Learning using privileged information with logistic regression on acute respiratory distress syndrome detection

Area of Science:

• Machine Learning
• Medical Informatics
• Clinical Decision Support

Background:

• Learning Using Privileged Information (LUPI) is an advanced paradigm.
• LUPI leverages training data unavailable during prediction.
• This study applies LUPI to acute respiratory distress syndrome (ARDS) detection.

Purpose of the Study:

• Develop and evaluate privileged logistic regression (PLR) models for ARDS detection.
• Investigate the impact of privileged information on model performance.
• Assess the interpretability and explainability of PLR models.

Main Methods:

• Developed PLR models within the LUPI framework.
• Utilized mechanical ventilation variables and chest X-ray features as privileged information.
• Employed electronic health records as the base domain data.
• Incorporated a specific objective function to encourage knowledge transfer.

Main Results:

• PLR models achieved superior classification performance for ARDS detection compared to standard logistic regression.
• Performance gains were observed even when privileged information was only partially available.
• PLR models matched or exceeded the performance of existing state-of-the-art LUPI models.
• Asymptotic analysis provided conditions for improved convergence rates with privileged information.

Conclusions:

• PLR models are effective for ARDS detection using the LUPI paradigm.
• The models offer enhanced performance, interpretability, and explainability.
• Proposed models are suitable for clinical applications with partially available privileged information.