Computational approaches to support comparative analysis of multiparametric tests: Modelling versus Training

Area of Science:

• Oncology
• Genomics
• Biostatistics

Background:

• Multiparametric assays are crucial for breast cancer risk stratification, with expanding applications in other cancers.
• Discrepancies in risk estimates between different tests necessitate reliable comparison methods.
• Existing multigene tests like Oncotype DX®, Prosigna™, and MammaPrint® utilize distinct computational approaches.

Purpose of the Study:

• To develop and evaluate cost-effective methods for comparing the performance of different multigene cancer risk stratification tests.
• To assess the accuracy of computational modeling versus a training approach for deriving risk classifications.
• To provide a framework for robust comparisons between multigene tests in clinical practice.

Main Methods:

• Computational modeling of existing multigene test results using published algorithms.
• A training approach utilizing reference results from commercially available tests.
• Comparison of derived risk classifications against actual test results in a breast cancer cohort.

Main Results:

• Computational modeling without real-world data can introduce errors in risk estimation.
• The training approach demonstrated a cost-effective solution for real-world comparisons of multigene signatures.
• The training approach achieved a close approximation to the true signature results.

Conclusions:

• A training approach is superior to purely computational modeling for accurate multigene test comparison.
• This study presents a viable, cost-effective method for comparing multigene tests, aiding healthcare providers and researchers.
• Robust comparison of multigene tests is essential for reliable cancer risk stratification and management.