Radiol Oncol 2025; 59(4): 607-616. doi: 10.2478/raon-2025-0028 607 research article Are there clinically relevant prognostic factors in diffuse large B-cell lymphoma beyond International Prognostic Index? Milica Miljkovic¹,², Vita Setrajcic Dragos²,³, Gorana Gasljevic4,5, Srdjan Novakovic²,³, Lucka Boltezar¹,², Barbara Jezersek Novakovic¹,² 1 Department of Medical Oncology, Institute of Oncology Ljubljana, Ljubljana, Slovenia 2 Faculty of Medicine, University of Ljubljana, Ljubljana, Slovenia 3 Department of Molecular Diagnostics, Institute of Oncology Ljubljana, Ljubljana, Slovenia 4 Department of Pathology, Institute of Oncology Ljubljana, Ljubljana, Slovenia 5 Faculty of Medicine, University of Maribor, Maribor, Slovenia Radiol Oncol 2025; 59(4): 607-616. Received 3 March 2025 Accepted 28 March 2025 Correspondence to: Prof. Barbara Jezeršek Novaković, M.D., Ph.D., Department of Medical Oncology, Institute of Oncology Ljubljana, Zaloška 2, SI-1000 Ljubljana. E-mail: bjezersek@onko-i.si. Disclosure: No potential conflicts of interest were disclosed. This is an open access article distributed under the terms of the CC-BY license (https://creativecommons.org/licenses/by/4.0/). Background. Diffuse large B-cell lymphoma (DLBCL) has variable prognosis, with only 50 to 60% of patients cured by standard first line treatment. Identifying patients unlikely to benefit from standard first line therapy is therefore crucial. Schmitz’s study identified four molecular subtypes of DLBCL with differing prognoses: MCD, BN2, N1, and EZB, with BN2 and EZB showing more favorable outcomes. This study aimed to evaluate the effectiveness of the Archer FusionPlex Lymphoma Assay in identifying the newly defined genetic subtypes of DLBCL, while also exploring the association between immunohistochemical (IHC) and next-generation sequencing (NGS) methods for classifying the cell of origin (COO) and assessing their predictive value for patient survival. Materials and methods. We classified 131 DLBCL patients using Hans algorithm into GCB (germinal center B-cell- like) and ABC (activated B-cell-like) subtypes, and with NGS applying Archer FusionPlex lymphoma assay into ABC, GCB, unclassified, and into Schmitz’s novel genetic subtypes. A mutational analysis of just 7 genes (MYD88L265P, CD79B, EZH2, NOTCH1, NOTCH2, BCL2, and BCL6) was used for genetic classification. Various statistical models were applied to assess survival differences between subtypes. Finally, STRATOS analysis was conducted to validate our preliminary statistical findings. Results. 35.9% of patients were successfully classified into new genetic subtypes, with acceptable consistency be- tween IHC and NGS method for COO determination. However, the new genetic subtype classification by NGS did not correlate with overall survival, nor did the COO classifications by IHC or NGS. The inclusion of these classifications also did not improve the predictive value of models compared to the basic model based on the International Prognostic Index (IPI) only. Conclusions. The Archer FusionPlex Lymphoma assay showed a somewhat lower detection rate of novel genetic subtypes compared to reports based on exome sequencing, yet identified novel genetic subtypes in over one-third of patients. However, an in-depth STRATOS statistical analysis did not confirm its predictive value for DLBCL prognosis, likely due to factors like patient selection and sample size limitations. Key words: diffuse large B-cell lymphoma; next generation sequencing; new genetic types; prognostic factors Introduction Diffuse large B-cell lymphoma (DLBCL) is the most common subtype of non-Hodgkin’s lympho- mas (NHL), accounting for approximately 30% of all NHL.1 It is a heterogeneous disease in terms of clinical presentation, as well as in terms of its biological and pathological features. Diffuse large Radiol Oncol 2025; 59(4): 607-616. Miljkovic M et al. / Diffuse large B-cell lymphoma and International Prognostic Index608 B-cell lymphoma, not otherwise specified (DLBCL, NOS), is the most common subtype, representing 80-85% of all cases.1-3 Still, even the DLBCL, NOS, is not a homogeneous entity and can be subdivided into several morphological, immunohistochemical and molecular subgroups.1-3 The addition of rituximab to standard chemo- therapy CHOP (cyclophosphamide, doxorubicin, vincristine and prednisone) has substantially im- proved the outcomes of patients with the DLBCL, yet to a variable extent among different patients. Only between 50 to 60% of patients will be cured with R-CHOP, while up to 10% will be refractory to this treatment and another 30% will relapse af- ter achieving their first complete remission.4,5 It is therefore important to identify upfront those pa- tients who will not be cured with R-CHOP in order to tailor their individual first line treatment. One of the widely accepted robust prognostic tools to categorize the DLBCL patients in risk groups is the International Prognostic Index (IPI), introduced in the pre-rituximab era and validated also in the rituximab era.6,7 However, treatment outcomes can vary signifi- cantly, even within the same IPI risk group, and the IPI alone is not sufficient for the unequivocal identification of patients who may not be cured with R-CHOP. This outcome can be attributed to the important genetic and molecular heterogene- ity of DLBCL, NOS, highlighting the need for the identification of additional prognostic markers. Gene-expression profiling (GEP) studies have identified different molecular subtypes (“germinal center B-cell-like” – GCB, “activated B-cell-like” – ABC (non-GCB), and non-classified types of DLBCL) related to the cell of origin (COO), which are supposed to be of