基于人工智能的营养干预系统应用探索

Exploration of the Application of Artificial Intelligence-Based Nutrition Intervention Systems

摘要: 目的：评估智能营养干预系统在群体膳食管理中的实施效能及其对代谢健康指标的影响。方法：采用三阶段前瞻性队列设计(2023~2025)，依托广西壮族自治区人民医院营养健康食堂示范项目，构建纵向干预研究模型。实验组(n = 356)依次接受：I期(2023年)基础营养宣教模式、II期(2024年)餐前营养素警示系统、III期(2025年)基于机器学习的个性化营养推荐系统。对照组(n = 215)来自同区域采用传统管理模式的企事业单位。通过智能餐饮平台获取连续膳食日志数据，结合年度健康体检指标(包括总胆固醇、LDL-C、空腹血糖等8项参数)，采用混合效应模型分析营养干预效果。结果：1) 实验组III期日均总能量摄入较I期降低18.7% (95% CI: 15.3%~22.1%)，宏量营养素供能比更趋近DRIs推荐范围(P < 0.05)；2) 含糖饮料选择频次呈现显著时间–干预交互效应(P < 0.05)，III期较基线下降63%；3) 代谢相关指标大部分有改善，其中总胆固醇、低密度脂蛋白胆固醇、同型半胱氨酸和高尿酸的异常率均有明显下降(P < 0.05)，而对照组各指标无显著变化(P > 0.1)。结论：基于人工智能的多模态营养管理系统通过实时膳食监测、个性化反馈及预测性健康预警，与营养摄入模式的显著优化及代谢紊乱状态的改善存在统计学关联，为群体性营养干预提供了循证实践方案。研究提示，将深度学习算法整合至膳食管理流程，可能将成为慢性病防控的新型技术。

Abstract: Objective: To evaluate the implementation efficacy of an intelligent nutrition intervention system in population dietary management and its impact on metabolic health indicators. Methods: This study adopted a three-phase prospective cohort design (2023~2025) based on a nutrition-focused healthy cafeteria demonstration project at the People’s Hospital of Guangxi Zhuang Autonomous Region, establishing a longitudinal intervention model. The experimental group (n = 356) underwent sequential interventions: Phase I (2023) with basic nutrition education, Phase II (2024) with a pre-meal nutrient alert system, and Phase III (2025) with a machine learning-based personalized nutrition recommendation system. The control group (n = 215) comprised employees from local enterprises adhering to traditional dietary management. Continuous dietary log data were collected via an intelligent dining platform, combined with annual health metrics (including total cholesterol, LDL-C, fasting glucose, and 5 additional parameters). Mixed-effects models were employed to analyze intervention outcomes. Results: 1) In the experimental group, Phase III demonstrated an 18.7% reduction in daily total energy intake compared to Phase I (95% CI: 15.3%%~22.1%), with macronutrient distribution aligning closer to Dietary Reference Intakes (DRIs) (P < 0.05). 2) Sugar-sweetened beverage consumption exhibited significant time-intervention interaction effects (P < 0.05), showing a 63% decline from baseline in Phase III. 3) Metabolic improvements were observed, with significant reductions in abnormal rates of total cholesterol, LDL-C, homocysteine, and uric acid (P < 0.05). No significant changes occurred in the control group (P > 0.1). Conclusion: The artificial intelligence-based multimodal nutrition management system, through real-time monitoring of dietary habits, personalized feedback, and predictive health alerts, is statistically associated with a significant improvement in nutritional intake patterns as well as better management of metabolic disorders; it thus provides evidence-based approaches for collective nutrition interventions. Research suggests that integrating deep learning algorithms into dietary management processes could represent a new technology for the prevention and control of chronic diseases.