Background
The growing availability of nationwide health data and advanced analytical methods has created new opportunities to understand long-term outcomes of cancer survivors at scale. As survival in non-metastatic gastrointestinal cancers improves, non-cancer causes of death and comorbidities increasingly determine long-term prognosis. Leveraging population-level big data with competing risks analytics is essential to move beyond cancer-specific outcomes and inform data-driven survivorship care.
Methods
We utilized nationwide, population-based cancer registry and health claims data in Korea to construct large-scale longitudinal cohorts of patients diagnosed with localized or regional gastric cancer (GC) and colorectal cancer (CRC), including approximately 189,000 GC and 95,000 CRC patients with up to 10 years of follow-up. Impact of type 2 diabetes and cardiovascular disease on competing mortality were modeled as time-varying covariates to capture dynamic comorbidity trajectories. All-cause mortality was analyzed using Cox models, while cancer-specific and non-cancer mortality were evaluated using Fine–Gray competing risks models, enabling scalable risk stratification across age, disease stage, and treatment groups.
Results
Long-term survival was high in non-metastatic disease, with five-year overall survival exceeding 85% in localized GC. In GC, non-cancer mortality surpassed cancer-specific mortality within 3–4 years after diagnosis and dominated long-term outcomes, particularly among older adults (≥65 years; 10-year non-cancer mortality ~26% vs ~10%) and patients treated with curative intent. In CRC, approximately 25% of patients had type 2 diabetes and 19% had cardiovascular disease at diagnosis. These comorbidities were associated with increased all-cause mortality (HR ~1.4) and substantially higher non-cancer mortality (e.g., cardiovascular disease sHR >2), with the strongest effects observed among patients younger than 50 years.
 
Conclusions
This large-scale, longitudinal analysis demonstrates how big data–driven competing risks models can reveal clinically meaningful patterns in cancer survivorship that are not captured by conventional survival metrics. Integrating such analytic frameworks into digital health and AI-enabled cancer surveillance systems could support personalized survivorship care, proactive chronic disease management, and population-level risk stratification to promote healthy aging among cancer survivors.
 
 
This work was supported by the National Cancer Center Grant (No. 24H1100-3)