Syndemic analysis of stroke in Indonesia

Summary

Background

Stroke is the first leading cause of mortality in Indonesia, yet no national study has applied a comprehensive syndemic approach that integrates co-occurring diseases and socio-economic conditions to assess the determinant of this cerebrovascular disease. To address this gap, we analyzed the ten most prevalent clinical conditions and major socio-economic indicators to identify the interacting factors that contribute to stroke incidence.

Methods

We analyzed national health insurance data from over 12 million patients, integrated with provincial socio-economic indicators. All variables were aggregated and analyzed at the provincial level from 2018 to 2023. Multiple advanced statistical approaches were leveraged to identify syndemic patterns.

Findings

Stroke, diabetes mellitus (DM), chronic kidney disease (CKD), and poverty showed consistent increases across Indonesian provinces. Correlation analysis identified strong associations between stroke and CKD, DM, and cirrhosis, with poverty demonstrating a moderate correlation. Bayesian Gaussian Network Analysis indicated that stroke was the most probable downstream outcome (51.85%) within this interconnected system. Path analysis showed that DM had strong direct associations with CKD (75.88%) and cirrhosis (61.27%), indicating a major upstream role of diabetes in the network. CKD in turn showed substantial direct associations with both poverty (48.67%) and stroke (19.18), while poverty and cirrhosis also demonstrated indirect associations with stroke. Geographically and Temporally Weighted Regression revealed marked spatial and temporal heterogeneity in the strength of these determinants across provinces. Principal Component Analysis consistently clustered several provinces in high-burden syndemic profiles.

Interpretation

Stroke in Indonesia emerges as the convergent outcome of interconnected metabolic, hepatic, and socio-economic factors, rather than isolated clinical conditions. These findings underscore the importance of a syndemic framework to effectively reduce the national stroke burden.

Funding

This study was funded by Indonesian Ministry of Higher Education, Science and Technology's BIMA Research Program under grant number 050/E5/PG.02.00.PL/2024 and 02035/UN4.22.2/PT.01.03/2024.

Research in context.

Evidence before this study

We searched PubMed using the terms (“syndemic” AND (“stroke” OR “chronic kidney disease” OR “diabetes mellitus” OR “cirrhosis” OR “poverty”)). Existing research demonstrates increasing use of syndemic frameworks for non-communicable diseases (NCDs), showing that co-occurring metabolic, infectious, and social conditions synergistically amplify disease severity, particularly in socio-economically disadvantaged populations. However, evidence from low- and middle-income countries remains limited.

In Indonesia, only two published studies have applied a syndemic framework, focusing on HIV populations and non-fatal drug overdose. We found no prior studies that have examined stroke within a syndemic structure integrating CKD, diabetes, cirrhosis, and poverty, nor any that have mapped their spatiotemporal dynamics nationally or applied Bayesian network modelling, path analysis, GTWR, and PCA to uncover their joint determinants.

Added value of this study

This study is the first to combine national health insurance data from more than 12.5 million patients with provincial socio-economic indicators to characterise the syndemic architecture of stroke across all 34 Indonesian provinces (2018–2023). Through correlation analysis, we identified CKD, DM, and cirrhosis as the strongest clinical correlates of stroke, with poverty emerging as the key structural determinant. Using Bayesian Gaussian Network Analysis, we show that stroke is the most probable downstream outcome within this interconnected system. Path analysis further reveals the hierarchical organisation of these relationships: DM drives CKD and cirrhosis, CKD contributes to poverty, and CKD, cirrhosis, and poverty each exert independent effects on stroke.

Geographically and Temporally Weighted Regression uncovers substantial spatial and temporal heterogeneity in these relationships, including provinces where structural disadvantage outweighs clinical factors. PCA identifies stable multi-year clusters of provinces with shared syndemic profiles, separating high-burden provincial groupings (Domination of DM–CKD–cirrhosis–poverty interactions) from low-burden regions with more favourable structural and metabolic characteristics.

Implications of all the available evidence

The interconnected rise of metabolic disease, chronic liver disease, and poverty underscores the need for a syndemic-oriented stroke prevention strategy in Indonesia. Policies focused solely on single risk factors are unlikely to shift population-level stroke burden without addressing upstream structural drivers such as poverty and regional inequities in healthcare access.

Our findings demonstrate the value of incorporating probabilistic modelling and spatially explicit regression into national surveillance systems to detect high-burden clusters, anticipate evolving risk patterns, and guide resource allocation. This approach is applicable to other low- and middle-income countries undergoing similar epidemiological and socio-economic transitions, where multi-disease interactions and structural vulnerability shape cerebrovascular outcomes.

