The SEER database is the authoritative source from the National Cancer Institute (NCI) that tracks cancer incidence and survival rates in the population (23). It has been available since 1973 and is updated every year. Data for related patients diagnosed with GI-nets between 2010–2015 were extracted from the SEER 17 registry database by SEER*Stat 8.4.4 software. Referring to previous research (24–26), they were assigned randomly to the training group or the internal-validation group at a 7:3 ratio. This ratio balance optimizes both model training capacity and validation credibility. 86 GI-net patients treated at the Fourth Hospital of Hebei Medical University from 2010 to 2020 served as the external-validation group. The training group was used to select variables and build the model, whereas the validation group was used to validate the model.

The inclusion criteria were as follows: (1) International Classification of Diseases (ICD) codes O–3 morphology: 8240/3 and 8249/3; (2) primary site codes: C16.0–C16.6, C17.0, C17.1–C17.2, C17.3–C17.9, C18.0, C18.1, C18.2–C18.9, C19.9, C20.9, and C25.0–C25.9; (3) staging on the basis of the seventh edition of the AJCC; (4) age at diagnosis > 18 years; and (5) NET as the only confirmed tumor.

The exclusion criteria were as follows: (1) not pathologically confirmed; (2) survival time < 1 month; (3) unknown age, sex, size, T stage, N stage, M stage, treatment methods, etc; and (4) reporting source was autopsy or death certificate only. The specific screening process is shown in Figure 1 .

The clinicopathological data of the patients included patient information (age and sex), tumor data (site, size, grade, T, N, and M), treatment methods (surgery, radiotherapy, and chemotherapy), and outcome variables (survival status and survival time (months)). X-tile software was used to evaluate the optimal cutoff values for age and tumor size. The primary endpoint of our study was OS, with a secondary focus on CSS. OS was defined as the interval from the date of initial diagnosis to the date of last follow-up or death from any cause. CSS was defined as the interval from the date of initial diagnosis to the date of last follow-up or death specifically due to neuroendocrine causes.

This study was approved by the Ethics Committee of the Fourth Hospital of Hebei Medical University (Approval No: 2024KS116), and obtained waiver for informed consent to participate. Our research was conducted in accordance with the Declaration of Helsinki.

The categorical variables are presented as percentages, and the continuous variables are presented as the means ± standard deviations (SDs). Univariate Cox regression was used to calculate the hazard ratio (HR) and 95% confidence interval (CI) for each risk factor. Significant factors identified in the univariate analysis were included in the multivariate Cox regression. This inclusion is based on the Akaike information criterion (AIC). In the multivariate analysis, significant clinical variables served as independent predictive factors. These factors were used to create nomograms that predict 3-year and 5-year OS and CSS. The C-index and time‐dependent receiver operating characteristic (ROC) curve were used to evaluate the predictive ability of the nomogram. The C-index and AUC ≥ 0.7 indicate that the nomogram has good predictive discrimination ability. The calibration curve was used to evaluate the difference between the predicted results and the actual results of the nomogram. The closer the predicted calibration curve is to the standard curve, the greater the predictive ability of the nomogram. Decision curve analysis was used to assess the clinical benefit of the nomogram, quantifying the net benefit at different threshold probabilities. KM curves and the log-rank test were used to analyze differences in OS and CSS among the different groups. The p value was used for all the statistical analyses, with p<0.05 defined as statistically significant.

Variance inflation factor (VIF) values were computed to assess the extent of multicollinearity present among the characteristics (27). VIF of >5 indicates that the characteristics are highly correlated and should be considered for removal. Additionally, Pearson’s correlation coefficients were determined to identify collinearity between the characteristics (28). Specifically, a correlation coefficient of <1 between any two independent characteristics was interpreted as an absence of multicollinearity.

Statistical analysis and plotting were completed using R 4.4.1 (http://www.R-project.org/) and the RStudio environment.

