Extracting and communicating insights about geographical patterns from spatiotemporal disease models

Spatiotemporal modelling is increasingly employed to understand how geographic patterns of disease vary over time; however, most studies report minimal information, often neglecting uncertainty altogether. This represents a missed opportunity to understand changes in disease risk at a fine granularity and to provide a valuable, actionable evidence base. This study describes generalisable methods for extracting key metrics from the outputs of Bayesian relative risk spatiotemporal models for epidemiological applications, such as the magnitude of relative change between defined periods and the level of evidence that any change was non-zero. Analytical approaches are introduced for identifying types of cancers or areas with particularly notable changes in rates and for visualising areas where both high rates and large changes in rates coincide. These metrics and analytical approaches enhance the interpretability of complex spatiotemporal patterns and can inform targeted interventions and support evidence-based prioritisation. To demonstrate these methods, we present a novel, flexible Bayesian hierarchical spatiotemporal disease model and describe how different implementations of the model can be used to understand either temporal changes by geographic area or shifts in geographic patterns over time. Variations on model specifications are described to support different temporal and space-time interaction models. Finally, we discuss practical considerations and challenges encountered in implementing spatiotemporal models. Overall, this work demonstrates how spatiotemporal modelling outputs can be systematically interpreted to generate actionable insights for public health planning and evaluation.