Look at the patient as a whole, not just the disease
At the State University of Campinas in Brazil, researchers have found that the body itself carries a story about cancer's trajectory — one that tumor staging alone cannot tell. By combining measurements of fat and muscle density from routine CT scans into a single marker called VMD, the team revealed a fourfold difference in median survival between high- and low-risk gastric cancer patients. The work reflects a deepening conviction in oncology that the terrain in which a disease grows matters as much as the disease itself, and that precision medicine may already be visible in images we already take.
- Two gastric cancer patients with identical tumor stages can face median survival times of 13.8 months versus 58.5 months — a gap that current clinical tools cannot reliably predict.
- Oncologists have long been frustrated by the limits of tumor staging, which describes the cancer but says little about how the patient's body is responding to it at a metabolic and inflammatory level.
- The UNICAMP team used machine learning to sift through CT imaging, clinical, and laboratory data from 461 patients, arriving at a formula that captures opposing signals in fat and muscle density as a unified risk portrait.
- A built-in calibration safeguard — using the difference between tissue densities rather than absolute values — makes the marker potentially transferable across hospitals with different equipment.
- Because VMD is derived from scans already part of standard care, adoption would require no new tests, lowering the barrier to clinical use if prospective multicenter validation confirms the findings.
- Researchers are now asking whether nutritional therapy during treatment could shift a patient's VMD profile — a question that could open an entirely new front in personalized cancer care.
Researchers at the State University of Campinas in São Paulo have developed a new prognostic marker for gastric cancer that looks not at the tumor itself, but at what the disease does to the body carrying it. The marker, called VMD, combines measurements of fat and muscle radiodensity drawn from CT scans patients already receive as part of routine care. Its predictive power is striking: patients with unfavorable VMD values had a median survival of 13.8 months, while those with better readings survived a median of 58.5 months.
Gastric cancer is the fifth most common cancer worldwide, yet treatment decisions have long depended almost entirely on tumor staging — a system that leaves oncologists unable to explain why two patients with identical diagnoses can experience vastly different outcomes. The UNICAMP team, spanning radiology, oncology, and physics, chose to reframe the question: rather than asking only what the cancer looks like, they asked what it does to the person.
Analyzing scans from 461 patients treated over nearly a decade, the researchers used machine learning to test combinations of measurements until a reliable risk-stratification formula emerged. VMD captures how cancer alters the radiodensity of both fat and muscle tissue — changes that reflect the inflammatory and metabolic disruption the disease causes. The logic is counterintuitive: in fat, higher radiodensity signals worse prognosis; in muscle, lower radiodensity predicts poorer outcomes. Together, the opposing signals form an integrated picture of the patient's physiological state that neither measurement alone could provide.
The team also built in a technical safeguard, using the difference between tissue densities rather than absolute values, which cancels out calibration variation between CT machines and makes the marker more portable across clinical settings. This design choice matters if the tool is ever to move beyond a single institution.
The researchers are measured in their claims. The study is retrospective and requires prospective multicenter validation before it can guide treatment decisions. Whether nutritional intervention during treatment could improve a patient's VMD profile — and whether that would alter prognosis — remains an open question. Still, the work aligns with oncology's broader turn toward precision medicine, and because it draws on imaging already embedded in standard care, it asks nothing extra of patients. Early tests in other cancer types suggest the approach may not be limited to gastric cancer alone.
A team of researchers at the State University of Campinas in São Paulo has identified a new way to predict how gastric cancer will progress in individual patients—not by looking at the tumor alone, but by examining what the disease does to the body itself. The marker, called VMD, combines measurements of fat and muscle density visible on CT scans, images that gastric cancer patients already receive as part of routine care. The difference is striking: patients with unfavorable VMD values had a median survival of 13.8 months, while those with better readings survived a median of 58.5 months.
Gastric cancer ranks as the fifth most common cancer worldwide, yet treatment decisions have long relied almost entirely on tumor staging—assessing the size of the growth and whether it has spread. Two patients with identical tumor stages can experience vastly different outcomes, a reality that has frustrated oncologists for years. The UNICAMP team, working across the Department of Radiology and Oncology and the Institute of Physics, decided to shift their focus. Rather than asking only what the cancer looks like, they asked what it does to the person carrying it.
The researchers analyzed CT scans from 461 gastric cancer patients treated at UNICAMP over nearly a decade. Using machine learning, they tested different combinations of measurements until they found a formula that could reliably separate patients into risk groups. What emerged was VMD—a variable that captures how cancer alters the radiodensity of both adipose tissue and muscle. Radiodensity is simply a measure of how much tissue blocks X-rays on a scan, visible as different shades of gray. Changes in these patterns reflect inflammatory and metabolic shifts triggered by the disease.
The logic is counterintuitive but revealing. In fat tissue, higher radiodensity signals worse prognosis and suggests inflammation. In muscle, the opposite holds true: lower radiodensity predicts poorer outcomes. By combining these opposing signals, the marker captures something deeper than either measurement alone—an integrated picture of the patient's metabolic and inflammatory state. A nutritionist and co-advisor on the study explained that this difference between fat and muscle ultimately reveals the patient's phenotype, the physical expression of how cancer is reshaping their body at a cellular level.
The team used artificial intelligence not to replace expert judgment but to accelerate it. Rather than manually examining one variable at a time, machine learning allowed researchers to evaluate vast amounts of imaging, clinical, and laboratory data simultaneously, testing combinations at a scale impossible by hand. They also built in a safeguard: since CT machines can have slight calibration differences, they used the difference between fat and muscle density rather than absolute values, a choice that cancels out technical noise and makes the marker more reliable across different hospitals and equipment.
From a clinical standpoint, the implications are substantial. Gastric cancer treatment today is guided almost entirely by tumor characteristics. If VMD proves valid in larger, prospective studies, it could help oncologists identify which patients would genuinely benefit from aggressive chemotherapy and which might avoid its toxicity. The marker reflects something oncologists cannot currently measure with precision: the patient's own metabolic and inflammatory condition, the terrain in which the cancer grows.
Yet the researchers are careful about what they claim. The study is retrospective, analyzing data already collected, and it requires external validation in different populations before it can guide clinical decisions. They also do not yet know whether a patient's body composition can be improved through nutritional therapy during treatment, or whether such changes would alter prognosis. These are open questions that will require further investigation.
What makes this work significant is its alignment with a broader shift in oncology toward precision medicine—treatment tailored to individual characteristics rather than applied uniformly. Because VMD comes from scans already part of standard care, implementing it requires no new tests, no additional burden on patients. The researchers have begun testing the approach in other cancer types, with early results suggesting the method may extend beyond gastric cancer. The work points toward a future where doctors treat not just the disease but the person, using information already at hand to make smarter, more personalized choices.
Citas Notables
Cancer treatment is still very tumor-centric. Our proposal is to look at the patient as a whole.— Jun Takahashi, co-advisor on the study
The marker reflects the patient's metabolic and inflammatory condition, which can guide decisions about whether chemotherapy would truly benefit them.— José Barreto Campello Carvalheira, clinical oncology professor and study leader
La Conversación del Hearth Otra perspectiva de la historia
Why does it matter that you're looking at the patient's body composition rather than just the tumor?
Because two people with the same tumor can have completely different outcomes. Tumor staging tells you about the cancer. Body composition tells you about the person fighting it—their metabolic state, their inflammation level. That's what actually predicts survival.
So the radiodensity measurements are picking up inflammation?
Exactly. When cancer progresses, it changes how fat and muscle look on a scan. Higher density in fat suggests inflammation, which is bad. Lower density in muscle suggests weakness and metabolic dysfunction, also bad. Together, they paint a picture of the patient's condition.
Why did machine learning matter here? Couldn't a doctor just look at the scans?
A doctor could look at one or two measurements. But we tested hundreds of combinations to find which ones actually predicted survival. Machine learning let us do that at scale, then we validated it against actual patient outcomes.
The survival difference is enormous—13.8 months versus 58.5 months. How confident are you in that?
Confident enough to publish, but not confident enough to change treatment yet. This was retrospective data. We need prospective studies in different hospitals with different populations before doctors should rely on it for decisions.
What's the next step?
External validation, first. Then we need to know if we can actually change the marker through treatment—if nutritional therapy or other interventions can improve a patient's body composition and survival. Right now we have the question but not the answer.
Could this work for other cancers?
We're already testing it. Early results suggest the approach might apply broadly. If it does, this becomes a tool for precision medicine across oncology, not just gastric cancer.