Introduction

Indonesia faces a growing and complex disease burden shaped by the simultaneous rise of non-communicable (NCDs) and communicable diseases (CDs), socio-economic disparities, and uneven regional development.¹ As the fourth most populous country in the world, Indonesia’s vast geographic and demographic diversity presents significant challenges for health system planning and disease prevention.^2,3 Indonesia reached a GDP of US$1.37 trillion in 2023, ranking 16th globally,⁴ while its life expectancy at birth remained at 71.3 years (in 2019), placing it 115th in the world.⁵ The interplay of these overlapping threats, epidemiological, social, and structural, requires a more integrated framework to understand how co-occurring diseases emerge, interact, and concentrate within vulnerable populations and regions.⁶

Stroke is the leading cause of mortality and disability nationwide, and its burden has risen sharply over the past decade and varies widely across provinces.⁷ Yet, despite its substantial impact, stroke is often examined through the lens of single risk factors, such as hypertension or diabetes mellitus (DM), without considering the broader constellation of interacting metabolic, infectious, hepatic, and socio-economic drivers that may jointly shape cerebrovascular vulnerability.⁸ This narrow perspective limits the ability to design interventions that effectively address the upstream determinants contributing to Indonesia’s persistent stroke burden.

A syndemic refers to the synergistic interaction of multiple epidemics that co-occur within specific populations and contexts, driven by harmful social conditions.⁹ While individual behaviors are often seen as direct causes of disease, syndemic theory emphasizes that more effective public health interventions must address the underlying social determinants that fuel both the clustering and interaction of diseases.¹⁰ In this regard, a syndemic approach offers a powerful tool to analyze this complexity, emphasizing how co-occurring diseases interact biologically and socially within contexts shaped by structural inequities such as poverty, inequality, and limited access to care.^6,11

For instance, one of the most widely cited syndemic study involves the interaction of HIV infection, tuberculosis, and substance use, where biological vulnerability is compounded by social marginalization and poverty, resulting in amplified morbidity and mortality.⁹ Similarly, a study utilizing a syndemic approach revealed how NCDs including obesity, hypertension, and DM are compounded by social vulnerabilities such as caste, education, and wealth disparities, leading to amplified disease burden beyond what would be expected from any single risk factor.^12,13 Furthermore, a recent conceptual work applying a syndemic framework emphasizes that disease clustering is not merely biological but shaped by broader social and structural conditions. Health problems interact within contexts of poverty, unequal access to care, limited institutional support, and social disadvantage. In such syndemic perspective, social structures do not simply coexist with disease but actively influence how health burdens accumulate and reinforce one another within vulnerable populations.¹⁴

However, both in Indonesia and global level, the interplay between clinical conditions and socio-economic determinants remains insufficiently integrated into national disease surveillance and analysis.¹ This gap limits understanding of how structural vulnerabilities shape disease emergence, progression, and clustering, particularly for high-burden conditions such as stroke.¹¹ There is a pressing need for analytical approaches capable of revealing how disease burdens evolve over time, intersect with social and economic marginalization, and manifest in regionally distinct patterns.

To address this gap, we conducted a multi-year analysis using national health insurance data covering over 12 million patients across all Indonesian provinces from 2018 to 2023. We integrated epidemiological and socio-economic indicators using a combination of advanced statistical approaches including correlation analysis, Bayesian Gaussian Network Analysis (BGNA), pathway analysis, Geographically and Temporally Weighted Regression (GTWR), and Principal Component Analysis (PCA). Through this syndemic-oriented framework, we aimed to uncover how co-occurring clinical and structural conditions interact, cluster, and evolve across Indonesia’s diverse regions to shape stroke incidence, thereby generating system-level insights to inform more targeted and equitable public health strategies.

Methods

Data sources and study population

This study was reviewed and approved by the Ethics Committee of Medical Research, Faculty of Medicine, Hasanuddin University (Approval Number: 857/UN4.6.4.5.31/PP36/2024). The requirement for individual patient consent was waived, as the study used de-identified administrative data obtained from BPJS Kesehatan and involved no direct contact with participants. All data were analyzed in aggregated form at the provincial level and analyzed as province-year units from 2018 to 2023. Health-related data were obtained from the Social Health Insurance Administration Body (BPJS Kesehatan), which manages Indonesia’s universal health coverage system and covers approximately 95% of the national population. Individuals were identified using the BPJS patient identification number, which allowed repeated healthcare visits and multiple claims to be linked to the same person.

BPJS Kesehatan collects administrative claims data from public and private healthcare facilities across all provinces, including primary care centres, district hospitals, and tertiary referral hospitals. Diagnoses are recorded using standardized clinical coding systems in accordance with national reporting guidelines, which enhances consistency across regions. As an administrative claims database, BPJS provides large-scale, routinely collected data with broad national coverage. However, healthcare utilization, diagnostic capacity, and provider availability may vary across provinces, particularly between urban and rural or remote regions. Such disparities may contribute to differential case detection and potential underdiagnosis or underreporting in underserved populations, especially in areas with limited access to healthcare services.

