Methodological aspects of estimating rare cancer prevalence in Europe: the experience of the RARECARE project.

This paper describes the usage and the performance evaluation of the completeness index method in the 'Surveillance of Rare Cancers in Europe project' (RARECARE) for estimating rare cancer prevalence in Europe. The 15-year prevalence at 1st January 2003 for 255 cancers is obtained from a pool of 22 RARECARE cancer registries (CRs). Incidence and survival models are applied to the RARECARE database to estimate the parameters from which the completeness indices are calculated. Complete prevalence is obtained adjusting the observed 15-year prevalence by the completeness index, to account for those cancer survivors diagnosed before the CR activity started. Main factors influencing the performance of the completeness index method for rare cancers are the same as for common cancers: age distribution of incidence and lethality of the cancer. For cancers occurring in the elderly, with low survival rates and consequently a restricted number of long-term survivors we obtained completeness indices higher than 0.9. Values lower than 0.7 correspond to those cancers with good prognosis and/or incidence more concentrated at the younger ages, indicating that 15 years of follow up are insufficient to detect all prevalent cases. Validation analysis shows that for a restricted subgroup of rare cancers with very low incidence and low survival, the completeness indices were not able to adequately correct the observed prevalence even considering a registration period of 20 years. On average, sensitivity analyses show a slight overestimation of complete prevalence for rare and common cancers whose increasing incidence is known in literature. RARECARE is the largest project on rare cancers conducted to date. Improving health care programs for cancer survivors is a public health priority and prevalence data which provides important information in this field should be regularly asked to Member States and included in the EU health statistics.