Divergent prognostic utility of metabolic risk scores in large B-cell lymphoma subtypes: a real-world analysis

Eur J Nucl Med Mol Imaging. 2025 Jun 16. doi: 10.1007/s00259-025-07368-y. Online ahead of print.

ABSTRACT

PURPOSE: Large B-cell lymphomas (LBCL) include diffuse large B-cell lymphoma, not otherwise specified (DLBCL, NOS), and subtypes such as transformed non-Hodgkin’s lymphoma (tNHL), primary mediastinal B-cell lymphoma (PMBL), and double/triple-hit lymphomas (DHL/THL). While metabolic risk scores based on metabolic tumor volume (MTV) have demonstrated prognostic value in DLBCL, NOS, their applicability to other LBCL subtypes remains unclear.

METHODS: Baseline [¹⁸F]FDG-PET/CT scans of LBCL patients treated with R-CHOP regimens at our institution were retrospectively analyzed. Metabolic parameters, including MTV, lesion dissemination (SDmax) and tumor surface volume ratio were calculated for each histological subgroup. Four metabolic risk scores-the international metabolic prognostic index (IMPI), MTV/WHO PS, MTV/SDmax and Clinical PET model- were applied to calculate progression risk in LBCL subtypes. Harrell’s C-index evaluated the prognostic performance. A multivariable model was developed for tNHL.

RESULTS: We included tNHL (n = 88), DHL/THL (n = 32), PMBL (n = 26) and others (n = 49), and compared them to a cohort of previously published DLBCL, NOS (n = 355). IMPI demonstrated the highest C-index amongst the metabolic risk scores in tNHL for progression-free survival (PFS), overall survival, time to progression and progression of disease within 12 months, but was outperformed by the IPI. For DHL/THL the highest C-indices were observed for MTV/WHO PS. For PMBL the clinical PET score showed the highest C-indices. SDmax improved prognostic predictions in PMBL and tNHL, but not in DHL/THL. Multivariate analysis identified independent predictors of PFS in tNHL, including IPI and SUV_mean.

CONCLUSION: Metabolic risk scores show variable prognostic value across LBCL subtypes. Subtype-specific metabolic models may enhance personalized risk stratification and guide treatment approaches.

PMID:40518458 | DOI:10.1007/s00259-025-07368-y