Statistical method for pooling categorical biomarker data from multi-center matched/nested case-control studies

Area of Science:

• Epidemiologic research
• Biostatistics
• Biomarker analysis

Background:

• Pooled analyses increase sample size and diversity in epidemiologic research.
• Biomarker data often analyzed categorically, yet pooling methods focus on continuous data.
• Between-study variability in biomarker measurements can bias pooled estimates.

Purpose of the Study:

• To develop a statistical method for pooling categorical biomarker data across studies.
• To address systematic measurement errors and between-study variability.
• To evaluate biomarker-disease relationships in matched/nested case-control studies.

Main Methods:

• Proposed a likelihood-based method for categorical biomarker analysis.
• Incorporated study-specific calibration processes to handle measurement variability.
• Utilized a sandwich variance estimator for valid asymptotic variances.

Main Results:

• Simulation studies evaluated the method's performance under various conditions.
• The proposed methods demonstrated validity in finite sample performance.
• Applied the methods to a vitamin D and colorectal cancer pooling project.

Conclusions:

• The developed statistical approach accurately evaluates biomarker-disease relationships using pooled categorical data.
• The method accounts for calibration uncertainties, yielding reliable regression parameter estimates.
• This work facilitates more robust meta-analyses of biomarker data in epidemiology.