National cancer incidence is estimated using the incidence/mortality ratio in countries with local incidence data: Is this estimation correct?
Menée en France, cette étude analyse la fiabilité des données nationales d'incidence et de mortalité par cancer issues des registres des cancers en les comparant avec les données hospitalières et celles de l'assurance maladie, pour 22 sites de cancer sur la période 2004-2006
Background In countries with local cancer registration, the national cancer incidence is usually estimated by multiplying the national mortality by the incidence/mortality (I/M) ratio from pooled registries. This study aims at validating this I/M estimation in France, by a comparison with estimation obtained using the ratio of incidence over hospital discharge (I/HD) or the ratio of incidence over health insurance data (long-duration diseases, I/LDD). Methods This comparison was performed for 22 cancer sites over the period 2004–2006. In France, a longitudinal I/M approach was developed relying on incidence and mortality trend analyses; here, the corresponding estimations of national incidence were extracted for 2004–2006. The I/HD and I/LDD estimations were performed using a common cross-sectional methodology. Results The three estimations were found similar for most cancers. The relative differences in incidence rates (vs. I/M) were below 5% for numerous cancers and below 10% for all cancers but three. The highest differences were observed for thyroid cancer (up to +21% in women and +8% in men), skin melanoma (up to +13% in women and +8% in men), and Hodgkin disease in men (up to +15%). Differences were also observed in women aged over 60 for cervical cancer. Except for thyroid cancer, differences were mainly due to the smoothing performed in the I/M approach. Conclusion Our results support the validity of I/M approaches for national estimations, except for thyroid cancer. The longitudinal version of this approach has, furthermore, the advantage of providing smoothed estimations and trend analyses, including useful birth-cohort indicators, and should thus be preferred.
Cancer Epidemiology , résumé, 2012