Sample-efficient learning with auxiliary class-label information.

Area of Science:

• Machine Learning
• Medical Informatics
• Clinical Decision Support

Background:

• Clinical data classification models require extensive expert labeling, which is time-consuming and costly.
• Reducing the number of labeled examples is essential for cost-efficiency in developing these models.
• Existing methods often lack ways to incorporate nuanced expert confidence during the learning process.

Purpose of the Study:

• To develop and evaluate a novel classification learning approach that utilizes auxiliary information reflecting expert confidence.
• To improve the cost-efficiency of building clinical classification models by reducing the need for extensive manual annotation.
• To enhance classification performance when working with limited labeled clinical datasets.

Main Methods:

• Developed a new classification approach integrating support vector machines (SVM) and learning to rank methodologies.
• Incorporated auxiliary information representing expert confidence levels alongside traditional class labels.
• Tested the approach on the specific task of learning an alert model for Heparin Induced Thrombocytopenia (HIT).

Main Results:

• The proposed method effectively utilizes expert confidence information to enhance classification learning.
• Models trained with the new approach demonstrated improved performance with a significantly smaller number of labeled examples.
• The approach proved beneficial for the HIT alert model, showing enhanced classification accuracy.

Conclusions:

• Incorporating expert confidence into the learning process is a valuable strategy for clinical data classification.
• This method offers a more cost-efficient and effective way to build accurate predictive models from electronic health records.
• The developed approach has practical implications for improving clinical decision support systems and reducing diagnostic burdens.