Multi-dimensional Spatio-temporal Prediction Network for Pre-hospital Emergency Care

Predictive analysis of pre-hospital care can optimize resource allocation, address challenges in assessing ambulance demand, and significantly improve emergency care efficiency and patient survival rates. However, the lack of high-quality publicly available datasets, combined with the influence of complex factors such as urban spatial patterns, weather, and time, makes predictive analysis of pre-hospital emergency care a challenge. To address this, we introduce the Pre-hospital Emergency Information Mart, a large-scale, publicly accessible repository of authentic emergency records. Given the complexity of such analysis, we propose a Multivariate Spatiotemporal Prediction Network. Specifically, we design a multifactor fine-grained sequence learning branch to model historical sequences, time factors, and external factors. The spatial distribution prediction branch, based on spatiotemporal attention, effectively captures spatial features with hyperconnections. The global knowledge transfer module transfers global spatial knowledge to the sequence prediction module, increasing spatial dependencies for reliable predictions under multidimensional constraints. Experiments on the Pre-hospital Emergency Information Mart dataset show that our proposed Multivariate Spatiotemporal Prediction Network outperforms six well-known methods.