89

Cause-of-mortality statistics are one of the oldest medical statistics available. Mortality data, which provide essential information for general knowledge on a population’s health, and are a tool for international comparisons, are also a growing challenge for the governance of public health, particularly in crisis situations. Based on a study of documents from the bodies that regulate and produce these data, the article examines the socio-technical trajectory of cause-of-mortality coding and analyses the context in which a deep-learning method was used during the COVID period. From a theoretical perspective combining the sociology of science and technology with the sociology of quantification, it interprets the integration of connectionist AI methodologies into the coding process as a further stage in the technical trajectory of the tool that began with automation. It also sheds light on the institutional context of the use of neural networks and the consolidation of that use, relating them to the recent functions of mortality statistics in the governance of health crises and issues around the availability and speed of their production.