Dynamic nomogram for predicting the overall survival and cancer-specific survival of patients with gastrointestinal neuroendocrine tumor: a SEER-based retrospective cohort study and external validation

Abstract

Background: Gastrointestinal neuroendocrine tumor (GI-net) is a rare heterogeneous tumor, and there is a lack of models to predict its prognosis. Our study aims to develop and validate two new nomograms to predict the overall survival (OS) and cancer-specific survival (CSS) of GI-net patients and investigate their application value.

Methods: SEER*Stat 8.4.4 software was used to download clinicopathological information of GI-net patients between 2010 and 2015 from the Surveillance, Epidemiology, and End Results (SEER) database. These patients were randomly divided into a training group (n=3007) and an internal-validation group (n=1289) at a 7:3 ratio. Patients from the Fourth Hospital of Hebei Medical University were enrolled in this study to form the external-validation group (n=86). Univariate and multivariate Cox analyses were performed to explore the independent prognostic factors and establish two nomograms. The concordance index (C-index), area under the time-dependent receiver operating characteristic curve (AUC), calibration curve, and decision curve analysis (DCA) were used to evaluate the nomograms. X-tile was used to divide GI-net patients into high-, medium-, and low-risk groups. Kaplan-Meier (KM) curves and log-rank tests were used to compare survival differences among the three groups.

Results: Seven variables (age, site, size, grade, M stage, surgery, and chemotherapy) were selected to establish the nomogram for OS, and 6 variables (age, size, grade, M stage, surgery, and chemotherapy) were selected for CSS. The C indices (0.785, 0.813, and 0.936 in the training, internal-validation, and external-validation groups for OS; 0.888, 0.893, and 0.930 for CSS, respectively) and AUCs (≥0.7) indicated that the nomograms had satisfactory discriminative ability. Calibration curve analysis and DCA revealed that the nomogram had a satisfactory ability to predict OS and CSS. KM curves indicated that each of the two nomograms clearly differentiated the high-, medium-, and low-risk groups. In addition, two online risk calculators were developed to predict the OS and CSS of these patients visually.

Conclusions: Our nomograms may play an important role in predicting 3- and 5-year OS and CSS for GI-net patients. Risk stratification systems and online risk calculators can be utilized in clinical practice to help doctors create personalized treatment plans.

Keywords: cancer-specific survival (CSS); gastrointestinal neuroendocrine tumor (GI-net); nomogram; overall survival (OS); surveillance epidemiology and end results (SEER) database.

Figure 1

The flowchart of patients inclusion among the SEER database. The flow diagram of selection of patients with GI-net in this study.

Figure 2

The optimal cut-off value of age and size calculated by X-tile. The age and size data were presented in a triangular grid pattern. Each pixel highlight represented the log-rank teat value [(A) age of OS, (C) size of OS, (E) age of CSS, (G) size of CSS], the distribution of the number of patients is shown in the histogram (B) age of OS, [(D) size of OS, (F) age of CSS, (H) size of CSS]. OS, overall survival; CSS, cancer-specific survival.

Figure 3

Nomograms for predicting 3- and 5-year OS and CSS of GI-net patients. (A) nomogram to predict 3-, and 5-year OS for GI-net patients; (B) nomogram to predict 3-, and 5-year CSS for GI-net patients. OS, overall survival; CSS, cancer-specific survival.

Figure 4

Time-dependent ROC curves of nomograms. (A) Time-dependent ROC curves of the OS nomogram showed that the AUCs in the training group were 0.7799,and 0.7993 for predicting 3-year and 5-year OS, respectively. (B) Time-dependent ROC curves of the OS nomogram showed that the AUCs in the internal-validation group were 0.8542, and 0.8493 for predicting 3-year and 5-year OS, respectively. (C) Time-dependent ROC curves of the OS nomogram showed that the AUCs in the external-validation group were 0.9812, and 0.9901 for predicting 3-year and 5-year OS, respectively. (D) Time-dependent ROC curves of the CSS nomogram showed that the AUCs in the training group were 0.8730, and 0.9051 for predicting 3-year and 5-year CSS, respectively. (E) Time-dependent ROC curves of the CSS nomogram showed that the AUCs in the internal-validation group were 0.9100 and 0.9098 for predicting 3-year and 5-year CSS, respectively. (F) Time-dependent ROC curves of the CSS nomogram showed that the AUCs in the external-validation group were 0.9812 and 0.9911 for predicting 3-year and 5-year CSS, respectively. AUC, area under the time‐dependent receiver operating characteristic curves; OS, overall survival; CSS, cancer-specific survival.

Figure 5

Calibration curves of nomograms. (A–C) Calibration curves of 3-year and 5-year OS for GI-net patients in training, internal-validation, and external-validation groups. (D–F) Calibration curves of 3-year and 5-year CSS for GI-net patients in training, internal-validation, and external-validation groups. GI-net, gastrointestinal neuroendocrine tumor; OS, overall survival; CSS, cancer-specific survival.

Figure 6

Decision curve analysis (DCA) of nomograms. (A–C) DCA of 3-year and 5-year OS for GI-net patients in training, internal-validation, and external-validation groups. (D–F) DCA of 3-year and 5-year CSS for GI-net patients in training, internal-validation, and external-validation groups. GI-net, gastrointestinal neuroendocrine tumor; OS, overall survival; CSS, cancer-specific survival.

Figure 7

The optimal cut-off value of nomograms score calculated by X-tile. The nomogram score data were presented in a triangular grid pattern. Each pixel highlight represented the log-rank teat value [(A) nomogram score of OS, (C) nomogram score of CSS], the distribution of the number of patients is shown in the histogram [(B) nomogram score of OS, (D) nomogram score of CSS]. OS, overall survival; CSS, cancer-specific survival.

Figure 8

Kaplan–Meier survival analyses for GI-net patients according to the risk stratification. Survival curves showed the OS [(A) and CSS (D) of the low-risk (1: red),medium-risk (2: green) and high-risk (3: blue) groups in the training group, the OS (B) and CSS (E) in the internal-validation group, and the OS (C)] and CSS (F) in the external-validation group. GI-net, gastrointestinal neuroendocrine tumor; OS, overall survival; CSS, cancer-specific survival.