An algorithm for identifying chronic kidney disease in the French national health insurance claims database

Published algorithms for identifying chronic kidney disease in healthcare claims databases have poor performance except in patients with renal replacement therapy. We propose and describe an algorithm to identify all stage chronic kidney disease in a French healthcare claims databases and assessed its performance by using data from the Renal Epidemiology and Information Network registry and the French Childhood Cancer Survivor Study cohort.

A group of experts met several times to define a list of items and combinations of items that could be related to chronic kidney disease. For the French Childhood Cancer Survivor Study cohort, information on confirmed chronic kidney disease cases extracted from medical records was considered the gold standard (KDIGO definition). Sensitivity, specificity, and positive and negative predictive value and kappa coefficients were estimated. The contribution of each component of the algorithm was assessed for 1 and 2 years before the start of renal replacement therapy for confirmed end-stage kidney disease in the Renal Epidemiology and Information Network registry.

The algorithm's sensitivity was 78%, specificity 97.4%, negative predictive value 98.4% and positive predictive value 68.7% in French Childhood Cancer Survivor Study cohort and the kappa coefficient was 0.79 for agreement with the gold standard. The algorithm 93.6% and 55.1% of confirmed incident end-stage kidney disease cases from the Renal Epidemiology and Information Network registry when considering 1 year and 2 years, respectively, before renal replacement therapy start.

The algorithm showed good performance among younger patients and those with end-stage kidney disease in the twol last years prior to renal replacement therapy. Future research will address the ability of the algorithm to detect early chronic kidney disease stages and to classify the severity of chronic kidney disease.