Handling missing values in population-based cancer registries: Application to female breast cancer with missing hormonal status

Inferences from population-based cancer registries are used for studying cancer epidemiology, patterns of diagnosis, treatment and clinical outcomes by providing valuable sources of data for health services research. We collaborated with the Florida Cancer Data System (FCDS) Registry and the Florida Agency for Health Care Administration (AHCA). We linked these two databases in order to enhance the information provided by FCDS. The enriched dataset provided us the means to predict survival for female breast cancer patients.

However, often registries have missing patient information on important demographics, clinical and clinicopathological characteristics. Ignoring missing information might lead to inaccurate inferences resulting in under or over estimating the real effect. Therefore, proper handling of missing values in the cancer registries during data capturing, analysis and reporting need further investigation to improve the quality of results.

Advanced statistical methods for handling missing values are available and have been used in population-based surveys such as NHANES but uptake in these methods in cancer registries has been limited and in need of improvement. Depending on the missingness mechanism several ad-hoc methods (simple and hot-deck imputation) and advanced statistical model-based approaches (regression and multiple imputation) for handling and accounting for missing values in cancer registries can be introduced. When these methods are applied to the missing hormonal status for female breast cancer patients, survival inferences can be more accurate in the FCDS and AHCA-linked dataset ; this methodology can be a model for other cancer registries.