JMIR Publications

ABSTRACT

Background:

Foodborne disease is one of the common threats to human health worldwide, leading to millions of deaths every year. Thus, the accurate prediction of future foodborne disease risk is very urgent and of great significance for public health management.

Objective:

We aimed to design a spatial temporal risk prediction model suitable for foodborne diseases to predict the future foodborne disease risks in various regions, so as to provide guidance for the prevention and control of foodborne diseases.

Methods:

We design a novel end-to-end framework to predict the foodborne disease risk by using a multi-graph structural LSTM neural network, which can utilize the encoder-decoder structure to achieve multi-step prediction. In particular, to capture the multiple spatial correlations, we divide regions by administrative area and construct the adjacent graph with different metrics, including the proximity of regions, the similarity of historical data, the similarity of regional function and the similarity of foodborne disease exposure food. Furthermore, we also integrate the attention mechanism in both spatial and temporal dimensions and external factors to refine the prediction accuracy.

Results:

We validate our model with extensive experiments on a long-term real-world foodborne disease dataset, ranging from 2015 to 2019 in multiple provinces of China, where the experimental results clearly demonstrate that our approach can outperform other state-of-the-art baselines with a significant margin.

Conclusions:

Our proposed spatial temporal risk prediction model of foodborne diseases can take into account the spatial temporal characteristics of foodborne disease data and provide a certain degree of precision for future disease spatial temporal risk prediction, thereby providing support for the prevention and risk assessment of foodborne disease.