Background – Country of birth (COB) is missing for 12% of people with a cancer diagnosis reported to the Victorian Cancer Registry (VCR). It is hypothesised that most of these people would not have been born in Australia, therefore giving rise to underestimates of cancer incidence in migrant populations when incomplete records are excluded.
Aim – Our aim was to evaluate and compare the accuracy in estimating cancer incidence in migrant groups of: (i) using complete records only and (ii) multiple imputation approaches to handle missing country of birth.
Methods – Using data from the VCR for 2008-2015, we used multiple imputation by chained equations in R to impute missing COB data based on year of diagnosis, year of birth, cancer type, age, sex, residential location and socioeconomic status. From 20 imputations, the mean incidence and standard error was calculated for each of 12 COB regions. We compared results to those obtained using additional data from the Victorian Admitted Episodes Database (VAED) to assess their accuracy.
Results – The mean incidences ranged between 200 and 2,000 per 100,000 population. The incidence for the VCR-only data was the lowest across all COB groups. When comparing with estimates using the VCR-only data to the equivalent complete-case dataset including additional COB data from the VAED, we found that multiple imputation gave accurate estimates.
Conclusion – Our results show that the complete-records approach underestimates cancer incidence in migrant groups as well as Australian/New Zealand born people. Multiple imputation provides accurate estimates of cancer incidence.