A total of 4,296 GI-net patients were selected from the SEER database and assigned randomly to the training group (n=3007) or the internal-validation group (n=1289). Additionally, 86 GI-net patients from our hospital were included in the external-validation group. The baseline demographics and clinicopathologic characteristics are listed in Table 1 . No significant differences in demographics and clinical characteristics were observed between the training and internal-validation groups (all P > 0.05). In the SEER group (n=4296), the average age of patients is 55.55 ± 14.43 years. There are more female patients than male (53.84% vs 46.15%). The most common site of GI-net is the small intestine (42.76%). A majority of patients are in the early stage (T0-2: 69.9%, N0: 68.92%, M0: 90.29%). More patients underwent surgery (95.68%), while fewer received radiotherapy (0.37%) and chemotherapy (2.58%). In the external-validation group (n=86), the average age of patients is 54.45 ± 11.60 years. The most common site of onset is the stomach (43.02%).

The optimal cutoff values for patient age and tumor size were calculated using X-tile software version 3.6.1 (https://x-tile.software.informer.com/) ( Figure 2 ). For OS, the optimal cutoff values for age are 58 and 72 years, and for tumor size are 9 and 23 mm. For CSS, the optimal cutoff values for age are 58 and 73 years, and for tumor size are 14 and 26 mm.

Nomogram variable screening a total of 11 clinical variables were included. According to the univariate Cox regression analysis, only sex was not associated with OS (P=0.08), and CSS was similarly unaffected (P=0.648). The remaining variables were included in the multivariate Cox regression analysis. The multivariate Cox regression analysis revealed that age, site, size, grade, M stage, surgery, and chemotherapy were independent prognostic factors for OS ( Table 2 ). For CSS ( Table 3 ), age, size, grade, M stage, surgery, and chemotherapy were found to be independent prognostic factors.

All the Pearson’s correlation coefficients between pairs of characteristics were < 1 and the VIF values were <5, indicating no collinearity among the independent characteristics ( Supplementary Figure 1 ).

Nomograms were constructed to predict the 3-year and 5-year OS and CSS rates using the selected clinical variables ( Figure 3 ). Each variable’s value corresponds to a specific individual score. By summing all the individual scores, we can determine the total score. The point where this total score’s vertical line intersects with the predicted probability line indicates the patient’s 3-year and 5-year survival rates. In the nomograms for OS and CSS, age was the most significant variable affecting survival outcomes.

The C indices for OS were as follows: training group, 0.785 (95% CI: 0.764–0.805); internal-validation group, 0.813 (95% CI: 0.785–0.841); and external-validation group, 0.936 (95% CI: 0.859–1.013). For CSS, the C indices were as follows: training group: 0.888 (95% CI: 0.866–0.911); internal-validation group: 0.893 (95% CI: 0.864–0.922); and external-validation group: 0.930 (95% CI: 0.849–1.011). All the C indices are greater than 0.70. The AUCs of the nomogram for predicting 3- and 5-year OS were 0.780 and 0.799, respectively ( Figure 4A ). The AUCs of the nomogram for predicting 3- and 5-year CSS were 0.874 and 0.905, respectively ( Figure 4D ). The internal-validation group and the external-validation group yielded similar results (≥0.7) ( Figures 4B, C, E, F ). These findings indicate that the two nomograms we established were accurate. The calibration curves ( Figure 5 ) revealed high consistency between the predicted OS and CSS probabilities and the actual occurrence probabilities in the three groups. Moreover, DCA ( Figure 6 ) revealed that the nomograms had good net clinical benefits. Therefore, our nomograms showed good predictive ability for GI-net patients.

Risk stratification is very important for guiding doctors in patient management. X-tile software was used to determine the optimal cutoff value for risk stratification on the basis of the overall nomogram score of patients ( Figure 7 ). Patients were divided into three risk groups: low-risk (OS: points ≤ 73.6; CSS: points ≤ 84.6), medium-risk (OS: 73.6 < points ≤ 134.6; CSS: 84.6 < points ≤ 175.9), and high-risk (OS: points > 134.6; CSS: points > 175.9) groups. KM survival curves and log-rank tests revealed significant differences among the three risk groups (p < 0.001) in the training cohort ( Figure 8 ). The internal-validation group and the external-validation group yielded similar results. These findings indicate the effectiveness of the nomogram risk stratification system.

To facilitate clinical application, we created two online risk calculators based on the shinyapp.io platform ( Supplementary Figure 2 ). These calculators allow doctors to quickly understand patients’ survival probabilities by selecting clinical variables. The web-based OS online risk calculator can be accessed at https://liexiantusc.shinyapps.io/DynNomapp_OS/; the web-based CSS online risk calculator can be accessed at https://dynnom-sc.shinyapps.io/DynNomapp_CSS/.