What is it about?

In this paper, we show how infectious disease surveillance can be done through careful analysis of mobile phone records. We draw on a rich dataset of mobile phone metadata linked to health records on influenza infection, and develop a new anomaly detection method to identify those users that are likely to be showing symptoms.

Featured Image

Why is it important?

Infectious disease surveillance remains challenging. Mobile phones are near ubiquitous across the world, and normal use generates large quantities of data each second. We show how these data need to be analysed to provide insights into the spread of influenza by learning the characteristic changes to individual calling, movement, and other behaviour that is consistent with the onset of the disease. Unlike previous efforts, which were hampered by a lack of ability to directly validate, we are able to validate our model predictions for disease against clinically confirmed health records. Our methodology exploits recent advances in graph neural networks (GNN) to overcome the noisy, sparse, and complex high-dimensional call detail records gathered from the mobile phones. From the graph-based model of user routine behaviour, we count up the number of people behaving differently from normal, allowing us to estimate the fraction of the population believed to be infected. After training, our methods are able to provide public health officials with estimates for the intensity and trajectory of an outbreak.

Perspectives

The ability to analyse mobile phone records for public health purposes and understand the uncertainty associated with the output is a step towards the goal of better passive surveillance for a variety of health conditions, infectious or otherwise. Other approaches are similarly promising, for example sampling sewage for fragments of infectious agents to track disease levels – but are highly sensitive to the existence of appropriate infrastructure which varies greatly between settings. Mobile phone use is ubiquitous, mobile infrastructure is largely uniform, and the mobile data arising from normal use of phones is similar between settings. Work such as the one presented in our paper is needed to unlock the potential of mobile phone use for public health globally. While infectious surveillance through mobile phone data is likely to be complementary to traditional surveillance in high income settings, low income settings where traditional surveillance is underfunded and patchy stand to benefit the most. In tandem with the ongoing dialogue surrounding data privacy in arenas that benefit the public, realising such systems requires close collaborations between researchers and the telecom industry as well as further investment into technological advancements such as ours.

Leon Danon
University of Bristol

Read the Original

This page is a summary of: Dynamic Network Anomaly Modeling of Cell-Phone Call Detail Records for Infectious Disease Surveillance, August 2022, ACM (Association for Computing Machinery),
DOI: 10.1145/3534678.3542678.
You can read the full text:

Read

Resources

Contributors

The following have contributed to this page