Obesity, a major global public health challenge, is strongly associated with several metabolic disorders, such as insulin resistance (IR), type 2 diabetes mellitus, and cardiovascular disease (Czech, 2017; Tian et al., 2022; Powell-Wiley et al., 2021). Traditionally, obesity assessment has relied primarily on body mass index (BMI), which is weight (in kilograms) divided by height squared (in meters). Despite its widespread clinical use, BMI does not reflect how specific distributions of adipose tissue affect obesity severity (Prillaman, 2023). Therefore, over time, researchers have realized the limitations associated with relying solely on BMI to assess obesity and have attempted to redefine obesity in terms of multidimensional indicators, which include not only BMI but also neck circumference (NC), waist circumference (WC), NC-to-height ratio (NHtR), and WC-to-height ratio (WHtR) (Lv et al., 2024; Cai et al., 2023; Busetto et al., 2024; Zhao et al., 2024; Ren et al., 2024). These indicators can more accurately reflect the potential health effects of different fat distribution.

Notably, obstructive sleep apnea (OSA) and hypertension are two common diseases that are closely related to and often coexist with obesity and aggravate IR and metabolic disorders in patients (Aminian et al., 2024; Zhou et al., 2024). Patients with OSA often experience intermittent hypoxemia and decreased sleep quality (Aminian et al., 2024; Zhou et al., 2024; Xu et al., 2022; Xu et al., 2024; Yeghiazarians et al., 2021). Moreover, repeated apnea and hypopnea can lead to sympathetic hyperactivity, which leads to blood pressure fluctuation and increased challenges with blood pressure control (Yeghiazarians et al., 2021; He et al., 2020). Therefore, the comorbidity of OSA and hypertension is not limited to a simple concurrent relationship but involves the interaction of various complex physiological and pathological pathways that jointly aggravate the occurrence and development of cardiovascular disease (Aminian et al., 2024; Zhou et al., 2024; Xu et al., 2022; Xu et al., 2024; Yeghiazarians et al., 2021; He et al., 2020; Floras, 2015; Javaheri et al., 2017).

Notably, a growing body of research evidence demonstrates that machine learning can effectively elucidate the association between the aggregation of multiple disease factors and long-term risks (Johnson et al., 2018; Ye et al., 2023; Wang J. et al., 2025). This capability facilitates a more nuanced analysis of how various risk factors interact and contribute to disease progression, thereby enhancing our ability to predict outcomes and tailor interventions for patients (Johnson et al., 2018; Ye et al., 2023; Wang J. et al., 2025). By leveraging machine learning techniques, we can uncover intricate patterns within large datasets, leading to improved risk stratification and more personalized approaches to disease management (Johnson et al., 2018; Chen et al., 2024).

While recent studies have demonstrated the superiority of alternative obesity indices and metabolic biomarkers individually, there remains a critical gap in the development of comprehensive risk stratification frameworks specifically designed for OSA-hypertension populations (Cai et al., 2023; Zhao et al., 2024; Aminian et al., 2024). Despite the documented limitations of traditional risk assessment tools and the emerging evidence supporting novel metabolic indices, limited investigation has been conducted into their integration with sophisticated computational methodologies for enhanced predictive accuracy. This knowledge gap represents a significant barrier to optimal clinical decision-making and personalized therapeutic interventions in this high-risk patient population. Therefore, to address the inherent limitations of conventional BMI-based obesity assessment and bridge this critical knowledge gap, this study aimed to develop and validate a comprehensive risk stratification framework. We hypothesized that effective cardiovascular risk prediction in patients with OSA and hypertension requires an integrated evaluation of metabolic, anthropometric, demographic, and clinical parameters, rather than reliance on any single measurement. This approach addresses the dual challenge of metabolic complexity and predictive accuracy by leveraging both multiple types of data and multiple machine learning methodologies.

This paradigm shift, driven by integrating medical insights with machine learning methodologies, stems from recognizing that BMI alone provides an inadequate diagnostic criterion for obesity-related health risks. Incorporating multidimensional data into predictive models enables more precise evaluation of individual health risks, facilitating the development of personalized prevention and treatment strategies tailored to specific metabolic profiles. This approach enhances our understanding of obesity and its associated health consequences while ultimately improving patient outcomes.

