Integrated partially linear model for multi-center studies with heterogeneity and batch effect in covariates

Area of Science:

• Biostatistics
• Data Science
• Collaborative Research

Background:

• Multi-center studies leverage multiple research groups for robust findings.
• Conventional analysis struggles with nonlinear predictors, batch effects, and heterogeneity in collaborative research.
• Ignoring these complexities leads to biased estimates and unreliable outcomes in multi-center settings.

Purpose of the Study:

• To propose an integrated partially linear regression model (IPLM) for multi-center studies.
• To simultaneously account for predictor nonlinearity, batch effects, group heterogeneity, high-dimensional covariates, and measurement error.
• To provide a unified analysis framework for complex multi-center data.

Main Methods:

• Utilizes local linear regression for nonlinear component estimation.
• Employs a regularization procedure for identifying homogeneous or heterogeneous predictor effects.
• The IPLM model simplifies to a single parsimonious model when predictor effects are homogeneous across centers.

Main Results:

• The proposed IPLM method demonstrates asymptotic estimation and variable selection consistency, even with high-dimensional covariates.
• The method effectively handles nonlinearity, batch effects, and heterogeneity simultaneously.
• Numerical simulations and an Alzheimer's disease project illustrate the method's effectiveness and computational efficiency.

Conclusions:

• The integrated partially linear regression model (IPLM) offers a powerful solution for complex multi-center data analysis.
• IPLM provides accurate and reliable regression estimates by addressing multiple data complexities.
• This approach enhances the applicability and reproducibility of findings in large-scale collaborative studies.