CT-Based VMD Biomarker Predicts Gastric Cancer Survival, Offering Path to Personalized Treatment
核心洞察
Researchers at UNICAMP in Brazil developed VMD (搜索), a novel CT-based biomarker combining visceral fat and muscle radiodensity to predict gastric cancer (搜索) prognosis.
Patients with higher VMD (搜索) values had a median overall survival of 13.8 months compared to 58.5 months for those with lower values, a striking fourfold difference.
The marker was derived using machine learning analysis of routine CT scans from 461 patients and may complement traditional tumor staging for therapeutic stratification.
Researchers at the State University of Campinas (UNICAMP) (搜索) in São Paulo, Brazil, have identified a novel CT-based biomarker that sharply stratifies survival outcomes in gastric cancer (搜索) patients, potentially enabling clinicians to tailor treatment intensity based on an individual's metabolic and inflammatory profile rather than tumor characteristics alone.
The biomarker, termed VMD (搜索), combines radiodensity measurements of visceral adipose tissue and skeletal muscle obtained from routine computed tomography scans. In a retrospective analysis of 461 patients treated at UNICAMP over nearly a decade, those with higher VMD values — indicating a worse prognosis — had a median overall survival of just 13.8 months, compared to 58.5 months for patients with lower VMD values.
"Today, cancer treatment is still very tumor-centric. Our proposal is to look at the patient as a whole," said Jun Takahashi, full professor at the Gleb Wataghin Institute of Physics (IFGW) at UNICAMP and co-advisor of the study. "It isn't enough to treat the disease; you have to treat the patient."
A composite marker rooted in tissue biology
The VMD (搜索) marker emerged from the observation that radiodensity changes in fat and muscle carry opposite prognostic implications. "In adipose tissue, higher radiodensity values are associated with a worse prognosis and may indicate inflammation. However, in muscle, the opposite is true: the lower the radiodensity, the worse the prognosis," explained Maria Carolina Santos Mendes, a nutritionist and co-advisor on the study.
By combining these two measurements into a single variable, the team captured what José Barreto Campello Carvalheira, full professor of clinical oncology and study leader, described as "an integrated patient phenotype in which characteristics associated with metabolism and an inflammatory state associated with higher clinical risk emerge."
The study, conducted at the Department of Radiology and Oncology of the Faculty of Medical Sciences (FCM) at UNICAMP in partnership with IFGW and supported by the São Paulo Research Foundation (FAPESP) (搜索), was published in the journal Clinical Nutrition Espen.
Machine learning accelerates biomarker discovery
Rather than manually testing one variable at a time, the researchers deployed machine learning techniques to evaluate large volumes of imaging, clinical, and laboratory data. "I taught the machine to look in the same direction that experts were already looking, but with greater speed and scale. The technology allowed us to test various combinations until we arrived at the formula that best identified patients with the worst prognosis," Takahashi said.
The team also addressed a practical concern: calibration variability across different CT scanners. By using the difference between fat and muscle radiodensity, the researchers effectively canceled out machine-specific technical variations, enhancing the marker's reproducibility. This design consideration was highlighted by researcher Maria Emília Seren Takahashi.
Toward therapeutic stratification
Currently, gastric cancer (搜索) — the fifth most common cancer worldwide — is treated according to tumor staging, which considers characteristics such as tumor size and the presence of metastases. Yet patients at identical stages can experience markedly different disease trajectories.
"The goal of this line of research, and of this study in particular, is to expand staging beyond the tumor by incorporating an assessment of the patient," Carvalheira said. He noted that VMD (搜索) could eventually aid in therapeutic stratification by identifying which patients would truly benefit from chemotherapy and which might be spared toxic and aggressive treatment following surgery.
Validation and next steps
The authors emphasized that the study is retrospective and requires external validation in different populations — ideally through prospective, multicenter studies with larger patient cohorts — before the marker can guide clinical decisions.
Additionally, it remains unknown whether the body composition profile captured by VMD (搜索) can be modified during treatment. "We believe that nutritional therapy can help improve the patient's condition, but that wasn't evaluated in the study. We still don't know if it's possible to change that profile and impact the prognosis by doing so," Mendes said.
Despite these open questions, the approach aligns with the broader shift toward precision medicine in oncology. Because VMD (搜索) is derived from CT scans already performed as part of routine care, it could expand clinically actionable information without requiring additional testing. The researchers have begun evaluating the marker in other cancer types, with early results suggesting the approach may prove broadly applicable.