Machine Learning Predicts Peripherally Inserted Central Catheters-Related Deep Vein Thrombosis Using Patient Features...

Area of Science:

• Medical Informatics
• Clinical Data Science
• Biostatistics

Background:

• Peripherally inserted central catheters (PICCs) are widely used but associated with deep vein thrombosis (DVT).
• Predicting PICCs-DVT is crucial for patient safety and effective clinical management.
• Identifying key patient and catheterization features can improve risk stratification.

Purpose of the Study:

• To develop and evaluate machine learning (ML) models for predicting PICCs-DVT.
• To analyze the importance of patient and catheterization features in PICCs-DVT prediction.
• To provide data-driven insights into the formation mechanisms of PICCs-DVT.

Main Methods:

• A systematic literature review identified 21 relevant patient and catheterization features.
• Data from 1,065 PICCs patients were used to train and test ML models (SVC, RF, ANN).
• Permutation Importance was employed for feature importance analysis, evaluating prediction performance using precision, recall, accuracy, and AUC.

Main Results:

• ML models demonstrated strong performance in predicting PICCs-DVT, with mean metrics of P=0.92, R=0.95, ACC=0.88, and AUC=0.81.
• The Random Forest model achieved the highest performance (P=0.95, R=0.96, ACC=0.92, AUC=0.86).
• Catheter-to-vein rate, Zubrod-ECOG-WHO score, and insertion attempt were consistently identified as the top three most important features across models.

Conclusions:

• Machine learning models are effective tools for predicting PICCs-DVT.
• Specific patient and catheterization features significantly contribute to PICCs-DVT risk.
• These findings can assist clinicians in understanding PICCs-DVT formation and implementing preventative strategies.