The algorithm, called “Life2vec,” is based on medical, social and economic data from six million people and is able to predict a person’s death conditions 78.8% of the time.
Between 2008 and 2016, 100,000 profiles aged 35 to 65 were tested, half of whom died four years later. The goal? Predict what will happen in the future based on past events.
To develop this algorithm, scientists had access to a lot of data: history of doctor's appointments with diagnoses, income, employment, education, etc. Information updated daily from 2008 to 2020. According to the Technical University of Denmark, the results are not surprising: men with the lowest Income dies more often. Likewise, a person who leads a poor lifestyle and has lived in polluted places all his life is more likely to die in the next four years.
“Advancing science”
Despite these predictions, the algorithm cannot predict a person's exact age at death. “Our goal is not to alarm people by telling them their expected age of death,” one of the study’s authors, Sune Lehmann, told Franceinfo. “We just want to advance science.”
For comparison: In France, researchers have access to 4% of the population's data. In this sense, “the size and precision of the database used represent a major advance in demographic research,” emphasizes Florian Bonnet, researcher at the National Institute for Demographic Studies, to Franceinfo.
And while life expectancy in France is “calculated according to a specific variable: gender, socio-professional category, geographical origin,” Danish researchers integrate all of these variables and add temporality. “The algorithm created represents a break with the classic statistical analysis of demography,” explains the researcher.