This study focused on the ten most reported diseases: stroke, chronic kidney disease (CKD), DM, schizophrenia, tuberculosis, anemia, hepatitis cirrhosis, hypertension, hepatitis B, and heart failure between 2018 and 2023. Socio-economic indicators were sourced from the Central Bureau of Statistics (Badan Pusat Statistik, BPS), Indonesia’s national agency for demographic and economic data. These included the percentage of the population living in poverty, percentage of smokers aged 15 years and older, realized deconcentration budget, and average daily per capita calorie intake. The detailed descriptions and calculation formulas for these variables are provided in the Supplementary Information.

These socio-economic indicators were selected based on their theoretical relevance to the syndemic framework and their consistent availability as standardized provincial-level data across the study period. Poverty rate was included to represent structural deprivation; smoking prevalence reflects behavioral cardiovascular risk; average daily per capita calorie intake captures nutritional and metabolic transition; and realized deconcentration budget serves as a proxy for provincial-level public resource allocation and governance capacity. All variables are available for each consecutive year (2018–2023) and were analyzed at the provincial level, resulting in a nationwide ecological study design. All statistical analyses were conducted using R software (R Foundation for Statistical Computing, Vienna, Austria).

Correlation study

To examine the dynamics of change, we first calculated the percentage variation in each variable from the baseline year (2018) to the final year (2023) for all provinces. These temporal patterns were visualized using a heat map to illustrate the relative increases or decreases across regions.⁷ To identify which factors were potentially involved in the syndemic structure of stroke, we subsequently performed a correlation analysis using pooled province–year observations from 2018 to 2023. Stroke incidence was correlated with all clinical and socio-economic variables at the provincial level. Disease variables that demonstrated strong correlations with stroke, and socio-economic indicators that showed moderate correlations, were selected for inclusion in the subsequent probabilistic network modelling and spatiotemporal analyses within the syndemic framework.

Syndemic analysis

Clinical variables that demonstrated strong correlations with stroke (CKD, DM, and cirrhosis) and socio-economic indicators with moderate correlations (particularly poverty) were selected as the core components of the syndemic framework. We then performed BGNA using the bnlearn package.¹⁵ BGNA is a probabilistic graphical modelling approach that estimates conditional dependencies among continuous variables and infers the most likely directional network structure linking them. By comparing alternative network configurations, this method allows identification of hierarchical relationships and determination of the most probable downstream (outcome) node within an interconnected system. In this analysis, alternative network configurations were constructed in which each variable (stroke, CKD, DM, cirrhosis, and poverty) was sequentially treated as the potential outcome node. The BGNA algorithm evaluates these competing structures and assigns a posterior probability to each configuration based on how well the network explains the observed data. A higher probability indicates that the corresponding configuration is more frequently supported by the data, allowing identification of the most plausible downstream node within the network.

To further quantify the structural (direct and indirect) associations among variables, path analysis was conducted with the lavaan package.¹⁶ This analysis is a structural equation modelling technique that estimates direct and indirect effects among predefined variables, thereby quantifying the magnitude of each predictor’s contribution to stroke and clarifying the plausible pathways of association linking metabolic, renal, hepatic, and socio-economic factors.

Following identification of the optimal response variable, we applied GTWR to evaluate spatial and temporal heterogeneity in predictor–outcome relationships across the 34 provinces from 2018 to 2023. GTWR is a local regression technique that extends traditional global models by allowing regression coefficients to vary simultaneously across geographic location and time. This approach estimates province-specific and year-specific parameter values, thereby capturing localized and dynamic variations in the strength of associations between CKD, DM, cirrhosis, poverty, and stroke. Analyses were conducted using the GWmodel package,¹⁷ in combination with sf, tmap, and classInt for spatial processing and visualization.

Dimensionality reduction was undertaken through PCA using the FactoMineR package,¹⁸ with visualization by factoextra. PCA is a multivariate statistical technique that transforms correlated variables into a smaller set of uncorrelated components (principal components) that capture the maximum variance within the dataset. In this study, PCA was applied to CKD, DM, cirrhosis, poverty, and stroke to identify latent dimensions representing the joint syndemic burden across provinces. The resulting principal component scores were subsequently used as input for K-means clustering to group provinces according to similarities in their multidimensional syndemic profiles.

Role of the funding source

The funding source had no role in the study design, data collection, data analysis, data interpretation, or writing of the report. The corresponding author had full access to all data in the study and had final responsibility for the decision to submit for publication.

