Development and internal validation of a SEER-based prediction model for survival in poorly and undifferentiated colorectal neuroendocrine neoplasms.
Poorly and undifferentiated colorectal neuroendocrine neoplasms (CR-NENs) are rare and highly aggressive tumors with limited individualized prognostic assessment. Conventional staging systems are insufficient to capture heterogeneity, and dedicated prognostic models for this subgroup remain lacking. This study aimed to develop and internally validate a multivariable prediction model for survival in patients with CR-NENs.
Patients diagnosed with poorly and undifferentiated CR-NENs were identified from the Surveillance, Epidemiology, and End Results (SEER) database spanning 2000-2019. Eligible cases were randomly divided into training and validation cohorts (7:3). Overall survival (OS) and cancer-specific survival (CSS) were defined as study endpoints. Candidate variables including demographic and clinicopathological factors were analyzed using Cox regression to identify independent predictors, which were incorporated into nomograms. Model performance was evaluated by concordance index (C-index), time-dependent receiver operating characteristic (ROC) analysis with area under the curve (AUC), calibration plots, and decision curve analysis (DCA). Patients were stratified into low- and high-risk groups based on predicted risk scores, and Kaplan-Meier survival analysis was performed.
A total of 904 patients were included. The cohort was characterized by a high proportion of advanced disease, with approximately 80% of patients presenting with T3-4 tumors, about 75% having lymph node involvement, and over 40% with distant metastases. Multivariate analysis identified lymph node metastasis and distant metastasis as independent risk factors, while female sex, married status, and surgical treatment were protective factors (all P<0.05). The nomograms demonstrated moderate discriminative ability, with C-indexes of 0.70 [95% confidence interval (CI): 0.67-0.72] for OS and 0.71 (95% CI: 0.68-0.73) for CSS in the training cohort, and comparable performance in the validation cohort. For OS prediction, the 1-, 3-, and 5-year AUC values were 0.76, 0.82, and 0.85 in the training cohort and 0.72, 0.80, and 0.80 in the validation cohort; for CSS prediction, the corresponding AUC values were 0.76, 0.81, and 0.84, and 0.76, 0.86, and 0.88, respectively. Calibration curves showed good agreement, and DCA indicated clinical net benefit. Risk stratification based on the nomograms effectively distinguished patients with significantly different survival outcomes (P<0.001).
We developed and internally validated a SEER-based prediction model for OS and CSS in poorly and undifferentiated CR-NENs. The model may assist in individualized risk stratification, although further external validation is required before clinical application.
Patients diagnosed with poorly and undifferentiated CR-NENs were identified from the Surveillance, Epidemiology, and End Results (SEER) database spanning 2000-2019. Eligible cases were randomly divided into training and validation cohorts (7:3). Overall survival (OS) and cancer-specific survival (CSS) were defined as study endpoints. Candidate variables including demographic and clinicopathological factors were analyzed using Cox regression to identify independent predictors, which were incorporated into nomograms. Model performance was evaluated by concordance index (C-index), time-dependent receiver operating characteristic (ROC) analysis with area under the curve (AUC), calibration plots, and decision curve analysis (DCA). Patients were stratified into low- and high-risk groups based on predicted risk scores, and Kaplan-Meier survival analysis was performed.
A total of 904 patients were included. The cohort was characterized by a high proportion of advanced disease, with approximately 80% of patients presenting with T3-4 tumors, about 75% having lymph node involvement, and over 40% with distant metastases. Multivariate analysis identified lymph node metastasis and distant metastasis as independent risk factors, while female sex, married status, and surgical treatment were protective factors (all P<0.05). The nomograms demonstrated moderate discriminative ability, with C-indexes of 0.70 [95% confidence interval (CI): 0.67-0.72] for OS and 0.71 (95% CI: 0.68-0.73) for CSS in the training cohort, and comparable performance in the validation cohort. For OS prediction, the 1-, 3-, and 5-year AUC values were 0.76, 0.82, and 0.85 in the training cohort and 0.72, 0.80, and 0.80 in the validation cohort; for CSS prediction, the corresponding AUC values were 0.76, 0.81, and 0.84, and 0.76, 0.86, and 0.88, respectively. Calibration curves showed good agreement, and DCA indicated clinical net benefit. Risk stratification based on the nomograms effectively distinguished patients with significantly different survival outcomes (P<0.001).
We developed and internally validated a SEER-based prediction model for OS and CSS in poorly and undifferentiated CR-NENs. The model may assist in individualized risk stratification, although further external validation is required before clinical application.