The global prevalence of diabetes is increasing due to changes in diet and lifestyle. In South Korea, 15.5% of adults aged ≥ 30 years had diabetes and 41.1% had prediabetes in 2021~2022, highlighting a critical window for early lifestyle-based interventions [1]. Prediabetes is clearly distinct from normal glycemic states and is characterized by increased insulin resistance and impaired glucose regulation, which already heighten the risk for both microvascular and macrovascular complications [2]. Without proactive management at the prediabetes stage, individuals may experience decreased quality of life and increased personal and societal healthcare expenditures [3,4].

For individuals with prediabetes, lifestyle modifications such as dietary control, physical activity, and weight reduction are considered the most effective preventive strategies, rather than pharmacological interventions [5]. The American Diabetes Association (ADA) recommends that adults with prediabetes engage in at least 150 minutes of moderate-intensity aerobic exercise per week and achieve weight loss, as such behavioral improvements enhance insulin sensitivity and directly contribute to better glycemic control [6]. Meanwhile, the practice of self-care behaviors is closely linked to individual capacities, such as educational attainment and health literacy [7]. In fact, individuals with higher health literacy tend to acquire necessary information for diabetes management more effectively and translate it into consistent self-care behaviors [8]. Therefore, nursing strategies aimed at diabetes prevention should extend beyond general education and instead focus on identifying levels and patterns of self-care among individuals, enabling the development of tailored interventions suited to specific needs.

While the significance of self-care in diabetes prevention has been well established in prior studies, a clear theoretical framework is crucial for a systematic understanding of self-care behaviors. Orem's Self-Care Deficit Nursing Theory offers a comprehensive explanation of health management, focusing on the interaction between self-care demands, self-care agency, and self-care behaviors. The theory defines a self-care deficit as a mismatch between an individual's ability to perform self-care and the necessary actions required to maintain health [9]. Research using this theory in chronic disease management, including diabetes and hypertension, has shown that individuals with higher self-care agency are more likely to adopt health-promoting behaviors, such as maintaining a healthy diet and engaging in regular physical activity, which ultimately lead to improved health outcomes [10,11].

Although prediabetes is a stage preceding the onset of disease, it is marked by increasing insulin resistance and a heightened need for self-care [2]. During this critical period, self-care behaviors among adults with prediabetes can be conceptualized not merely as the presence or absence of action, but as typologies emerging from the complex interplay of various factors, as explained by Orem's Self-Care Theory. Classifying these behavioral patterns provides insight into why certain subgroups, despite being at high risk, fail to adequately engage in self-care—reflecting a mismatch between self-care demands and behaviors. Moreover, such typological classification facilitates the design of tailored nursing interventions that are aligned with the specific characteristics of each subgroup [12]. Thus, Orem's theory serves as an appropriate and robust conceptual framework for analyzing and typifying the multifaceted self-care behaviors observed in adults with prediabetes.

Recently, interest in prediabetes prevention and management has gained momentum, as evidenced by a bibliometric analysis highlighting escalating global research trends in this area [13]. However, existing studies predominantly rely on traditional statistical methods, providing fragmented analyses of influencing factors [14]. Such approaches often fail to uncover latent heterogeneity in complex health behavior data or to elucidate individualized self‑care typologies. Moreover, there is a notable lack of theory-based investigations that could contribute to the systematic conceptualization of self‑care behaviors. Consequently, a conceptually grounded framework is needed to advance our understanding and classification of self‑care patterns in prediabetes.

This study aims to scientifically classify self-care typologies among Korean adults with prediabetes by applying an unsupervised machine learning technique that captures nonlinear interactions within complex data, using the large-scale, nationally representative Korea National Health and Nutrition Examination Survey (KNHANES) dataset [15]. By integrating Orem's Self-Care Theory into the exploration of typologies, this research enhances its theoretical validity and seeks to provide evidence for personalized nursing interventions applicable in clinical practice. Therefore, the objective of this study is to identify and interpret self-care patterns among adults with prediabetes using exploratory clustering techniques based on machine learning, and to contextualize the results within the framework of Orem's theory. The findings are expected to serve as foundational data for developing tailored diabetes prevention strategies according to distinct self-care typologies.

Using data from the 9th cycle, 2nd year of the KNHANES (2023), this study conducted unsupervised machine learning-based clustering analysis on prediabetic adults, focusing on Orem's self-care components—self-care demand, self-care agency, and self-care behaviors. Based on the relative levels of these attributes, three distinct self-care profiles were identified and labeled as the High Self-Care Performance Group, the Latent Self-Care Risk Group, and the Self-Care Vulnerable Group. The following discussion outlines tailored nursing intervention strategies for each group.