Results

Spatiotemporal patterns of major diseases and socio-economic indicators across Indonesia provinces

In this study, we first examined temporal trends of the 10 most common diseases and key socio-economic indicators across Indonesian provinces from 2018 to 2023. This analysis was based on national health insurance data covering 12,540,322 patients who received care at healthcare facilities. As illustrated in Fig. 1, the incidence of CKD, DM, and stroke markedly increased over the six-year period in nearly all provinces. Notably, the magnitude of increase was generally greater in several provinces outside Java, whereas provinces in eastern Indonesia showed stable or slightly declining trends. Jakarta, the capital city, demonstrated only modest changes over the study period. Other diseases that showed cumulative increases nationwide included schizophrenia, tuberculosis, and anemia, although the magnitude of increase generally remained below 15%. Meanwhile, the incidence of hypertension, hepatitis B, and heart failure declined substantially in the majority of provinces.

Fig. 1.

Spatiotemporal percentage change in major diseases and socio-economic indicators across Indonesian provinces. Heatmap illustrating the percentage change from 2018 to 2023 for the ten most prevalent diseases (stroke, CKD, DM, schizophrenia, tuberculosis, anemia, cirrhosis, hypertension, hepatitis B, and heart failure) and key socio-economic indicators (poverty, smokers ≥15 years, realized deconcentration budget, and daily calorie intake) across all provinces. Red shades indicate increases, blue shades indicate decreases, with intensity proportional to magnitude of change. This figure highlights substantial regional heterogeneity, with particularly pronounced increases in CKD, DM, stroke, and poverty in several Sumatra, Kalimantan, and Sulawesi provinces, contrasted with more stable or declining patterns in parts of Java and eastern Indonesia.

In terms of socio-economic indicators, the percentage of the population living in poverty demonstrated a consistent upward trend at the national level, increasing by approximately 20% over the study period. This trend was particularly pronounced in provinces outside of Java, including Bengkulu, Lampung, West Kalimantan, Central Kalimantan, and East Nusa Tenggara. In parallel, average per capita calorie intake exhibited a declining trend across most provinces. Additionally, both the realization of deconcentration budget funds and the proportion of smokers aged 15 years and over showed modest decreases across the country, although the reductions were relatively limited and inconsistent across regions. To provide additional context on absolute trends, province-level temporal trajectories of disease incidence and socio-economic indicators from 2018 to 2023 are presented in Supplementary Fig. S1.

Correlation analysis of CKD, DM, stroke, and poverty

Prior to conducting a syndemic analysis of stroke, we first evaluated how closely each major disease and socio-economic factor was associated with this disease at the provincial level. For this reason, we performed a comprehensive correlation analysis (Fig. 2) involving all diseases and socio-economic indicators to determine which conditions showed the strongest relationships with stroke and were therefore most suitable for probabilistic network modelling and spatiotemporal analyses.

Fig. 2.

Correlation matrix of major diseases and socio-economic indicators across Indonesian provinces. Heatmap showing pairwise Pearson correlation coefficients among the ten most prevalent diseases and key socio-economic indicators. The strength of association is represented by color intensity, ranging from very low to very strong correlations. Significant correlations are marked with asterisks (∗∗∗∗p < 0.0001; ∗∗∗p < 0.001; ∗∗p < 0.01; ∗p < 0.05). This matrix informed variable selection by identifying clinical and socio-economic factors most closely associated with stroke, which were then carried forward into subsequent probabilistic network and spatiotemporal syndemic analyses. Correlation strength classification: 0.00–0.19 = very low. 0.20–0.39 = low. 0.40–0.59 = moderate. 0.60–0.79 = strong. 0.80–1.00 = very strong.

The correlation matrix demonstrated a distinct clustering of interrelated conditions across provinces. Stroke showed its strongest correlations with CKD (r ≈ 0.73), cirrhosis (r ≈ 0.63, strong), and DM (r ≈0.66), indicating a shared metabolic–inflammatory burden linking vascular, renal, and hepatic dysfunction. Heart failure exhibited a moderate correlation with stroke (r ≈ 0.59), while hypertension demonstrated only a low association (r ≈ 0.33). Other diseases, including anemia, tuberculosis, and hepatitis B, showed low to moderate correlations, reflecting peripheral relationships within the broader disease network.

Among socio-economic indicators, poverty emerged as the most relevant structural determinant (r ≈ 0.47, moderate), whereas smoking prevalence, daily caloric intake, and budget realization showed low or very low correlations with stroke.

Guided by these empirical patterns, we selected disease variables with strong correlations to stroke (CKD, DM, and cirrhosis) as the core clinical components of the syndemic structure. Among the socio-economic indicators examined, only poverty demonstrated a moderate correlation with stroke, whereas smoking prevalence, calorie intake, and budget realization showed low or very low correlations. Therefore, poverty was the sole socio-economic variable incorporated to represent the structural dimension in subsequent modelling. These selected variables were subsequently examined through further analysis.

