Dealing With Missing, Imbalanced, and Sparse Features During the Development of a Prediction Model for Sudden Death...

Area of Science:

• Emergency Medicine
• Data Science
• Clinical Informatics

Background:

• Emergency department (ED) data present challenges like missing, imbalanced, and sparse features, hindering the development of effective disease prediction models.
• Early diagnosis and timely rescue in EDs are crucial for patient survival but are often compromised by data quality issues.

Purpose of the Study:

• To propose a systematic approach for addressing missing, imbalanced, and sparse features in ED data.
• To develop robust sudden-death prediction models using emergency medicine data.

Main Methods:

• A 3-step data preprocessing strategy was employed: Random Forest (RF) for missing values, k-means for imbalanced data, and Principal Component Analysis (PCA) for sparse features.
• Performance was evaluated using R-squared (R²) and kappa (κ) coefficients for variable interpolation, and Area Under the Receiver Operating Characteristic Curve (AUROC) and Area Under the Precision-Recall Curve (AUPRC) for model estimation.
• A logistic regression (LR) model was built using processed ED data from Hainan Hospital of Chinese PLA General Hospital.

Main Results:

• Data preprocessing significantly improved model performance compared to using initial diagnostic data.
• The logistic regression model achieved a recall of 0.746, an F1-score of 0.73, and an AUROC of 0.708 after data processing.
• The proposed method demonstrated effectiveness in handling data quality issues, leading to a more reliable prediction model.

Conclusions:

• The systematic approach is effective for building reliable prediction models for emergency patients.
• Addressing data quality issues in ED data is critical for improving patient outcomes and survival rates.