Letter to the Editor: clinical translation considerations regarding deep learning-derived CT body composition as a prognostic biomarker for locally advanced gastric cancer

Dear Editor, We read with great interest the article by Lai et al, published in your journal, titled “Deep learning-derived CT body composition enhances survival risk stratification of the TNM staging system in locally advanced gastric cancer: a multi-modal cohort study”[1]. The authors successfully integrated automated CT body composition (CTBC) analysis with plasma metabolomics and tumor immunometabolic phenotypes, demonstrating that the visceral-to-subcutaneous adipose tissue ratio (VAT/SAT) is a prognostic factor independent of TNM stage and proposing potential immunometabolic mechanisms. This research skillfully employs artificial intelligence technology, providing significant biological insights into understanding survival heterogeneity in gastric cancer patients and representing an exciting direction in precision oncology. However, before advancing this biomarker into routine clinical application, we believe several critical issues require clarification and resolution. We offer the following constructive criticisms aimed at fostering more robust and translationally impactful research in this field. First, the study established an association between a high VAT/SAT ratio and poor prognosis and discovered a U-shaped risk curve. However, it failed to define a clear, clinically actionable cutoff value. The use of multiple thresholds (e.g., 0.6 for survival analysis, 0.97 for metabolic analysis, and 1.4 for tissue analysis) created confusion. For clinicians, the key question is: For a specific stage III gastric cancer patient, how high a VAT/SAT ratio warrants a change in treatment strategy? Should it prompt intensified adjuvant chemotherapy, earlier use of immune checkpoint inhibitors, or aggressive metabolic/nutritional intervention? The study did not link the biomarker to specific treatment response data (e.g., comparing outcomes of patients with high VAT/SAT ratios under different adjuvant regimens or immunotherapies), leaving a substantial “clinical translation gap.” An ideal prognostic biomarker study should aim to develop a “risk-adapted” treatment algorithm, not merely risk stratification. Second, while dichotomizing the continuous VAT/SAT ratio simplified the model, it may have lost important biological gradient information and introduced bias risk from arbitrary threshold selection. More importantly, although the cohort received relatively uniform treatment (D2 gastrectomy, predominantly with capecitabine/S-1 plus platinum-based chemotherapy), analysis stratified by specific chemotherapy regimens or cycles was not provided. In the current era of increasingly diversified gastric cancer treatment, especially with immune checkpoint inhibitors (ICI) used in advanced and adjuvant settings (e.g., ATTRACTION-5, CheckMate 577 trials)[2], understanding the prognostic and predictive value of the VAT/SAT ratio across different treatment modalities (e.g., chemotherapy alone vs chemotherapy plus ICI) is crucial. For instance, does the hypothesized immunosuppressive microenvironment associated with a high VAT/SAT ratio predict a poorer response to ICIs? The lack of such stratified analyses significantly limits the generalizability of the findings to current clinical practice. Third, although the study focused on a homogeneous stage III population and standardized surgery, several host factors that strongly influence survival and body composition were not measured or included in the multivariable model. These include: specific systemic inflammatory markers (e.g., neutrophil-to-lymphocyte ratio, platelet-to-lymphocyte ratio), detailed nutritional status assessment (e.g., Patient-Generated Subjective Global Assessment, PG-SGA), comorbidity indices (e.g., Charlson index), and postoperative weight change trajectories. It is well-known that cancer cachexia, associated with severe fat and muscle loss[3], may correlate with very low VAT/SAT ratios (the left end of the U-shaped curve) and independently contribute to poor prognosis. Failure to adequately control for these factors may mean the observed association between VAT/SAT ratio and survival is partly driven by these unmeasured confounders. Finally, the discussion primarily framed the VAT/SAT ratio within the specific context of gastric cancer mechanisms. However, immunometabolic dysregulation related to visceral obesity (e.g., IDO1/AHR axis activation, regulatory T-cell enrichment) is a common phenomenon across multiple cancer types. For example, in colorectal and endometrial cancers, visceral fat has similarly been shown to promote immunosuppression and therapy resistance through analogous pathways[4]. This study missed the opportunity to situate its findings within this broader “obesity-cancer-immunity” interdisciplinary landscape. Furthermore, while the targeted metabolomics panel identified differential metabolic pathways[5], it failed to cover some emerging key mediators, such as specific lipid-derived mediators (e.g., prostaglandins, resolvins) or adipokines (e.g., leptin, adiponectin), which may be direct messengers linking fat distribution to tumor phenotype. To advance the VAT/SAT ratio as a clinical biomarker, we recommend validating its prognostic value in diverse, multi-center cohorts using advanced statistical methods for refined risk stratification. Prospective studies or subgroup analyses of existing trials should assess its predictive value for treatment selection. Future models should integrate CTBC metrics, inflammatory markers, nutritional status, and genomic features to enhance prognostic and predictive robustness[6]. Additionally, exploring the efficacy of IDO1-AHR axis inhibitors in preclinical or early-phase clinical trials for gastric cancer patients with high VAT/SAT ratios is warranted. In summary, the study by Lai et al is an important contribution, using advanced technology to unveil a neglected host factor in gastric cancer prognosis. By addressing the aforementioned limitations and striving to connect the biomarker to concrete therapeutic decisions and broader scientific understanding, subsequent research has the potential to truly realize its translational promise, ultimately improving personalized treatment and survival outcomes for gastric cancer patients.