Probabilistic and path analytic structure of the stroke syndemic

Building on the close intercorrelations identified among stroke, CKD, DM, cirrhosis, and poverty observed in the correlation analysis, we next applied BGNA to ascertain their conditional dependencies and determine the most probable endpoint within this interconnected system. We constructed alternative network schemes in which each variable, stroke, CKD, DM, cirrhosis, and poverty, was sequentially modelled as the outcome node (Fig. 3). The probability values represent the posterior probability that a given variable functions as the terminal node within the inferred network structure. Across these configurations, the model treating stroke as the final outcome demonstrated the highest probability (mean 51.85%; median 52.12%). This indicates that the network structure in which stroke occupies the most downstream position within the system was most frequently supported by the BGNA algorithm. Importantly, these probabilities reflect relative support for alternative network structures rather than the epidemiological probability of stroke occurrence in the population.

Fig. 3.

Bayesian Gaussian Network Analysis (BGNA) used to identify the most probable final outcome and the directional dependencies among CKD, DM, cirrhosis, poverty, and stroke. (A–E) Five network schemes were constructed by modelling each variable, stroke (A), CKD (B), DM (C), poverty (D), and cirrhosis (E), as the potential final outcome node. Directed arrows indicate the most probable directional relationships, while undirected lines represent interaction pathways. Predictor variables are shown in grey, and the outcome variable in each scheme is highlighted in red. (F) Summary table of mean, median, minimal, and maximal posterior probabilities for each scheme. The configuration treating stroke as the final outcome demonstrated the highest probability (mean 51.85%; median 52.12%), indicating that stroke is the most plausible downstream end-point within the interconnected system of metabolic, hepatic, and socio-economic factors.

In contrast, when CKD was modelled as the final outcome, the resulting probability was notably lower (mean 49.96%), and although relationships with DM, stroke, cirrhosis, and poverty were retained, the directionality was less coherent than in the stroke-focused model. Schemes positioning DM or poverty as outcomes yielded even weaker support (mean probabilities 49.91% and 49.98%, respectively), with networks characterized primarily by undirected associations rather than stable directional influence. These configurations suggest that neither DM nor poverty functions as the downstream endpoint within this constellation of interrelated factors. When cirrhosis was modelled as the final outcome, the probability was similarly lower (mean 49.98%), with only DM and stroke contributing directionally, reflecting a narrower and less plausible coherent network structure compared with the stroke-oriented model.

To further quantify the magnitude and structure of the relationships suggested by the probabilistic model, we conducted a path analysis incorporating DM, CKD, cirrhosis, and poverty as predictors of stroke (Fig. 4). This framework allowed us to decompose each pathway into its direct and indirect components, providing a more granular understanding of how metabolic, hepatic, and socio-economic factors jointly shape stroke risk at the provincial level.

Fig. 4.

Path analysis quantifying direct and indirect effects of DM, CKD, cirrhosis, and poverty on stroke. This structural path model decomposes each predictor’s contribution into direct effects (solid arrows), indirect effects via a single mediator (dashed arrows), and indirect effects via two sequential mediators (dash-dot arrows). DM exerts strong upstream effects on CKD and cirrhosis, which in turn influence poverty and stroke. CKD, cirrhosis, and poverty each contribute direct effects to stroke, while DM additionally affects stroke through multiple indirect pathways. Percentage values represent standardized path coefficients.

DM exerted the strongest upstream influence in the network, with a very large direct effect on CKD (75.88%) and cirrhosis (61.27%), consistent with the well-established progression from chronic hyperglycemia to renal and hepatic metabolic dysfunction. CKD also showed a substantial direct effect on poverty (48.67%), highlighting the economic consequences of deteriorating renal health. Poverty, in turn, had a notable direct influence on stroke (19.18%), underscoring the structural vulnerability imposed by socio-economic disadvantage.

With respect to stroke, direct associations were observed for cirrhosis (40.43%), poverty (32.55%), CKD (19.18%), and DM (7.08%). Although DM showed a relatively small direct effect, its indirect contributions were substantial, primarily mediated through CKD (24.70%) and cirrhosis (24.78%), as well as through sequential pathways such as DM → CKD → poverty → stroke (7.08%). Moreover, our data showed that CKD was directly (32.55%) and indirectly (9.33%) associated with stroke. Poverty, on the other hand, was directly associated with stroke at 19.18%.

