Multiple imputation in a large-scale complex survey: a practical guide

Area of Science:

• Health Services Research
• Biostatistics
• Cancer Epidemiology

Background:

• The Cancer Care Outcomes Research and Surveillance (CanCORS) Consortium is a large, population-based study examining lung and colorectal cancer care quality.
• Observational studies like CanCORS frequently encounter complex missing data patterns, posing significant analytical challenges.
• Multiple imputation, a common technique for handling missing data, has seen limited adoption in large-scale, complex datasets.

Purpose of the Study:

• To implement and demonstrate the feasibility of sequential regression multiple imputation for addressing missing data in the CanCORS study.
• To construct a centralized, complete database for use by investigators across multiple sites.
• To provide a practical example and discussion of multiple imputation for complex survey data.

Main Methods:

• Sequential regression multiple imputation was employed to handle non-response in CanCORS surveys.
• Publicly available software was utilized for the implementation of the imputation methods.
• A centralized, completed database was constructed from the imputed data.

Main Results:

• The study successfully illustrated the feasibility of using multiple imputation in a large-scale, multi-objective survey.
• The implemented method demonstrated the capacity to handle complex missing data patterns effectively.
• A centralized database was created, facilitating easier data access for consortium investigators.

Conclusions:

• Multiple imputation is a viable and effective strategy for managing complex missing data in large observational studies like CanCORS.
• The detailed implementation process serves as a valuable guide for practitioners facing similar data challenges.
• Further research is warranted to address remaining challenging issues in handling missing data in complex survey research.