitivity 76.2%, with positive and negative predictive values of 98.4% and 82.8% respectively. The facial recognition algorithm correctly matched 249 (99.2%) vaccinated and microchipped dogs (true positives) and failed to match two (0.8%) vaccinated dogs (false negatives). Operators correctly identified 186 (74.1%) vaccinated dogs (true positives), and 280 (98.9%) unvaccinated dogs (true negatives), but incorrectly classified 58 (23.1%) vaccinated dogs as unmatched (false negatives). Reduced application sensitivity resulted from poor quality photos and light-associated color distortion. With development and operator training, this technology has potential to be a useful tool to identify dogs and support research and intervention programs.