Performance of administrative databases for identifying individuals with multiple sclerosis

Administrative databases are an alternative to disease registries as a research tool to study multiple sclerosis. However, they are not initially designed to fulfill research purposes. Therefore, an evaluation of their performance is necessary. Our objective was to assess the performance of the French administrative database comprising hospital discharge records and national health insurance databases in identifying individuals with multiple sclerosis, in comparison with a registry that exhaustively compiles resident multiple sclerosis cases in Lorraine, northeastern France, as reference. We recorded all individuals residing in the Lorraine region who were identified by the administrative database or the registry as having multiple sclerosis from 2011 to 2016. We calculated the Matthews correlation coefficient and other concordance indicators. For identifying individuals with multiple sclerosis, the Matthews correlation coefficient by the administrative database was 0.79 (95% CI 0.78-0.80), reflecting moderate performance. The mean time to identification was 5.5 years earlier with the registry than the administrative database. Administrative databases, although useful to study multiple sclerosis, should be used with caution because results of studies based on them may be biased. Our study highlights the value of regional registries that allow for a more exhaustive and rapid identification of cases.