prognostic significance.8-12 Immunohistochemical (IHC) algorithms, such as the one proposed by Hans et al., have also been introduced as rapid and inexpensive alternatives to GEP that are readily available and have demon- strated reasonable concordance to gene expression profiling.13 Nonetheless, clinical interest extends to other prognostic markers, including the determinants of molecular heterogeneity in DLBCL, NOS, as indicated by the new molecular subtypes intro- duced by Schmitz et al., Chapuy et al., and other authors.14-17 The study of Schmitz et al. identified four molecular subtypes of DLBCL: MCD (based on the co-occurrence of MYD88 and CD79B altera- tions), BN2 (based on BCL6 fusions and NOTCH2 mutation), N1 (based on NOTCH1 mutation), and EZB (based on EZH2 mutation and BCL2 translo- cations), that were determined to be of prognos- tic significance.14 A more favorable prognosis has been predicted for BN2 and EZB in comparison to other two subtypes.14 Chapuy et al., on the other hand, identified five molecular subtypes showing certain overlapping with subtypes identified by Schmitz: C1 (resembling BN2), C2 (TP53 mutation and TP53BP1 alteration), C3 (resembling EZB), C4 (RHOA mutations, TET2, ZFP36L1 alterations) and C5 (resembling MCD). If the genetic driver could not be identified, the DLBCL was categorized as C0.15 In the mentioned studies, in addition to ge- netic changes useful for genetic classification, gene expression signatures related to the tumor micro- environment, rearrangements of BCL2 and MYC and other markers (such as TP53 mutations, high proliferative activity, CD5, and CD30 expression) appear to play an important role in the prognosti- cation process of high grade lymphomas. To the best of our knowledge, there are cur- rently no commercial gene panels available on the market specifically designed to define the genetic subtypes of DLBCL as determined in Schmitz’s or Chapuy’s classification. Our retrospective study aimed to evaluate the effectiveness of the Archer FusionPlex Lymphoma Kit in identifying the new genetic subtypes of DLBCL as defined by Schmitz et al.14 Additionally, we examined the association between the IHC (Hans algorithm) and next generation sequencing (NGS) methods for classifying the COO and as- sessed their ability to predict survival. Patients and methods Patients One hundred and thirty-one patients with DLBCL, NOS, were enrolled in this retrospective clinical study. The inclusion criteria were as followed: all patients were older than 18 years, were diagnosed with DLBCL at least 5 years prior to beginning of this study, and were (except of one patient) treated with R-CHOP/RCHOP-like therapy between 2011 and 2017 at the Institute of Oncology Ljubljana (OIL), Slovenia. This study included only patients with DLBCL, NOS, and excluded patients with tes- ticular lymphoma, primary central nervous system lymphoma or plasmablastic lymphoma. Patients with HIV positive lymphomas were also excluded from this study. All clinical data were obtained from medical records available in hospital’s infor- mation system (patients’ age at diagnosis, clinical stage of the disease, data for IPI score, treatment Radiol Oncol 2025; 59(4): 607-616. Miljkovic M et al. / Diffuse large B-cell lymphoma and International Prognostic Index 609 protocols applied and number of treatment cycles, treatment outcomes - overall response rate [ORR], progression-free survival [PFS], and overall sur- vival [OS]). Survival data were retrieved from the Cancer Registry of the Republic of Slovenia and survival status was censored for all patients on 20th of June, 2023. The study was approved by the National Medical Ethics Committee of the Republic of Slovenia (Approval Number 0120-103/2020/4) and by Institutional Review Board (Approval number KSOPKR-0012/2020) as well as the Institutional Medical Ethics Committee (Approval number EK- 0120-103/2020/4). The requirement for individual informed consent was waived, as this was a ret- rospective database analysis. Additionally, the institutional informed consent form for treatment included permission to use patients’ data, materi- als, and/or test results for research purposes. The study was conducted according to the Declaration of Helsinki. Pathological examination For all patients included in the study, paraffin blocks and corresponding hematoxylin and eosin- stained slides were retrieved from the archive of Department of Pathology of the Institute of Oncology Ljubljana, and diagnoses were reviewed. The tissue microarrays (TMA) were constructed and IHC staining as well as interpretations were performed as already described by Boltezar et al.18 to classify patients according to the Hans algo- rithm into the GCB and ABC (non-GC) types.13 At the same time, material for genetic analysis was cut from each paraffine block. The review, TMA evaluations and the classification of patients ac- cording to the Hans algorithm, were performed by a skilled hematopathologist who was blinded to all clinical data. Next generation sequencing (NGS) The Archer FusionPlex Lymphoma kit was select- ed for the NGS procedure due to its commercial availability and its targeting of 125 lymphoma- related genes, which we considered potentially advantageous for classifying samples into novel genetic subtypes. RNA was isolated using the MagMAXTM FFPE DNA/RNA Ultra Kit (ThermoFisher, Waltham, MA, USA). A total of 250 ng of RNA was reverse tran- scribed into cDNA, and NGS was performed using the Archer FusionPlex Lymphoma Kit, following the manufacturer’s protocol (Invitae ArcherDX, San Francisco, CA, USA). The quality of the start- ing material was evaluated by assessing the quality of the cDNA synthesized from the