[945] Global Health Epidemiology

Global Health Epidemiology gathers the efforts of the Institute of global health to participate in epidemiological developments in the transdisciplinary field of global health. Our research group is dedicated to research and development in Global Health using advanced modelling, epidemiological and statistical techniques. We collaborate with the Swiss Data Science Center (EPFL and ETHZ). Antoine Flahault has been appointed by the Regional Director of WHO-Europe as chair of its High-Level European Expert Group on stabilization of COVID-19.

CURRENT PROJECTS

Epidemiological study of DEngue virus infections in MAdagascar and REunion Island (DE-MA-RE)

DE-MA-RE is a project funded by SNSF in which Antoine Flahault is the Principal Investigator and Dr Olga De Santis the Study Coordinator.

Dengue is the most globally prevalent arboviral disease. Incidence of dengue follows an increasing trend that is expected to continue due to the increase in urbanization, population size, air traffic and climate change. The situation of dengue in the Indian Ocean region is poorly known. Aedes albopictus, usually considered as a secondary vector for dengue virus (DENV) has spread since the eighties to more than 25 European countries, North America and to Central Africa. Madagascar and Reunion Island are representative of the islands in the Indian Ocean. On both islands, dengue has emerged in the past few years and transmission is due to A.albopictus. Many questions remain to be answered regarding the low transmission rate and to understand how to predict the future evolution of this emergent disease in the Indian Ocean region.

In this project we aim:

To estimate the incidence rate of Reverse Transcriptase - Polymerase Chain Reaction (RT-PCR) DENV infections among household members and in the neighborhood of dengue index cases in Toamasina (Madagascar) and on Reunion Island.

To estimate the prevalence of all disease patterns of dengue, from asymptomatic to severe illnesses.

To estimate the incidence rate of RT-PCR confirmed DENV infections collecting the sample pads of malaria RDT performed for fever cases management in primary health care centers (Madagascar).

To assess the performance of blood DENV Rapid Diagnostic Test (Reunion Island)

To assess the capacity of asymptomatic DENV-infected cases to transmit DENV to Aedes albopictus.

4000 community participants are to be recruited into geographical clusters around dengue index cases (2000 on each island). Demographic and clinical data as well as biological samples are to be collected for investigations such as DENV RT-PCR, serology and DENV RDT. Malaria RDTs are to be collected in primary health care centers in Madagascar in order to perform dengue RT-PCR. A.albopictus DENV infection rate after indirect feeding with blood from dengue asymptomatic cases are to be assessed.

The data collected in this study will provide an estimation of the prevalence and incidence of DENV infections of all disease patterns on two islands in the Indian Ocean. Looking for the presence of DENV in malaria RDT could offer a new way for emergent diseases prevalence estimation based on fever detection. Moreover, refining the dengue-like syndrome definition in these settings and assessing the performance of dengue RDT on Reunion Island could improve public health measures for surveillance. Finally, assessing the transmission rate from asymptomatic cases to A.albopictus is highly relevant has this vector is spreading almost all over the world; half or more of DENV infections are assumed to be asymptomatic. The present study aims at completely inversing the paradigm and tries to understand why transmission is low while other factors that would favor transmission are present. Tackling the question of emergent diseases with this different perspective will bring new knowledge on transmission of dengue and may allow to better predict future dengue outbreaks.

COVID-19 Daily Epidemic Forecasting (COVID-19 Dashboard)

The Covid-19 Dashboard is a joint initiative funded by the Fondation Privée des HUG, launched and steered by the Institute of Global Health (UNIGE) and the Swiss Data Science Center (EPFL-ETHZ). The two Co-PIs of the project are Antoine Flahault and Christine Choirat. Collaboration and implication from Guillaume Obozinski, Sun Tao, Ekaterina Krimova, Oksana Riba Grognuz, Kristen Namigai, Adeline Dugerdil, Elisa Manetti.

Providing daily 7-day predictions of COVID-19 for all WHO Member States and territories is key for three major reasons: 1) people want to know; 2) in case our predictions which are based on the past recent trend overestimate the subsequent observations, it may signal that control measures break the epidemic dynamic faster than expected; 3) in case our predictions underestimate observations, it may trigger an early warning signal that new hot zones are bursting out or that control measures may prove ineffective enough.

Background:

- Outbreak of SARS-CoV2 since December 2019, starting in China, spreading in 209 countries and territories

- It comes from an animal coronavirus, with no immunity in human

- Clinical forms: more than 80% benign self-limiting disease with upper respiratory symptoms and fever, 15% severe respiratory disease, 5% critical, 0.5-0.7% mortality rate with huge gradient with age and comorbidities.

- Incubation time below 14 days

- R0 (basic reproductive number): 2-3

- This is the first pandemic in history which leads to an open real-time access of data at a worldwide level. Currently these data are published in various website (e.g. World Health Organization, Johns Hopkins University, Worldometers, OurWorldinData), however there was no release of daily predictions available when we started our project. The Institute of Global Health, from the early beginning of this pandemic published on the microblogging Twitter each evening an 7 day-prediction for a set of selected countries, in Europe, America, Latin America and Africa. These tweets became very popular and appreciated, and in collaboration with SDSC (ETHZ-EPFL), we decided to deliver a dashboard, updated on a daily basis, covering all countries and territories and providing 7-day predictions from official reported data on confirmed cases and deaths.

Design and methodology:

- The initial model used is estimating a geometric growth rate corresponding to the initial exponential growth of the number of cases, but which is itself allowed to evolve geometrically over time to capture the inflexion of the curve. This model has been dramatically improved by leveraging state-of-the-art statistical modelling and machine learning techniques. Indeed, as we transition from the initial exponential growth of the epidemic and reach a plateau, and/or as some of the confinement and prophylactic measures impact the R0, modelling the growth factor becomes complicated. It should be stressed that the modelling problem to tackle here is an extrapolation problem, which is harder than the typical forecast problem in which one can rely on a history of similar patterns to overcome a simple approximation-estimation trade-off. Also, beyond the stochastic nature of the data, the task of fitting accurate forecasting models is here complicated by the presence of seasonal October 01, 2020 – COVID19 Daily Epidemic Forecasting - Phase II effects (modulation by the day of the week), by the influence of exogeneous variables such as (sometimes difficult to document or detect) modifications in testing or case and death reporting policies, etc. COVID-19 Dashboard Epidemic Forecasting:

- We have worked on models that minimize the absolute error in the forecast of the number of daily new cases detected on a horizon of one to two weeks. We have built and fit generalized linear, polynomial and auto-regressive models with loss functions that are well matched to the problem considered, and robust to outliers and seasonality patterns. We envision to construct finer models which can potentially treat the seasonal effect explicitly as a nuisance parameter, and to use estimator aggregation techniques to stabilize more the predictions. The forecast will also include error estimates on the expected cases or deaths forecast themselves, or forecast of quantiles, in order to produce confidence bands around the predictions. This should be achievable by the same families of models.

- Once state-of-the-art forecast will be obtained for this horizon, we will consider building models on longer horizons that leverage time varying parameters in SEIR models, structured parameters, or both.