Mediation analysis using incomplete information from publicly available data sources.

Area of Science:

• Epidemiology
• Biostatistics
• Cancer Research

Background:

• Racial disparities in colorectal cancer (CRC) incidence persist in the U.S.
• Existing mediation analysis methods are limited by data requirements, particularly for diverse populations.
• Novel methods are needed to integrate multiple datasets with incomplete information for mediation analysis.

Purpose of the Study:

• To develop and validate a new statistical method for mediation analysis using multiple, diverse datasets.
• To accommodate complex survey and registry data, and allow for multiple mediators.
• To rigorously evaluate the proportion of racial disparity in CRC risk mediated by specific modifiable risk factors.

Main Methods:

• Proposed a novel mediation analysis approach integrating multiple datasets with incomplete information.
• The method accommodates complex survey and registry data, and allows for multiple mediators.
• Validated the approach through simulations for unbiased causal effect estimates and confidence intervals.

Main Results:

• The developed method yields unbiased causal effects estimates and confidence intervals with nominal coverage in simulations.
• Applied the method to U.S. cancer registry and survey data.
• Evaluated the mediation of CRC risk differences between non-Hispanic Whites and Blacks by screening history, BMI, and aspirin use.

Conclusions:

• The new mediation analysis method effectively integrates diverse data sources for complex epidemiological questions.
• The approach provides a robust framework for understanding the mediators of racial disparities in diseases like CRC.
• This work lays the groundwork for more accurate assessments of risk factor contributions to health inequities.