Development and validation of a competing risk model for second primary pancreatic ductal adenocarcinoma: A population-based study

BACKGROUND: With advances in early diagnosis and treatment, the number of cancer survivors continues to grow, and more and more cancer survivors face the threat of second primary cancer (SPM). Second primary pancreatic ductal adenocarcinoma (spPDAC) is an important subclass of SPM, but its prognostic characteristics are poorly understood. METHODS: A total of 5,439 spPDAC samples and 67,262 primary pancreatic ductal adenocarcinoma (pPDAC) samples were extracted from the SEER database for this study. Survival differences between spPDAC and pPDAC samples were compared using Kaplan–Meier curves and log-rank tests. The Fine and Gray proportional subdistributed hazard method was used to analyze potential associations between clinical variables and pancreatic ductal adenocarcinoma-specific death (PDACSD) and death from other causes. After that, the clinical variables significantly related to PDACSD were screened out to construct a competing risk nomogram, which was used to evaluate the probability of the occurrence of PDACSD. The C-index was used to evaluate the discriminative ability of the model. The area under the curve (AUC) was used to verify the discrimination of the model. The calibration curve was used to verify the calibration of the model. Decision curve analysis (DCA) was used to validate the clinical utility of the model. RESULTS: Compared with patients with spPDAC, the pPDAC sample had a better prognosis (p = 0.0017). Across all spPDAC samples, the three most common sites of first-present cancer were the prostate, breast, and digestive system. Age (p < 0.001), race (p = 0.006), interval (p = 0.016), location (p < 0.001), T stage (p = 0.003), M stage (p < 0.001), chemotherapy (p < 0.001), and radiotherapy (p = 0.006) were the clinical variables associated with PDACSD screened by multivariate competing risks analysis. The concordance index values for the training and validation sets were 0.665 (95% CI, 0.655, 0.675) and 0.666 (95% CI, 0.650, 0.682), respectively. AUC, calibration curve, and DCA indicated that the model we constructed had good discrimination, calibration, and clinical utility. CONCLUSIONS: In conclusion, we first analyzed the impact of previous cancer history on prognosis. We then constructed a competing risk model that can predict the probability of developing PDACSD in spPDAC. This model has good discriminative ability, calibration, and clinical practicability and has certain guiding value for clinical decision-making.