Classification models for the prediction of clinicians' information needs

Area of Science:

• Medical Informatics
• Clinical Decision Support

Background:

• Clinicians have many unmet information needs during patient care.
• Infobuttons are tools within electronic medical record (EMR) systems that link to online health information.
• These tools aim to help clinicians find information to meet their needs.

Purpose of the Study:

• To develop classification models using medication infobutton usage data.
• To predict medication-related content topics (e.g., dosage, adverse effects, patient education) clinicians are likely to select.
• To improve information retrieval within EMR systems.

Main Methods:

• A dataset of 3078 infobutton sessions with 26 attributes was created.
• Automatic attribute selection was used to reduce dataset complexity.
• Nine classification models were generated using machine learning algorithms.

Main Results:

• Model performance was evaluated using Area Under the ROC Curve (AUC) and kappa agreement.
• The best model achieved an AUC ranging from 0.73 to 0.99.
• High performance was observed in predicting topics such as adult dose, pediatric dose, patient education, and pregnancy category.

Conclusions:

• Classification models using infobutton data show promise for predicting clinician information needs.
• This approach can enhance the utility of EMR systems in supporting patient care.
• Accurate prediction of content topics can streamline access to crucial clinical information.