Dynamic Prediction of Venous Thromboembolism in Gynecological Surgery Using Perioperative Biomarkers and an Extremely...

Area of Science:

• Medical Science
• Machine Learning Applications
• Surgical Outcomes

Background:

• Hospital-associated venous thromboembolism (HA-VTE) is a leading preventable cause of death in gynecological surgery.
• Current risk assessment tools, like the Caprini score, inadequately address gynecological surgery-specific risks.
• There is a need for improved VTE risk stratification in this patient population.

Purpose of the Study:

• To develop and validate a preliminary machine learning (ML) framework for predicting VTE in gynecological surgery patients.
• To compare the performance of the ML model against the established Caprini score.

Main Methods:

• Retrospective case-control study involving 75 VTE cases and 225 controls from gynecological surgery.
• Development of an Extremely Randomized Trees (Extratrees) ML classifier using 22 predictors.
• Model training and validation using a 70:30 split, with performance evaluated by Area Under the Curve (AUC) and accuracy.

Main Results:

• The ML model achieved an AUC of 0.907 (95% CI: 0.833-0.972), significantly outperforming the Caprini score (AUC 0.731, 95% CI: 0.660-0.803).
• The model demonstrated strong discriminative ability with an accuracy of 0.833.
• Consistent performance was observed across validation groups.

Conclusions:

• The developed ML framework offers enhanced risk stratification for VTE in gynecological surgery.
• This specialized ML model shows superior predictive performance compared to the traditional Caprini score.
• The findings support the potential of ML in improving patient safety and outcomes in gynecological procedures.