Multimodal AI for Predicting Response to Neoadjuvant Immunotherapy in Gastric Cancer (PRISM-GC)

Condition(s) studied

Investigational drug(s) / intervention(s)

Standard of Care PD-1 Inhibitors: Patients receive standard neoadjuvant chemotherapy (e.g., SOX or XELOX regimen) combined with any NMPA-approved PD-1 inhibitor (including but not limited to Sintilimab, Tislelizumab, Camrelizumab, etc.) as determined by the treating physician in real-world practice.

Multimodal AI Assessment: Non-invasive assessment using a multimodal deep learning system (DeepComp) to analyze preoperative contrast-enhanced CT images and pathological slides. The AI model predicts the probability of pathological complete response (pCR) but does not alter the clinical treatment plan.

Study summary

Gastric cancer is a major global health challenge. Currently, a combination of chemotherapy and immunotherapy (PD-1 inhibitors) is frequently used before surgery to shrink tumors, a strategy known as neoadjuvant therapy. While this approach is effective for many patients, responses vary significantly, and there are currently no reliable tools to predict which patients will benefit the most before treatment begins.

The PRISM-GC study aims to develop and validate a novel Artificial Intelligence (AI) system to address this need. This is a prospective, observational study that will collect data from patients diagnosed with locally advanced gastric cancer who are scheduled to receive standard neoadjuvant chemotherapy combined with immunotherapy in a real-world clinical setting. The specific choice of immunotherapy drug is determined by the treating physician and is not dictated by the study.

Researchers will analyze standard preoperative CT scans and pathological tissue slides using advanced deep learning algorithms. The goal is to create a "multimodal" AI model that can accurately predict how well a tumor will respond to treatment (specifically, whether the tumor will disappear or shrink significantly). If successful, this AI tool could help doctors personalize treatment plans in the future, ensuring that each patient receives the most effective therapy while avoiding unnecessary side effects.

Eligibility

Primary outcome measure(s)

• Predictive Accuracy of the Multimodal AI Model for Pathological Complete Response (pCR) — From baseline assessment to postoperative pathological evaluation (approximately 5 months)
  The performance of the DeepComp AI model in predicting pCR will be evaluated using the Area Under the Receiver Operating Characteristic Curve (AUC). The model's predictions (based on preoperative baseline CT and pathology slides) will be compared with the ground truth postoperative pathological results. Secondary metrics including sensitivity, specificity, accuracy, positive predictive value (PPV), and negative predictive value (NPV) will also be calculated.
• Pathological Complete Response (pCR) Rate — At the time of postoperative pathological evaluation (approximately 1 month after surgery)
  Defined as the complete absence of viable tumor cells in the resected specimen (primary tumor and lymph nodes, ypT0N0), assessed according to standard pathological guidelines (TRG 0). This outcome measures the real-world efficacy of neoadjuvant chemo-immunotherapy across the cohort.

Trial sites (9)

More Qun Zhao trials in China

Other trials for the same condition