Identifying Heterogeneous Effect using Latent Supervised Clustering with Adaptive Fusion.

Area of Science:

• Biomedical Informatics
• Machine Learning
• Statistical Genetics

Background:

• Precision medicine aims to tailor treatments to individual patients.
• Identifying patient subgroups with heterogeneous treatment effects is crucial.
• Existing methods often require strong assumptions or lack predictive power.

Purpose of the Study:

• To introduce latent supervised clustering, a novel exploratory machine learning tool.
• To identify heterogeneous subpopulations for personalized treatment strategies.
• To achieve competitive estimation and prediction accuracy with minimal prior assumptions.

Main Methods:

• Formulating the problem as a regression with subject-specific coefficients.
• Employing adaptive fusion to cluster coefficients into subpopulations.
• Developing an efficient accelerated proximal gradient algorithm for parameter estimation.

Main Results:

• The proposed method demonstrates competitive estimation and prediction accuracy.
• Numerical studies validate the effectiveness of latent supervised clustering.
• The approach yields interpretable clustering results for heterogeneous effects.

Conclusions:

• Latent supervised clustering offers a flexible and powerful approach for precision medicine.
• The method effectively identifies patient subpopulations based on treatment response.
• This tool enhances personalized treatment strategies and clinical decision-making.