Cachexia—a syndrome of muscle wasting and weight loss—is one of the most serious complications of advanced pancreatic cancer and a major contributor to death. Despite this, routine assessment of body composition is rarely incorporated into treatment planning. Most oncologists focus on tumor characteristics while body composition goes unmeasured.
Sarcopenia (low muscle mass) and myosteatosis (fat infiltration into muscle) are two measurable aspects of cachexia that can be extracted from routine CT scans using AI. This multicenter study investigated whether these automatically measured body composition markers could predict survival in advanced pancreatic ductal adenocarcinoma (PDAC) patients.
The study analyzed CT scan data from 601 patients treated at three German cancer centers. Researchers developed a deep learning model to automatically measure two body composition markers: the muscle-to-bone ratio (MBR), reflecting sarcopenia, and the ratio of inter- and intramuscular fat to muscle volume, reflecting myosteatosis.
These measurements were extracted automatically from abdominal CT images—the same scans already taken as part of standard cancer care—without any additional imaging. Statistical analyses included univariable and multivariable models to determine whether body composition independently predicted overall survival after accounting for other known factors like cancer stage.
In multivariable analysis, both sarcopenia (low MBR, hazard ratio 0.70, p=0.019) and myosteatosis (hazard ratio 3.11, p=0.02) were independently associated with worse overall survival, even after adjusting for cancer stage and other clinical factors. This means body composition added prognostic information beyond what tumor characteristics alone could predict.
Among patients treated with standard chemotherapy regimens, the same body composition markers remained significant predictors of survival. This suggests the findings hold across different treatment contexts, not just in a specific subgroup.
Looking deeper into the biology, researchers found that the muscle-to-bone ratio was significantly associated with several blood markers including albumin, hemoglobin, C-reactive protein (CRP), and total protein. Low muscle mass correlated with higher inflammation (CRP) and lower nutritional markers (albumin), consistent with cachexia biology.
Similarly, myosteatosis was associated with lower albumin and higher CRP, linking fatty muscle infiltration to systemic inflammation. These findings reinforce that the CT-measured body composition markers reflect real underlying biological processes related to poor outcomes—they are not just incidental measurements.
The key practical takeaway is that body composition can be assessed automatically from CT scans that are already part of standard cancer care—no extra procedures needed. This makes it feasible to incorporate AI-derived sarcopenia and myosteatosis assessments into routine oncology workflows without added cost or patient burden.
Patients identified as sarcopenic or myosteatotic could be prioritized for nutritional interventions, exercise programs, or dose adjustments in chemotherapy. Proactively addressing muscle health may improve treatment tolerance and potentially extend survival, though prospective interventional trials are needed to confirm this.
This multicenter study demonstrates that deep learning can reliably extract prognostically meaningful body composition data from routine CT imaging in advanced pancreatic cancer. The consistency of findings across three different cancer centers strengthens the evidence for clinical applicability.
Integrating body composition assessment into pancreatic cancer staging and monitoring could help personalize treatment decisions and identify patients at high risk of rapid deterioration. Future work should test whether treating sarcopenia and myosteatosis can improve outcomes.