A Disease-Agnostic Data Model and Pluggable Module Interface for Spatio-Temporal Risk Analytics: An Open Reference Implementation Evaluated on Veterinary Surveillance Data

Spatio-temporal risk analytics for animal-health surveillance relies on established methods—maximum entropy (MaxEnt) suitability modeling, space–time scan statistics, principal component analysis (PCA), and Bayesian prevalence estimation—usually run as isolated, disease-specific workflows. This paper presents a disease-agnostic event schema and a typed plug-in interface, with an open-source Python reference implementation. The full registry-ingestion and modeling workflow was tested with raw anthrax and rabies registries containing 4183 and 363 georeferenced records, respectively. The PCA and Bayesian components, however, used aggregates reconstructed from earlier publications, so those tests assessed interface conformance and numerical reproduction rather than independent validation. With the same inputs, the scan component reproduced the primary SaTScan clusters, yielding Jaccard indices of 0.91 and 1.00, although it matched 8 of 13 significant anthrax clusters overall. Spatial block cross-validation produced produced area under the receiver operating characteristic curve (AUC) values values of 0.74–0.77 for anthrax, compared with 0.67–0.70 for the elapid MaxEnt implementation; the exact figures depended on fold assignment and regularization. The roughly two-orders-of-magnitude runtime gap also applies only to those particular implementations, feature sets, and datasets. Adding a fifth disease required a single 37-line source loader, with no edits to the 1034 lines of shared code. These results show that the system can run on one workstation, but database-backed deployment, operational scaling, and prospective forecasting were not tested.