The Clustering of Lifestyle Risk Factors in the Serbian Adult Population and Association with Self-Rated Health

Non-communicable diseases (NCDs), including cardiovascular diseases, cancers, chronic respiratory diseases and diabetes, have been the leading causes of death and disability worldwide. The four major lifestyle risk factors (LRFs) for NCDs are smoking, unhealthy diet, physical inactivity and harmful use of alcohol (1). Since these LRFs are modifiable, an understanding of their determinants and interrelationships is important from a health promotion perspective. However, NCDs are also shaped by a broader set of determinants, including genetic predisposition, environmental exposures and climatic influences, which interact with behavioural risks (2, 3). While these factors are not directly addressed in this study, their relevance should be acknowledged as they provide a more comprehensive framework for understanding the multifactorial origins of NCDs.

If a pattern of LRFs occurs more frequently than would be expected based on marginal prevalence rates, it results in a clustering of these risk factors (4). The presence of multiple LRFs is associated with a significantly increased likelihood of adverse health outcomes, such as the development of chronic conditions, comorbidities, disabilities and premature mortality (5, 6). Furthermore, the clustering of health behaviours is socially patterned, with education, socio-economic and demographic characteristics including rural-urban differences and psychosocial stressors shaping lifestyle trajectories (7, 8). This underlines the importance of examining sociodemographic correlates of clusters, not only as a practical means of targeting interventions but also as a way to illuminate broader mechanisms of health inequalities.

While cluster analysis itself does not “identify risk factors,” it enables recognition of patterns of co-occurrence among existing behaviours, offering insights into population subgroups. Better insight into the socio-demographic characteristics of clusters could be useful in identifying vulnerable groups that would benefit most from public health interventions (9). The latest research evidence from Europe, including the Netherlands (10), Poland (11), Spain (12, 13), Norway (14), and Bosnia and Herzegovina (15), has consistently shown that lifestyle risk behaviours cluster differently across cultural and socioeconomic contexts. Such international variation highlights the importance of studying these phenomena in diverse national settings.

Besides this, LRFs are often associated with self-rated health. Self-rated health (SRH) is a validated measure of overall health that closely aligns with objectively assessed health status (16) and serves as a strong predictor of morbidity and mortality (17, 18). It reflects individuals’ overall perception of their physical, mental and social well-being (19, 20). SRH is extensively utilised in population-based surveys as a marker of current health status and functional capability (20). SRH correlates closely with multidimensional measures of health-related quality of life (HRQoL) and has been employed across multicountry and multicentre studies and reviews to monitor population health and identify vulnerable groups (21). However, interpreting SRH requires careful consideration, as cultural factors can significantly affect how individuals evaluate and report their health (22). Previous research found that co-occurrence of multiple LRFs was associated with worse SRH (12, 23, 24).

When examining contextual factors in the context of Serbia, it is essential to note that there are significant regional disparities not only in GDP per capita but also in other areas, e.g. educational attainment, between Belgrade and Vojvodina and the other two regions: Šumadija and Western Serbia, as well as Southern and Eastern Serbia (25). Rural population share is considerably higher in Šumadija and Western Serbia (~55%), Southern and Eastern Serbia (~52.5%) and Vojvodina (43.3%) compared to Belgrade (~19%) (26).

A better understanding of the risk factors for NCDs is essential, as these conditions represent a significant public health concern in Serbia. According to data from the 2019 National Health Survey, half of the Serbian population reported having at least one NCD. The most common conditions were arterial hypertension (29.6%), hyperlipidemia (10.8%) and coronary heart disease/angina pectoris (9.8%) (27). In addition, multimorbidity presents a significant public health issue, affecting 12% of the working-age population, with higher prevalence among vulnerable groups, including women, individuals with lower levels of education and those with lower incomes (28). Additionally, the substantial prevalence of smoking, alcohol consumption, physical inactivity and unhealthy diet, with regional differences, was reported (27). Yet, to date, there has been limited empirical investigation of how these risks cluster within the Serbian population, despite comparable evidence from other European and middle-income countries (23, 24, 29,30,31). Exploring lifestyle patterns in Serbia is particularly relevant given the country’s high burden of NCDs and ongoing health system reforms aimed at prevention.

As of the latest assessments, there is no currently valid strategic document in the Republic of Serbia that comprehensively addresses tobacco control, the prevention of non-communicable diseases, or national programmes targeting biological risk factors such as obesity, nor behavioural risk factors such as insufficient physical activity and excessive alcohol consumption (32, 33). Moreover, Serbia has not fully implemented a comprehensive national tobacco control strategy since 2016; according to the WHO report, despite drafting the Strategy for Tobacco Control 2016–2025, it was never formally adopted and remains without a cohesive action plan at the national level (34, 35). This is particularly concerning given that in regions such as Vojvodina, the adult smoking prevalence now approaches 36%, among the highest in Europe, especially in rural areas, lower-educated populations and among younger adults (33).

