Development of a forecasting model for brucellosis spreading in the Italian cattle trade network aimed to prioritise the field interventions

Abstract

Brucellosis caused by Brucella abortus is an important zoonosis that constitutes a serious hazard to public health. Prevention of human brucellosis depends on the control of the disease in animals. Livestock movement data represent a valuable source of information to understand the pattern of contacts between holdings, which may determine the inter-herds and intra-herd spread of the disease. The manuscript addresses the use of computational epidemic models rooted in the knowledge of cattle trade network to assess the probabilities of brucellosis spread and to design control strategies. Three different spread network-based models were proposed: the DFC (Disease Flow Centrality) model based only on temporal cattle network structure and unrelated to the epidemiological disease parameters; a deterministic SIR (Susceptible-Infectious-Recovered) model; a stochastic SEIR (Susceptible-Exposed-Infectious-Recovered) model in which epidemiological and demographic within-farm aspects were also modelled. Containment strategies based on farms centrality in the cattle network were tested and discussed. All three models started from the identification of the entire sub-network originated from an infected farm, up to the fifth order of contacts. Their performances were based on data collected in Sicily in the framework of the national eradication plan of brucellosis in 2009. Results show that the proposed methods improves the efficacy and efficiency of the tracing activities in comparison to the procedure currently adopted by the veterinary services in the brucellosis control, in Italy. An overall assessment shows that the SIR model is the most suitable for the practical needs of the veterinary services, being the one with the highest sensitivity and the shortest computation time.

Fig 1. Graphic description of the static transmission network.

The figure shows the graphic description of TN and the geographical node distribution. The holdings are nodes (green circles in the figure) and the movements of animals from holding to holding are arcs. The arc is oriented from the holding of origin to the holding of destination of the animal movement. The node shape characterizes the type of holding: diamonds represent pastures, squares represent markets and staging points, green dots represent herds. Moreover the blue dots represent the real brucellosis infected herds and the red dot represents the seed (the first brucellosis outbreak) detected in Sicily during 2009.

Fig 2. Graphic illustration of population subdivision role on moved animals.

The node population consists of two sub-populations: the sub-population "r" of resident animals, and the sub-population "e" of animals introduced in the node. The panel ‘a’ represents the case of the number of moved animals from A to B is less than animals in “r” of A (m_AB ≤ Ar). Panel ‘b’ shows the case of the number of moved animals from A to B is more than animals in “r” of A (m_AB > Ar).

Fig 3. Results of the six scenarios proposed in terms of TN reduction.

Fragmented networks resulting from the blocking the movements of the holdings with the highest values of the centrality measures (≥ 98° percentile): a. authority, b. hub, c. cut-point, d. degree, e. out-degree and f. in-degree.

Fig 4. ROC curves.

ROC curves for the methods considered to prioritise the field activities of outbreak tracing. DFC (blue), SIRa (green), SIRb (dashed green), SEIRa (red) and SEIRb (dashed red). Differences in performances when using the entire TN (panel I) and the TN of Sicily only (panel II) are shown. The black dot represents the Sensitivity and Specificity of the veterinary services approach.

Fig 5. Outbreaks detected for a set of 100 checked farms.

Number of outbreaks detected by each of the models (DFC, SIR and SEIR) on a set of 100 checked farms. SIR model is the most efficient detecting 7 secondary outbreaks, against the only one identified by veterinary services method.

Fig 6. Pasture density map and outbreaks not detected by SEIR model.

All outbreaks not detected by the SEIR model (9 on 25) fall in the high density zones of pastures of Sicily region.