The spread of influenza can be modelled and forecast using a machine-learning-based analysis of anonymized mobile phone data. The mobility map, presented in Nature Communications this week, is shown to accurately forecast the spread of influenza in New York City and Australia.
The spread of viral diseases through a population is dependent on interactions between infected people and uninfected people. Building models to predict how diseases will spread across a city or country currently makes use of data that are sparse and imprecise, such as commuter surveys or internet search data.
To gain a richer dataset Adam Sadilek and colleagues collected anonymized tracking data from Android phones that have the Location History setting enabled, and used machine learning to break the data down into individual ‘trips’ to build a map of human mobility. They use this mobility map to forecast influenza activity in and around New York city in 2016–2017, using a model for infectious disease transmission calibrated against known data from hospital visits and laboratory tests. They find that their model performs better than standard forecasting models in common use, and is comparable to using commuter survey data, which are more expensive to obtain. They also forecast influenza spread across Australia for the 2016 influenza season. Despite a sparser population and different influenza dynamics, the model still accurately predicts the ebb and flow of the disease.
Existing high-resolution mobility data are based on phone call records, which are provider specific and do not typically capture cross-border or international movement. Location data is not limited in this way and so has greater potential to monitor long-distance disease spread. However, there are limitations in the completeness of these data, as they exclude mobility data for children and the elderly who are less likely to use smart phones. Despite these limitations, the authors have shown that the use of data gathered from mobile phones has the potential to forecast the spread of epidemics. Whether this can also be applied to other viral infections such as SARS-CoV-2 requires further investigation.