A mixed linear model controlling for case underascertainment across multiple cancer registries estimated time trends in survival

Abstract

Objectives: Large temporal and geographical variation in survival rates estimated from epidemiological cancer registries coupled with heterogeneity in death certificate only (DCO) notifications makes it difficult to interpret trends in survival. The aim of our study is to introduce a method for estimating such trends while accounting for heterogeneity in DCO notifications in a cancer site-specific manner.

Study design and setting: We used the data of 4.0 million cancer cases notified in 14 German epidemiological cancer registries. Annual 5-year relative survival rates from 2002 through 2013 were estimated, and proportions of DCO notifications were recorded. "DCO-excluded" survival rates were regressed on DCO proportions and calendar years using a mixed linear model with cancer registry as a random effect. Based on this model, trends in survival rates were estimated for Germany at 0% DCO.

Results: For most cancer sites and age groups, we estimated significant positive trends in survival. Age-standardized survival for all cancers combined increased by 7.1% units for women and 10.8% units for men.

Conclusion: The described method could be used to estimate trends in cancer survival based on the data from epidemiological cancer registries with differing DCO proportions and with changing DCO proportions over time.

Keywords: Cancer registries; Cancer survival trends; Death certificate only (DCO); Germany; Mixed linear regression; Survival analysis.