The interest of health databases in anesthesia is no longer to be demonstrated. The aim of this research was to develop a natural language processing approach to establish a classification of adverse events observed during the perioperative period and to facilitate their analysis:

The main objective of the study was to identify what a "naïve" unsupervised model would discover based on Adverse Event (AE) descriptions. Our second goal was to identify apparently unrelated events whose combination could favor the occurrence of an AE

Inclusion Criteria:

• Minors and adults having had an allergic reaction associated with care
• Having had an adverse event reported by an anesthetist between January 01, 2009 and June 30, 2020

Exclusion Criteria:

- Patient not meeting the inclusion criteria

Minors and adults having had an allergic reaction associated with care, and having had an adverse event reported by an anesthetist between January 01, 2009 and June 30, 2020