As a global health problem, obesity is not merely a simple consequence of being overweight but also a manifestation of chronic metabolic disorders, which are closely related to IR (Czech, 2017). Notably, current BMI-based measures of obesity may both underestimate and overestimate adiposity, thereby providing insufficiently nuanced information about health at the individual level (Busetto et al., 2024). The increase in adipose tissue mass in people with obesity is not limited to its function as an energy reservoir but involves a range of related endocrine and metabolic activities (Ghaben and Scherer, 2019). Moreover, central obesity, which is an excess of abdominal fat, is considered to be a key trigger of IR (Czech, 2017). Adipocytes in people with obesity can release large amounts of adipokines, which in turn interfere with the insulin signaling pathway, leading to the progression of IR (Czech, 2017). Additionally, obesity is often accompanied by systemic chronic inflammatory response, with inflammatory factors like tumor necrosis factor-α and interleukin-6 exacerbating IR and promoting cardiovascular disease (Liu et al., 2020). To better assess IR, the TyG index has been widely used to reflect triglyceride and FBG metabolism, and it provides a simpler and more accurate assessment tool for IR (Simental-Mendía et al., 2008; Muniyappa et al., 2008; Mohd Nor et al., 2016). However, the current TyG index calculation formula fails to fully capture the role of obesity in IR and cardiovascular disease. Although traditional obesity indicators like BMI are simple and easy to use, they do not factor in the distribution of fat in different parts of the body (e.g., the neck and waist), which differently affects the metabolism and cardiovascular diseases (Er et al., 2016). Therefore, effective assessment of IR and redefining obesity are crucial for reducing cardiovascular disease risk.

Importantly, insulin resistance severity is significantly influenced by adipose tissue distribution and total adiposity load (Pulsford et al., 2025; Packer et al., 2025). Visceral adipose tissue, in particular, contributes to systemic insulin resistance through multiple mechanisms including increased free fatty acid release, pro-inflammatory cytokine secretion, and decreased adiponectin production (Pulsford et al., 2025; Packer et al., 2025). Based on this pathophysiological understanding, we introduced modified TyG indices in our study, including TyG-BMI, TyG-NC, TyG-WC, TyG-NHtR, and TyG-WHtR indices, to more comprehensively evaluate the metabolic risk profile in OSA patients with hypertension. By multiplying TyG by anthropometric indices, these composite markers capture what can be conceptualized as an “adiposity-amplified insulin resistance phenotype”, where the degree of insulin resistance is proportionally weighted by the adiposity burden that drives and perpetuates the metabolic dysfunction (Lee et al., 2023; Kuang et al., 2023). The clinical utility of this approach has been extensively validated through multiple large-scale studies. Er et al. first demonstrated that TyG-BMI showed superior predictive capacity for insulin resistance compared to TyG alone, establishing it as a simple and clinically useful surrogate marker for insulin resistance in nondiabetic individuals (Er et al., 2016). Building on this foundation, subsequent cardiovascular outcome investigations confirmed that TyG-BMI provides significantly enhanced discrimination, with studies consistently showing that modified TyG indices outperform the traditional TyG index in predicting insulin resistance by accounting for body fat distribution and obesity’s contributory role (Ramírez-Vélez et al., 2019; Li et al., 2024). Recent prospective research by Xuan et al. demonstrated the superior performance of TyG-WHtR over TyG, WHtR, TyG-WC and TyG-BMI in predicting diabetes development (Xuan et al., 2022), while Park et al. further validated that TyG-BMI, TyG-WC, and TyG-WHtR were associated with increased risk of new-onset cardiovascular disease in participants without diabetes (Park et al., 2024). The prominence of TyG-BMI in our SHAP analysis actually validates rather than contradicts this comprehensive assessment strategy. Unlike standalone BMI, which provides only static anthropometric information, TyG-BMI represents a composite biomarker that integrates metabolic functionality (insulin resistance as reflected by triglyceride-glucose interaction) with adiposity burden (BMI). This combination captures the dynamic interplay between metabolic dysfunction and body composition that characterizes high-risk OSA-hypertension phenotypes. Mechanistically, our study demonstrates that TyG-BMI functions as a “metabolic amplifier” in OSA-hypertension patients, where the combination of insulin resistance and adiposity creates exponentially increased cardiovascular risk rather than simply additive effects. The superior predictive performance of TyG-BMI over isolated BMI in our models reflects this synergistic relationship: OSA-induced intermittent hypoxia simultaneously exacerbates insulin resistance (captured by elevated TyG values) and promotes visceral adiposity accumulation (reflected by BMI), creating a multiplicative cardiovascular risk profile that traditional anthropometric measures fail to capture. This mechanistic insight translates directly into clinical practice through our comprehensive risk stratification framework. TyG-BMI, integrated alongside our multi-dimensional feature set including sex, age, metabolic biomarkers and sleep parameters, enables identification of high-risk patient subgroups that would be missed by BMI-only assessment strategies. Patients with elevated TyG-BMI values identified by our predictive model represent a distinct high-risk phenotype requiring comprehensive metabolic assessment and intensive cardiovascular risk reduction strategies, moving beyond traditional BMI-based obesity management toward personalized metabolic risk stratification. This validates our study’s core premise: effective cardiovascular risk prediction in OSA-hypertension patients requires integrated evaluation of metabolic, anthropometric, demographic, and AHI rather than reliance on any single measurement, thereby supporting the superiority of our holistic machine learning-driven approach over traditional BMI-centric risk stratification methods.

