Machine learning-based prediction model of lower extremity deep vein thrombosis after stroke

Area of Science:

• Medical Informatics
• Neurology
• Thrombosis Research

Background:

• Post-stroke lower extremity deep vein thrombosis (DVT) is a significant complication.
• Accurate risk prediction models are crucial for timely intervention and improved patient outcomes.

Purpose of the Study:

• To develop and validate machine learning (ML) models for predicting post-stroke lower extremity DVT.
• To identify key predictors of DVT in stroke patients using ML techniques.
• To compare the performance of different ML algorithms and select the optimal model for clinical application.

Main Methods:

• Retrospective analysis of 620 stroke patients.
• Development and evaluation of eight ML models: logistic regression, SVM, random forest, decision tree, neural network, XGBoost, Bayesian, and KNN.
• Performance assessment using ROC curves, AUC, PR curves, PRAUC, accuracy, sensitivity, specificity, and clinical decision curves (DCA).
• Feature importance analysis using Shapley's additive explanation (SHAP).

Main Results:

• The random forest (RF) algorithm demonstrated superior performance with high AUC (0.74/0.73) and accuracy (0.75/0.77).
• SHAP analysis identified D-dimer, age, Brunnstrom stage (lower limb), prothrombin time (PT), and mobility ability as significant DVT predictors.
• The RF model achieved the highest clinical net benefit according to DCA analysis.

Conclusions:

• Machine learning, particularly the RF algorithm, offers a robust approach for predicting post-stroke lower extremity DVT.
• Key clinical and laboratory indicators can be effectively utilized by ML models to guide DVT prevention strategies.
• The developed RF model shows potential for enhancing clinical decision-making in stroke patient management.