Mortality Prediction in Diffuse Large B-Cell Lymphoma Using Supervised Machine Learning Models-A Retrospective Study.

Area of Science:

• Oncology
• Biostatistics
• Computational Biology

Background:

• Diffuse large B-cell lymphoma (DLBCL) is a heterogeneous cancer with unpredictable patient outcomes.
• Accurate prognostic models are crucial for tailoring treatment and follow-up plans.
• Predicting patient survival at diagnosis is essential for effective clinical management.

Purpose of the Study:

• To evaluate the predictive performance of various machine learning (ML) models for 26-month mortality in DLBCL patients.
• To compare ML model efficacy against the traditional Cox proportional hazards model.
• To identify optimal modeling approaches for risk stratification in DLBCL.

Main Methods:

• Retrospective analysis of 412 DLBCL patients diagnosed and treated between 2015-2023.
• Utilized baseline clinical and paraclinical data for model training and testing.
• Compared six ML models (logistic regression, RF, SVM-RBF, MLP, RSF, XGBoost) with the Cox model.

Main Results:

• Random Forest (RF) demonstrated the highest predictive accuracy (AUC=0.9060, Accuracy=0.833, F1=0.902).
• XGBoost and MLP also showed strong performance, outperforming the Cox model (AUC=0.5561).
• RF and logistic regression exhibited the best model calibration.

Conclusions:

• Machine learning frameworks, especially RF, significantly outperform classical statistical models for DLBCL outcome prediction.
• These ML models offer a promising tool for enhancing risk assessment in DLBCL patients.
• The findings support the integration of advanced computational methods in clinical oncology.