The EUROCARE-5 study on cancer survival in Europe 1999-2007 : database, quality checks and statistical analysis methods

Abstract

Background: Since 25 years the EUROCARE study monitors the survival of cancer patients in Europe through centralised collection, quality check and statistical analysis of population-based cancer registries (CRs) data. The European population covered by the study increased remarkably in the latest round. The study design and statistical methods were also changed to improve timeliness and comparability of survival estimates. To interpret the EUROCARE-5 results on adult cancer patients better here we assess the impact of these changes on data quality and on survival comparisons. Methods: In EUROCARE-5 the survival differences by area were studied applying the complete cohort approach to data on nearly nine million cancer patients diagnosed in 2000–2007 and followed up to 2008. Survival time trends were analysed applying the period approach to data on about 10 million cancer cases diagnosed from 1995 to 2007 and followed up to 2008. Differently from EUROCARE-4, multiple primary cancers were included and relative survival was estimated with the Ederer II method. Results: EUROCARE-5 covered a population of 232 million resident persons, corresponding to 50% of the 29 participating countries. The population coverage increased particularly in Eastern Europe. Cases identified from death certificate only (DCO) were on average 2.9%, range 0–12%. Microscopically confirmed cases amounted to over 85% in most CRs. Compared to previous methods, including multiple cancers and using the Ederer II estimator reduced survival estimates by 0.4 and 0.3 absolute percentage points, on average. Conclusions: The increased population size and registration coverage of the EUROCARE-5 study ensures more robust and comparable estimates across European countries. This enlargement did not impact on data quality, which was generally satisfactory. Estimates may be slightly inflated in countries with high or null DCO proportions, especially for poor prognosis cancers. The updated methods improved the comparability of survival estimates between recently and long-term established registries and reduced biases due to informative censoring.