Sguanci Marco, Palomares Sara Morales, Cangelosi Giovanni, Petrelli Fabio, Sandri Elena, Ferrara Gaetano, Mancin Stefano
A.O. Polyclinic San Martino Hospital, Genova, Italy.
Department of Pharmacy, Health and Nutritional Sciences (DFSSN), University of Calabria, Rende, Italy.
Adv Nutr. 2025 May 5:100438. doi: 10.1016/j.advnut.2025.100438.
Malnutrition is a critical complication among cancer patients, affecting ≤80% of individuals depending on cancer type, stage, and treatment. Artificial intelligence (AI) has emerged as a promising tool in healthcare, with potential applications in nutritional management to improve early detection, risk stratification, and personalized interventions. This systematic review evaluated the role of AI in identifying and managing malnutrition in cancer patients, focusing on its effectiveness in nutritional status assessment, prediction, clinical outcomes, and body composition monitoring. A systematic search was conducted across PubMed, Cochrane Library, Cumulative Index to Nursing and Allied Health Literature, and Excerpta Medica Database from June to July 2024, following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses guidelines. Quantitative primary studies investigating AI-based interventions for malnutrition detection, body composition analysis, and nutritional optimization in oncology were included. Study quality was assessed using the Joanna Briggs Institute Critical Appraisal Tools, and evidence certainty was evaluated with the Oxford Centre for Evidence-Based Medicine framework. Eleven studies (n = 52,228 patients) met the inclusion criteria and were categorized into 3 overarching domains: nutritional status assessment and prediction, clinical and functional outcomes, and body composition and cachexia monitoring. AI-based models demonstrated high predictive accuracy in malnutrition detection (area under the curve >0.80). Machine learning algorithms, including decision trees, random forests, and support vector machines, outperformed conventional screening tools. Deep learning models applied to medical imaging achieved high segmentation accuracy (Dice similarity coefficient: 0.92-0.94), enabling early cachexia detection. AI-driven virtual dietitian systems improved dietary adherence (84%) and reduced unplanned hospitalizations. AI-enhanced workflows streamlined dietitian referrals, reducing referral times by 2.4 d. AI demonstrates significant potential in optimizing malnutrition screening, body composition monitoring, and personalized nutritional interventions for cancer patients. Its integration into oncology nutrition care could enhance patient outcomes and optimize healthcare resource allocation. Further research is necessary to standardize AI models and ensure clinical applicability. This systematic review followed a protocol registered prospectively on Open Science Framework (https://doi.org/10.17605/OSF.IO/A259M).
营养不良是癌症患者的一种关键并发症，根据癌症类型、阶段和治疗方法的不同，影响着≤80%的患者。人工智能（AI）已成为医疗保健领域一种很有前景的工具，在营养管理方面具有潜在应用价值，可改善早期检测、风险分层和个性化干预。本系统评价评估了人工智能在识别和管理癌症患者营养不良方面的作用，重点关注其在营养状况评估、预测、临床结局和身体成分监测方面的有效性。按照系统评价和Meta分析的首选报告项目指南，于2024年6月至7月在PubMed、Cochrane图书馆、护理及相关健康文献累积索引和医学文摘数据库中进行了系统检索。纳入了调查基于人工智能的干预措施用于肿瘤学中营养不良检测、身体成分分析和营养优化的定量初级研究。使用乔安娜·布里格斯研究所的批判性评价工具评估研究质量，并采用牛津循证医学中心框架评估证据的确定性。11项研究（n = 52,228名患者）符合纳入标准，并分为3个总体领域：营养状况评估和预测、临床和功能结局以及身体成分和恶病质监测。基于人工智能的模型在营养不良检测方面显示出较高的预测准确性（曲线下面积>0.80）。包括决策树、随机森林和支持向量机在内的机器学习算法优于传统筛查工具。应用于医学成像的深度学习模型实现了较高的分割准确性（骰子相似系数：0.92 - 0.94），能够早期检测恶病质。人工智能驱动的虚拟营养师系统提高了饮食依从性（84%），并减少了非计划住院次数。人工智能增强的工作流程简化了营养师转诊流程，将转诊时间缩短了2.4天。人工智能在优化癌症患者的营养不良筛查、身体成分监测和个性化营养干预方面显示出巨大潜力。将其整合到肿瘤营养护理中可改善患者结局并优化医疗资源分配。需要进一步研究以规范人工智能模型并确保其临床适用性。本系统评价遵循了在开放科学框架（https://doi.org/10.17605/OSF.IO/A259M）上预先注册的方案。