RNA. For this purpose, the Archer PreSeq RNA QC assay was employed (InvitaeArcherDX, San Francisco, CA, USA). The library was quantified using the qPCR Library Quantification Kit (KAPA Biosystems, Wilmington, MA, USA) and sequenced on the MiSeqDx system (Illumina, San Diego, CA, USA). Data were analyzed using the Archer Analysis ver- sion 6.0.3.2. A genetic variant was considered true positive if the allele fraction was at least 10% and the coverage depth was at least 100x. Fusions were considered true positive if covered by five or more unique reads and represented over 10% of reads. Variants listed in the GnomAD database were ex- cluded as germline. Only previously identified pathogenic variants were used for patient sub- grouping. Variants were considered pathogenic if listed in Schmitz et al.14 Supplementary Table or the OncoKB database as oncogenic.19 Variants of un- certain significance and benign variants were ex- cluded. CD79B gene amplification was considered true positive when its relative expression exceeded 8 on a 0-9 scale calculated by the Archer analysis software. Cases were classified by gene expres- sion patterns into the ABC, GCB, and unclassified subgroups according to the COO classification. We used a simplified approach of mutational analysis of just 7 genes, as alterations in MYD88L265P, CD79B, EZH2, NOTCH1, NOTCH2, BCL2, and BCL6 were employed for genetic classification into novel sub- types. This decision was based on literature data indicating that hallmark genetic alterations for each subtype include MYD88 (66.2% prevalence) and CD79B (50.0% prevalence) for the MCD sub- type; EZH2 (44.7% prevalence) and/or BCL2 (68.4% prevalence) for the EZB subtype; BCL6 (72.8% prev- alence) and/or NOTCH2 (41.8% prevalence) for the BN2 subtype; and NOTCH1 (100% prevalence) for the N1 subtype. Other alterations used by Chapuy and Schmitz to define specific genetic subtypes were mostly reported with a lower prevalence.8,14,15 Cases with CD79B and MYD88 alterations were therefore classified as “MCD,” with EZH2 and/ or BCL2 as “EZB,” with BCL6 and/or NOTCH2 as “BN2” and with NOTCH1 as “N1.” Remaining cas- es were genetically unclassified.14 Statistical analysis The median age, stage at the time of diagno- sis, bone marrow infiltration, IPI score, number Radiol Oncol 2025; 59(4): 607-616. Miljkovic M et al. / Diffuse large B-cell lymphoma and International Prognostic Index610 and complete response (CR). OS was defined as the time from the date of diagnosis to the time of death from any cause. The PFS and OS were esti- mated using the Kaplan-Meier method and differ- ences were compared using the log-rank test. The IPI score was used as a categorical variable: low (score of 0 or 1), low-intermediate (score 2), high- intermediate (score 3) and high-risk group (score of 4 or 5) for the purpose of survival analyses and as a numeric variable for STRATOS initiative anal- yses. P value ≤0.05 was considered to indicate a sta- tistically significant difference. GraphPad Prism software 9.0.0 (GraphPadSoftware, Boston, MA, USA) was used for analyses. Additional statistical analysis was performed – namely the STRATOS analysis, to check our basic statistic. The analysis was run in R (v4.3.2) in RStudio (v2023.12.1+402) using the packages boot (v1.3.28.1), dplyr (v1.1.4), forcats (v1.0.0.), ggplot2 (v3.4.4), ggpubr (v0.6.0), gridExtra(v2.3), gtsummary(v1.7.2), Hmisc (v5.1.1.), kableExtra (v1.3.4), knitr (v1.45), lubridate (v1.9.3), readxl (1.4.3), pacman (v0.5.1), purr (v1.0.2), readr (v2.1.4), readaxl (v1.4.3), rio (v1.0.1), rms (v6.7.1), rsample (v1.2.0), stringr (v1.5.1), survival (v3.5.7) and survminer (v0.4.9), tibble (v3.2.1), tidyr (v1.3.0), tidyverse (v2.0.0), timeROC (v0.4), webshot (v0.5.5) and their dependencies. According to the guide- lines of the STRATOS initiative, we also compared the calibration, discrimination, Brier scores and clinical utility of various tested Cox models in or- der to test our model more profoundly.20-22 Lacking an additional dataset for external validation, we validated our models using optimism-corrected internal validation with bootstrapping. Results Demographic data A total of 131 patients with DLBCL were included in the study. The patients’ characteristics are sum- marized in Table 1. There was a slight female pre- dominance, median age was 65 years (range 28-89). More than half of our group had stage IV disease (50.4%) and elevated serum lactate dehydrogenase level (LDH) (58.8%), while nearly half had B symp- toms (49.6%). The highest percentage of patients were in the high-intermediate risk group (28.2% of patients), other three risk groups were quite evenly distributed. All patients except one (130 patients - 99.2%) received first line systemic treat- ment. Treatment regimen was R-CHOP (rituximab, cyclophosphamide, doxorubicin, vincristine and prednisolone) +/- middle dose of methotrexate (500 TABLE 1. Patients’ characteristics N % Number of patients: 131 Gender Male 59 45.0% Female 72 55.0% Age Age range 28−89 / Median age 65 / Stages (Ann Arbor): Stage I 19 14.5% Stage II 25 19.0% Stage III 21 16.0% Stage IV 66 50.4% Median stage: 4 (range 1−4) Other characteristics Bone marrow involvement 41 31.3% Elevated LDH level 77 58.8% B symptoms 65 49.6% IPI group: Low risk 32 24.4% Low-intermediate risk 31 23.7% High-intermediate risk 37 28.2% High risk 31 23.7% Median IPI value: 3 (range 0-5) Treatment R-CHOP/R-CHOP like 128 97.7% R-COEP 2 1.5% Palliative care 1 0.8% Treatment response 130 CR 65 50.0% PR 48 36.9% SD 1 0.8% PD 16 12.3% CR = complete response; IPI = International Prognostic Index; PD =progressive disease; PR = partial response; R-CHOP = rituximab, cyclophosphamide, doxorubicin, vincristine and prednisone; R-CHOP like = R-CHOP +/- middle dose methotrexate; R-COEP = rituximab, cyclophosphamide, etoposide, vincristine and prednisone; SD = stable disease of treatment cycles, subtype by IHC (ABC and GCB), subtype by NGS (ABC and GCB and non- classified) and new genetic subtypes according to Schmitz’s classification were determined. PFS was defined as the time from the end of first sys- temic treatment to disease progression or death from any cause for patients achieving partial (PR) Radiol Oncol 2025; 59(4): 607-616. Miljkovic M et al. / Diffuse large B-cell lymphoma and International Prognostic Index 611 mg/m²) in 128 patients (97.7%). Two patients (1.5%) received R-COEP (rituximab, cyclophosphamide, vincristine, etoposide and prednisolone) therapy due to their cardiac conditions, while one patient underwent just palliative treatment (and was ex- cluded from cohort analysis). The median follow- up time was 61 months (range 2–152 months). For each patient, response to first line treatment was defined as CR, PR, stable disease (SD) or pro- gressive disease (PD) based on the revised criteria of Cheson et al.23 The ORR for the entire group was 86.9% (with the CR at 50% and the PR at 36.9%). Classification of diffuse large B-cell lymphoma (DLBCL) according to the cell of origin (COO) According to the COO determined by IHC method, 54 patients (41.2%) were classified as ABC subtype and 77 patients (58.8%) as GCB subtype. According to the NGS, 42 patients (32.0%) were classified as ABC subtype and 62 patients (47.3%) as GCB sub- type, while 12 patients (9.2%) remained unclassi- fied, and for 15 patients (11.5%) the NGS method could not provide a clear result (QC failed). NGS served as the reference method and concordance between the determined COO subtypes accord- ing to the IHC and the NGS method is presented in Table 2. The overall concordance between NGS determined COO and IHC determined COO in the whole study group was 61.8% (81 of total 131 pa- tients). Genetic classification of diffuse large B-cell lymphoma (DLBCL) by next generation sequencing (NGS) according to the Schmitz’s classification Of the entire group of 131 patients, 47 patients (35.9%) were successfully categorized into one of the new genetic subtypes of DLBCL, 70 patients remained unclassified (53.4%), while 14 patients (10.7%) were categorized as QC failed. Among those 47 patients, 17 patients had MCD (13%), none had N1, 12 had BN2 (9.2%), and 18 patients had EZB (13.7%) subtype. New genetic subtypes in the con- text of COO groups ABC, GCB and unclassified group are presented in Figures 1 and 2. When focusing only on the QC failed (as deter- mined by NGS) group of 15 patients of the entire 131 patient cohort, 1 patient was categorized with the EZB subtype (6.7%). The remaining 14 patients remained unclassified to the new genetic sub- types. Overall response rate according to the immunohistochemical (IHC) and next generation sequencing (NGS) determination of cell of origin (COO) and NGS determination of new genetic subtypes According to the IHC classification of COO into ABC and GCB subtype, the ORR for ABC subtype was 84.9 % and 88.3% for the GCB subtype. Based on the NGS classification of COO, the ORR for the ABC subtype was 82.9%, 88.7% for the GCB sub- type and 91.7 % for the unclassified group. The ORR for MCD genetic subtype was 76.5%, for BN2 genetic subtype 91.7%, for EZB genetic subtype 94.4%, for unclassified group 87.1%, and for QC failed group 85.7%. There was no association be- tween the ORR and classification to the abovemen- tioned subgroups regarding COO by IHC, COO by NGS and classification to new genetic subtypes by NGS (p = 0.59, p = 0.76, and p = 0.80, respectively). TABLE 2. Concordance between the IHC and NGS method in determination of the COO. The reference method was NGS. A pairwise comparison between the results of both methods was performed for each patient. Each patient who was subclassified into the same (ABC or GCB) subtype by both NGS and IHC was considered concordant. Patients who were subclassified differently by IHC and NGS were considered discordant COO by NGS COO by IHC Concordance (%) ABC subtype 42 31 73.8 GCB subtype 62 50 80.6 ABC = activated B-cell like; COO = cell of origin; GCB = germinal center B-cell like; IHC = immunohistochemical determination; NGS = next generation sequencing FIGURE 1. New genetic subtypes in the context of COO groups ABC and GCB, as determined by NGS (N1 group is not included in the Figure since there were no patients with this subtype). ABC = activated B-cell; COO = cell of origin; GCB = germinal center B-cell; NGS = next generation sequencing Radiol Oncol 2025; 59(4): 607-616. Miljkovic M et al. / Diffuse large B-cell lymphoma and International Prognostic Index612 Overall survival according to the immunohistochemical (IHC) and next generation sequencing (NGS) determination of cell of origin (COO) and NGS determination of new genetic subtypes, and according to the International Prognostic Index (IPI) Five-year overall survival (OS) of the entire group was 67.8% (Figure 3). According to the IHC clas- sification of COO into ABC and GCB subtype, the 5-year OS for ABC subtype was 62.5% and 71.4% for GCB subtype. Survival was not significantly different between the two groups, (p = 0.27, HR = 1.36 [95% CI 0.76–2.42]) Supplementary Figure 1. Based on the NGS classification, the five-year OS for ABC subtype was 54.8% and 74.2% for GCB subtype, 64.2 % for the unclassified group and 80% for the QC failed group, and, again, there was no statistically significant difference between groups (p = 0.06) – Supplementary Figure 2. The 5-year OS for patients diagnosed with the new genetic subtypes according to the Schmitz’s clas- sification was 66.7% for the BN2 subtype, 77.8% for the EZB subtype, 64.7% for the MCD subtype, 78.6% in the QC failed group, and 63.9% for the unclassified group (p = 0.61) - Supplementary Figure 3. The 5-year OS was 87.5% in low risk IPI group, 87.0% in low-intermediate risk IPI group, 70.3% in high-intermediate risk IPI group and 25.8% in high risk IPI group (p < 0.0001, HR = 0.117 [95% CI 0.06–0.245]) (Figure 4). Cox models We are not giving the PFS data as we consider them to be only of predictive and not of a prog- nostic significance. Even though we are aware of the potential influence of subsequent therapies to the OS, we chose to evaluate the prognostic significance of the three classifications that had been applied in the study – namely the COO by IHC classification, COO by NGS classification and the new genetic types by NGS classification. We used the Cox proportional model to investigate the association between the OS and the particu- lar classification (COO by IHC, COO by NGS, and new genetic types by NGS, respectively) as well as the IPI score. The following Tables summarize the four models used - Supplementary Tables 1, 2 and 3. The base model uses only IPI as an independ- ent variable and the three classifications available to IPI as a second independent variable. However, when combining IPI score and any of classifica- tions (COO by IHC, COO by NGS, and new ge- netic types by NGS) into one model, none of the classifications had a significant prognostic impact on patients’ survival and only IPI remained prog- nostic for OS. So, regardless of the classification used, the survival of patients was not statistically significantly different between the ABC and GCB subtype (and unclassified subtype) or between the new genetic subtypes in our data set. STRATOS statistical analyses The STRATOS initiative produced guidelines for comparing Cox models. For technical reasons, IPI is included as a numerical variable (i.e. the number of risk factors present). As shown previously, re- gardless of the model/classification used, only the IPI remains statistically significantly associated with survival. Given that each individual classification (COO by IHC, COO by NGS, and new genetic types by NGS) was not associated with survival, we ex- pected the subsequent analysis to show that inclu- sion of any one of the classifications into the model would not improve the quality of the model or its clinical usefulness. Discrimination and calibration Corrected internal discriminations of the com- pared models at a fixed time point 5 years after diagnosis are given in Supplementary Table 4 and calibrations at a fixed time point 5 years after di- agnosis are given in Supplementary Table 5. The addition of any classification did not improve the quality of the model, which was expected, given that none of them was significantly associated with overall survival. FIGURE 2. New genetic subtypes in the context of COO unclassified group, as determined by NGS (N1 group is not included in the Figure since there were no patients with this subtype). COO = cell of origin; NGS = next generation sequencing Radiol Oncol 2025; 59(4): 607-616. Miljkovic M et al. / Diffuse large B-cell lymphoma and International Prognostic Index 613 Overall quality The Brier score and the scaled Brier score (Index of Prediction Accuracy, IPA) estimate the overall model quality. Brier score estimates the average difference between predicted and observed val- ues at time t (and a lower value indicates a better model). IPA improves the interpretability of the score and estimates the reduction in Brier score when using a more complex model compared to the base model (a higher value therefore indicates a better model). The Brier score and IPA are given in Supplementary Table 6. We can conclude that the additional information about the classifica- tion does not improve the model quality (the ad- dition of IPI compared to the null model improves the Brier score by 20%, while the extended models (with COO by IHC, COO by NGS and new genetic subtypes by NGS) improve it by 17% to 19%). Calibration, discrimination and overall qual- ity were assessed using bootstrap (B = 1000) and optimism corrected internal validation, since an additional data source for external validation was not available. As expected, we demonstrated that the addition of any one of the available classifica- tions did not improve the performance of the base model that only included the IPI score. Clinical usefulness Discrimination and calibration are necessary to assess the model quality or to compare different models, but they are not enough to evaluate their clinical usefulness. When an additional marker is available (in this case the subtype classification, whether histological or genetic), the evaluation of clinical usefulness should tell us whether the use of the additional marker leads to improved clinical decision making. We assessed the clini- cal usefulness of the extended models (with COO by IHC, COO by NGS and new genetic subtypes by NGS) compared to the base model at various possible risk thresholds. We have shown that the extended models do not improve the net benefit to patients when used in clinical decision making, regardless of the threshold or extended model chosen (Supplementary Figure 4, Supplementary Table 7). Discussion DLBCL is highly heterogeneous in its genetic char- acteristics, making accurate sub-classification of patients based on tumor genetic changes essen- tial for optimal treatment approaches. The clas- sifications proposed by Schmitz and Chapuy are currently the most effective in grouping patients according to disease prognosis and treatment out- comes.14,15 However, no commercial tool (gene pan- el) is at present available that could classify patients into the genetic groups proposed by these authors. Therefore, the aim of our study was to evaluate the utility of the Archer FusionPlex Lymphoma Kit in classifying DLBCL into the groups outlined FIGURE 3. Overall survival (Kaplan-Meier) of all patients (N = 131). FIGURE 4. Overall survival (Kaplan-Meier) of patients according to International Prognostic Index (IPI) risk groups; (p < 0.0001). Radiol Oncol 2025; 59(4): 607-616. Miljkovic M et al. / Diffuse large B-cell lymphoma and International Prognostic Index614 by Schmitz et al.14 Furthermore, we specifically as- sessed the impact of IHC classification of COO, NGS classification of COO, and NGS classification of new genetic subtypes based on Schmitz’s pro- posal on overall survival in patients with diffuse large B-cell lymphomas. In fact, similar approaches for the genetic cat- egorization of lymphomas have been used by oth- er authors. Crotty et al. conducted a smaller study involving 41 DLBCL patients using the Archer FusionPlex Lymphoma platform to analyze a pan- el of 125 lymphoma-related genes and evaluate its concordance with the IHC Hans algorithm, ob- serving an 80.5% concordance for COO.24 Another group, led by Scott, used the Lymph2Cx assay, to test 20 genes and compare it to GEP by NanoString to define the COO. This assay provided con- cordant COO definitions in 96% of their cases.25 Multivariate analyses in their report showed that COO defined by Lymph2Cx was independently prognostic of survival, regardless of the IPI score.25 However, in the present study, the overall con- cordance between NGS determined COO and IHC determined COO was lower, at only 61.8%, likely because we did not exclude the QC-failed samples and on account of the NGS determined unclassi- fied group. Specifically, we observed a 73.8% con- cordance in the ABC group and 80.6% in the GCB group. In contrast, the study by Crotty et al. includ- ed far fewer patients (41) compared to ours (131), and while the proportions of ABC, GCB, and un- classified subgroups in Crotty’s study were evenly distributed, similar to our findings, they did not report any failed results.24 Regarding Schmitz’s proposed classification of the new genetic subtypes of DLBCL, we were able to subclassify 35.9% of our patients using the Archer FusionPlex Lymphoma kit. This rep- resents a lower proportion of classified cases compared to Schmitz’s study, where 46.6% of pa- tients were categorized into the new genetic sub- types.14 Considering the fact that Schmitz’s study performed exome and transcriptome sequenc- ing, while we conducted a limited panel of RNA sequencing, the difference in the proportion of successfully classified cases is relatively small. Our cohort included a larger number of samples with mutations that were not helpful to classify correctly patients into subgroups, suggested by Schmitz. The clinical impact of these mutations is, nevertheless, still unknown. Still, in other stud- ies using a classification similar to Schmitz’s, the proportion of successfully classified patients was also less than 100%. Lacey performed a whole-ex- ome sequencing on tumor samples of 928 patients (including primary central nervous system lym- phomas and plasmablastic lymphomas), in a pan- hematological malignancy panel of 293 genes, and found some overlapping groups over Schmitz’s and Chapuy’s classification. They identified 5 genomic clusters and had a 27% rate of “unclassi- fied cases”.16 Wright and his colleagues created the LymphGen algorithm that provided a probabilis- tic classification of the tumor from an individual patient into a genetic subtype and with a similar methodology as Schmitz they managed to subclas- sify 63.1% of patients.17 A closer comparison of our results with those of other authors indicates that with our simplified approach, we detected a relatively low proportion of the EZB subtype – namely, 21.8% in Schmitz’s study, 18.9% in Lacey’s study, and only 13.7% in our study.14,16 The EZB subgroup is primarily included within the GCB cases. Since the GCB subgroup is more prevalent in our study (58.8% of GCB by IHC and 47.3% by NGS) compared to Schmitz’s study (28.2% of GCB cases), the relatively low EZB detec- tion rate in our study remains unexplained.14 Schmitz et al. reported the predicted 5-year over- all survival rates for the MCD, BN2 and EZB sub- types of 26%, 65%, and 68%, respectively, while in our study they were numerically superior - 64.7%, 66.7%, and 77.8%.14 Lacey’s genomic cluster MYD88, which overlaps with the MCD subtype, showed a 5-year OS of 62.8% in R-CHOP treated population.16 Their MYD88 group included testicular and prima- ry CNS lymphomas, while in our study those pa- tients were not included. Lacey’s BCL-2 cluster that overlaps with EZB subtype, and whose 5-year OS of 69.5% adjusts with the one reported in Schmitz’s study, was, compared to the EZB survival of our group, inferior.16 But, as stated previously, the EZB group was relatively weakly represented in our study in comparison with other studies. Furthermore, Schmitz’s study investigated the survival data of only 119 patients (treated with R-CHOP/CHOP like chemotherapy) diagnosed with new genetic subtypes out of all 257 patients classified into novel subgroups, so their survival data deficiently cover only half of the new genetic subtypes’ population.14 To the contrary, our study reports survival data of all included patients. In Lacey’s study, only two thirds of patients were treated with R-CHOP, however, they reported re- sults of survival for patients treated with R-CHOP separately.16 In some of the studies, genetic sub- classification essentially had a prognostic impact on survival of their patients.14,16,17 Finally, Zhang Radiol Oncol 2025; 59(4): 607-616. Miljkovic M et al. / Diffuse large B-cell lymphoma and International Prognostic Index 615 et al. conducted a randomized phase II trial of addition of a targeted therapy to R-CHOP in pa- tients with DLBCL, driven after the first cycle of R-CHOP by newly determined genetic subtypes. Their study was not powered to show survival differences, but it did meet its primary endpoint by achieving higher complete response rates with novel therapeutic approaches. This indicates that the spectrum of possible future decisions in choos- ing of an optimal first line therapy might have to be based on gene expression analyses.26 The classification of the new genetic types by NGS used in our study, however, was not associat- ed with overall survival, as were also not the other two classifications of COO determined by IHC or by NGS. Similarly, the inclusion of any of the three classifications (COO by IHC, COO by NGS and new genetic types by NGS according to Schmitz’s proposal) improved neither the calibration and the discrimination nor the clinical utility of the tested models, when compared to the basic model includ- ing only IPI values. One of the strengths of our study is the use of advanced statistical methods,20,21,22 as well as the thorough histopathological evaluation of all samples by a skilled hematopathologist who was blinded to the clinical data. The STRATOS analysis of Cox regression, a novel and advanced statisti- cal method to analyse the potential differences be- tween classifications, has, to the best of our knowl- edge, never been done in the setting of the DLBCL. This advanced methodology disclosed no differ- ence in survival regardless of the classification (COO by IHC, COO by NGS, and new genetic types by NGS) used. Based on the data and subanalysis of this study, the only factor with valid prognostic significance for overall survival was IPI, which re- mained significant regardless of the classification method applied (COO by IHC, COO by NGS, or new genetic subtypes by NGS per Schmitz’s pro- posal). Schmitz et al. and Chapuy et al. also showed IPI’s prognostic significance for overall survival in a multivariate model.14,15 However, their stud- ies reported also significantly different survival outcomes based on COO subtypes, which was not confirmed in our study. Additionally, the prognos- tic value of COO subtyping has been questioned in a Hungarian study of 247 DLBCL patients, where the COO subtype failed to predict prognosis.27 Thus, the prognostic impact of COO subtypes ap- pears more complex than initially suggested by Hans et al.13 The disadvantages of this study are its retrospec- tive nature, and when compared to the Schmitz’s, Chapuy’s and Lacey’s study, a smaller number of patients included (they included 574, 304 and 928 patients, respectively).14-16 Yet, the number is still higher than in Crotty’s and Scott’s study.24,25 Another limitation of this study is the relatively high number of samples that failed quality con- trol (QC failed category). The sequencing quality control likely failed for several reasons. The most common issue was the contamination of sample with DNA, as indicated by an imbalanced ratio of RNA to DNA reads. In a few cases, the final library concentration was low, leading to insufficient cov- erage for meaningful analysis. On the other hand, the limited number of just 7 genes analyzed in our study can be either regarded as a limitation of the study on classification capabilities into novel ge- netic subtypes due to the reduced impact of altera- tions in other genes (mutated in lymphoma) or as the strength of the study by offering a simplified approach to this classification in clinical practice. Conclusions The Archer FusionPlex Lymphoma assay tested in our study showed a somewhat lower detec- tion rate of novel genetic subtypes compared to reports based on exome sequencing. An in-depth statistical analysis of patients’ survival across the groups defined by our approach did not confirm its value in predicting outcome of DLBCL patients. However, the difference in proportion of success- fully categorized patients within novel genetic subgroups, as proposed by Schmitz et al., with Archer’s FusionPlex Lymphoma assay compared to exome sequencing was relatively small, making our simplified approach to classifying of DLBCL patients potentially useful in everyday practice. Acknowledgments This study was partially supported by the Slovenian Research Agency [grant number P3- 0321]. We would also like to extend our gratitude to Mrs. Violet Ruparchic for her generous donation to the Institute of Oncology Ljubljana, which par- tially funded the next generation sequencing kits used in this study. Technical assistance with genetic analysis of Mrs. Simona Traven is acknowledged, as well as assistance with STRATOS analysis of Assist. Maša Kušar, MD, PhD. Radiol Oncol 2025; 59(4): 607-616. Miljkovic M et al. / Diffuse large B-cell lymphoma and International Prognostic Index616 References 1. Tilly H,da Silva GM, Vitolo U, Jack A, Meignan M, Lopez-Guillermo A, et al. Diffuse large B-cell lymphoma: ESMO Clinical Practice Guidelines. Ann Oncol 2015; 26: 116-25. doi: 10.1093/annonc/mdv304 2. National Comprehensive Cancer Network. Clinical Practice Guidelines in Oncology (2025). B-cell lymphoma (version 1.2025) [Internet]. [cited 2025 Feb 15]. Available at: https://www.nccn.org/professionals/physician_gls/ pdf/b-cell.pdf 3. Khoury JD, Solary E, Abla O, Akkari Y, Alaggio R, Apperley JF, et al. The 5th edition of the World Health Organization Classification of Haematolymphoid Tumours: myeloid and histiocytic/dendritic neoplasms. Leukemia 2022; 36: 1703-19. doi: 10.1038/s41375-022-01613-1 4. Sarkozy C, Sehn LH. New drugs for the management of relapsed or re- fractory diffuse large B-cell lymphoma. Ann Lymphoma 2019; 3: 10. doi: 10.21037/aol.2019.09.01 5. Crump M, Neelapu SS, Farooq U, Van Den Neste E, Kuruvilla J, Westin J, et al. Outcomes in refractory diffuse large B-cell lymphoma: results from the international SCHOLAR-1 study. Blood 2017; 130: 1800-8. doi: 10.1182/ blood-2017-03-769620 6. International Non-Hodgkin’s Lymphoma Prognostic Factors Project. A predictive model for aggressive non-Hodgkin’s lymphoma. New Engl J Med 1993; 329: 987-94. doi: 10.1056/NEJM199309303291402 7. Ziepert M, Hasenclever D, Kuhnt E, Glass B, Schmitz N, Pfreundschuh M, et. al. International prognostic index remains a valid predictor of outcome for patients with aggressive CD20+ B-cell lymphoma in the rituximab era. J Clin Oncol 2010; 28: 2373-80. doi: 10.1200/JCO.2009.26.2493 8. Shimikus G, Nonaka T. Molecular classification and therapeutics in diffuse large B-cell lymphoma. Front Mol Biosci 2023; 10: 1124360. doi: 10.3389/ fmolb.2023.1124360 9. Weber T, Schmitz R. Molecular subgroups of diffuse large b cell lymphoma: biology and implications for clinical practice. Curr Oncol Rep 2022; 24: 13- 21. doi: 10.1007/s11912-021-01155-2 10. Alizadeh AA, Eisen MB, Davis RE, Ma C, Lossos IS, Rosenwald A, et al. Distinct types of diffuse large B-cell lymphoma identified by gene expression profil- ing. Nature 2000; 403: 503-11. doi: 10.1038/35000501 11. Rosenwald A, Wright G, Chan WC, Connors JM, Campo E, Fisher RI, et al. The use of molecular profiling to predict survival after chemotherapy for diffuse large-B-cell lymphoma. N Engl J Med 2002; 346: 1937-47. doi: 10.1056/ NEJMoa012914 12. López AB, Villambrosía GS, Francisco M, Malatxeberria S, Sáez A, Montalban C, et al. Stratifying diffuse large B-cell lymphoma patients treated with chemoimmunotherapy: GCB/non-GCB by immunohistochemistry is still a robust and feasible marker. Oncotarget 2016; 7: 18036-49. doi: 10.18632/ oncotarget.7495 13. Hans CHP, Weisenburger DD, Greiner TC, Gascoyne RD, Delabie J, Ott G, et al. Confirmation of the molecular classification of diffuse large B-cell lym- phoma by immunohistochemistry using a tissue microarray. Blood 2004; 103: 275-82. doi: 10.1182/blood-2003-05-1545 14. Schmitz R, Wright GW, Huang DW, Johnson CA, Phelan JD, Wang JQ, et al. Genetics and pathogenesis of diffusse large B-cell lymphoma. N Engl J Med 2018; 378: 1396-1407. doi: 10.1056/NEJMoa1801445 15. Chapuy B, Stewart CH, Dunford AJ, Kim J, Kamburov A, Redd RA, et al. Molecular subtypes of diffuse large B cell lymphoma are associated with distinct pathogenic mechanisms and outcomes. Nature Med 2018; 24: 679- 90. doi: 10.1038/s41591-018-0016-8 16. Lacy SE, Barrans SHL, Beer PHA, Painter D, Smith AG, Roman E, et al. Targeted sequencing in DLBCL, molecular subtypes, and outcomes: a Haematological Malignancy Research Network report. Blood 2020; 135: 1759-71. doi: 10.1182/blood.2019003535 17. Wright GW, Huang DW, Phelan JD, Coulibaly ZA, Roulland S, Young RM, et al. A Probabilistic classification tool for genetic subtypes of diffuse large b cell lymphoma with therapeutic implications. Cancer Cell 2020; 37: 551-68.e14. doi: 10.1016/j.ccell.2020.03.015 18. Boltežar L, Kloboves Prevodnik V, Pohar Perme M, Gašljević G, Jezeršek Novaković B. Comparison of the algorithms classifying the ABC and GCB subtypes in diffuse large B-cell lymphoma. Oncol Lett 2018; 15: 6903-12 19. Suehnholz SP, Nissan MH, Zhang H, Kundra R, Nandakumar S, Lu C, et al. Quantifying the expanding landscape of clinical actionability for patients with cancer. Cancer Discov 2024; 14: 49-65. doi: 10.1158/2159-8290.CD- 23-0467 20. Crowson CS, Atkinson EJ, Therneau TM. Assessing calibration of prog- nostic risk scores. Stat Methods Med Res 2016; 25: 1692-706. doi: 10.1177/0962280213497434 21. Austin PC, Harrell Jr FE, Klaveren D. Graphical calibration curves and the integrated calibration index (ICI) for survival models. Stat Med 2020; 39: 2714-42. doi: 10.1002/sim.8570 22. McLernon DJ, Giardiello D, Calster BV, Wynants L, Geloven N, Smeden M, et al. Assessing performance and clinical usefulness in prediction models with survival outcomes: Practical Guidance for Cox Proportional Hazards Models. Ann Intern Med 2023; 176: 105-14. doi: 10.7326/M22-0844 23. Cheson BD, Pfistner B, Juweid ME, Gascoyne RD, Specht L, Horning SJ, et al. Revised response criteria for malignant lymphoma. J Clin Oncol 2007; 25: 579-86. doi: 10.1200/JCO.2006.09.2403 24. Crotty R, Hu K, Stevenson K, Maggie Y, Pontius MY, Sohani AR, et al. Simultaneous identification of cell of origin, translocations, and hotspot mutations in diffuse large B-cell lymphoma using a single RNA-sequencing assay. Am J Clin Pathol 2021; 155: 748-54. doi: 10.1093/ajcp/aqaa185 25. Scott DW, Mottok A, Ennishi D, Wright GW, Farinha P, Ben-Neriah S, et al. Prognostic significance of diffuse large B-cell lymphoma cell of origin deter- mined by digital gene expression in formalin-fixed paraffin-embedded tissue biopsies. J Clin Oncol 2015; 33: 2848-56. doi: 10.1200/JCO.2014.60.2383 26. Zhang MCH, Tian SH, Fu D, Wang L, Cheng SH, MeiYi H, et al. Genetic subtype-guided immunochemotherapy in diffuse large B cell lymphoma: the randomized GUIDANCE-01 trial. Cancer Cell 2023; 41: 1705-16.e5. doi: 10.1016/j.ccell.2023.09.004 27. Baliko A, Szakacs Z, Kajtar B, Ritter Z, Gyenesei A, Farkas N, et al. Clinicopathological analysis of diffuse large B-cell lymphoma using mo- lecular biomarkers: a retrospective analysis from 7 Hungarian centers. Front Oncol 2023; 3: 1224733. doi: 10.3389/fonc.2023.1224733