Machine learning algorithm as a prognostic tool for venous thromboembolism in allogeneic transplant patients

Area of Science:

• Hematology
• Oncology
• Medical Informatics

Background:

• Venous thromboembolism (VTE) is a serious complication after allogenic hematopoietic stem cell transplantation (allo-HSCT).
• VTE significantly increases nonrelapse mortality, highlighting the need for risk stratification.
• Identifying high-risk patients is crucial for targeted therapeutic management and improved survival.

Purpose of the Study:

• To establish a machine learning-based prognostic model for predicting the risk of death in post-transplantation VTE patients.
• To identify key predictors of mortality in this patient population.
• To develop a tool for clinical decision-making to improve patient survival.

Main Methods:

• Retrospective evaluation of 256 VTE patients who underwent allo-HSCT.
• Data divided into derivation (80%) and test (20%) cohorts.
• Least Absolute Shrinkage and Selection Operator (LASSO) regression for feature selection.
• Eight machine learning classifiers evaluated, with logistic regression selected as the best-performing model.
• 10-fold cross-validation and external validation using the test cohort.

Main Results:

• The study identified 8 potential predictors from 54 candidate variables using LASSO regression.
• The developed BRIDGE model, based on logistic regression, accurately predicted 2-year overall survival (OS).
• Area under the curve (AUC) values for the BRIDGE model were 0.883 (training), 0.871 (validation), and 0.858 (test).
• The model demonstrated high agreement between predicted and observed outcomes (Hosmer-Lemeshow test).

Conclusions:

• The BRIDGE model precisely predicts the risk of all-cause death in VTE patients post-allo-HSCT.
• Clinical application of the BRIDGE model can aid physicians in advance treatment strategies.
• Early identification and management of high-risk VTE patients can significantly improve survival rates.