Although obesity affects approximately one in eight individuals globally, no universal consensus has yet been reached regarding its classification and definition (Rubino et al., 2025). Moreover, the incidence of OSA in people with obesity can be as high as 30% (Rubino et al., 2025). OSA is a prevalent sleep disorder, wherein patients experience intermittent hypoxemia and sleep disruption due to recurrent upper airway obstructions during sleep (Yeghiazarians et al., 2021). Hypoxemia may lead to excessive activation of the sympathetic nervous system, thereby elevating blood pressure and impairing blood pressure control (He et al., 2020). Concurrently, fragmented sleep can induce hormonal imbalances, further exacerbating insulin resistance (IR) and metabolic disorders (Yeghiazarians et al., 2021). While multidimensional data offer a more comprehensive insight into patient status, information overload may complicate clinical decision-making. Currently, machine learning techniques can effectively elucidate the relationships between potential risk factors and prognosis by leveraging multidimensional data (Krittanawong et al., 2017; Wang et al., 2024a; Wang et al., 2024b; Yao et al., 2024; Zhu et al., 2025; Zhang and Li, 2024). This approach not only enhances the precision of risk assessment but also enables the visualization of associations between individual variables and long-term outcomes. By applying machine learning methods, we can identify critical determinants of adverse long-term outcomes and quantify their impact with greater accuracy (Wang J. et al., 2025; Krittanawong et al., 2017; Wang et al., 2024a; Wang et al., 2024b; Yao et al., 2024; Zhu et al., 2025; Wang Y. et al., 2025). Such visualization aids in a deeper understanding of the roles of various variables, thereby assisting clinicians in developing more personalized treatment plans that account for each patient’s unique circumstances. Furthermore, machine learning can uncover complex relationships that traditional statistical methods may overlook, providing a more robust foundation for clinical decision-making (Wang J. et al., 2025; Krittanawong et al., 2017; Wang et al., 2024a; Wang et al., 2024b; Yao et al., 2024; Zhu et al., 2025; Wang Y. et al., 2025). To more effectively evaluate cardiovascular disease risk and address obesity in patients with OSA and hypertension, our study employs various machine learning techniques to elucidate the relationships between potential risk factors and prognosis. Our model enhances the precision of risk assessment and enables the visualization of associations between individual variables and long-term outcomes.

Notably, obesity is recognized as a chronic, systemic disease characterized by excessive fat accumulation (Rubino et al., 2025). This new perspective underscores the intricate and multifaceted nature of obesity, highlighting its profound impact on individual health. It involves a complex interplay of various determinants that collectively contribute to its development (Rubino et al., 2025). In our study, leveraging advanced machine learning technologies, we can more accurately analyze and predict health risks associated with obesity and metabolic status, identify potential comorbidities, and develop personalized intervention strategies. Based on this comprehensive approach, we aim to develop novel and evidence-based strategies and solutions for the prevention and management of obesity. In clinical practice, these indices are easy to calculate and reproducible and can be obtained from routine physical examination data. Regular monitoring of these indices facilitates the early identification of high-risk patients, enabling timely personalized interventions. Moreover, our data provide a more nuanced understanding of obesity than the current definition, emphasizing the significance of identifying subtle differences between individuals and guiding more precise treatment for patients having OSA with hypertension.

We developed a novel predictive model that integrates diverse machine learning algorithms with multidimensional datasets to identify individuals at high risk of obesity-associated long-term MACCEs. Furthermore, we redefined the concept of obesity through a TyG index that incorporates body shape and fat distribution parameters, thereby enabling a more comprehensive assessment of insulin resistance and cardiovascular risk stratification.