Background: Since 2001, the French national case mix program is allowed by law to use an enciphering algorithm named "FOIN" to produce a unique anonymous identifier in order to crosslink, within and across hospitals, discharge abstracts from a given patient. This algorithm "thrashes" the person's health insurance number, date of birth and gender. Before using information produced by the case mix program, either for case mix payment or for epidemiology research or for assessing care approaches, the quality of linkage must be evaluated.
Methods: Foin error flags were first assessed in the 2002 Rhône-Alpes regional case mix database. Second, for the two university hospitals of Lyon and Saint-Etienne, double identifiers (two or more Foin identifiers for the same patient) and collisions (a single Foin identifier for at least two patients) were compared with others identifiers: administrative identifier and an anonymous identifier produced by Anonymat software from name, forename and date of birth. Third, Foin error flags are crossed with Foin double identifier or collision mistakes.
Results: First, among 1,668,971 hospital discharge abstracts from the regional case mix database, 206,710 (12.4%) had at least one Foin error flag. The most frequent error flag (93026 [5.5%] stays) was due to the lack of the three identifying variables. The greatest number for error flags concerned the stays of newborns (38.5%) and those of public hospitals (17.3%). Second, Foin created a few double identifiers: 1.2% among 137,236 patients from university hospital of Lyon and 0.3% among 39512 patients from university hospital of Saint-Etienne. The collisions concerned 7776 (5.7%) patients from Lyon and 460 (1.2%) from Saint-Etienne. The identifier produced by Anonymat performed better than the one produced by Foin: 99.6% from the two university hospitals. Third, less than 3% of stays without Foin error flag nevertheless had mistakes on Foin when compared with others identifiers.
Conclusion: The overall assessment is not in favour of a quality threshold using the Foin identifier on a routine basis except in some areas and if certain activities like neonatology are excluded. There are several ways to improve the linkage of health data.