Multi‑institutional development and validation of a nomogram to predict prognosis of early-onset gastric cancer patients.

Background: Early-onset gastric cancer (EOGC, ≤45 years old) is characterized with increasing incidence and more malignant phenotypes compared with late-onset gastric cancer, which exhibits remarkable immune cell infiltration and is potential immunotherapeutic population. Till now, restricted survival information of EOGC is available due to limited case numbers. This study established a novel nomogram to help evaluate cancer-specific survival (CSS) of EOGC patients who underwent gastrectomy, and may provide evidence for predicting patients' survival. Methods: We retrospectively enrolled a cohort containing 555 EOGC cases from five independent medical centers in China, among which 388 cases were randomly selected into a training set while the other 167 cases were assigned into the internal validation set. Asian or Pacific Islander (API) patients diagnosed with EOGC during 1975-2016 were retrieved from the SEER database (n=299) and utilized as the external validation cohort. Univariate and multivariate analyses were conducted to test prognostic significances of clinicopathological factors in the training set. Accordingly, two survival nomogram models were established and compared by concordance index (C-index), calibration curve, receiver operating characteristics (ROC) curves and decision curve analyses (DCA). Results: The 5-year CSS rate of training cohort was 61.3% with a median survival time as 97.2 months. High consistency was observed on calibration curves in all three cohorts. Preferred nomogram was selected due to its better performance on ROC and DCA results. Accordingly, a novel predicative risk model was introduced to better stratify high-risk EOGC patients with low-risk patients. In brief, the 5-year CSS rates for low-risk groups were 92.9% in training set, 83.1% in internal validation set, 89.9% in combined NQSQS cohort, and 85.3% in SEER-API cohort. In contrast, the 5-year CSS rates decreased to 38.5%, 44.3%, 40.5%, and 36.9% in the high-risk groups of the four cohorts above, respectively. The significant survival difference between high-risk group (HRG) and low-risk group (LRG) indicated the precise accuracy of our risk model. Furthermore, the risk model was validated in patients with different TNM stages, respectively. Finally, an EOGC web-based survival calculator was established with public access, which can help predict prognosis. Conclusions: Our data provided a precise nomogram on predicting CSS of EOGC patients with potential clinical applicability.