Multichannel qualitative harmonic analysis for two-step patient pathway
Résumé
The SNIIRAM (Systeme National d Information Inter-Regimes de l Assurance Maladie) database, collected by the French National Insurance, is a very rich source of information on the drugs prescribed to patients: type, date of delivery and quantity are recorded. However, some pieces of information are missing, e.g. when drugs are actually taken and what disease is treated. A patient pathway is defined as a sequence of drugs delivered to the same person. A treatment consists of a drug, or a set of drugs, prescribed to a patient for a specific purpose (intention to cure), either as first treatment or for continuation of the treatment. A two-step patient pathway is assuming. An unsupervised multichannel qualitative harmonic analysis clustering is performed to identify patterns of first fertility treatment intention. Our approach uses a fuzzy coding framework to take into account the two steps. Prior incomplete knowledge indirectly deduced from administrative data in the design of the analysis is incorporated. As the database lacks labelled information, a pseudo labelled dataset to allow for an accurate evaluation of our method according to medical practices is retrieved and enhanced.