Peripherally inserted central-related upper extremity deep vein thrombosis and machine learning.

Area of Science:

• Medical Informatics
• Oncology
• Vascular Medicine

Background:

• Peripherally inserted central catheters (PICCs) are widely used in cancer patients.
• Upper extremity deep vein thrombosis (UEDVT) is a significant complication associated with PICC use.
• Accurate risk prediction for PICC-related UEDVT is crucial for patient management.

Purpose of the Study:

• To develop and evaluate machine learning (ML) models for predicting UEDVT in patients with PICCs.
• To compare the predictive performance of ML models against the established Seeley scale.
• To assess the effectiveness of ML in identifying patients at high risk for UEDVT.

Main Methods:

• A cohort of 452 cancer patients with PICCs was analyzed.
• Machine learning models, including LASSO regression (ML-LASSO and ML-Seeley-LASSO), were constructed.
• Models were trained and tested using patient data, with UEDVT diagnosis confirmed by ultrasound.

Main Results:

• ML models demonstrated superior predictive performance compared to the Seeley scale.
• The ML-LASSO model achieved an area under the curve of 0.856 in the test set.
• Both ML-Seeley-LASSO (0.799) and ML-LASSO models showed significant effectiveness in predicting PICC-related UEDVT.

Conclusions:

• Machine learning modeling provides accurate risk estimation for PICC-related UEDVT.
• Implementing ML-based prediction can help mitigate the incidence of UEDVT.
• This study highlights the potential of ML in improving patient outcomes in oncology care.