STATISTICAL METHODS FOR ANALYSIS OF COMBINED CATEGORICAL BIOMARKER DATA FROM MULTIPLE STUDIES

Area of Science:

• Biostatistics
• Epidemiology
• Biomarker Research

Background:

• Biomarker measurements often vary between laboratories, necessitating calibration for pooled data analysis.
• Existing methods assume reference laboratories are the gold standard, which may not reflect true underlying values.
• Addressing laboratory variability is crucial for accurate biomarker-disease relationship estimation.

Purpose of the Study:

• To develop and compare two new statistical methods, exact calibration and cut-off calibration, for analyzing aggregated categorical biomarker data.
• To evaluate the performance of these methods in estimating biomarker-disease relationships.
• To assess the impact of different calibration designs (random sample vs. controls-only) on estimation accuracy.

Main Methods:

• Developed exact calibration and cut-off calibration statistical methods.
• Analyzed aggregated categorical biomarker data.
• Compared method performance under random sample and controls-only calibration designs.
• Applied methods to evaluate vitamin D levels and colorectal cancer risk.

Main Results:

• The exact calibration method yielded significantly less biased estimates and more accurate confidence intervals.
• The cut-off calibration method showed minimal bias and valid confidence intervals with small measurement errors or exposure effects.
• Controls-only calibration designs introduced some bias, which was minimal under small exposure effects or disease prevalences.

Conclusions:

• The exact calibration method is recommended for its superior performance in biomarker data analysis.
• The cut-off calibration method can be a viable alternative under specific conditions of low error and effect size.
• Careful consideration of calibration design is necessary to minimize bias in biomarker-disease association studies.