PhD opportunity at Sorbonne University: Transfer learning to inform the spread of other respiratory viruses : Application to Influenza using COVID19 and drug sales

In high-income countries, the COVID-19 pandemic fostered the generation of surveillance data at spatial and temporal resolution unseen before, providing comprehensive and accurate estimates of cases, detection capability, hospitalizations and deaths. At the same time, data describing behavioral response, mobility, mixing and compliance to public health measures have also become available with similar level of detail. Such an exhaustive picture of the unfurling of a pandemic was a first in human history, made possible because we live in the digital age. It does not imply that epidemiological surveillance will remain this way in the future. As COVID-19 becomes less virulent with vaccination and acquired immunity, political pressure is shifting away from comprehensive detection of cases, and individual willingness to get tested may also be declining. At the same time, corporate commitment to make proprietary data on human behavior available to scientific research (e.g., mobile phone data) is waning. This underpins the main scientific goal of this project: can we use the experience of “wartime” COVID-19 surveillance during years2020-2022 to improve epidemic understanding in the future “peacetime” period ? Typical data available for surveillance in peacetime is scarcer, for example syndromic surveillance for influenza and other respiratory viruses as reported in networks of general practitioners (GP), with limited virological confirmation. Other data sources, including participatory surveillance and drug sales, may complement such reports, but are less specific. Importantly, during the first 2 years of COVID-19, the aforementioned high-resolution data and the scarcer traditional data sources were observed together. We wish to exploit this overlap to build statistical and mathematical models that will extract more and better information from peacetime surveillance data. Specifically, we aim at generating estimates of incidence, severe cases, reproductive number that are better than those previously available in terms of spatial resolution, temporal resolution, predictive power (ability to make short-term forecasts and mid-term projections of epidemic activity). We will make use of AI/ML techniques to come up with models with which transfer of knowledge, for example from the dynamics of COVID-19 to that of Influenza, or from drug sales data to influenza, from mobility to infectious spread will make it possible to improve accurate estimation of influenza incidence and short term prediction. The impact of this project will be thus twofold. First, we will improve the knowledge and predictability of seasonal epidemic waves of airborne, directly transmitted pathogens. Second, we will provide with policymakers with new tools to inform public health response to seasonal acute respiratory illness.

More at: