Patient characteristics

This study included 402 patients with newly diagnosed stage III/IV DLBCL. Table 1 shows the clinical characteristics of 281 patients with stage III/IV DLBCL in the training group. The median survival time in the training group was 55 months. A total of 263 (65.4%) patients survived and 139 (34.6%) died, 87 of whom died from disease progression and 37 died from tumor-related complications, including 9 from infection-related disease and 6 from respiratory failure due to tumor compression of the mediastinum. Ten patients died of massive gastrointestinal hemorrhage due to tumor rupture, and 12 died due to cachexia. The final 15 patients died of non-tumor related diseases, including 3 patients who died of cerebral infarction, 4 patients who died of heart failure, 5 patients who died of severe pneumonia, and 3 patients who died of natural causes after recovery. According to MaxStat method, the optimal cut-off values of Age, LDH, Alb, PLT, HB, FER, and β2 microglobulin were 60 years old, 245U/L, 40 g/L, 100*10^9/L, 120 g/L, 130 μg/L, and 3.05 mg/L, respectively. There were 152 males (54.1%) and 129 females (45.9%). A total of 145 cases (51.6%) were less than 60 years old, and 136 cases (48.4%) were more than 60 years old at the first diagnosis. Of all patients in the training group who received first-line treatment, 202 (71.8%) received chemotherapy with R-CHOP, 35 (12.4%) received chemotherapy with CHOP, 13 (4.6%) received chemotherapy with R-DA-EPOCH, and 31 (11.2%) received other treatments. There were 260 cases (92.4%) with ECOG score 0–1 and 21 cases (7.6%) with ECOG score greater than 1. The number of extranodal involvement was 0–1 in 138 (49.1%) patients and 1 or more in 143 (50.9%) patients. There were 79 cases (28.1%) with large tumors at initial diagnosis. There were 134 cases (51.2%) of GCB subtype and 128 cases (48.9%) of non-GCB subtype. A total of 202 patients (72.7%) were treated with R-CHOP as first-line treatment. At initial diagnosis, 215 (76.5%) patients were positive for BCL-2, 232 (82.6%) patients were positive for BCL-6, 216 (76.9%) patients were positive for C-MYC, and 86 (30.6%) patients were positive for CD10. The 260(92.4%) patients had an ECOG score of 0–1. CD5 expression was positive in 92 (32.7%) patients, Ki-67 was low expressed in 75 (26.7%) patients and high expressed in 206 (73.3%) patients. Hepatitis B surface antigen was positive in 41 (14.6%) patients, and LDH was higher than normal value in 148 (52.7%) patients. After first-line treatment, 86 patients (30.6%) achieved CR (Complete remission),71 patients (25.2%) achieved PR(Partial remission), 55 patients (19.6%) achieved SD(Stable disease), and the remaining 69 patients (24.6%) achieved PD(Progressive disease).

Table 1 Clinical characteristics of 402 patients with advanced DLBCLSurvival analysis, nomogram construction and internal validation

The patients in the nomogram development cohort (n = 402) were divided into training cohort (n = 281) and validation cohort (n = 121) according to the ratio of 7:3. Univariate analysis was performed to identify potential prognostic factors in the training cohort: age (≥ 60 vs < 60,p < 0.01), BCL-2 (positive vs negative, p = 0.02), C-MYC (positive vs negative, p < 0.001), CD5 (positive vs negative, p = 0.003), KI-67 (high expression vs low expression, p < 0.001), LDH (elevated vs normal, p = 0.02), PLT (elevated vs normal, p = 0.28), FER (elevated vs normal, p < 0.001), β2 microglobulin (elevated vs normal, p < 0.001), HB (elevated vs normal, p = 0.04), and Alb (elevated vs normal, p = 0.05). In multivariate analysis, Ki-67 (p < 0.001), LDH (p = 0.05), FER (p < 0.001) and β2 microglobulin (p < 0.001) were the independent risk factors related to the prognosis of patients (Table 2). Subsequently, ki-67, LDH, FER, and β2 microglobulin were used for nomogram construction (Fig. 1). The values on the variable axis attributed to a single case were located and a vertical line was drawn upward from the variable axis to determine the total number of points assigned to the patient, enabling an estimate of the OS rate on the survival axis. Based on the constructed nomogram, the total score was used to identify three discrete risk groups according to the X-tile: low-risk, intermediate risk, and high-risk. There was no crossover in the KM curve drawn according to the risk groups scored by the nomogram. The 5-year OS rates of low-risk group, intermediate high-risk group and high-risk group were 81.6%, 44.2% and 6%, respectively. (Fig. 2). Internal validation showed good agreement between the predicted values of the nomogram and the actual 3-year OS rate in the calibration curve (Fig. 3a). In the internal validation cohort, the C-index was 0.76 and the AUC was 0.828 (Fig. 3b).

Table 2 Univariate and multivariate prognostic analysis of OS in patientsFig. 1

Nomogram model based on patients in the training group

Fig. 2

Kaplan–Meier survival curves for risk groups of the nomogram model in the training group

Fig. 3

a Clibration curve for predicting 3-year OS of patients with advanced DLBCL in the training group. b The ROC curve of the nomogram model to predict the 3-year OS rate of patients with advanced DLBCL in the training group

Nomogram external validation

The nomogram was externally validated by the calibration plot in Fig. 3a and by calculating bootstrap C-index in an independent validation cohort of 121 patients. In the external validation step, the C-index of the nomogram for predicting 3-year OS was 0.74, indicating that it is a model with good discrimination. The calibration curve indicated a good nomogram (Fig. 4a), and the AUC was 0.803 (Fig. 4b).

Fig. 4

a Calibration curve for predicting 3-year OS of advanced DLBCL patients in validation group. b ROC curve of the nomogram model predicting 3-year OS of patients with advanced DLBCL in the validation group.

Comparison of OS predictive accuracy between the nomogram and current staging or prognostic scoring systems

Both IPI low-risk and high-risk patients had a good prognosis stratification level (Fig. 5). For patients with stage III/IV DLBCL, the IPI score had poor stratification ability for patients with low-risk and low intermediate risk, and the 5-year survival rates were 61.5% and 55.8%, respectively. However, when the low-risk and low intermediate risk were included in our nomogram, the nomogram showed better stratification ability, and the 5-year survival rates were 93%, 34% and 6%, respectively (Fig. 6a). Similarly, for patients with stage III/IV DLBCL, the IPI score had poor stratification ability for high-intermediate and high-risk groups, with 5-year survival rates of 36.9% and 13.4%, respectively. However, when high-intermediate and high-risk groups were included in our nomogram, the nomogram showed better stratification ability. The 5-year survival rates of low, intermediate and high-risk patients were 52.9%, 30% and 6% (Fig. 6b), respectively. The C-index of the nomogram in the training cohort (0.72) was higher than that of the IPI (0.70).

Fig. 5

Kaplan–Meier survival curves for IPI risk groups in the training group

Fig. 6

a The IPI low and low-intermediate risk groups in the training group were included in the Kaplan–Meier survival curve of the nomogram model. b The IPI high intermediate and high-risk groups in the training group were included in the Kaplan–Meier survival curve of the nomogram model