A Machine Learning-Based Prognostic Model for Lower Extremity Deep Vein Thrombosis Following Acute Stroke.

Area of Science:

• Neurology
• Medical Informatics
• Vascular Medicine

Background:

• Deep vein thrombosis (DVT) is a significant complication following acute stroke.
• Accurate prediction of DVT risk is crucial for timely intervention and improved patient outcomes.
• Existing prognostic models may not fully incorporate detailed limb functional assessments.

Purpose of the Study:

• To develop and validate a machine learning (ML)-based prognostic model for predicting lower extremity DVT risk post-acute stroke.
• To emphasize the role of limb functional assessments in DVT risk prediction.
• To compare the performance of different ML models against traditional statistical methods.

Main Methods:

• Retrospective analysis of 225 acute stroke patients.
• Variable selection using LASSO regression.
• Comparison of Gradient Boosting Machine (GBM), Random Survival Forest (RSF), and Generalized Linear Model (GLM) survival models.
• Performance evaluation using concordance index (C-index), C/D AUC, and integrated Brier score (IBS).

Main Results:

• Six key predictors identified: age, stroke type, gender, muscle tension, Brunnstrom stage (lower limb), and sitting balance.
• The Random Survival Forest (RSF) model demonstrated superior performance (IBS: 0.081, C-index: 0.841).
• Age and lower limb Brunnstrom stage were the most influential predictors.

Conclusions:

• An effective ML-based prognostic model for predicting lower extremity DVT risk after acute stroke was developed.
• The model highlights the significance of age and lower limb functional status (Brunnstrom stage) in DVT risk.
• This tool offers potential for enhanced clinical risk stratification in stroke patients.