Unfortunately, there is a lack of research providing information about lifestyle patterns among Serbian adults and the association between SRH and multiple LRFs. Aiming to prepare an evidence-based foundation for the implementation of effective public health measures, the following objectives were set: 1) to identify clusters of risk behaviours related to NCDs in the Serbian adult population, 2) to examine the association between socio-demographic characteristics and these clusters, and 3) further assess the association of the identified clusters with SRH.

Our objective was to identify population groups exhibiting multiple LRFs to determine where targeted public health interventions might be most needed. According to the findings, participants were classified into five different clusters: a cluster with no LRFs (cluster 1), two clusters with one or two LRFs (cluster 2, cluster 3), and two multi-risk groups (cluster 4, cluster 5). These clusters are similar to those identified in previous studies (10, 29, 38). Researchers utilising cluster analysis have identified patterns that closely match ours, clustering of smoking, unhealthy diet and physical inactivity (9, 10, 23, 38), and clustering of risky drinking, smoking, unhealthy diet and physical inactivity (9, 10, 23, 29, 30, 38, 39).

In line with existing literature, cluster 1 (Healthy lifestyle) comprised mostly females and individuals with higher levels of education, as they tend to live a healthier lifestyle concerning smoking and alcohol consumption (10). Two high-risk groups (cluster 4, cluster 5) exhibit specific socio-demographic characteristics (males, young and middle-aged, living alone, less educated), in line with existing literature (7, 11, 38, 39).

Consistent with previous studies (9, 15, 23, 24, 31, 40), younger adults were more likely to be in the unhealthy lifestyle clusters. These results may be attributed to a lack of awareness regarding NCDs in young people or selection bias, as individuals with a higher risk of exhibiting two or more risk factors may not reach older age due to premature death (31).

Single individuals had a higher probability of showing one LRF (cluster 3) or multiple LRFs (cluster 4, cluster 5) than married individuals. These findings were consistent with previous research showing an association between living alone and less favourable health behaviours (7, 9, 40). A possible explanation for this association could be that marital status has a protective effect on health status by maintaining a healthy lifestyle through social, psychological and economic support between married people (41).

Compared to the healthy lifestyle cluster, adults in cluster 2 (Low fruit and vegetable intake) and multi-risk clusters 4 and 5 were more likely to have a lower education level (primary or secondary school) than those with college or university education. These results go beyond previous reports, showing that lower education was significantly associated with higher odds of one or multiple LRFs (7, 11, 14, 15, 30, 40, 42). People with lower educational backgrounds are more likely to have a smoking lifestyle (9) and risky alcohol consumption, which was partly linked to a higher level of stress due to material limitations, although there was no evidence suggesting that these groups cope with stressful situations less effectively (43). Education is linked to future employment and income opportunities, as well as to improvements in health literacy and knowledge that contribute to better health-related behaviours (44).

In line with previous research, men were more likely to be in an unhealthy diet lifestyle cluster (15, 38), whereas women were more likely to be in a physically inactive cluster (8, 15). Individuals in multi-risk clusters 4 and 5 were more likely to be men, which is supported by the other studies (7, 45). In addition, women were more likely to engage in healthier lifestyle behaviours and more frequently adhered to recommendations regarding diet, tobacco use, and alcohol consumption compared to men, and this may be a result of social role differentiation (14).

Additionally, we identified a significant association between geographic regions and lifestyle clusters. Individuals in cluster 2 (Low fruit and vegetable intake) and multi-risk clusters 4 and 5 were more likely to live in one of the other three regions of Serbia than in the Belgrade region. Compared to the other three regions of Serbia, Belgrade has the smallest percentage of its population living in rural areas. In addition, the Belgrade region has the highest employment rate, the highest gross domestic product and the highest proportion of well-educated working-age people. At the same time, these indicators are lowest in Southern and Eastern Serbia. This may partly explain the clustering of unhealthy behaviours outside the capital, as confirmed by Serbian studies showing that sufficient intake of fruit and vegetables is strongly associated with higher education and better socioeconomic status, more common in urban areas (46, 47). Our results are similar to the findings of Cissé et al., who demonstrated that lower consumption of fruits and vegetables and higher alcohol consumption are more prevalent in rural areas (45).

