Fiche publication
Date publication
janvier 2014
Auteurs
Membres identifiés du Cancéropôle Est :
Pr BINQUET Christine
,
Pr QUANTIN Catherine
Tous les auteurs :
Quantin C, Benzenine E, Hagi M, Auverlot B, Abrahamowicz M, Cottenet J, Fournier E, Binquet C, Compain D, Monnet E, Bouvier AM, Danzon A
Lien Pubmed
Résumé
BACKGROUND: Cancer registries are a reliable source of data to estimate national cancer incidence rates, but they are not always available in all regions. This study assessed the value of medical information systems (PMSI) to identify incident cases of colorectal cancer METHODS: Two algorithms were elaborated to identify these incident cases in the PMSI database. The first algorithm was based on diagnosis and medical procedure codes and the second algorithm was based exclusively on diagnoses and the absence of diagnoses over the last five years. The results obtained for two departments were compared with those derived from two cancer registries, constituting the reference data. We then elaborated two multivariate logistic regression models to correct the national number of incident cases estimated by the algorithm adopted after evaluation of the results. RESULTS: The first algorithm provided results that were very close to those derived from the regional registries (646 vs 645 cases) with a good sensitivity and positive predictive value of 75%. The second algorithm overestimated the incidence by about 50% with a positive predictive value of 60% and was therefore not adopted for the national estimation. By applying the first algorithm to the national PMSI MCO database (medicine, surgery, obstetrics), and after modelling, the estimated incidence differed by only 2.34% compared to that observed by all 14 registries. The national estimation of cancer incidence was 39,122 [37,020; 41,224] cases for 2005, which is consistent with the figure published by the Francim national registry network (37,413). CONCLUSION: This study demonstrates the value of PMSI data for estimation of national incidence rates for certain cancers in the absence of cancer registries. However, raw data must be corrected and can be achieved by the method proposed here.
Référence
Sante Publique. 2014 Jan-Feb;26(1):55-63.
