“Gastro” is presented as a predictive tool that engages users through its interactive features, thereby improving the understanding of gastric cancer survival statistics. Future development will focus on incorporating additional variables to enrich the tool’s predictive capabilities.

This work was supported by the National Cancer Center Grant (NCC 24H1100-1). In addition, one of the authors, Anujin Rentsentavkhai was supported by the “International Cooperation & Education Program” (NCCRI·NCCI 52210-52211, 2024)” of National Cancer Center, Korea.

1. Data Source

Data were obtained from the Cancer Public Library Database (CPLD), established under the Korean Clinical Data Utilization Network for Research Excellence (K-CURE) project. The CPLD integrates four nationwide population-based databases:

• Korea National Cancer Incidence Database (KNCI DB)
• Cause-of-death data from Statistics Korea
• National Health Insurance Service (NHIS)
• Health Insurance Review & Assessment Service (HIRA)

The database encompasses 96.7% of all cancer incidence cases in Korea, with an estimated completeness of 98.3% (as of 2020).

2. Patient Selection

Primary stomach cancer patients diagnosed between 2012 and 2020 were identified from the CPLD (initial pool: n = 248,501). The following exclusion criteria were applied:

• Unknown or duplicated SEER stage (n = 12,292)
• Death Certificate Only cases or diagnosis date errors (n = 9)
• Age 30 or younger at diagnosis (n = 681)
• Diagnosed in 2012–2013 (n = 55,057; excluded to ensure sufficient follow-up)

The final analytic cohort comprised 180,462 patients diagnosed between 2014 and 2020, with follow-up through December 31, 2021 (maximum follow-up: ~8 years).

3. Input Variables

Base Variables

Variable	Categories / Range
Age at diagnosis	Continuous, 30–85 years
Sex	Male / Female
SEER Stage	Localized / Regional / Distant / Unknown
Tumor Location	Upper / Middle / Lower / Overlapped / Unknown
Comorbidity	None / Mild / Moderate / Severe Based on 13 conditions (Charlson Comorbidity Index adapted for gastric cancer): myocardial infarction, congestive heart failure, cerebrovascular disease, dementia, diabetes, liver disease, paraplegia, chronic pulmonary disease, renal disease, rheumatologic disease, peptic ulcer disease, peripheral vascular disease, mild/severe liver disease. Weights were derived from Cox regression on other-cause mortality in this cohort.

Health & Lifestyle Variables (optional)

Variable	Categories
BMI	Underweight (<20) / Normal (20–24) / Pre-obese (25–29) / Obese (≥30)
Metabolic Syndrome	Yes / No Defined as ≥3 of: abdominal obesity, high blood pressure, elevated triglycerides, low HDL, elevated fasting glucose.
Smoking	Never / Ex-smoker / Current smoker
Drinking	Non-drinker / Low-risk / High-risk Classified by daily alcohol intake (sex-specific WHO thresholds).
Physical Activity	None / Low / High Based on weekly MET-minutes (IPAQ protocol).

Treatment Variables (optional)

Variable	Categories
Initial Surgery	Endoscopic / Subtotal Gastrectomy / Total Gastrectomy / Others / None
Chemotherapy	Yes / No
Radiotherapy	Yes / No

4. Statistical Models

Overall Survival (OS) — Cox Proportional Hazards

All-cause mortality is modelled using the Cox proportional hazards model. Every patient who remains alive at end of follow-up is censored.

$$S(t\mid x) = e^{-H_0(t)\, e^{\beta x}}$$

where $H_0(t)$ is the baseline cumulative hazard and $\beta x$ is the linear predictor.

Cancer-Specific Survival (CSS) — Cause-Specific Hazard

A second Cox model is fitted with gastric cancer death (GCD) as the sole event; deaths from other causes are censored alongside survivors. This isolates the hazard attributable to gastric cancer under the assumption of independent competing risks.

$$\lambda_{cs}(t) = \lim_{\Delta t \to 0} \frac{P(t \le T < t+\Delta t,\; D=\text{GCD} \mid T \ge t)}{\Delta t}$$

Competing Risks — Fine-Gray Subdistribution Hazard

The cumulative incidence of each cause of death is estimated via the Fine-Gray model, which keeps patients who experienced a competing event in the risk set. The outcome variable (crisk) is coded as:

• 0 — Alive / censored
• 1 — Gastric cancer death (GCD)
• 2 — Other-cause death (OCD)

Separate models are fitted for GCD (failcode = 1) and OCD (failcode = 2).

$$\lambda_{sd}(t) = \lim_{\Delta t \to 0} \frac{P(t \le T < t+\Delta t,\; D=k \mid T \ge t \text{ or } (T < t \text{ and } D \ne k))}{\Delta t}$$ $$F_k(t\mid x) = 1 - e^{-\Lambda_{sd,k}(t)\, e^{\gamma_k x}}$$

Derived Outcomes

Restricted Mean Survival Time (RMST) — expected survival time up to horizon $\tau$:

$$\text{RMST}(\tau) = \int_0^{\tau} S(t)\, dt$$

Conditional Survival (CS) — probability of surviving an additional $t$ years given survival to landmark time $\tau$:

$$CS(t \mid \tau) = \frac{S(\tau + t)}{S(\tau)}$$