Send to

Choose Destination
Rev Epidemiol Sante Publique. 2014 Aug;62(4):257-66. doi: 10.1016/j.respe.2014.04.006. Epub 2014 Jul 17.

[Record linkage of hospital discharge data and first health certificates: a test in the Val d'Oise].

[Article in French]

Author information

Périnat-ARS-IDF, agence régionale de santé d'Île-de-France, 2, rue Saint-Martin, 75004 Paris, France. Electronic address:
Observatoire régional de santé d'Île-de-France, Paris, France.
Périnat-ARS-IDF, agence régionale de santé d'Île-de-France, 2, rue Saint-Martin, 75004 Paris, France; CHI André-Grégoire, 93100 Montreuil, France.
Périnat-ARS-IDF, agence régionale de santé d'Île-de-France, 2, rue Saint-Martin, 75004 Paris, France; Hôpital Louis-Mourier, AP-HP, HUPNVS, 92700 Colombes, France.
Observatoire de la direction prévention santé, Conseil général du Val d'Oise, Cergy Pontoise, France.
Inserm UMR 1153, équipe de recherche en épidémiologie obstétricale, périnatale et pédiatrique (EPOPé), centre de recherche épidémiologie et statistique, Sorbonne Paris Cité, DHU risques et grossesse, université Paris Descartes, 75006 Paris, France.



Our objective was to test the feasibility of an indirect linkage of data on births from health certificates (HC) with hospital discharge (HD) data.


The linkage was carried out for live births between April 1st and June 30th, 2011 in six of the nine maternity units in the district of Val d'Oise. The HC and HD had 3284 and 3550 births registered during this period, respectively. Linkage was conducted using variables available in both sources: number of fetuses, baby's birth date, gender, maternity unit of birth, maternal age, municipality of residence, gestational age and birth-weight. Two linkage methods were tested: a deterministic and a semi-deterministic method and a probabilistic approach. The latter method calculates a probability estimate for the strength of the relationship between two linked observations related to the discriminatory power and the error rate of the matched variables. For cases that were linked despite discordance on some of the matching variables, random samples of observations were checked against both HC and HD records to compute rates of false matches.


The deterministic and semi-deterministic method linked 92.5% and 85.6% of observations in the HC and HD, respectively. The probabilistic method achieved a linkage rate of 99.6% for HC and 92.7% for HD. Cases only linked by the probabilistic method were more often preterm and had low birth-weight. Cases matched using the probabilistic method only had an error rate of 0.4% with 95% CI [0.2-0.6%].


Common items in the HC and HD make it possible to achieve a high rate of linkage. The probabilistic method links more births and, in particular, those at higher risk, and error rates were low.


Chaînage d’enregistrement de données médicales; First health certificate; Hospital discharge data; Live births; Medical record linkage; Naissances vivantes; PMSI; Premier certificat de santé

[Indexed for MEDLINE]

Supplemental Content

Full text links

Icon for Masson (France)
Loading ...
Support Center