Integration of [18F]FDG-PET radiomics with liquid biopsy improves outcome prediction in newly diagnosed DLBCL patients

The relapsed/refractory fraction of diffuse large B-cell lymphoma (DLBCL) patients remains an unmet clinical need and is not easily identified a priori. Baseline circulating tumor DNA (ctDNA) levels and PET/CT radiomics allow to identify DLBCL patients at early relapse, but their integration has not been extensively evaluated. On these grounds, a real-life cohort of 120 newly diagnosed DLBCL treated with R-CHOP were included in the present study. Samples of ctDNA were analysed with CAPP-seq, thus obtaining ctDNA levels measurement and LymphGen molecular clusters classification. Using the standardized uptake value (SUV) threshold of 4.0, the following parameters were collected: i) maximum SUV of the hottest lesion (SUVmax), ii) total metabolic tumor volume (tMTV), iii) total tumor lesion glycolysis (tTLG), and iv) distance between the two farthest hypermetabolic lesions (Dmax). The maximally selected rank statistics was used to identify the best cut-offs both for PET/CT parameters and ctDNA levels. tMTV, tTLG and Dmax retained independent prognostic value when adjusted for ctDNA levels and were therefore grouped together in variable named high-risk PET (N=67). By multivariate analysis, ctDNA-high (HR 2.04, 95% CI 1.10-3.78, p=0.024) and high-risk PET (HR 3.84, 95% CI 1.78-8.28, p<0.001) independently predicted progression-free survival (PFS) and were combined into a 2-factor prognostic model (C-indices of 0.712 for PFS and 0.696 for OS). In addition, molecular clustering by capturing high-risk intrinsic DLBCL biological features further improved outcome prediction. Consistently, BN2/EZB/ST2 molecular clusters (HR 0.27, 95% CI 0.12-0.62, p=0.02) maintained an independent association with shorter PFS when adjusted for the variables of the 2-factor prognostic model and were therefore included in a 3-factor prognostic score (C-indices of 0.745 for PFS and 0.746 for OS). The latest 3-factor prognostic model allowed to identify a high risk group of patients (N=22, 40-month PFS of 12.1%) which should be prioritized for an early disease evaluation and for the access to novel agents, such as bispecific antibodies and/or CAR-T cell therapy. The integration of baseline PET/CT variables (tMTV, tTLG and Dmax), ctDNA levels and molecular clusters improved the outcome prediction of newly diagnosed DLBCL compared to PET/CT and ctDNA levels alone.