Change surface regression for nonlinear subgroup identification with application to warfarin pharmacogenomics data.

Area of Science:

• Genetics and Genomics
• Pharmacology
• Biostatistics

Background:

• Pharmacogenomics is key to personalized medicine, optimizing drug efficacy and reducing adverse effects by studying genetic variations.
• Drug metabolism complexity and nongenetic factors create heterogeneity in drug response across populations.
• Existing methods struggle with complex, high-dimensional datasets like the International Warfarin Pharmacogenetic Consortium (IWPC) data.

Purpose of the Study:

• To develop a novel change surface model for identifying patient subgroups with distinct drug-gene associations.
• To capture and model between-patient heterogeneity in drug dosing requirements.
• To provide a clearer understanding of dynamic drug-gene associations in complex datasets.

Main Methods:

• Formulation of a novel change surface model for multiple subgroup identification.
• Accommodation of nonlinear subgroup divisions and handling of high-dimensional data via a doubly penalized approach.
• An iterative 2-stage method combining change point detection and smoothed local adaptive majorize-minimization for surface regression.

Main Results:

• The proposed model effectively identifies nonlinear subgroup structures in complex data.
• Extensive numerical studies demonstrate the method's performance.
• Application to the IWPC dataset identified 3 distinct patient subgroups with unique pharmacogenomic relationships.

Conclusions:

• The change surface model offers a powerful approach for subgroup identification in pharmacogenomics.
• This method enhances understanding of drug-gene associations and patient heterogeneity.
• Findings contribute valuable insights for personalized medicine, particularly in warfarin dosing.