On Monday April 27, a team of mathematicians, statisticians and epidemiologists (Sorbonne University, CNRS, Collège de France, Oxford University) launched a survey of French households that have experienced the presence of the new coronavirus during the containment period. The project has been dubbed "Alcov2" - a hybrid of alcove and SARS-Cov2.
The team's idea is as follows: the stories of households where at least one person has experienced symptoms linked to Covid-19 (fever, cough, headache or sore throat, fatigue, diarrhea, aches and pains, chills, nausea, loss of taste or smell, respiratory discomfort, chest pain or tightness...) are all independent repetitions of the same micro-epidemic process of transmission within the household.
This stochastic process, i.e. of a probabilistic nature, will be finely modelled using parameters such as the daily infection rate per person in the household, the decrease in this infection rate as a function of the time elapsed since infection, the dependence of this infection rate on the severity of symptoms and the probability of being asymptomatic.
By questioning French households about the number of people in the household who have or have not experienced symptoms potentially linked to Covid-19, their risk factors, the precise clinical picture of each case and, above all, for each of these cases, the date of onset of the first symptoms, this team will be able to reconstruct the parameters of this micro-epidemic process.
The first stage of this reconstruction involves the development of an unsupervised classification algorithm that classifies cases according to their clinical picture and risk factors, and assigns them a score quantifying the probability that they are actually infected. The second step consists in inferring the parameters of the epidemic model, integrating the uncertainty about the presence of the virus resulting from the previous step (using a Bayesian approach) and the uncertainty about the presence of asymptomatic or pauci-symptomatic infected individuals (using hidden Markov chains).
This work at the interface between modeling mathematics, statistics and medicine brings together complementary skills, including those of data scientists from a wide variety of backgrounds (Sorbonne University, Ekimetrics) and federated within the Club datacraft. This collaborative effort will make two new tools available to the scientific community:
1) an algorithm capable of calculating the probability of having been ill with Covid-19 based on symptoms experienced and risk factors;
2) a fine-tuned estimate of the crucial parameters of virus transmission, in particular variations in the rate of infection over the lecture period since infection, and the frequency of asymptomatic patients.
For the general public, this work could eventually lead to the distribution of a diagnostic questionnaire enabling everyone to calculate the risk for their household of having been exposed or not to the coronavirus, according to the symptoms experienced and their chronology.
It is being distributed in parallel to a representative national panel of 10,000 households, thanks to the gracious cooperation of the Bilendi group and the BVA research institute. The survey is also intended to be replicated in several other countries.
The greater the number of responses, the more accurate the results will be: so don't hesitate to take part in this survey if you're concerned, and to spread the word around you if you're not.
