Cancer cachexia is one of the most challenging complications of cancer care. Characterised by progressive, involuntary weight loss and skeletal muscle wasting, this complex metabolic syndrome can profoundly affect patients’ physical function, treatment tolerance, quality of life, and survival. Yet, despite its clinical importance, cachexia is often recognised only after substantial weight loss has occurred, when opportunities for effective intervention may already be limited. What if the information needed to identify at-risk patients were already available in their medical records, long before nutritional deterioration became clinically evident?
New research published in Nature Medicine in October 2026 reveals how large-scale integration of clinical and genomic data could help identify cancer-related weight loss earlier, opening new perspectives for predictive oncology and personalized nutritional care. In their research, entitled “A health-system-scale, episode-resolved multimodal atlas of cancer cachexia,” Castillon and colleagues examined how routinely collected healthcare data could help explain the development and progression of cancer-related weight loss. By integrating longitudinal clinical measurements, laboratory findings, disease characteristics, and tumour genomic information, the researchers developed one of the largest and most comprehensive datasets dedicated to understanding cachexia in patients with cancer.
The study included 52,236 patients treated at Memorial Sloan Kettering Cancer Centre between 2010 and 2024, with more than 1.7 million body mass index measurements analysed over time. Rather than treating cachexia as a static condition, the researchers used an innovative approach, identifying distinct episodes of clinically significant weight loss throughout each patient’s cancer trajectory. This perspective revealed a far more dynamic picture of nutritional deterioration than conventional assessments based on isolated measurements. Almost half of the patients (46.6%) experienced at least one qualifying weight-loss episode, with 30,938 episodes identified overall. Notably, approximately one in five affected patients experienced multiple episodes, suggesting that cancer-related weight loss may follow fluctuating trajectories, with periods of deterioration, stabilisation, and, occasionally, recovery.
The findings also highlighted important differences across cancer types. Weight-loss episodes were particularly frequent among patients with pancreatic and gastric cancers, affecting 72.7% and 71.7% of patients, respectively. Moreover, the researchers observed a significant temporal relationship between these episodes and disease progression. Across nearly all cancer types investigated, episodes of substantial weight loss occurred around periods of radiologically documented tumor progression more frequently than expected by chance. Weight-loss episodes were also associated with poorer overall survival. Although these findings do not establish that weight loss directly reflects tumor progression, they reinforce the importance of considering longitudinal nutritional changes as potentially valuable clinical signals.
This research is particularly interesting because it goes beyond body weight alone. By examining routinely collected laboratory measurements, the investigators identified recurring biological patterns associated with weight-loss episodes, including lower albumin and haemoglobin concentrations and increased inflammatory markers, such as C-reactive protein and neutrophils. Importantly, some of these changes occurred even when individual laboratory values remained within conventional reference ranges. This suggests that tracking biological parameters over time may yield clinically meaningful information that would otherwise be overlooked when interpreting isolated test results. Integrating tumor genomic data added another dimension, revealing associations between specific molecular alterations, including TP53 mutations, and the risk of weight-loss episodes in selected cancer populations.
Could these findings eventually help clinicians anticipate cachexia rather than respond to it? To explore this possibility, the researchers developed statistical prediction models that combined clinical characteristics, disease stage, baseline nutritional indicators, and selected genomic information. Tested in patients with lung, colorectal, and prostate cancers, these models showed modest predictive performance, with concordance indices ranging from approximately 0.64 to 0.67. External evaluation of a reduced lung cancer model provided further evidence of feasibility, although its performance remained insufficient for immediate clinical implementation. Nevertheless, the results show how longitudinal patient data could help identify individuals who may benefit from closer nutritional surveillance and earlier supportive care.
From an artificial intelligence and digital medicine perspective, the implications are particularly relevant. Although the study did not develop or validate an AI-based clinical decision-support system, it provides an important foundation for future data-driven healthcare applications. The ability to synthesise thousands of fragmented clinical observations into meaningful longitudinal patterns is a crucial step towards more sophisticated predictive tools. In the future, AI-powered systems could integrate these information sources with additional indicators, such as body composition, dietary intake, physical performance, and treatment-related factors, to generate more comprehensive and individualised risk assessments. Such developments could be particularly valuable in oncology, where timely nutritional interventions may influence treatment tolerance, functional status, and quality of life.
Important limitations remain. The study’s retrospective design, limited availability of direct skeletal muscle measurements, and lack of systematic information on nutritional interventions mean its findings cannot yet be translated into validated clinical recommendations. Prospective studies will be needed to determine whether prediction-based monitoring can genuinely improve nutritional and clinical outcomes.
Nevertheless, the study offers a compelling example of how the future of personalised medicine may depend not only on discovering new biomarkers or developing increasingly complex technologies, but also on extracting greater value from information already collected during routine care. In cancer cachexia, this could eventually support a shift from reactive management, initiated after substantial weight loss, to earlier recognition of nutritional vulnerability and more proactive, patient-centred care.
Reference. Castillon, V. J., Xie, A. X., Boscenco, S., et al. (2026). A health-system-scale, episode-resolved multimodal atlas of cancer cachexia. Nature Medicine. https://doi.org/10.1038/s41591-026-04689-9
