Fused Lasso Approach in Regression Coefficients Clustering - Learning Parameter Heterogeneity in Data Integration

Area of Science:

• Statistics
• Data Science
• Biostatistics

Background:

• Combining data sets increases sample size and statistical power.
• Data heterogeneity (population, design, coordination) poses a significant challenge in data integration.
• Ignoring heterogeneity can lead to biased estimation and misleading inferences.

Purpose of the Study:

• To propose a novel regularized fusion method for handling data heterogeneity in regression analysis.
• To identify and merge homogeneous parameter clusters across studies without hypothesis testing.
• To develop a computationally efficient procedure for large-scale integrated data analysis.

Main Methods:

• Development of a regularized fusion method utilizing the fused lasso.
• Incorporation of estimated parameter ordering within the fused lasso for computational efficiency.
• Comparison with conventional methods through extensive simulation studies and an application example.

Main Results:

• The proposed fused lasso method effectively identifies and merges inter-study homogeneous parameter clusters.
• The method demonstrates computational efficiency for large-scale integrated data.
• The approach achieves statistical power comparable to traditional methods without compromising interpretation.

Conclusions:

• The regularized fusion method offers a powerful and efficient solution for data heterogeneity in integrated studies.
• This approach enhances statistical power and provides meaningful interpretation, particularly beneficial for numerous smaller data sets.
• The fused lasso provides a robust alternative to traditional methods for analyzing complex, integrated datasets.