A study on a real-world data-based VTE risk prediction model for lymphoma patients

Area of Science:

• Oncology
• Hematology
• Data Science

Background:

• Malignant tumors increase venous thromboembolism (VTE) risk, impacting patient prognosis.
• No reliable predictive models currently exist for thrombosis risk in lymphoma patients.
• This study addresses the need for a dependable risk assessment tool for early VTE identification in this population.

Purpose of the Study:

• To develop and validate a machine learning model for predicting VTE risk in lymphoma patients.
• To leverage real-world data for creating a dependable risk assessment tool.
• To facilitate early identification and risk stratification of VTE in clinical practice.

Main Methods:

• Retrospective analysis of 605 hospitalized lymphoma patients (January 2019 - June 2024).
• Inclusion of demographic, comorbidity, tumor-related, treatment-related, and laboratory predictors.
• Development involved multiple imputation, sampling, feature selection strategies, and nine machine learning algorithms.

Main Results:

• 243 models were generated by combining different data processing and machine learning techniques.
• The optimal model (Simp-SMOTE_rf_GBM) achieved the highest predictive performance with an AUC of 0.954.
• Key predictors identified include anticoagulant use, D-dimer, LDH, CVC, CEA, ECOG score, TP, TC, and infectious disease.

Conclusions:

• A validated machine learning model demonstrates excellent VTE risk prediction in lymphoma patients (AUC = 0.954).
• The model supports clinical decision-making for anticoagulation timing and strategy.
• Implementation can improve patient outcomes through early VTE screening and risk stratification.