A novel performance scoring quantification framework for stress test set-ups.

Area of Science:

• Biomedical Engineering
• Machine Learning
• Clinical Diagnostics

Background:

• Stress tests are crucial clinical tools for detecting pathology by measuring physiological reserves.
• Developing new stress tests is complex, time-consuming, and requires extensive domain expertise.
• Existing methods often struggle with the disconnect between underlying pathology and clinical manifestation.

Purpose of the Study:

• To introduce a novel machine learning framework, Stress Test Performance Scoring (STEPS), for modeling expected performance in stress tests.
• To provide a distributional-free approach that simplifies and accelerates the creation of new stress tests.
• To enhance the accuracy and efficiency of clinical screening tools.

Main Methods:

• Developed a machine learning framework (STEPS) to model stress test performance.
• Trained a performance scoring function using task performance data, stress test parameters, and subject medical information.
• Investigated various aggregation methods for performance scores across different stress levels via simulation.
• Applied the STEPS framework to a real-world dataset for neurodegeneration screening.

Main Results:

• The STEPS framework demonstrated effectiveness in modeling expected performance during stress tests.
• An Area Under the Curve (AUC) of 84.35 was achieved in distinguishing neurodegeneration patients from controls in a real-world data example.
• The framework successfully integrated domain knowledge and clinical measures to improve screening capabilities.

Conclusions:

• STEPS offers a novel, efficient, and data-driven approach to developing and validating stress test screening tools.
• The framework has the potential to significantly reduce the time and complexity associated with creating new stress tests.
• STEPS improves clinical screening by leveraging machine learning and existing medical data.