Clinical · Molecular biology · Bioinfo & AI
Multimodal features and prognostic risk assessment in locally advanced gastric cancer patients following neoadjuvant therapy based on machine learning algorithms: a multicenter study
Summary
This multicenter study developed a multimodal model (MuMo) to assess the prognosis of patients with locally advanced gastric cancer following neoadjuvant therapy. The model integrates radiomic features from preoperative CT images, post-neoadjuvant therapy transcriptomic sequencing data, and clinicopathological information. Machine learning algorithms identified independent radiomic and molecular risk factors. MuMo demonstrated superior performance compared to single-modality models and ypTNM staging in predicting overall and disease-free survival, suggesting the involvement of immune-inflammatory pathways in therapy efficacy.
Critical appraisal
Clinical. This large-scale multicenter study (505 patients) proposes a multimodal prognostic model for locally advanced gastric cancer after neoadjuvant therapy. Radiomic (HR: 4.0, P < 0.001) and molecular (HR: 7.1, P < 0.001) risks were identified as independent prognostic factors. The MuMo model, with C-indexes of 0.855 for OS and 0.786 for DFS, outperforms current staging methods. This could enable better patient stratification for personalized adjuvant treatments or intensified follow-up, with a potential clinical impact in the medium term (3-5 years) by justifying clinical trials to validate its routine use.
Molecular biology. The discovery of three molecular features significantly associated with neoadjuvant therapy response, obtained through RNA sequencing on post-NAT specimens, is noteworthy. Mechanistic analysis suggests that neoadjuvant therapy efficacy is enriched in immune-inflammatory pathways, opening avenues for new therapeutic targets or predictive biomarkers of response. Validation of these markers in independent cohorts and their integration into clinical NGS panels could follow.
Bioinfo & AI. The study leverages machine learning algorithms to integrate heterogeneous data (radiomic, transcriptomic, clinical) into a multimodal model. The MuMo model was trained on 168 cases and validated on internal (72 cases) and external (804 cases from TCGA/GEO) cohorts, demonstrating robustness. Performance is evaluated by the C-index, showing significant improvement over single-modality models. This type of model could be clinically deployed to automate and refine prognostic assessment, assisting multidisciplinary teams in making informed decisions.
The summary and the appraisal were written by a language model from the title and abstract of the original article. A curator decided to publish it. Methodology