Spatiotemporal modeling of regional short-term pertussis transmission risk using a propagation-enhanced prediction...

Accurate forecasting of regional pertussis transmission is essential for public health surveillance, yet remains challenging because of complex spatial heterogeneity, implicit cross-city diffusion pathways, and nonlinear temporal dynamics.

A spatiotemporal risk prediction framework was developed to model fine-grained disease spread across cities. The framework combines three-dimensional convolutional embeddings for local spatial pattern extraction, a Mamba-based state space architecture for temporal dependency modeling, and a propagation-aware spatial module constructed from administrative adjacency relations to characterize potential cross-city diffusion. In addition, a spatiotemporal mixed output head was introduced to enhance structural alignment and gradient continuity through state fusion and sequence permutation.

Experiments conducted on real pertussis heatmap sequences from Zhejiang Province showed that the proposed framework outperformed existing methods, achieving an MSE of 0.000930, an MAE of 0.020062, a PSNR of 31.057, and an SSIM of 0.9753. Ablation experiments further demonstrated that each module contributed consistently to performance improvement, while significance analysis confirmed the statistical reliability of the gains. Robustness evaluation under incomplete data conditions indicated that the model maintained strong predictive consistency even when the missing rate reached 15%.

The results demonstrate that the proposed framework can effectively capture spatial transmission structure and temporal evolution patterns of pertussis risk. Its strong accuracy, robustness, and stability indicate practical value for supporting regional infectious disease monitoring and public health decision-making.