The High Self-Care Performance Group showed low levels of self-care demand, yet high levels of education, income, and health literacy. This group consistently engaged in recommended self-care behaviors, including aerobic exercise, muscle-strengthening exercise, and dietary control. Interventions for this group should focus on maintaining current behavior while preventing burnout or motivational fatigue. Garner et al. [26] found that volunteers who participated in a diabetes lifestyle mentoring program experienced improved dietary habits, reduced sedentary time, and significantly enhanced diabetes-specific self-efficacy. Therefore, members of this group may also benefit from serving as peer mentors, which can reinforce self-care among mentees while simultaneously helping to prevent burnout in mentors.

The Latent Self-Care Risk Group was characterized by low self-care demand, relatively high educational attainment, but lower income, and moderate health literacy. Despite having adequate informational capacity, individuals in this group exhibited low adherence to self-care behaviors, particularly in muscle-strengthening exercise. These findings suggest that merely providing knowledge is insufficient; interventions should also include practical training in behavioral skills. Rahayu [27] highlighted the mediating role of self-efficacy between diabetes knowledge and self-care behavior, underscoring the need to enhance self-efficacy. Effective strategies may include goal-setting, practice and tracking of self-management skills, peer modeling, and positive reinforcement [28].

The Self-Care Vulnerable Group had the highest self-care demand, but also the lowest levels of education, income, health literacy, and behavioral adherence. These individuals may struggle with self-care due to limited information processing capacity and constrained social resources. Educational interventions tailored to their level of health literacy, hands-on training to facilitate behavior change, and support systems involving family and community networks are essential [29,30]. Given that this group is also affected by social determinants of health, such as low income and limited access to care [31], structural support through community-based and national policy-level interventions is crucial.

Taken together, the findings demonstrate that self-care levels among prediabetic adults are shaped not by a single variable, but by the interaction of multiple factors. Notably, general demographic variables such as sex, age, and marital status did not significantly differ across groups, indicating that self-care is more directly influenced by health literacy, educational attainment, behavioral capability, and self-care demand. The limited self-care capacity observed in the vulnerable group may be due to restricted access to information and education, which reinforces Orem's assertion that nursing plays a critical role in supporting individuals in meeting their self-care demands [32]. This study contributes to theoretical development by demonstrating the need for group-specific interventions at the prediabetic stage.

Methodologically, this study employed unsupervised clustering, a machine learning technique that identifies hidden patterns in data without pre-labeled outcomes. This allowed for the discovery of latent self-care types that may be overlooked in traditional variable-driven analyses, enabling early identification of at-risk individuals in clinical practice [33]. Because the core variables used in the clustering process can be collected via simple questionnaires, the resulting model has the potential to be operationalized as a quick screening tool in clinical and community settings. From an academic perspective, this study contributes to the development of refined, tailored preventive nursing interventions by classifying individuals based on integrated dimensions of self-care demand, agency, and behavior.

Nevertheless, the study has some limitations. The analysis did not incorporate the complex sampling design of the KNHANES dataset, and, as an exploratory clustering study, it did not utilize advanced machine learning features, such as hyperparameter tuning. Still, the Orange software used is known for its accuracy and user-friendly interface, allowing non-programmers to perform intuitive clustering analysis suitable for exploratory research [34]. Future research should aim to replicate and validate these findings, develop predictive tools for cluster classification, evaluate the effects of interventions across identified groups, and compare alternative machine learning algorithms to enhance generalizability and practical application.

Grounded in Orem's Self-Care Theory, this study examined the self-care characteristics of adults with prediabetes, focusing on three core components: self-care demand, self-care agency, and self-care behavior. Using an unsupervised learning approach—specifically PCA-K-means clustering—participants were categorized into three distinct groups: the High Self-Care Performance Group, the Latent Self-Care Risk Group, and the Self-Care Vulnerable Group. The findings revealed that differences in self-care agency and behavior were more pronounced than those based on sociodemographic characteristics. These results support Orem's theoretical assertion that nursing plays a facilitative role in enhancing individuals' capacity to meet their own self-care demands. Moreover, the findings demonstrate that self-care capacity is not determined by a single factor, but rather by the complex interaction of multiple dimensions, underscoring the need for tailored nursing interventions. Importantly, by applying an unsupervised machine learning technique, the study was able to uncover latent subgroups that may have been overlooked in prior research, highlighting the potential for early intervention in both clinical and community health settings. Based on these findings, it is recommended that adults with prediabetes be clustered using simple biological indicators such as blood glucose and HbA1c in combination with basic self-care variables, to facilitate the development and application of tailored nursing interventions for each subgroup. Nonetheless, this study has limitations. The complex sampling design of the KNHANES was not fully accounted for, which may limit the generalizability of the findings. In addition, the clustering analysis remained exploratory, without extending to model validation or optimization. Future research should aim to replicate these findings using diverse datasets, develop predictive indicators for self-care typologies, evaluate the effectiveness of targeted nursing interventions, and explore the comparative utility of various machine learning techniques to enhance both practical applicability and generalizability.