Comparing the performance between static and dynamic populations in COVID-19 hotspot prediction

ABSTRACT

Objectives: This study aimed to set up the prediction model of COVID-19 hotspot areas by using the census data and human mobility from telecommunication data in Taipei and New Taipei City. The comparison between their accuracy and limitations can provide the relevant insights for future epidemic control. Methods: The spatio-temporal resolution is fixed at the village level in two cities in May 2021. The static and dynamic data are used to construct the mobility network. The former applies gravity model to mimic human flow, and the latter uses telecommunication data as the measure of mobility. We propose the footprints similarity by structural equivalence of spatial networks and integrate it with the number of confirmed cases for computing the risk level of the villages. The performance of the models is evaluated using ROC curves and logistic regression under different thresholds for the confirmed cases. Results: The mobility derived from the telecommunication data provided better prediction performance than that from the census data;they have an average AUC of 0.75 and 0.69, respectively. Besides, the telecommunication data had a tendency to identify a further village as high-risk zone compared to the gravity model. According to the results of logistic regression, the odds ratio (OR) of exceeding the cases' threshold estimated from the telecommunication data is 1.45 on average, while the one estimated from the census data is 1.10. Conclusions: Telecommunication data can be beneficial in identifying the potential high-risk areas and enhancing situational awareness in advance. © 2022, Taiwan Public Health Association. All rights reserved.