Spatiotemporal dynamics and syndemic clustering of stroke determinants

We next applied GTWR to assess how the influence of CKD, DM, cirrhosis, and poverty on stroke varied across both space and time from 2018 to 2023 (Fig. 5). In several Sumatra and Java provinces, DM and CKD alternated as dominant predictors over time. In Kalimantan and Sulawesi, all four predictors contributed variably, with no single determinant consistently prevailing throughout the study period. Provinces with lower development indices more frequently exhibited stronger associations with poverty and cirrhosis. In eastern Indonesia, associations were generally weak or non-significant across most years. To further illustrate these temporal dynamics, supplementary plots showing the year-by-year GTWR regression coefficients for representative provinces are provided in Supplementary Fig. S2.

Fig. 5.

Geographically and Temporally Weighted Regression (GTWR) showing spatiotemporal variability in stroke predictors across Indonesian provinces. This matrix summarizes the dominant predictor(s) of stroke for each province in each study year, based on locally weighted regression coefficients. Colored cells indicate the predictor with the strongest association at that province–year combination: (A) CKD, (B) DM, (C) cirrhosis, or (D) poverty. Combinations (e.g., AB, BCD) denote years in which multiple predictors contributed comparably. The figure highlights marked heterogeneity in determinant patterns, reflecting geographically variable and time-dependent influences on stroke risk across Indonesia.

To further synthesize the multidimensional patterns identified in the GTWR analysis, we conducted a PCA using CKD, DM, cirrhosis, poverty, and stroke to derive a latent structure capturing the joint syndemic burden across Indonesian provinces (Fig. 6).

Fig. 6.

Spatiotemporal clustering of the syndemic burden of stroke across Indonesian provinces, 2018–2023. Principal Component Analysis (PCA) integrating CKD, DM, cirrhosis, poverty, and stroke identified distinct province-level clusters reflecting shared metabolic, hepatic, structural, and cerebrovascular characteristics. A subset of provinces including Jakarta, Yogyakarta, East Kalimantan, Bali, and North Sulawesi, consistently occupied the high-risk cluster across all six years, indicating a persistent syndemic pattern characterized by the co-occurrence and interaction of multiple determinants that collectively amplify stroke vulnerability. Other provinces demonstrated intermittent transitions between medium- and high-risk clusters, while low-risk provinces remained relatively stable over time. Full cluster trajectories, inter-annual stability indices, and province-level transitions are provided in Supplementary Fig. S1.

The PCA-derived clustering revealed clear and temporally stable groupings of provinces based on their shared syndemic profiles. Jakarta, Yogyakarta, East Kalimantan, Bali, and North Sulawesi consistently occupied the high-risk cluster from 2018 to 2023 (Fig. 6), indicating a persistent syndemic pattern in which metabolic (DM, CKD), hepatic (cirrhosis), and structural (poverty-related) determinants co-occurred and interacted to amplify stroke vulnerability. In contrast, several other provinces exhibited intermittent mobility between medium- and high-risk clusters, reflecting fluctuating combinations of metabolic progression and structural exposure across years. Provinces in the low-risk cluster remained comparatively stable, suggesting the presence of protective demographic or socioeconomic baselines that limited syndemic reinforcement (Supplementary Fig. S3).

Discussion

This study provides a comprehensive assessment of the syndemic architecture underlying stroke across Indonesia by integrating temporal trends, inter-variable associations, probabilistic network modelling, pathway analysis, and spatiotemporal examination. We found a marked national rise in CKD, DM, and stroke, with disproportionate increases in several provinces outside Java, accompanied by worsening structural indicators such as poverty. These early patterns highlight broad metabolic and socio-economic pressures shaping population vulnerability and set the foundation for the deeper syndemic relationships uncovered in the subsequent correlation, probabilistic, and spatial analyses.

Our BGNA and path analysis further elucidated the hierarchical structure of interactions among metabolic (DM and CKD), hepatic (cirrhosis), and structural determinants (poverty) of stroke. Across all alternative network configurations, stroke emerged as the most probable downstream outcome, indicating that it serves as the convergent end-point of multiple interconnected pathways rather than an isolated clinical entity. Within this configuration, DM emerged as the most central upstream correlate within the network structure, exerting large direct association with CKD and cirrhosis, reflecting well-established mechanisms through which hyperglycaemia accelerates renal and hepatic metabolic injury.19, 20, 21

CKD showed a strong association with stroke within the path model. This likely reflects the cumulative vascular and hemodynamic consequences of renal impairment, including endothelial dysfunction, accelerated atherosclerosis, hypertension, and the neurotoxic effects of retained uremic metabolites, all of which are recognized pathways linking CKD to both ischaemic and hemorrhagic stroke.22, 23, 24 These findings reinforce the critical role of diabetic and kidney disease as a major cerebrovascular risk amplifier in populations with high metabolic burden.^25,26

