The goal of this observational study is to develop and validate a deep learning model to dynamically assess postoperative bleeding risk and assist in decision-making for re-operation in adult patients (≥18 years) diagnosed with primary gastric cancer undergoing radical gastrectomy. The main question[s] it aims to answer [is/are]:

Can an AI model based on perioperative dynamic physiological parameters and precise intraoperative blood loss accurately predict the risk of postoperative bleeding requiring re-operation? Does the application of this AI model improve clinical decision-making (e.g., earlier warning time, optimal intervention timing) and patient outcomes (e.g., mortality, length of stay)? Since there is no comparison group (this is a pure observational study without intervention arms), researchers will not compare different treatment groups. Instead, the investigators will evaluate the model's performance (sensitivity, negative predictive value, AUC, calibration) using retrospective data for training and prospective multi-center data for external validation.

Participants will:

Undergo standard radical gastrectomy and routine postoperative care as per clinical practice (no study-specific interventions).

Have their perioperative data collected, including demographics, medical history, vital signs, laboratory tests (blood gas analysis), surgical details, and precise intraoperative blood loss measurements.

(For prospective participants only) Provide informed consent and complete follow-up assessments up to 30 days post-surgery.

This study employs a hybrid design, collecting both retrospective and prospective data.

Inclusion Criteria:

1. Age: Patients aged ≥ 18 years.
2. Diagnosis: Histologically confirmed primary gastric cancer.
3. Surgical Procedure: Underwent radical gastrectomy (including proximal, distal, or total gastrectomy).
4. Consent: Provision of written informed consent (required specifically for the prospective phase).
5. Data Completeness: Availability of complete preoperative clinical data and postoperative follow-up records covering at least the first 15 days post-surgery.
6. Oncological History: No history of other primary malignant tumors.

Exclusion Criteria:

1. Surgical Type: Patients who underwent non-radical resection or emergency surgery.
2. Data Quality: Missing rate of key data fields exceeds 20%.
3. Preoperative Condition: Presence of severe preoperative infection or organ failure.
4. Follow-up Compliance: Unwillingness to participate in prospective follow-up or inability to complete the follow-up schedule (applicable only to the prospective phase).

This study includes adults (≥18y) with confirmed primary gastric cancer undergoing elective radical gastrectomy (proximal/distal/total) with D1+/D2 lymphadenectomy at [Center]. The cohort comprises retrospective ([2015.6]-[2026.2]) and prospective ([2026.3]-present) arms. Emergency, palliative, or multi-organ resections are excluded to ensure homogeneity.

The investigators anticipate enrolling 7000 patients. The primary outcome is postoperative hemorrhage requiring surgical re-intervention within 30 days. Estimated incidence is 0.5%-2.0%. To address this class imbalance, the AI model will employ stratified sampling and cost-sensitive learning.This population represents standard candidates for curative surgery in tertiary centers. By excluding extreme cases, the model is optimized for risk stratification in routine elective settings, where early warnings prevent catastrophic outcomes. Prospective data will validate real-time generalizability.