Évaluation de la qualité du chaînage des séjours hospitaliers pour l’étude des variations spatiales de santé à partir des données du PMSI. Étude de faisabilité dans trois régions françaises

Background. - Since 2001, the French hospital stay databases (Programme de medicalisation des systemes d'information, PMSI) have included a unique and anonymous identifier in order to cross-link discharge abstracts from a given patient, within and across hospitals. These data could be used to estimate prevalence for some diseases at a territorial level provided that linkage quality is good enough. Few morbidity data are available at this scale. This study analyzes the link between linkage quality and hospitalization rates in three French regions (Picardy, Brittany and Provence-Alpes-Cote d'Azur-Paca). Methods. - We studied short stays in medicine-chirurgical-obstetrical units for the 2004-2005 period (all stays, and stays with mention of cancer or asthma). To study linkage quality, the percentage of linkable stays (no error during the production of the anonymous identifier) was calculated at regional and territorial levels (areas used by regional health authorities). The interquartile range (IQR = third quartile - first quartile) of the percentage of linkable stays was calculated and the link between this percentage and standardized rates of people hospitalized at least once in 2004 or 2005 tested by Spearman correlation coefficients. Results. - For all stays, percentages of linkable stays were 94.4%, 96.6% and 97.0% in Picardy, Paca and Brittany respectively in 2004-2005. Geographical variation at the territorial level was higher in Picardy (IQR between 4 and 6) than in the two other regions (IQR between 1 and 2). The percentage of linkable stays was positively and significantly associated with the hospitalization rate for all stays and those with mention of cancer in Picardy only. Conclusion. - According to these results, PMSI data earlier than 2006 should be used with precaution; linkage quality should be analyzed before making geographical or time comparisons of hospitalization rates. Comparisons cannot always be made. Other studies should be carried out in other regions, and to analyze recent trends in linkage quality. (C) 2011 Elsevier Masson SAS. All rights reserved.