Résumé
Chronic kidney disease (CKD) represents a heavy global health burden associated with increased mortality and morbidity and high economic impact. Chronic kidney disease, which is largely asymptomatic and is diagnosed based on laboratory tests, is particularly difficult to identify in medical-administrative databases in the absence of laboratory results and no specific medications or procedures. The aim of this paper is to describe the progressive stages of constructing and validating an algorithm for targeting chronic kidney disease in the French medical administrative databases SNDS . A consortium of experts in nephrology, kidney epidemiology and healthcare claims databases, referred to as group \"REDSIAM Kidney Disease\", collaborated to design a practical algorithm for assessing the probability of chronic kidney disease cases likelihood through a combination of items associated with the CKD care pathway. The performance of the RENALGO-EXPERT algorithm differs significantly depending on the population and the databases used. Sensitivity tends to improve in more at-risk populations. However, at this stage, the results are not very satisfactory. To improve case detection performance and in the hope of capturing weak signals overlooked by experts, a project using machine learning methods was devised, RENALGO-IA.