From 2010 to 2015, 1807 patients of GNEC were enrolled in the SEER database, while 32 patients from Hangzhou TCM Hospital were identified as the external validation cohort. Of all patients, 1292 (71.5%) and 515 (28.5%) patients had DGNEC and PGNEC, respectively. The two groups demonstrated significant differences across multiple parameters (all p < 0.05), including age, gender, race, marital status, tumor grade, AJCC stage, T stage, N stage, tumor size, regional nodal evaluation (RNE), and treatment modalities such as radiation, surgery, and chemotherapy. The DGNEC patients typically exhibited a greater proportion of Grade I (18.0% vs. 14.2%), Grade II (44.5% vs. 42.5%), AJCC I (35.9% vs. 27.2%), AJCC II (27% vs. 22.5%), tumor size ≤2cm, those who received radiation (81.5% vs. 65.8%), those aged >60 years (79.9% vs. 67.8%), and those who didn’t receive chemotherapy (63.6% vs. 42.5%), and who received surgery (86.4% vs. 65.2%). The PGC group presented a high percentage of males (70.1% vs. 57.4%), married patients (61.6% vs. 57.8%), Grade Ⅳ (4.7% vs. 1.5%), AJCC III stage (31.3% vs. 23.7%), AJCC IV (19.0% vs. 13.4%), T2 stage (20.6% vs. 16.7%), T3 stage (35.7% vs. 28.1%), N1 (31.3% vs. 22.2%), those who received radiation (34.2% vs. 18.5%), who received chemotherapy (57.5% vs. 36.4%), and who didn’t received surgery (34.8% vs. 13.6%). As the characteristics of the two groups didn’t match, a 1:1 PSM was employed to minimize the impact of confounding variables. The standard deviations (SDs) of the majority of the variables after PSM was < 0.1, demonstrating a good balance (Fig. S1). Ultimately, 820 individuals were divided into two groups: 410 patients in the DGNEC group and 410 patients in the PGNEC group. Table 1 illustrates the demographic characteristics of both groups before and after PSM.
Survival analysis
Before PSM, patients in the DGNEC group showed significantly better Cancer-Specific Survival (CSS) and Overall Survival (OS) compared to those in PGNEC group (p < 0.001). (Fig. 2A, B). After PSM, significant differences in OS and CSS were observed between the two groups likewish. (Fig. 2C, D). When the competing risk variables were considered, cumulative incidence plots were created, and the DGNEC group showed a significantly reduced CSD (P < 0.001). Additionally, compared with patients in the PGNEC group, patients in DGNEC group experienced reduced 1, 3, and 5 year CIF of the CSD (17.68% vs. 26.75%; 34.69% vs. 46.90%; 5-year: 39.45% vs. 54.18%, P < 0.001) (Table 2). Then, subgroup analyses of AJCC stage, chemotherapy, sex, grade stage, surgery, tumor size, T stage, and N stage were performed. It turned out that DGNEC group had a reduced CSD in cohorts of male gender, female gender, tumor size (≤2 cm, 2< tumor size ≤5 cm, >5 cm and unkown), grade I-II stage, and AJCC I-Ⅲ stage, chemotherapy yes or none, surgery yes or none groups. (Fig. S2)
Similar results can be observed after PSM that significant differences existed between DGNEC and PGNEC in the 1-, 3-, and 5-year CIF of CSD (1-year: 18.24% vs. 25.02%; 3-year: 33.52% vs. 44.05%; 5-year: 37.31% vs. 51.36%, P<0.01) (Table 2). We conducted the subgroup analyses again, which showed that patients in the PGNEC group experienced more CSD in the following areas: male gender, female gender, tumor size (≤2 cm, 2< tumor size ≤5 cm, >5 cm and unkown), grade I-II stage, and AJCC I-Ⅲ stage, chemotherapy yes or none, surgery yes or none groups (Fig. 3).
Univariate and multivariate analysis.
Further independent analyses were conducted where all patients were randomly divided into two groups for model development: a training group comprising 70% of the patients (n = 1265), and a validation group comprising of the remaining 30% of the patients (n = 542). There were no apparent distinctions in the clinical baseline characteristics between the two groups. In the multivariate competing risks regression analysis, the PGNEC on CSD had a significant connection (HR, 1.40; 95% CI, 1.13–1.73; P = 0.002) (Table 3). Table S1 show the results of the multivariate subdistribution hazards model on OCD before and after PSM. The CIF values at 1, 3, and 5 years for the CSD were calculated through univariate analysis. Important factors (with a statistical significance level of P < 0.05) were identified through multivariate analysis using the Fine-Gray proportional subdistribution hazards model. According to the results of the multivariate competing risk analysis, primary site, gender, age, tumor size, grade stage, AJCC stage, T stage, N stage, and surgery were all independent predictors of CSD in patients with GNEC (Table 4). Taking into account the information about the T and N stages was included in the variable AJCC stage, we did not include it in our following analyses.
Constructing and verifying the nomogram
A nomogram for assessing the probabilities of experiencing CSD at 1, 3, and 5 years was developed based on variables determined in the multivariate analysis. (Fig. 4). Clinicians can calculate the cumulative score by adding up the points corresponding to each patient's prognostic characteristics. The score means the probability of CSD in individual patients at various time intervals. The nomogram derived from the training cohort was tested by the validation cohort and external validation cohort. The 1-, 3-, and 5-year AUC of the training group were 0.788, 0.819, and 0.833, respectively, whereas those of the validation group were 0.804, 0.824, and 0.834, respectively, and those of external validation group were 0.809, 0.860, 0.879 respectively. These results demonstrated a high discriminating capacity of the model (Fig. 5A-C). In addition, we tested the prediction accuracy of the model by using calibration plots (Figs. 5D-F), which showed a strong concordance in all datasets. These findings proved the high reliable and extreme precise in our nomogram.