“We propose a multi-agent architecture for nutritional oncology governed by a graduated autonomy model, and discuss the evidentiary, regulatory, and equity barriers that must be addressed before clinical deployment.”
Nutrition plays a critical role in cancer care, yet it is often overlooked. Many patients experience weight loss, muscle wasting, treatment-related side effects, and changes in appetite that can affect their ability to tolerate therapy and maintain quality of life. Although nutrition specialists can help manage these challenges, access to specialized nutritional care remains limited in many healthcare settings.
An editorial published in Volume 17 of Oncotarget, titled “Artificial intelligence in nutritional oncology: From isolated screening tools to agentic intervention systems,” explores how advances in artificial intelligence (AI) could help address this gap. The editorial was written by Arnab Sarkar and corresponding author Yashbir Singh-Wolkenhauer, who is affiliated with the Department of Radiology, Mayo Clinic, Rochester, Minnesota. Rather than presenting new clinical trial data, the authors outline a future vision in which AI systems move beyond isolated tasks to continuously support nutritional care throughout a patient’s cancer journey.
The Hidden Challenge of Cancer-Related Malnutrition
Cancer-related malnutrition is far more common than many people realize. Depending on the type and stage of cancer, it affects between 40% and 80% of patients and is estimated to contribute to 10% to 20% of cancer-related deaths. Cancer cachexia can develop in up to 80% of patients with advanced disease and is an independent predictor of mortality regardless of body mass index.
Professional organizations such as the European Society for Clinical Nutrition and Metabolism (ESPEN) and the American Society for Parenteral and Enteral Nutrition (ASPEN) recommend routine nutritional screening for patients with cancer. However, implementing these recommendations remains difficult. The authors note that only about one-quarter of oncology clinicians report having nutrition specialists integrated into their multidisciplinary teams, while the estimated ratio of registered dietitians to oncology patients in U.S. outpatient settings is approximately 1 to 2,308.
The authors argue that this is not simply a lack of nutritional knowledge, but a healthcare systems challenge that technology may help address.
AI Is Already Helping—but Only One Task at a Time
Artificial intelligence is already being applied to several aspects of nutritional oncology.
Machine learning models can identify patients at high risk for malnutrition with accuracy comparable to, or exceeding, traditional screening tools. Deep learning algorithms can analyze routine CT scans to automatically measure skeletal muscle and body composition, helping clinicians detect sarcopenia without requiring additional imaging. AI-powered virtual dietitians have also shown encouraging results, with patients reporting that they used the guidance to inform their diets and better manage treatment-related symptoms. However, these findings came from an observational deployment of a commercial product rather than a randomized trial and should be interpreted cautiously.
Despite these advances, the authors point out that today’s AI applications generally perform only one specific task at a time. One system may identify malnutrition risk, another may analyze body composition, while another provides dietary recommendations. These tools typically do not communicate with one another or continuously adapt as a patient’s condition changes during treatment.
From Individual Tools to Intelligent Care Partners
The editorial proposes moving beyond isolated AI applications toward agentic AI.
Unlike conventional AI models that respond to individual requests, agentic AI systems are designed to reason through complex problems, use multiple information sources, plan future actions, and remember previous interactions. Rather than answering a single question such as whether a patient is malnourished, an agentic system could continuously work toward the broader goal of optimizing a patient’s nutritional status throughout cancer treatment.
According to the authors, these systems could integrate information from electronic health records, laboratory results, medical imaging, dietary records, wearable devices, and clinical guidelines while adapting recommendations as treatments and symptoms evolve.
A Vision for AI-Assisted Nutritional Oncology
The authors describe a proposed multi-agent architecture in which several specialized AI agents work together under coordinated human oversight. The diagram presented in the editorial illustrates four specialized agents connected through a central coordination agent.
Within this framework:
-A Nutritional Screening Agent could monitor laboratory values, weight changes, and CT-derived body composition to detect early signs of malnutrition or cachexia.
-A Dietary Planning Agent could generate personalized meal plans that account for treatment side effects, cultural preferences, and individual dietary needs.
-A Treatment–Nutrition Interaction Agent could evaluate potential interactions between medications, nutritional supplements, and chemotherapy schedules.
-A Patient Engagement Agent could provide ongoing coaching through mobile applications or text messaging while tracking food intake and wearable health data.
-A central Coordination Agent would integrate information from these specialized systems, resolve conflicting recommendations, and alert clinicians whenever human review is needed.
Rather than replacing healthcare professionals, the proposed system is intended to support oncologists and dietitians by organizing complex information and providing timely recommendations.
Keeping Humans in Control
The authors emphasize that AI should not be allowed to make all clinical decisions independently.
Instead, they propose a graduated autonomy model, in which the level of AI independence depends on the clinical risk involved. Low-risk tasks, such as providing recipe suggestions or educational materials, could operate with relatively little supervision. Moderate-risk decisions, such as adjusting calorie targets or recommending referral to a dietitian, would require additional safeguards and clinician review. High-risk interventions, including decisions about enteral or parenteral nutrition, would always require explicit clinician authorization.
This framework aims to balance the efficiency of AI with the need for patient safety and clinical oversight.
Challenges That Must Be Addressed
Although the concept is promising, the authors stress that significant challenges remain before agentic AI can become part of routine cancer care.
Current AI systems are not perfect and can still generate inaccurate recommendations. No autonomous AI agent has received clearance from the U.S. Food and Drug Administration for independent clinical decision-making, and no randomized controlled trials have yet evaluated AI-driven nutritional interventions against major oncology outcomes such as survival, treatment completion, or maintenance of treatment dose intensity.
The editorial also highlights concerns related to patient privacy, clinical responsibility, and algorithmic bias. Because dietary habits vary across cultures, religions, and socioeconomic groups, AI systems must be validated in diverse populations to ensure equitable care and avoid reinforcing existing healthcare disparities.
Looking Ahead
The authors conclude that many of the technological building blocks needed for agentic AI in nutritional oncology already exist. They argue that nutritional care represents an especially promising area for future AI development because of the high prevalence of cancer-related malnutrition, shortages of specialized nutrition professionals, and the growing ability of AI systems to integrate complex clinical information.
However, they also emphasize that rigorous clinical validation remains essential before these systems can be widely adopted. Future research will need to determine whether AI-guided nutritional interventions improve meaningful patient outcomes while ensuring safety, fairness, and appropriate human oversight.
As artificial intelligence continues to evolve, the authors envision a future in which intelligent clinical support systems help oncologists and dietitians deliver more personalized, timely, and coordinated nutritional care—making nutrition a more integrated part of comprehensive cancer treatment rather than an afterthought.
Click here to read the full editorial published in Oncotarget.
_______
Oncotarget is an open-access, peer-reviewed journal that has published primarily oncology-focused research papers since 2010. These papers are available to readers (at no cost and free of subscription barriers) in a continuous publishing format at Oncotarget.com.
Oncotarget is indexed and archived by PubMed/Medline, PubMed Central, Scopus, EMBASE, META (Chan Zuckerberg Initiative) (2018-2022), and Dimensions (Digital Science).
Click here to subscribe to Oncotarget publication updates.
For media inquiries, please contact [email protected].
