Through the case study of the multiple myeloma, we will see how to re-identify treatment lines in the SNDS data (French national claim databases). First of all, why did we need an AI algorithm? The main challenge was to find the matches (= similarity) between theoretical treatment cycles and real patient treatment sequences, despite inevitable deviations between them. How did we use the AI algorithm? How did we combine Data science and clinical knowledge to build an algorithm custom-made for the disease?