Development and validation of a frailty assessment-based competing risk model for predicting cancer-specific survival in elderly patients with gastric cancer: a dual-center cohort study

Frailty serves as a critical determinant of clinical outcomes in elderly cancer patients. Nevertheless, there remains a paucity of ‌rigorously validated prognostic tools‌ that account for frailty heterogeneity in older adults with gastric cancer (GC). This multicenter investigation enrolled 1,278 consecutive GC patients aged ≥ 70 years treated between January 2015 and December 2022 across two geographically distinct Chinese cohorts: the Xi'an Cohort (n = 934) and Yulin Cohort (n = 344). ‌Using stratified randomization‌, the Xi'an cohort was partitioned into training (n = 653) and internal validation (n = 281) sets at a 7:3 ratio, while the Yulin cohort provided ‌external validation‌. Frailty status was systematically assessed using the validated 5-Item Modified Frailty Index. ‌Competing risks regression analysis‌ identified significant predictors of cancer-specific survival (CSS) in the training cohort. Based on the risk factors, predictive models to predict patients’ 1-, 3-, and 5-year CSS were constructed. ‌Model performance‌ was rigorously evaluated through multiple metrics: the area under the receiver operating curve (AUC), Harrell's concordance index (C-index), calibration plots, and decision curve analysis (DCA). Multivariable analysis revealed ‌severe frailty‌ as an independent CSS predictor (subdistribution hazard ratio = 2.84, 95% CI: 2.04–3.96; P  < 0.001). ‌The frailty-incorporated competing risk model demonstrated superior predictive performance‌, achieving AUC values of 0.838 (training), 0.846 (internal validation), and 0.841 (external validation) - ‌surpassing conventional prognostic models‌. Concordantly, the model attained the highest C-indices across all validation sets (training: 0.771; internal: 0.784; external: 0.784). Calibration plots showed ‌excellent agreement‌ between predicted and observed outcomes, while DCA confirmed ‌enhanced clinical utility‌ across relevant risk thresholds. These findings establish that ‌frailty-adapted prognostic modeling‌ significantly improves survival prediction accuracy in elderly GC patients. Our results underscore the imperative of ‌comprehensive frailty assessment‌ in clinical decision-making and risk stratification for this vulnerable population.