Workplace gaslighting in nursing practice: a concept analysis.

Metabolic Syndrome (MetS), characterized by abdominal obesity, hypertension, hyperglycemia, and dyslipidemia [], is a critical driver of cardiovascular disease, type 2 diabetes, and all-cause mortality []. Its prevalence exhibits marked age dependence and geographic variability: 41.8% of US adults are affected, escalating to 60.9% among individuals aged ≥ 60 years [], while China reports 31.1% overall prevalence rising to 36.9% in elderly populations [, ]. This accelerating global burden necessitates a paradigm shift toward data-driven precision prevention through digital health technologies. Effective MetS management depends on modifying core health behaviors, including physical activity, nutrition, and stress management [, ]. However, these behaviors do not occur in isolation. They arise from a complex interplay of biomedical, psychological, cognitive, and socioecological factors, as described in theoretical frameworks such as the Health Belief Model (HBM) [–]. This multifactorial nature leads to substantial interindividual heterogeneity, which limits the effectiveness of conventional one-size-fits-all interventions due to poor adherence and low sustainability []. In current practice, risk assessment relies primarily on traditional parametric models such as logistic regression. These models are constrained by their inability to capture nonlinear relationships and high-dimensional interactions inherent in integrated data sources, thereby impeding the development of practical digital decision-support tools capable of delivering individualized predictions []. Machine learning (ML) has emerged as a powerful approach for health behavior prediction, demonstrating superior performance in processing large-scale, high-dimensional, and heterogeneous data []. For example, Sirapangi and Gopikrishnan [] developed a Medical Internet-of-Things (MIoT)-based framework that integrates physiological, behavioral, and environmental data from wearable sensors to enable real-time forecasting. Similarly, Aguilar et al. [] applied ensemble learning to multi-omics data from over 488,000 UK Biobank participants, showing that multimodal integration improves latent risk detection. Notably, metabolomic features contributed modestly to model performance, highlighting the importance of evaluating the incremental value of each data domain.