Better tools for better estimates: improving approaches to handling missing data in Swiss cancer registries

Area of Science:

• Epidemiology
• Biostatistics
• Cancer Registry Science

Background:

• Missing vital status data is a common challenge in cancer registries.
• Various statistical methods exist to address this data gap.
• The accuracy of these methods for cancer registry analytics is not well-established.

Purpose of the Study:

• To compare different approaches for handling missing vital status data in cancer registries.
• To identify methods that yield the least biased estimates for typical cancer registry analyses.
• To evaluate the performance of imputation techniques in survival and incidence calculations.

Main Methods:

• A simulation study using Swiss National Agency for Cancer Registration data for six tumor types.
• Artificial introduction of 5%, 10%, and 15% missing vital status.
• Comparison of estimates (five-year overall survival, relative survival, standardized incidence ratio) using no imputation, single imputation, and multiple imputation against true values.

Main Results:

• Multiple imputation produced the least biased standardized incidence ratio estimates for colorectal cancer.
• Single imputation (-0.32) was more biased than no imputation (-0.21).
• A similar bias pattern was observed for overall survival and relative survival estimates.

Conclusions:

• Single imputation techniques for missing vital status data are likely too biased for practical use in cancer registries.
• Multiple imputation methods demonstrated the least bias for standardized incidence ratio, overall survival, and relative survival estimates.
• Multiple imputation shows promising, generalizable performance for cancer registry data analysis.