Context of the challenges

The impact of environmental factors, including pollution and socio-economic develoment, on the incidence of cancer remains unknown. Countries that have adopted similar socio-economic development models and exposed to comparable environmental situations have different cancerincidences. This underlines the need to refine our models of understanding of the epidemiology of cancers. The consequences of a better understanding of the the evolution of cancers and their causal factors are important both from a medical and a public health point of view, as a result of prioritizing the public health actions that would be deduced therefrom, and better targeting of possible prevention campaigns.

The public recent and massive availability of socio-economic data (economic development, mass consumption), demographic and environmental factors (pollution, exposure to carcinogens, urban planning data) and the development of new approaches to Statistics (machine learning) makes it possible to envisage better predictions of the incidence of the cancers.

By making large data sets widely available and covering the usual epidemiological themes, Epidemium calls on collective intelligence to build models for visualizing and anticipating the spatio-temporal evolution of cancers and thus opens the possibility for all to contribute to the knowledge of the dynamics of cancers and their determinants.


See and Foresee cancers

Visualization challenge

Constructing a Data-Visualization of the incidence of cancers by exposing the epidemiological factors associated with their dynamics.

The challenge consists in developing data visualization tools aimed at the general public and the medico-scientific community at large. Particular attention should be given to design with regard to user experience. These tools will have to integrate the variables and results of the "prediction challenge", highlighting the dynamic changes in time and space of potential risk factors.

Prediction challenge (the incidence of cancers)

Developing a predictive tool for the progression of cancer in time and space,depending on the known or supposed factors that determine its evolution.

The challenge consists in developing of a predictive tool for the incidence of cancers.The data used to learn the models are detailed below.

The "Training set" will be composed of the data covering the period from 1950 to 2003.The validation set will be composed of the data covering the period from 2003 to 2007.Note that the data from 2007 to 2012 in the process of being obtained and will be Integrated into the validation set.

This challenge is divided into two distinct parts: 1/ prediction in the world, 2/ prediction by country

The winning team will be the one whose prediction will be the most efficient on the validation set. A leaderboard will be set up with real-time updating.

This challenge run from October 2017 to March 2018

Predicting cancer mortality in developing countries

The most frequent cancers will be studied. According to GLOBOCAN 2012, the three most common cancers are lung cancer (1.8 million cases, 13.0% of total cancers), breast cancer (1.7 million cases, 11.9% of the total) and colorectal cancer (1.4 million cases, 9.7% of the total). Of course, these figures are an average, and there may be disparities in the incidence of cancers in developing countries. An approach by continent and possibly continents will be appreciated.

Participants will articulate their analysis from 3 datasets:

Main dataset: an open world IARC cancer mortality dataset for the World Health Organization (WHO) broken down by type of cancer, country, year, gender. On average, the depth of accessible data is about 25 years. Among the countries represented are the developing countries, with data available from the African continent and, for comparative purposes, the mortality data of the industrialized countries.

Complementary datasets: 1) an open dataset of the incidence of cancer in the world derived from WHO IARC, broken down by type of cancer, country, year, gender. The data have a minimum depth of 15 years. 2) an open dataset of population indicators, produced by the World Bank: there are data from economic, social, educational, agro-environmental indicators, etc.