A Shared Decision-Making System for Diabetes Medication Choice Utilizing Electronic Health Record Data.

Area of Science:

• Medical Informatics
• Clinical Decision Support
• Diabetes Management

Background:

• Shared decision-making (SDM) is crucial for managing complex conditions like type-2 diabetes mellitus (T2DM).
• Current clinical guidelines lack personalized patient health condition considerations in antihyperglycemic medication decisions.
• Existing decision aids do not adequately integrate patient-specific data.

Purpose of the Study:

• To propose a novel SDM system framework for T2DM patients.
• To integrate guideline knowledge with patient-specific electronic health record (EHR) data.
• To develop a multilabel classification model for recommending antihyperglycemic medications.

Main Methods:

• Utilized a multilabel classification model trained on class-imbalanced EHR data from 2542 T2DM inpatients.
• Engineered 77 features from EHR data to predict eight antihyperglycemic medication labels.
• Validated the recommendation model's performance on real-world data.

Main Results:

• The multilabel recommendation model demonstrated stable performance across single-label classifications.
• The model successfully predicted minority positive cases, achieving an average recall of 0.9898 across eight classes.
• As a multilabel classifier, the model achieved high performance with an F_exam score of 0.8269 and Recall_exam of 0.9664.

Conclusions:

• The proposed SDM system framework effectively integrates guideline knowledge with EHR data for personalized T2DM treatment.
• EHR-driven decision support enhances physician-patient understanding and tailors medication recommendations.
• This approach facilitates more effective shared decision-making conversations for antihyperglycemic medication strategies.