Data

Livestock and human case data for the 2018–19 Mayotte RVF epidemic were provided by the Veterinary Services of Mayotte, the CoopADEM (Coopérative des Eleveurs de Mayotte), and Santé Publique France. Livestock referred to cattle, sheep, and goats. Farmers were sensitized to the disease and encouraged to report clinical signs suggestive of RVFV infection (e.g. abortions or pyrexia) to the animal disease surveillance system (SESAM, Système d’épidémiosurveillance animale à Mayotte). Suspected livestock identified through this passive surveillance system were tested for RVFV by RT-PCR [20], and those that tested positive were defined as livestock cases of RVF. The proportion of farms with a declared veterinarian was only slightly higher in central (30.7%) than outer communes (26.6%) [21]. Also, due to the small size of the island (374km²), all farms in Mayotte were located less than one hour away, by car, from the nearest private veterinary practice (Dommergues, personal communication). This suggests that the coverage of veterinary services, and the level of underreporting were likely to be similar across the island. Human cases of RVF were individuals who tested positive by RT-PCR [20] following a consultation with a general practitioner for a dengue-like syndrome. For each case, animal or human, the sampling date and the commune where the farm was located, or where the person was a resident, were recorded. While samples were collected from humans on the day of their medical consultation, samples were collected from animals within two weeks following the onset of clinical signs suggestive of an RVFV infection (Dommergues, personal communication).

We used livestock movement records for the period of 2007–14, since such data were not available for 2018–19. A previous analysis of these data showed that the yearly networks were strongly correlated with one another, suggesting that the overall structure of the Mayotte livestock movement network was stable during that period [11]. Also, while the size of livestock populations in each commune likely varied over the past decade, the ranking of those communes according to their number of livestock was similar between the 2010 agricultural census [22] and a survey conducted in 2015 [23] (p = 0.015, Spearman’s rank correlation ρ = 0.61). Moreover, there was no livestock movement restriction during the epidemic. Therefore, assuming that the network structure remained stable over the following six years including the 2018–2019 epidemic, we computed the average daily number of livestock movements between communes in 2007–2014 and used it as an estimation of livestock movement patterns in 2018–2019. We also classified communes into one of two clusters (i.e. central or outer), based on their level of structural equivalence [24] (Fig 1B). Briefly, two communes were considered structurally equivalent if they received and sent the same number of animals from, and to, the same communes. We described the temporal changes in the number of livestock cases in each cluster. The size of the livestock population per commune was provided by the latest agricultural census conducted in 2010 [22], and for any two communes, the Euclidean distance between their geometric centroids was used as their geographical distance. There were on average 1,751 livestock per commune (IQR: 1,019–1,876, Fig A in S1 Text), and the average livestock density was 87.4/km² per commune (IQR: 41.8–202.3/km²), according to the 2010 agricultural census. Finally, we assessed variations in Normalized Difference Vegetation Index (NDVI) across the island, during the study period as a proxy of mosquito abundance since it reflects the level of vegetation and presence of water (i.e. conditions promoting mosquito proliferation) [25].