Our data showed that cirrhosis also emerged as an independent association with stroke in the model. This is in line with analyses from the Korean National Health Insurance Service, which showed that individuals with cirrhosis have an elevated risk of stroke, likely driven by coagulopathy, thrombocytopenia, altered cerebral autoregulation, and increased susceptibility to cerebral microbleeds.²⁷ This association is further reinforced by findings from the American National Inpatient Dataset, which demonstrated that patients with hepatic cirrhosis experience a markedly higher risk of stroke-related mortality, with an estimated 70 percent increase compared with non-cirrhotic patients.²⁸

Poverty also contributed meaningfully to the overall stroke risk structure in our model, being strongly associated with CKD and stroke. The strong direct influence of CKD on poverty in our analysis reflects well-established economic burdens associated with renal disease, including high out-of-pocket treatment costs, reduced productivity, and long-term disability, all of which can drive households into financial hardship.^29,30 In turn, poverty was associated with an increased risk of stroke, a relationship widely documented in global epidemiological studies.³¹ Individuals living in poverty experience a disproportionate exposure to behavioral and environmental risk factors, including limited access to preventive care, lower treatment adherence, food insecurity, and higher prevalence of untreated cardiometabolic disease, which collectively heighten susceptibility to stroke.³²

The spatiotemporal patterns revealed by GTWR further emphasize that the stroke syndemic in Indonesia is highly context-dependent rather than spatially uniform. While the BGNA and path models describe national-level hierarchical relationships, the GTWR findings highlight that the relative influence of CKD, DM, cirrhosis, and poverty varies substantially across provinces and over time. In more developed regions, particularly Java, metabolic pathways dominated the risk structure, with a clear progression from DM to CKD and subsequently to stroke, a pattern consistent with stronger diagnostic penetration and more stable health system functioning.^33,34 In contrast, several provinces outside Java exhibited more heterogeneous and temporally shifting patterns.

In addition, the PCA results further reinforced these spatial patterns by revealing a distinct cluster of five provinces, Bali, Yogyakarta, North Sulawesi, Jakarta, and East Kalimantan, that consistently exhibited high stroke incidence alongside similar socio-epidemiological configurations across the six-year period. Rather than being defined solely by their economic status, these provinces demonstrated a persistent syndemic pattern, where metabolic (DM, CKD), hepatic (cirrhosis), and structural determinants co-occurred and interacted to sustain elevated cerebrovascular risk. This recurrent clustering suggests that the convergence of lifestyle transitions associated with urbanization, population aging, shifting metabolic profiles, and uneven prevention coverage may collectively reinforce a high-burden environment.³⁵ In these settings, stronger health infrastructure does not necessarily translate into lower disease burden; instead, it may increase detection while underlying behavioural and structural drivers continue to intensify chronic disease trajectories.^36,37

The term “syndemic,” originally introduced by medical anthropologist Merrill Singer, refers not merely to the co-occurrence of two or more diseases within an individual, but to the synergistic interaction of multiple epidemics within populations shaped by adverse social and structural conditions.¹⁰ In adapting this framework to population-level epidemiology, we operationalize syndemic theory to assess hierarchical interdependencies and structural amplification across regions rather than isolated predictor–outcome relationships. In this context, the convergence of metabolic disorders, hepatic disease, and structural vulnerability illustrates a syndemic configuration in which social and biological processes reinforce one another to shape stroke burden across provinces.

Although hypertension is globally recognized as the most important modifiable risk factor for stroke at the individual level, the ecological correlation observed in our provincial-level analysis was more modest than expected. In contrast, DM demonstrated a comparatively stronger spatial association with stroke in our dataset.⁸ This divergence does not contradict established clinical evidence, but reflects the distinction between individual-level risk and population-level distribution patterns. In ecological analyses, correlation coefficients depend on variability across regions rather than biological strength of effect.³⁸ Hypertension prevalence in Indonesia appears relatively widespread across provinces, which may reduce geographic variability and attenuate correlation estimates at the aggregate level.⁸ By comparison, DM prevalence varies more unevenly across provinces, thereby producing clearer alignment with stroke incidence in certain regions. Differences in screening intensity, diagnostic practices, treatment coverage, and control rates across provinces may further obscure the ecological relationship between hypertension and stroke.

