Machine learning enhances risk stratification and treatment failure prediction in diffuse large B-cell lymphoma

Area of Science:

• Hematology
• Oncology
• Medical Informatics

Background:

• Diffuse large B-cell lymphoma (DLBCL) is the most common aggressive non-Hodgkin lymphoma globally.
• Current prognostic models like the NCCN IPI have limitations, including small variable sets and discretized inputs.
• There is a need for advanced prognostic tools that incorporate longitudinal data and nonlinear relationships for improved patient stratification.

Purpose of the Study:

• To develop and validate a machine learning model (ML_All) for predicting treatment failure in DLBCL within two years of first-line therapy.
• To compare the performance of ML_All against the established NCCN IPI.
• To assess the model's ability to reduce age-related bias in risk stratification.

Main Methods:

• Utilized clinical and laboratory data from 14,832 patients in the Danish Lymphoid Cancer Research (DALY-CARE) resource (2005-2021).
• Developed ML_All, a machine learning model for fixed-time risk prediction of treatment failure.
• Evaluated ML_All's performance using recall, precision, precision-recall AUC, and survival analyses in a blinded control population.

Main Results:

• ML_All demonstrated significant improvements over NCCN IPI, with a 22% relative increase in recall and a 7% increase in precision.
• The precision-recall AUC for ML_All was 28% higher than NCCN IPI in a blinded control population.
• ML_All identified over four times more low-risk patients than NCCN IPI, accurately classifying older patients with favorable outcomes.

Conclusions:

• ML_All offers more accurate and individualized risk stratification for DLBCL patients compared to existing models.
• The model's performance was robust across DLBCL and broader aggressive lymphoma cohorts, indicating shared prognostic signals.
• ML_All provides a adaptable framework for registry-based prognostic modeling applicable to other lymphoma subtypes and healthcare systems.