Combining markers with and without the limit of detection

Area of Science:

• Biostatistics
• Proteomics
• Biomarker Discovery

Background:

• Measuring low-level proteomic markers is challenging due to the limit of detection (LOD).
• Existing methods may not accurately combine markers when LOD is present.
• Proteomic markers are crucial for distinguishing between patient groups, such as cancer patients and non-cancer patients.

Purpose of the Study:

• To develop a statistical method for combining proteomic markers that accounts for the limit of detection (LOD).
• To propose a marker selection and combination procedure considering marker correlations.
• To enhance the accuracy of biomarker panels for disease classification.

Main Methods:

• Proposed a method to estimate distribution parameters of multivariate normal markers, incorporating LOD.
• Utilized linear discriminant analysis for marker combination after LOD-adjusted parameter estimation.
• Developed a marker subset selection procedure based on marker correlations.

Main Results:

• The proposed method significantly improved ROC curve parameter estimates compared to methods ignoring LOD.
• Simulation studies demonstrated that marker correlations substantially impact combined marker accuracy.
• The developed procedure effectively selected and combined markers, outperforming selection based solely on individual marker accuracy.

Conclusions:

• Accounting for the limit of detection (LOD) is critical for accurate proteomic marker combination.
• Marker correlation plays a vital role in the performance of combined biomarker panels.
• The proposed statistical framework offers a robust approach for biomarker discovery and combination in clinical applications, particularly for cancer detection.