These findings have important implications for public health policy and planning in Indonesia. The emergence of stroke as a syndemic endpoint, driven by the interplay of biological conditions such as DM, CKD, and cirrhosis and compounded by structural vulnerabilities like poverty, calls for a paradigm shift in health system strategies. Rather than addressing diseases in isolation, policy efforts must target the upstream socio-economic and clinical conditions that co-produce poor health outcomes across regions. In provinces with pronounced disease burdens, strengthening provincial level health system capacity should be a priority. Expanding access to mobile health clinics and telemedicine platforms can mitigate barriers imposed by geography and health workforce shortages, enabling earlier detection and continuity of care for chronic conditions.³⁹

This study has several limitations. First, our analysis is inherently limited to individuals who are registered and actively utilize Indonesia’s national health insurance system. This may exclude populations who rely on private insurance, pay out-of-pocket, or face systemic barriers to accessing care, particularly in rural or remote areas with limited healthcare infrastructure, thereby potentially underestimating the true burden of diseases and masking the full extent of regional disparities. Second, all variables were analyzed as province–year aggregates, and therefore the associations identified reflect population-level patterns rather than individual-level relationships. The probabilistic network, path analysis, and GTWR models describe conditional dependencies and spatial–temporal heterogeneity, but do not establish causal effects. Third although we employed advanced analytical approaches to reveal complex associations and structural drivers, residual confounding remains possible due to unmeasured variables such as dietary habits, occupational exposures, genetic predisposition, or local environmental hazards.

In conclusion, our findings highlight the urgent need to reframe disease prevention and control in Indonesia through a syndemic approach, one that acknowledges the dynamic interplay between clinical comorbidities, structural inequities, and regional contexts. The identification of stroke as a convergence point of biological and socio-economic pressures underscores its role as a sentinel condition within Indonesia’s health landscape. This syndemic perspective offers a critical framework for developing responsive, integrated, and regionally tailored health policies capable of curbing the growing burden of diseases in Indonesia.

Contributors

Data curation: RRAK, NQ.

Formal analysis: HK, RRAK.

Methodology: AAZ, IY, HK.

Investigation: AAZ, RRAK, NQ.

Data access and verification: AAZ, RRAK, and HK.

Visualisation: RRAK.

Writing—original draft: AAZ, RRAK, NQ.

Writing—review & editing: AAZ, RRAK, NQ, HK, AAF, TF, HR, IY.

Supervision: IY, HR.

The corresponding author (AAZ) had final responsibility for the decision to submit for publication.

Data sharing statement

The datasets analyzed in this study are not publicly available due to contractual agreements with BPJS Kesehatan regarding the reporting and use of the national dataset.

Use of generative AI in manuscript writing

ChatGPT 5.1 was used to assist with language editing (grammar and spelling) in certain parts of the manuscript.

Declaration of interests

We declare no competing interests.

Appendix A. Supplementary data

Supplementary Fig. S1.

Supplementary Fig 1. Province-level temporal trends of disease incidence and socio-economic indicators across Indonesia, 2018–2023. Multi-panel line graphs illustrating the absolute trajectories of Crude Incidence Rate (per 100,000 population) for major diseases and the values of key socio-economic indicators across Indonesian provinces from 2018 to 2023. Each colored line represents one province. Disease panels include anemia, cirrhosis, chronic kidney disease (CKD), diabetes mellitus (DM), heart failure, hepatitis B, hypertension, schizophrenia, stroke, and tuberculosis. Socio-economic panels include budget realization, average calorie consumption (kcal per capita per day), poverty rate, and smoking prevalence among individuals aged 15 years or older. This figure complements Fig. 1 by presenting absolute values and temporal patterns for each province, allowing clearer visualization of regional trajectories over time.

Supplementary Fig. S2.

Supplementary Fig. 2. Heatmap of province-level GTWR regression coefficients for stroke predictors across Indonesia, 2018–2023. Heatmap illustrating the regression coefficients estimated from the GTWR model for DM, CKD, cirrhosis, and poverty across Indonesian provinces from 2018 to 2023. Each cell represents the localized regression coefficient for a given predictor in a specific province and year. Red colour indicate stronger positive associations with stroke incidence, whereas blue colour indicate weaker or negative associations. Provinces are grouped by major geographic regions (Sumatra, Java, Kalimantan, Bali and Nusa Tenggara, Sulawesi, and Maluku–Papua) to facilitate regional comparison. This figure provides a comprehensive visualization of the spatial and temporal variability underlying the GTWR results and complements the dominant predictor patterns summarized in Fig. 5.

Supplementary Fig. S3.

Supplementary Fig 3. Temporal trajectories and cluster stability of provincial syndemic profiles from 2018–2023. This figure presents the year-to-year transitions of each Indonesian province across the PCA-derived clusters shown in Fig. 6, illustrating the longitudinal stability and mobility of syndemic configurations. Jakarta, Yogyakarta, East Kalimantan, Bali, and North Sulawesi maintained uninterrupted membership in the high-risk cluster, confirming the presence of a persistent syndemic pattern characterized by co-occurring metabolic, hepatic, structural, and cerebrovascular burdens. Several provinces displayed intermittent transitions between medium- and high-risk clusters, reflecting fluctuating combinations of risk determinants across time, while low-risk provinces exhibited minimal mobility. Stability indices, transition matrices, and cluster trajectory lines together provide a detailed temporal map of how syndemic structures evolve or remain entrenched across Indonesia.