Regression calibration utilizing biomarkers developed from high-dimensional metabolites

Area of Science:

• Nutritional Epidemiology
• Biostatistics
• Chronic Disease Research

Background:

• Systematic measurement errors in self-reported dietary intake pose a significant challenge in studying diet-disease relationships.
• Existing regression calibration methods for error correction require objectively measured biomarkers, which are limited for most dietary components.

Purpose of the Study:

• To propose a novel approach using high-dimensional objective measurements to construct biomarkers for a wider range of dietary components.
• To estimate diet-disease associations more accurately by correcting for measurement errors.
• To investigate and address challenges in variance estimation within high-dimensional regression models.

Main Methods:

• Development of a method to construct dietary biomarkers from high-dimensional objective measurements.
• Application of variance estimation techniques including cross-validation, degrees-of-freedom corrected estimators, and refitted cross-validation (RCV).
• Extensive simulations to evaluate the finite sample performance of the proposed estimators.
• Analysis of the Women's Health Initiative cohort data.

Main Results:

• The proposed method effectively constructs biomarkers for numerous dietary components.
• The variance estimation techniques provide reliable estimates in high-dimensional settings.
• The application to the Women's Health Initiative data yields insights into sodium/potassium intake and cardiovascular disease associations.

Conclusions:

• The novel approach enhances the ability to study diet-disease associations by overcoming limitations of traditional biomarker availability.
• The developed variance estimation methods are crucial for robust inference in high-dimensional nutritional epidemiology.
• This work provides a valuable tool for researchers investigating the impact of dietary factors on chronic disease risk.