Data from two centers of HHLWG in this study served as the derivation cohort. The two centers are 1) The Affiliated Hospital of Xuzhou Medical University and 2) Huai’an First People’s Hospital. Data from seven centers of HHLWG in this study served as the external validation cohort. The seven centers are 1) The Affiliated Hospital of Jining Medical University, 2) The First People’s Hospital of Changzhou, 3) Yancheng First People’s Hospital, 4) The First Affiliated Hospital of Anhui Medical University, 5) Qilu Hospital of Shandong University, 6) The Second Affiliated Hospital of Soochow University, and 7) The General Hospital of Wanbei Coal-Electric Group. Study approval was obtained from the independent ethics committees of each participating center in HHLWG and met the Helsinki Declaration. All patients were retrieved from the above centers between March 1, 2010, and January 19, 2021.

The following variables were collected: age, gender, lactate dehydrogenase (LDH), platelet (PLT), lymphocyte count (LYC), albumin (Alb), ferritin (Fer), B symptoms, white blood cell count (WBC), red blood cell count (RBC), hemoglobin (HGB), neutrophil count (NE), cell-of-origin (COO), presence or absence of hepatosplenomegaly, therapeutic regimens, and immunological markers (MYC, BCL-2, BCL-6, and Ki-67). Cases were staged according to the Ann Arbor staging system, and IPI scores were also evaluated.

All cases were diagnosed according to the WHO criteria. The exclusion criteria were as follows: 1) primary central nervous system and testicular lymphoma, 2) primary mediastinal large B-cell lymphoma, 3) transformed large B-cell lymphoma, 4) infected with human immunodeficiency virus, and 5) combined with other malignant tumors.

Biopsy samples were fixed in formalin, embedded in paraffin, sliced, and stained with hematoxylin and eosin for morphological analysis. Morphologically, all cases of DLBCL were categorized into three types: centroblastic, immunoblastic, and anaplastic variants. The centroblastic variant was shown to have large-sized nuclei with two to four small nucleoli closed to the nuclear membrane. Large, bizarre, and irregular cells were observed among tumor cells. The immunoblastic subtype was characterized by a rich cytoplasm with nucleolus laid in the center, and a local cytoplasm-like differentiation could be observed. Large to very large cells with bizarre pleomorphic nuclei were observed in the anaplastic variant, which may resemble the neoplastic cells of anaplastic large cell lymphoma.

Expression of biomarkers CD5, MYC, BCL-2, BCL-6, and Ki-67 was assessed using respective antibodies. All the histopathology sections were confirmed by at least two expert pathologists. Cutoff points for MYC, BCL-2, and BCL-6 proteins were designated as 40%, 50%, and 50% staining positive on lymphoma cells. It could be considered as CD5-positive when large tumor B cells express CD5 (>20%) as well as pan B markers (CD20, Pax5, CD19, etc.). GCB or non-GCB phenotypes were determined by the Hans algorithm (26).

Follow-up was conducted by consulting inpatient medical records and making phone calls. We followed up all the patients until July 12, 2021, or the death of patients. Overall survival (OS) was calculated as the interval between the time of diagnosis and death from any cause or the last follow-up. The survival status of all patients was confirmed with death records or a telephone call to next of kin (if patient died during the follow-up) or to the patients themselves.

Data were presented as numbers (percentages) for categorical variables and median (interquartile range, IQR) for all continuous variables. Outliers were verified by the hospital medical record system. Differences in clinicopathologic factors were analyzed by using the Mann–Whitney U test and χ ² test. Continuous variables were transformed into categorical variables by MaxStat analysis (titled as maximally selected rank statistics).

The potential impact of interrelationships among independent variables was evaluated by collinearity analysis. The degree of collinearity can readily be assessed by variance inflation factor (VIF) statistic (27). To evaluate the distance between variables and the presence of clusters, an r-type clustering analysis was utilized. The Cox proportional hazard model was used to analyze the univariate association between prognostic factors and OS. All variables with P <0.05 in univariable analysis were kept in the multivariable analysis by using forward selection for the best predictor set and the Akaike information criteria (AIC) were used to evaluate the model. The model was internally validated using a bootstrap resampling procedure (500 iterations) with a relatively corrected Harrell’s C-statistics (C-index), and the calibration curve was calculated according to the regression results. Brier score is another score function that measures the accuracy of probabilistic prediction. The nomogram was constructed on the grounds of the Cox model parameter estimates in the derivation cohort. The visual format of the nomogram reflects a statistical prediction that can determine how many points are attributed for each variable value, and the relative importance of predictors can be judged by the length of each line within the nomogram (25). All statistical tests were two-sided and statistical significance was set at P <0.05. The statistical analyses were performed by SPSS statistics for Windows, Version 19.0 (Armonk, NY: IBM Corp.) and R software (version 4.0.3; http://www.Rproject.org).

A total of 195 CD5+ DLBCL cases were selected from 1,864 DLBCL patients. The derivation cohort consisted of 131 cases from two centers and the external validation cohort consisted of 64 patients from seven centers. At the end of follow-up, a total of 81 (41.54%) deaths occurred. The median follow-up was 37.8 months [95% CI (35.9–39.7)] and the median survival time was 55.2 months [95% CI (39.0–71.4)]. The 5-year OS of patients was 42.1%. Median age at diagnosis was 61 years (range: 24–90), of whom 104 (53.33%) were older than 60 years and 100 (51.28%) were males. Ann Arbor stage III/IV accounted for 57.95%. The characteristics of patients in the two cohorts are detailed in Table 1 .

In this study, patients received regimens of CHOP-like (n = 12), R-CHOP/R-CHOP-like (n = 122), and R-based intensive regimens (n = 15). Nine patients received BTK inhibitor (BTKi), nine patients received methotrexate (MTX), and 14 patients received autologous hematopoietic stem cell transplantation (auto-HSCT). Kaplan–Meier analysis indicated that there was no significant difference in therapeutic regimens on the prognosis of CD5+ DLBCL in global comparison (P = 0.250, Figure 1A ). Further analysis on specific regimens showed that patients in BTKi regimen showed a better survival than those treated without BTKi (P = 0.038, Figure 1C ). However, there was no statistical difference in the regimens of auto-HSCT and R-CHOP with MTX ( Figures 1B, D ).

Using cutoff values of 40%, 50%, and 50% positive tumor cells for MYC, BCL-2, and BCL-6, respectively, 108 (55.4%) were positive for MYC, 129 (66.2%) cases were positive for BCL-2, and 115 (58.9%) cases were positive for BCL-6. Thirty-four (17.4%) patients were with high Ki-67 score (≥0.9) and 119 (61.0%) were non-GCB. MYC and BCL-2 coexpression in CD5+ DLBCL had a significant adverse impact on patient survival ( Figure 2D ), while MYC and BCL-6 coexpression was not a significant factor for OS ( Figure 2E ). The 5-year OS of CD5+ DLBCL patients with MYC/BCL-2 coexpression was 23% (P < 0.018, Figure 2D ). When assessed separately, patients with BCL-2+ or MYC+ had significantly inferior OS ( Figures 2A, B ) compared with patients with BCL-2− or MYC−, respectively. However, patients with BCL-6+ was not an adverse factor for OS ( Figure 2C ). Similarly, COO subtypes and Ki-67 were not predictive factors for CD5+ DLBCL survival ( Figure 2F , P = 0.064; data not shown for Ki-67, P = 0.118).

According to the MaxStat method, 74 years, 39.5 g/L, 6 × 10⁹/L, 3.87 × 10⁹/L, 3.62 × 10¹²/L, 106 g/L, 139 × 10⁹/L, 483 U/L, and 0.9 × 10⁹/L were the optimal cutoff points for age, Alb, WBC, NE, RBC, HGB, PLT, LDH, and LYC that distinguished the two groups most effectively (P < 0.05, Figure 3 ).

The results suggested that there were no interrelationships among independent variables (VIF < 7, R = 0.695). After cluster analysis, the original variables were divided into four categories. Age, RBC, IPI, MYC, and other clinical characteristics were statistically significant (P < 0.05). Univariable analysis exhibited that IPI, RBC, Alb, CNS involvement, BM involvement, and hepatosplenomegaly were strong prognostic predictors (P < 0.001). Following the model iterations in multivariable analysis, the final prognostic index consisted of six factors, as shown in Table 2 . IPI, NE, MYC, hepatosplenomegaly, and age proved to be adverse factors for the survival of CD5+ patients. Nevertheless, CNS involvement in the current multivariable model was observed to be not predictive [P = 0.129, HR = 1.808, 95% CI (0.841–3.885)].

A prognostic nomogram was developed to predict 1-, 3-, and 5-year OS of CD5+ DLBCL patients based on multivariable analysis ( Figure 4 ). The C-index (0.809) and the Brier score (0.151) were calculated between the predicted outcome and the real outcome of the model for internal validation. We further validated this nomogram externally and computed the C-index and the Brier score in an independent validation cohort of 64 patients (C-index = 0.770 and Brier score = 0.241). The calibration curves were close to the ideal curves, suggesting that the predicted result and the actual outcome had a good consistency ( Figure 5 ). In this study, all cases had complete data for all the variables required to calculate the IPI score. DCA was used to calculate the clinical net benefit of each model compared with all or none strategies, and the result demonstrated that the nomogram showed the best net benefit for 5-year OS ( Figure 6 ). Taken together, these results indicated that compared with IPI, the nomogram was more appropriate for predicting the survival of CD5+ DLBCL patients.