Quantifying the burden of vampire bat rabies in Peruvian livestock

Abstract

Background: Knowledge of infectious disease burden is necessary to appropriately allocate resources for prevention and control. In Latin America, rabies is among the most important zoonoses for human health and agriculture, but the burden of disease attributed to its main reservoir, the common vampire bat (Desmodus rotundus), remains uncertain.

Methodology/principal findings: We used questionnaires to quantify under-reporting of livestock deaths across 40 agricultural communities with differing access to health resources and epidemiological histories of vampire bat rabies (VBR) in the regions of Apurimac, Ayacucho and Cusco in southern Peru. Farmers who believed VBR was absent from their communities were one third as likely to report livestock deaths from disease as those who believed VBR was present, and under-reporting increased with distance from reporting offices. Using generalized mixed-effect models that captured spatial autocorrelation in reporting, we project 4.6 (95% CI: 4.4-8.2) rabies cases per reported case and identify geographic areas with potentially greater VBR burden than indicated by official reports. Spatially-corrected models estimate 505-724 cattle deaths from VBR in our study area during 2014 (421-444 deaths/100,000 cattle), costing US$121,797-171,992. Cost benefit analysis favoured vaccinating all cattle over the current practice of partial vaccination or halting vaccination all together.

Conclusions: Our study represents the first estimate of the burden of VBR in Latin America to incorporate data on reporting rates. We confirm the long-suspected cost of VBR to small-scale farmers and show that vaccinating livestock is a cost-effective solution to mitigate the burden of VBR. More generally, results highlight that ignoring geographic variation in access to health resources can bias estimates of disease burden and risk.

Fig 1. Reporting and vaccination rates per community across the study area.

(A) Map of Peru with the location of surveys (B) Zoomed map showing surveys in the Ayacucho, Apurimac and Cusco regions. Yellow triangles show the location of the SENASA offices where cases are reported (C) Blue pie-charts show the percentage of farmers reporting cattle mortality from a suspected infectious disease in each community (N = 10 per community). Districts that previously reported one or more laboratory-confirmed cases of VBR are coloured grey. (D) Red pie-charts show the percentage of farmers vaccinating their cattle against rabies. Country, region and district maps were obtained from the GADM (http://www.gadm.org//) database using the getData function from the raster package of R.

Fig 2. Calculation of multiplication factors from questionnaire and national surveillance data.

Diagram illustrating the estimation of uncorrected (MF_uncorrected) and spatially-corrected (MF_corrected) under-reporting multiplication factors. (Left) Events occurring between an outbreak on a farm and its report and confirmation through the national surveillance system of Peru. The probability of each event is shown in parenthesis, based on results from our surveys, national surveillance records and the literature (for FAT test). (Right) The derived multiplication factors are calculated from these probabilities. For MF_corrected, the spatial correction is applied to each outbreak by calculating probability p(x) of a farmer reporting a case, which is a function of its distance to the nearest reporting office. We show the value of the average MF_corrected based on confirmed outbreaks in 2014, with MF_corrected = ${\sum }_i^N\frac{1}{p\left(x_i\right)}$, where N = total number of outbreaks and p(x_i) = predicted reporting probability of outbreak i as a function of its distance x_i to the reporting office.

Fig 3. The effect of geographic isolation on the probability of reporting cattle deaths due to suspected infectious diseases.

Dots show the responses of farmers in relation to reporting the mortality of a sick cow (1 = reporting, 0 = no reporting) as a function of their least-cost distance to the reporting office (estimated with the least-cost function, see methods). Farms were located either in districts with confirmed VBR outbreaks since 2013 (black dots) or in districts without any confirmed outbreaks (white dots). Only a single district (Chalhuanca) had suspected, but no confirmed rabies cases. Lines show the prediction of the glmmPQL model predicting reporting probability by the distance to the SENASA office using all districts (solid line) or only endemic districts (dashed line). The latest prediction was used to calculate a spatially-corrected under-reporting rate.

Fig 4. Spatial distribution of livestock rabies cases across districts in southern Peru after correcting for the effect of geographic isolation on reporting.

(A) Relative percentage of cases per district across Ayacucho, Apurimac and Cusco regions that were reported to SENASA in 2014 (B) Relative percentage of cases per district in 2014 estimated after correcting by the effect of distance to the reporting office on reporting rates. (C) Relative difference between the official and the corrected number of cases in each district. (D) (E) and (F) are equivalent to (A) (B), (C) respectively for the period of 2003–2014. Districts coloured in beige do not have a VBR case reported and confirmed. Two cases from ‘Echarate’ district in Cusco were excluded from the 2003–2014 analysis given their very large distance to the reporting office, which generated a very high number of corrected cases and mislead the estimated relative proportion of cases of all other districts. Region and district maps were obtained from the GADM (http://www.gadm.org/) database using the getData function from the raster package of R.