Statistical inference for high-dimensional generalized estimating equations

Area of Science:

• Biostatistics
• Genomics
• Proteomics

Background:

• Analyzing correlated data with many variables is challenging, especially with limited sample sizes.
• High-throughput omics data, like proteomics, often present this high-dimensional correlated structure.
• Existing methods struggle with inference for high-dimensional regression coefficients in generalized estimating equations.

Purpose of the Study:

• To develop a novel inference procedure for linear functionals of high-dimensional regression coefficients.
• To address the analysis of correlated omics data, motivated by COVID-19 studies.
• To introduce a data-driven method for selecting tuning parameters in high-dimensional settings.

Main Methods:

• Developed a novel inference procedure using projected estimating equations.
• Established asymptotic normality of the proposed estimator under mild conditions.
• Introduced a cross-validation technique for tuning parameter selection.

Main Results:

• The proposed estimator is asymptotically normally distributed.
• Demonstrated robust finite-sample performance in simulations, particularly in bias and coverage.
• Successfully applied the procedure to provide confidence intervals for protein-COVID risk associations.

Conclusions:

• The novel procedure offers a statistically sound approach for high-dimensional correlated data analysis.
• The method enhances confidence interval estimation for associations in omics studies.
• The data-driven cross-validation improves practical application and reliability.