Furthermore, we investigated the association between SRH and lifestyle clusters. Unlike disease-specific measures, SRH captures an individual’s overall perception of health, integrating physical, mental and functional dimensions. As such, it is particularly well-suited for examining associations with lifestyle-related patterns. Within the context of health inequalities, SRH also represents a useful indicator for identifying the need for targeted policies and interventions aimed at addressing disparities across population subgroups defined by sex, age, educational level, place of residence and other socio-demographic factors (27). Accordingly, the association between SRH and lifestyle clusters was adjusted for socio-demographic variables. Numerous studies conducted among the adult population in Serbia have used SRH to assess socio-economic inequalities in health (48), to examine changes in population health during the transition period (49), and to explore the association between various demographic, socio-economic factors and LRFs, such as smoking and alcohol consumption (50). Our results provide further evidence that clustering of simultaneous LRFs is associated with poor SRH. Adults belonging to unhealthy clusters (cluster 3–5) were more likely to report their health as poor than adults in the healthy cluster (adjusted regression model). The negative relationship between cluster 5 (Risky drinking and other LRFs) and poor SRH in the unadjusted model becomes positive in adjusted models, resulting from age and gender adjustment. Individuals in cluster 5 are mostly younger and male, probably less concerned about the harmful consequences of unhealthy habits, and the negative association in the unadjusted model was confounded by their better health, in line with a prior study (51). Some studies have further suggested that good health may actually predict alcohol consumption, rather than alcohol consumption predicting health outcomes (52). Some commonly used methods for examining risk factor patterns include co-occurrence analysis, composite score methods, counting the number of LRFs or unsupervised learning techniques (e.g., cluster analysis, latent class analysis and factor analysis), which are designed to identify latent structures within the data (42). Recent research comparing Latent Class Analysis and Two-Step Cluster Analysis has confirmed that Two-Step Cluster Analysis can generate stable and valid clusters, particularly when applied in fields with large sample sizes and heterogeneous data types (53). Additionally, the clustering of LRFs, obtained by using Two-Step Cluster Analysis, remained stable over time (10).

Our findings are consistent with recent studies highlighting substantial regional disparities in health behaviours and outcomes within Serbia and the Western Balkans. For example, a spatio-temporal analysis of hypertension in the Autonomous Province of Vojvodina identified significant regional variations, with certain settlements exhibiting higher rates, suggesting the need for targeted public health interventions in these areas (54). Similarly, a geographic overview of NCDs in Vojvodina, using cluster analysis and the Getis-Ord Gi* method, revealed that specific counties were more burdened by cardiovascular diseases and diabetes, emphasising the importance of region-specific health strategies (55). Beyond Serbia, cluster analysis of health indicators across Western Balkan countries has demonstrated the influence of healthcare infrastructure, socioeconomic conditions and regional disparities on health outcomes, grouping countries into clusters with distinct health profiles (56). These findings collectively underscore the value of employing advanced statistical methods to uncover regional patterns of health behaviour and disease burden, informing more efficient allocation of resources and tailored public health interventions.

The study has some limitations. Information about health behaviours was self-reported, which has well-known disadvantages (e.g., recall bias, social desirability, under-estimation, or over-estimation). Due to the cross-sectional nature of the data, no conclusions about the causality of the relationship between clusters of LRFs and SRH can be drawn from the current study. In addition, SRH may be influenced by an individual’s temporary or short-term health status (20) and other broader groups of contextual indicators. Our study included participants who provided complete information on the four LRFs included in cluster analysis. While complete-case analysis (CCA) may introduce bias in small-sample contexts with high missingness, existing evidence suggests that in larger samples, CCA can remain stable and yield results similar to other imputation methods (57), though the limitations must be acknowledged. Missing data were more prevalent among older participants and those with lower educational attainment; both were accounted for as covariates in the multivariable models, thereby enabling partial statistical adjustment for potential bias. Binary variables were used in the cluster analysis to ensure comparability with previous studies; however, this approach did not allow for capturing variation in the intensity or quantity of the behaviours (10). SRH was also dichotomised to ensure comparability with previous studies (12, 23, 20, 51).

However, the limitations of the study should be balanced against its strengths, such as the source of data: a large population-based sample. To the best of our knowledge, it was the first study focusing on the patterning of LRFs in the Serbian adult population by using cluster analysis and identifying high-risk population groups for NCDs. Finally, as far as we know, this is the first study examining the association between multiple LRFs and SRH in the Serbian adult population.

This research shows that in Serbia, the clustering of lifestyle risk factors is more common among men, younger adults, people living alone, those in regions outside of Belgrade, and individuals with lower education levels. People with multiple concurrent lifestyle risk factors were more likely to report worse self-rated health. These findings emphasise the need to go beyond single-risk-factor approaches in public health and instead prioritise multiple health behaviour change strategies. High-risk population groups identified in this study should be prioritised in designing and implementing such interventions. However, current evidence on the best ways to modify multiple LRFs at the same time remains limited. More research is necessary to develop integrated, inclusive, and context-specific strategies that can sustainably improve health behaviours and reduce health inequalities across different socio-demographic groups.