Machine learning predicts cancer-associated deep vein thrombosis using clinically available variables.

Area of Science:

• Oncology
• Medical Informatics
• Thrombosis Research

Background:

• Deep vein thrombosis (DVT) is a significant complication in cancer patients.
• Accurate risk prediction is crucial for timely intervention and management.
• Existing risk assessment tools may have limitations in predicting cancer-associated DVT.

Purpose of the Study:

• To develop and validate machine learning (ML) models for predicting cancer-associated DVT.
• To compare the predictive performance of ML models against the established Khorana score (KS).
• To provide a practical tool for clinical risk assessment.

Main Methods:

• Retrospective analysis of 2100 cancer patients' data.
• Application of five ML algorithms: LDA, LR, CT, RF, and SVM.
• Validation using cross-validation and comparison with KS based on AUC, sensitivity, specificity, and accuracy.

Main Results:

• Cancer-associated DVT incidence was 22.3%.
• Linear Discriminant Analysis (LDA) and Logistic Regression (LR) models showed superior performance (AUC=0.773 and 0.772, respectively) compared to KS (AUC=0.642).
• Key predictors included D-dimer level, age, Charlson Comorbidity Index, length of stay, and VTE history. A nomogram and web calculator were developed.

Conclusions:

• ML models, particularly LDA and LR, demonstrate strong potential for predicting cancer-associated DVT.
• The developed nomogram and web calculator offer a valuable tool for individualized risk assessment.
• External validation through prospective studies is recommended to confirm the model's generalizability.