Integrating information across datasets to improve prognostic models for patients with diffuse large B-cell lymphoma

Accurate survival prediction is vital for guiding clinical decisions in diffuse large B-cell lymphoma (DLBCL), especially for advancing precision medicine within its genomically heterogeneous subgroups. However, datasets that include both detailed clinical variables and genomic profiling are usually small. In contrast, large population-based datasets often lack multiomics data. This gap limits the development and validation of robust prognostic models in DLBCL. We developed a transfer-learning prognostic framework that leverages the external population-level information from the Surveillance, Epidemiology, and End Results program to the prognostic model for an individual-level clinical and genomic cohort. The proposed model is a Cox-model-based boosted regression with individual data for generalization and enhancement (BRIDGE-Cox) that increases predictive accuracy and reduces uncertainty. The BRIDGE-Cox model incorporates the International Prognostic Index (IPI) factors, with model training performed in the Reddy genomic dataset and external validation in an independent National Cancer Institute genomic dataset. Compared with conventional IPI-based and revised IPI-based Cox models, the BRIDGE-Cox model improved predictive performance for overall survival, achieving a higher concordance index (0.70 vs 0.66 and 0.63, respectively) and consistently higher time-dependent area under the receiver operating characteristic curves for 2-, 3-, and 5- year survival prediction by using the repeated cross-validation in the Reddy dataset. Importantly, these gains were maintained within genomic subgroups. External validation confirmed the robustness and generalizability of the BRIDGE-Cox approach. By borrowing large populationlevel data, this model addresses key limitations of existing tools and enables more reliable risk stratification for patients with DLBCL, particularly in genomically defined subpopulations.

Authors

Institutions

Publication Details

Journal
Haematologica
Published
2026-11-19
DOI
https://doi.org/10.3324/haematol.2026.300829
Primary Topic
Lymphoma Diagnosis and Treatment
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Integrating information across datasets to improve prognostic models for patients with diffuse large B-cell lymphoma

Robert Andrew Harkins, Can Xie, Ruosha Li, Christopher R. Flowers et al.
Haematologica
Lymphoma Diagnosis and Treatment
article

Integrating information across datasets to improve prognostic models for patients with diffuse large B-cell lymphoma

Robert Andrew Harkins, Can Xie, Ruosha Li, Christopher R. Flowers, Dai Chihara, Xuelin Huang
article en

Abstract

Accurate survival prediction is vital for guiding clinical decisions in diffuse large B-cell lymphoma (DLBCL), especially for advancing precision medicine within its genomically heterogeneous subgroups. However, datasets that include both detailed clinical variables and genomic profiling are usually small. In contrast, large population-based datasets often lack multiomics data. This gap limits the development and validation of robust prognostic models in DLBCL. We developed a transfer-learning prognostic framework that leverages the external population-level information from the Surveillance, Epidemiology, and End Results program to the prognostic model for an individual-level clinical and genomic cohort. The proposed model is a Cox-model-based boosted regression with individual data for generalization and enhancement (BRIDGE-Cox) that increases predictive accuracy and reduces uncertainty. The BRIDGE-Cox model incorporates the International Prognostic Index (IPI) factors, with model training performed in the Reddy genomic dataset and external validation in an independent National Cancer Institute genomic dataset. Compared with conventional IPI-based and revised IPI-based Cox models, the BRIDGE-Cox model improved predictive performance for overall survival, achieving a higher concordance index (0.70 vs 0.66 and 0.63, respectively) and consistently higher time-dependent area under the receiver operating characteristic curves for 2-, 3-, and 5- year survival prediction by using the repeated cross-validation in the Reddy dataset. Importantly, these gains were maintained within genomic subgroups. External validation confirmed the robustness and generalizability of the BRIDGE-Cox approach. By borrowing large populationlevel data, this model addresses key limitations of existing tools and enables more reliable risk stratification for patients with DLBCL, particularly in genomically defined subpopulations.

Haematologica
The University of Texas MD Anderson Cancer Center (US), The University of Texas Health Science Center at Houston (US)
Lymphoma Diagnosis and Treatment
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon