A nomogram for predicting the rapid progression of diffuse large B‐cell lymphoma established by combining baseline PET/CT total metabolic tumor volume, lesion diffusion, and TP53 mutations

Abstract Objectives This study aimed to integrate positron emission tomography/computed tomography (PET/CT) metrics and genetic mutations to optimize the risk stratification for diffuse large B‐cell lymphoma (DLBCL) patients. Methods The data of 94 primary DLBCL patients with baseline PET/CT examination completed in the Shandong Cancer Hospital and Institute (Jinan, China) were analyzed to establish a training cohort. An independent cohort of 45 DLBCL patients with baseline PET/CT examination from other hospitals was established for external validation. The baseline total metabolic tumor volume (TMTV) and the largest distance between two lesions (Dmax) standardized by patient body surface area (SDmax) were calculated. The pretreatment pathological tissues of all patients were sequenced by a lymphopanel including 43 genes. Results The optimal TMTV cutoff was 285.3 cm3 and the optimal SDmax cutoff was 0.135 m−1. TP53 status was found as an independent predictive factor significantly affecting complete remission (p = 0.001). TMTV, SDmax, and TP53 status were the main factors of the nomogram and could stratify the patients into four distinct subgroups based on their predicted progression‐free survival (PFS). The calibration curve demonstrated satisfactory agreement between the predicted and actual 1‐year PFS of the patients. The receiver operating characteristic curves showed this nomogram based on PET/CT metrics and TP53 mutations had a better predictive ability than the clinic risk scores. Similar results were identified upon external validation. Conclusions The nomogram based on imaging factors and TP53 mutations could lead to a more accurate selection of DLBCL patients with rapid progression, to increase tailor therapy.


| INTRODUCTION
Diffuse large B-cell lymphoma (DLBCL) is the most prevalent type of non-Hodgkin's lymphoma, accounting for 40% of all lymphomas.It is a highly heterogeneous disease entity with differing prognoses. 1Almost 40% of patients experience relapse/metastasis following first-line standard treatment, and survival is particularly poor for patients relapsing within 1 year after R-CHOP, with <15% achieving durable remission. 2,38][9] Therefore, new prognostic models are needed as a benchmark for determining the prognosis and guiding novel treatment regimens for DLBCL patients.
Fluorodeoxyglucose ( 18 F)-positron emission tomography/computed tomography (PET/CT) is currently recognized as the most accurate imaging tool for staging and evaluating the treatment response of DLBCL.Baseline total metabolic tumor volume (TMTV) is a good indicator of prognosis, reflecting the baseline tumor burden and metabolism. 10,11A high baseline TMTV results in significantly shorter progression-free survival (PFS) and overall survival (OS) in many lymphoma subtypes, including DLBCL. 12,13In addition, the largest distance between two lesions (Dmax) calculated and normalized by patient body surface area (SDmax), as a simple imaging feature measured on PET scans, is a prognostic factor independent of TMTV that reflects lesion dissemination. 14However, it is not comprehensive to use imaging indexes solely to predict the curative effect without considering molecular heterogeneity.
6][17] The emergence of next-generation sequencing (NGS) over the past decade has enabled highthrough put DNA sequencing, and the heterogeneity of DLBCL has been analyzed based on genetic alterations. 18right et al., 16 divided DLBCL into seven genomic subtypes to analyze the heterogeneity of DLBCL and aid the development of rationally targeted therapy.Multiple gene mutations, especially TP53 mutations, are important in guiding the selection and efficacy of drugs and are closely related to prognosis. 19Even so, genetic molecules have not been included in the current clinical risk scoring system, which may result in many genetic high-risk patients missing the opportunity to receive adequate treatment at an early stage.
No previous study had explored an integrated prognostic model that combines imaging and genetic molecular factors.This study aimed to establish and validate the nomogram based on PET/CT metrics and genetic mutations for optimizing the prediction of high-risk DLBCL population.

| Study population
We retrospectively collected the clinical data of 152 primary DLBCL adult patients (age ≥18 years) diagnosed between April 2019 and February 2022 in the Shandong Cancer Hospital and Institute (Jinan, China).The main endpoints were the complete remission (CR) rate and PFS after firstline chemotherapy.The inclusion criteria were as follows: (1) DLBCL confirmed in all patients by histopathological review of the baseline biopsy; (2) the pretreatment pathological tissues of all patients were sequenced by a lymphopanel including 43 genes; (3) the data of baseline 18 F-PET/ and actual 1-year PFS of the patients.The receiver operating characteristic curves showed this nomogram based on PET/CT metrics and TP53 mutations had a better predictive ability than the clinic risk scores.Similar results were identified upon external validation.

Conclusions:
The nomogram based on imaging factors and TP53 mutations could lead to a more accurate selection of DLBCL patients with rapid progression, to increase tailor therapy.

K E Y W O R D S
diffuse large B-cell lymphoma, nomogram, positron emission tomography/computed tomography, TP53 CT inspection available; (4) All patients were treated by R-CHOP (rituximab, cyclophosphamide, doxorubicin, vincristine, and prednisone) or R-CHOP-like chemotherapy, including 28 patients diagnosed with double expressor DLBCL who came from a prospective, single-arm, phase II clinical trial, and received zanubrutinib combined with R-CHOP regimen.The initial results of the prospective study had been presented by ASH in 2022.The main exclusion criteria were incomplete systemic chemotherapy for at least 4 cycles or only one lesion on PET/CT.The patients were followed up monthly by review of hospital electronic medical records and telephone calls.The last follow-up period was up to December 2022.Finally, 139 patients were included in the discussion and analysis.Among them, 94 patients with baseline PET/CT examination completed in the Shandong Cancer Hospital and Institute were allocated to establish a training cohort.An independent cohort of 45 DLBCL patients with baseline PET/CT examination from other hospitals was established for external validation.
This study was approved by the Medical Ethical Committee of Shandong Cancer Hospital and Institute (No. SDTHEC2022007008).Written informed consent for participation was not required for this retrospective study in accordance with the national legislation and the institutional requirements.The waiver of informed consent was approved by the Medical Ethical Committee of Shandong Cancer Hospital and Institute.The study was conducted in accordance with the Declaration of Helsinki (as revised in 2013).

| Variables and definitions
The clinical data obtained from all patients included: sex, age at disease onset, Eastern Cooperative Oncology Group (ECOG)-point scale (PS), Ann Arbor stage at diagnosis, bulky disease, IPI, aa-IPI, and NCCN-IPI at diagnosis, histological classification, BCL2/MYC double expression, lactate dehydrogenase (LDH) level, initial chemotherapy, treatment response, pretreatment PET/CT images, and 43 gene mutations based on NGS.PFS was calculated from the date of randomization to the date of death from any cause, disease relapse or progression, or the date of last follow-up.CR was defined based on PET/CT treatment response to at least four cycles of R-CHOP-like chemotherapy according to the Lugano criteria, as follows: Score 1, 2, or 3 with or without a residual mass on 5-PS (1, no uptake above background; 2, uptake ≤ mediastinum; 3, uptake > mediastinum but ≤ liver; 4, uptake moderately > liver; 5, uptake markedly higher than liver and/or no new lesions; X, new areas of uptake unlikely to be related to lymphoma), no new lesions and no evidence of FDGavid disease in marrow.

| Baseline PET metrics
TMTV was defined by two nuclear medicine physicians (blinded to patient outcomes) using a 41% maximum standardized uptake value (SUVmax) threshold on artificial intelligence-assisted medical image auto-delineation (AccuContour version 3.2; ManteiaTech).Bone marrow involvement was included in the volume measurement only if there was focal uptake.The spleen was deemed to be involved and included if there was focal uptake or diffuse uptake >150% of the liver background.From the 3-dimensional (3D) coordinates of the metabolic volume of each lymphoma lesion, the center of mass (centroid) of each lesion was automatically obtained using AccuContour and was taken as the lesion location.The distances between all pairs of lesions were calculated using the Euclidian formula.
Dmax was calculated in each patient and normalized by the patient's body surface area.

| Statistical analysis
The threshold to determine the TMTV and SDmax optimal cutoff values of the quantitative parameters for PFS prediction was tested by the receiver operating characteristic (ROC) curve analysis.The baseline characteristics of the 2 groups were analyzed using Pearson's χ 2 or Fisher's exact tests.Survival functions were estimated using the Statistical analysis was performed using the survival, survminer, ggplot2, ROC, and rmda packages in R, version 4.2.2 (http://www.r-project.org/).All p-values were 2sided and those less than 0.05 were considered statistically significant.

| Patient characteristics
The baseline characteristics of DLBCL patients in each cohort are listed in Table 1.The median follow-up duration of the training and validation cohort were 25.5 and 24.6 months, respectively.The TMTV of the 139 patients had a non-normal distribution and the median TMTV was 249.0 cm 3 (P25-P75, 121.2-610.6 cm 3 ).ROC curve analysis showed that the optimal TMTV cutoff was 285.2 cm 3 (Figure S1).The median Dmax was 0.31 m (P25-P75, 0.10-0.54m), and SDmax was 0.150 m −1 (P25-P75, 0.060-0.250m −1 ).ROC curve analysis showed that the optimal SDmax cutoff was 0.135 m −1 (Figure S1).The TMTV and SDmax were converted into binary variables according to the cutoff values.Patients in each cohort were divided into seven subtypes according to the NGS results based on Wright's study 16 : MCD-like subtype (MYD88 L265P and CD79B mutations); BN2-like subtype (NOTCH2 mutations or BCL6 fusion); EZB-like subtype (EZH2 mutations or BCL2 translocation); N1-like subtype (NOTCH1 mutations); A53-like subtype (biallelic TP53 mutations); ST2like subtype (SGK1 and TET2 mutations); others subtype.No correlation was observed between two cohorts of the seven subtypes (Table 1).The top five genes with the highest mutation rates were PIM1, TP53, MYD88, CD79B, and KMT2D, with their mutation rates 36.69%,33.81%, 33.09%, 25.18%, and 25.18%, respectively, in the total population.The 43 gene mutation frequencies in the training cohort and in the validation cohort are presented in Table S1, and no significant differences in the 43 gene mutation frequencies were observed between two cohorts.

| Independent prognostic factors affecting CR and PFS in the training cohort
The results of univariate and multivariate analysis for CR in the training cohort are listed in Table 2.At multivariate level, TP53 status was found as an independent  3).No significant difference for CR and PFS was observed based on the other gene mutations or subtypes (Table S2).

| DISCUSSION
Early identification of patients with DLBCL who are rapidly progressing under conventional therapy is needed to aid stratification for innovative treatment.Based on the data from the training cohort and validation cohort, we established and validated a nomogram that incorporated pretreatment TMTV, SDmax, and TP53 status to predict PFS of DLBCL patients, showing higher predictive performance than the clinical risk scores.TMTV, which represents metabolic tumor burden, is significantly related to PFS and OS in DLBCL. 24,25ifferent studies have proposed various calculation methods and determined numerous cutoff values ranging from 200 cm 3 to 300 cm 3 . 11,12,26 In the current study, baseline TMTV was determined using a semi-automatic method (41% maximum standardized uptake value threshold) for each patient.It has been revealed that patients with a high TMTV were more prone to early progression and high TMTV has been identified as an independent predictive factor for PFS.8][29] Although it was less affected by height according to the LNH073B study, 28 we think that it is more reasonable to use body surface area to standardize Dmax, considering the consistency of comparison.The cutoff point of the SDmax between studies remains controversial.Our research shows that the cutoff value of SDmax is shorter than that reported in the REMARC study. 15The reason may be due to race-related factors and the heterogeneity of the samples included, for the non-germinal center B-cell-like (GCB) group accounted for 65% of our cohort whereas it accounted for  52% in the REMARC study.Thus, a multicenter largescale cohort study is still needed for verification.
The gene expression profile of DLBCL reflects heterogeneity and is of therapeutic importance. 16,17,304][35] In this study TP53 mutations were significantly associated with poor early prognosis.Our study did not detect the TP53 deletion by fluorescence in situ hybridization, and we defined patients with >50% mutations as the A53-like subtype.Univariate analysis showed the prognosis of patients in A53-like subgroup was worse but considering the limitations of detection method and the limited number of patients, it was not included in multivariate analysis.
Several large multicenter studies have reported that MCD-like type DLBCL cases associated with old age, extranodal involvement, and activated B-cell-like origin had a poor prognosis. 36,37Our follow-up showed no significant difference during MCD-like subtype and non-MCD-like subtype.Of the 18 patients with the MCD-like subtype, 15 received first-line immunochemotherapy combined with a BTK inhibitor (BTKi).This confirms that standard immunochemotherapy combined with targeted therapy is worthy of further research to improve the prognosis of MCD-like subtype. 38,39No significant difference for CR and PFS was observed based on the gene mutations except TP53 and therefore, longer follow-up and larger cohort verification are still needed.
Compared with the existing clinical models, such as IPI, aa-IPI, and NCCN-IPI, the nomogram has a better predictive power for the rapid progression in each cohort.It is indicated the predictive model not only helps to predict prognosis but also contributes to future clinical trials design.It might be meaningful to explore novel therapies or intensive combined therapies with more effectiveness for these high-risk cases based on the new model.It is also warranted to explore the reduction in the chemotherapy cycles and the adjustment of the interval of reexamination for the low-risk population.In the future, prospective trials are needed to establish more individualized therapies as suitable treatment for patients classified into different risk groups based on the predictive model.
Although the nomogram showed good accuracy in predicting prognosis, it still had the following limitations.First, the cutoff points of TMTV and SDmax are still cohort-dependent, being generated by ROC analysis of different measurement methods.The lack of agreement on the optimal cutoff points limited the use of PET parameters in routine clinical practice.Researchers have tried to use a segmentation method with a fixed threshold instead of the widely used percentage threshold at 41% of SUVmax to solve this problem. 40Currently, it is unknown how to best use the parameters of PET/CT.Second, this retrospective study included a high proportion of patients with advanced stage, with more than 70% of patients with III-IV stage, which may affect the distribution of TP53 mutations.Finally, this retrospective study may lead to a certain degree of selection bias, thus the predictive ability of the nomogram should be further validated in larger and prospective studies.

F I G U R E 1
Nomogram established from the training cohort for predicting PFS of DLBCL patients by combining baseline TMTV, SDmax, and TP53 mutations.DLBCL, diffuse large B-cell lymphoma; Dmax, the largest distance between 2 lesions; high SDmax, SDmax >0.135 m −1 ; high TMTV, TMTV>285cm 3 ; low SDmax, SDmax ≤0.135 m −1 ; low TMTV, TMTV ≤285cm 3 ; SDmax, standardized Dmax; TMTV, total metabolic tumor volume.F I G U R E 2 Calibration curve for predicting PFS at 1 year in the training (A) and the validation cohort (B).PFS is plotted on the y-axis; prognostic model-predicted probability of 1-year PFS is plotted on the x-axis.PFS, progression-free survival.

F I G U R E 3
Nomogram to predict PFS in the training cohort (A) and in the validation cohort (B).Subgroups of patients with different nomogram scores (low-risk: 0-72; intermediate-risk: 83-100; high-risk: 155-183 and extremely high-risk: >183) showed distinct progressionfree survival (PFS) in the training (n = 94) and the validation cohort (n = 45).

F I G U R E 4
ROC curves and the AUCs at 1 year to assess the prediction performance of the prognostic model compared with IPI, aa-IPI, and NCCN-IPI in the training (A) and validation cohort (B).aa-IPI, age-adjusted-IPI; AUCs, areas under the curve; IPI, the International Prognostic Index; NCCN-IPI, National Comprehensive Cancer Network IPI; PFS, progression-free survival; ROC, receiver operating characteristic.
Meier (KM) method and compared by log-rank test.The median follow-up was estimated by the reverse KM method.Univariate and multivariate analyses were performed using Cox proportional hazards models and logistic regression models.In the training cohort, risk factors selected for univariate analyses were based on previous studies and were routinely available in clinical practice.Considering the limited number of patients by genotyping and significance in univariate analysis, variables selected for multivariable Cox regression included Ann Arbor stage, TMTV, SDmax, aa-IPI, and TP53 status.According to the results of multivariate Cox regression analysis, a nomogram prediction model of PFS was established.Calibration curves were derived based on regression analyses to determine whether the predicted probability was consistent with the actual survival of the patients.Comparisons of the predictive ability between the nomogram with IPI, aa-IPI, and NCCN-IPI were investigated by the area under the ROC curves (AUC).
√ (weight × height) ∕3600Kaplan- Patients' clinical characteristics according to the training cohort and validation cohort.
Univariate and multivariate analysis for CR in the training cohort.Univariate and multivariate analysis for PFS in the training cohort.