Combination of multiple functional markers to improve diagnostic accuracy

Area of Science:

• Biostatistics and Medical Informatics
• Biomarker Discovery and Validation

Background:

• Combining multiple biomarkers enhances diagnostic accuracy, a critical need in clinical practice.
• While scalar biomarker combinations are well-studied, methods for functional markers (curves, images) are limited.
• Existing scalar methods cannot be directly applied to high-dimensional functional markers.

Purpose of the Study:

• To develop a method for combining multiple functional markers to improve diagnostic accuracy.
• To address the challenge of applying scalar combination techniques to infinite-dimensional functional data.
• To provide a practical tool for researchers and clinicians using functional biomarker data.

Main Methods:

• Propose a one-dimensional scalar feature derived from square loss distance for functional markers.
• Utilize functional principal component decomposition to generate projection scores for the scalar feature.
• Apply existing scalar combination methods to the derived scalar features after dimension reduction.

Main Results:

• The proposed scalar feature effectively retains information from the original functional curves.
• Scalar combination methods applied to the reduced functional markers improve diagnostic accuracy.
• Performance is assessed using Area Under the Receiver Operating Characteristic Curve (AUC) and Youden index.
• Demonstrated application in analyzing Hong Kong respiratory disease hospital admissions using weather and media data.

Conclusions:

• The novel scalar feature enables the effective combination of multiple functional markers for enhanced diagnostics.
• This approach overcomes the dimensionality challenge of functional data, allowing application of established methods.
• An R function is provided for convenient implementation of the proposed methodology.