Nonparametric Biomarker Based Treatment Selection With Reproducibility Data.

Area of Science:

• Biostatistics
• Genomics
• Translational Medicine

Background:

• Evaluating biomarkers for cancer treatment selection is crucial.
• Assay modification, such as migrating gene expression data from Affymetrix to Illumina platforms, presents challenges.
• Existing methods for biomarker migration may rely on assumptions that do not hold in practice, potentially leading to suboptimal treatment decisions.

Purpose of the Study:

• To develop and evaluate a robust method for assessing biomarkers under assay modification.
• To ensure optimal treatment selection despite potential measurement errors introduced during biomarker platform migration.
• To address limitations of classical measurement error models in biomarker evaluation.

Main Methods:

• Utilized nonparametric logistic regression to model the relationship between event rates and biomarkers.
• Assumed a nonparametric relationship between original and migrated biomarkers.
• Employed B-spline approximation for estimation.
• Validated the approach through simulation studies and application to lung cancer data.

Main Results:

• Nonparametric modeling provides optimal marker-based treatment selection.
• Error-contaminated biomarkers, compared to error-free ones, lead to suboptimal treatment selection.
• The proposed method effectively handles deviations from classical measurement error models.

Conclusions:

• The developed nonparametric approach offers a more accurate and optimal strategy for biomarker-based treatment selection following assay modification.
• This method enhances the reliability of biomarker migration and reduces the risk of suboptimal clinical decisions.
• The findings are applicable to lung cancer and potentially other diseases requiring biomarker-guided therapy.