Time-varying prognosis and competing mortality in gastric cancer survivors: implications for age-stratified surveillance strategies.

[BACKGROUND] Gastric cancer (GC) is a leading cause of cancer-related mortality worldwide. With improved survival of patients with GC after curative resection, there is an urgent need for precise dynamic prognostic assessment to guide individualized follow-up. The American Joint Committee on Cancer TNM staging system and lymph node ratio (LNR) are static prognostic tools that fail to capture time-dependent risk changes and competing noncancer mortality, limiting their utility in long-term survival management.

[METHODS] A single-center retrospective cohort study was conducted on 2772 patients with stage I to III GCs who underwent R0 gastrectomy. Patients with distant metastasis (stage IV) or those receiving neoadjuvant therapy were strictly excluded. Patients were stratified by LNR (negative LNR, low LNR, and high LNR) and age (early-onset GC [EOGC], conventional-onset GC [COGC], and late-onset GC [LOGC]). Conditional survival and time-varying hazard ratios (HRs) were calculated using landmark Cox models. Competing risk analysis was performed to estimate cumulative incidence functions (CIFs) of GC-specific death (GCSD) and noncancer death (NCD) using the Gray test for group comparisons. Multivariate Fine-Gray models were used to identify independent predictors of NCD.

[RESULTS] The prognostic disadvantage of high LNR attenuated over time. At 5 years postoperatively, the 5-year conditional survival rates of the high LNR group showed a convergence trend, comparable with those of the negative LNR (98.7%) and low LNR groups (93.0%). The time-varying HRs for high LNR vs negative LNR decreased from 13.67 at the 1-year landmark to 2.94 at the 5-year landmark, indicating a substantial attenuation of prognostic impact over time. The HR attenuated substantially over time but remained clinically relevant. A mortality crossover occurred at approximately 2.17 years (26 months) postoperatively in patients with LOGC. Before this time point, the cumulative incidence of GCSD was slightly higher than that of NCD. By 3 years postoperatively, the CIF of NCD (8.98% [95% CI, 6.54%-11.42%]) exceeded that of GCSD (7.96% [95% CI, 5.62%-10.30%]) (Gray test P =.032). Multivariate analysis showed that age stratification (LOGC vs EOGC: subdistribution HR [SHR] = 16.508; P <.001), LNR (low LNR vs negative LNR: SHR = 1.638; P =.013; high LNR vs negative LNR: SHR = 2.649; P =.002), and adjuvant chemotherapy (SHR = 0.362; P <.001) were independent predictors of NCD.

[CONCLUSION] The prognostic value of LNR is transient, with a high LNR-associated risk attenuating in long-term survivors. A 3-year mortality crossover in patients with LOGC mandates a shift from cancer-focused surveillance to comorbidity and frailty management. High LNR was identified as a proxy for frailty in the elderly patients, warranting a shift from aggressive oncologic surveillance to holistic geriatric care.