Skip Navigation
Skip to contents

JPMPH : Journal of Preventive Medicine and Public Health

OPEN ACCESS
SEARCH
Search

Articles

Page Path
HOME > J Prev Med Public Health > Volume 59(5); 2026 > Article
Original Article
Physical Inactivity and Non-communicable Disease Burden in Five ASEAN Countries: A 12-Year Panel Analysis Using Structural Equation Modeling and Machine Learning Projections
Moch Yunus1orcid, Mahalul Azam2orcid, Endang Sri Wahjuni3orcid, Ronal Surya Aditya4orcid, Reem Lafi AlMutairi5orcid, Supriyadi Supriyadi1orcid
Journal of Preventive Medicine and Public Health 2026;59(5):542-554.
DOI: https://doi.org/10.3961/jpmph.26.543
Published online: August 29, 2026
  • 1,138 Views
  • 55 Download

1Department of Medicine, Faculty of Medicine, Universitas Negeri Malang, Malang, Indonesia

2Department of Public Health, Faculty of Medicine, Universitas Negeri Semarang, Semarang, Indonesia

3Department of Medicine, Faculty of Medicine, Universitas Negeri Surabaya, Surabaya, Indonesia

4Department of Nursing, Faculty of Nursing, Universitas Jember, Jember, Indonesia

5Department of Public Health, College of Public Health and Health Informatics, University of Hail, Hail, Saudi Arabia

Corresponding author: Moch Yunus, Department of Medicine, Faculty of Medicine, Universitas Negeri Malang, Malang 65145, Indonesia, E-mail: moch.yunus.fk@um.ac.id
• Received: June 5, 2026   • Revised: August 14, 2026   • Accepted: August 19, 2026

Copyright © 2026 The Korean Society for Preventive Medicine

This is an Open Access article distributed under the terms of the Creative Commons Attribution Non-Commercial License (https://creativecommons.org/licenses/by-nc/4.0/) which permits unrestricted non-commercial use, distribution, and reproduction in any medium, provided the original work is properly cited.

prev next
  • Objectives
    Non-communicable diseases (NCDs) are a leading cause of preventable mortality in Southeast Asia, but population-level evidence linking physical inactivity to NCD burden across the Association of Southeast Asian Nations (ASEAN) remains limited. This study quantified associations between physical inactivity prevalence and NCD-related disability-adjusted life years (DALYs), cardiovascular mortality, and diabetes prevalence across five ASEAN countries from 2010 to 2021 and projected 2025 NCD burden trajectories under an exploratory business-as-usual scenario.
  • Methods
    A longitudinal panel dataset of 60 country-year observations was analyzed using fixed-effects regression with Driscoll–Kraay standard errors, partial least squares structural equation modeling, and XGBoost machine learning with Shapley additive explanations. Missing data (2.1–6.4%) were handled using multiple imputation by chained equations.
  • Results
    Physical inactivity was independently associated with NCD DALYs (β=0.421; 95% confidence interval, 0.251 to 0.591; p<0.001). A significant attenuating interaction with the universal health coverage (UHC) index was observed (β=−0.183; 95% CI, −0.358 to −0.008; p=0.041). Because the UHC index was rescaled to 10-point units, this indicates that for every 10-point increase in the UHC index, the regression coefficient for physical inactivity decreased by 0.183 units (DALYs per 100 000 population), demonstrating a significant buffering effect of health system coverage on the inactivity–NCD association. The structural equation model explained 78.2% of the variance in NCD burden and identified a significant exploratory longitudinal indirect association through health system resilience, with 19.8% of the total association mediated through this pathway. Under the exploratory business-as-usual scenario, NCD DALYs were projected to increase by 7.8–20.9% by 2025, with Indonesia and the Philippines showing the highest projected increases.
  • Conclusions
    UHC expansion may attenuate inactivity-related NCD burden. For ASEAN countries undergoing rapid epidemiological transition, coordinated investment in health system capacity and multisectoral physical activity promotion may help limit projected DALY increases and support progress toward Sustainable Development Goal 3.
Non-communicable diseases (NCDs) now account for 74% of global deaths, with a disproportionate burden in low-income and middle-income countries in Southeast Asia, where epidemiological transitions are advancing faster than many health systems can respond [1–3]. In this context, physical inactivity has been identified as a major modifiable risk factor for mortality, accounting for an estimated 3.2 million deaths annually and contributing substantially to cardiovascular disease, type 2 diabetes, several cancers, and common mental health disorders [2,4]. The Association of Southeast Asian Nations (ASEAN) region is an important setting for examining these relationships because rapid urbanization, motorization, and occupational sedentarization have reduced population-level physical activity faster than current surveillance and intervention systems can address [4,5]. Despite this growing epidemiological pressure, the longitudinal relationship between population-level physical inactivity and NCD burden across ASEAN member states has not been adequately characterized using multi-country data. This evidence gap limits policymakers’ ability to prioritize investment and design proportionate public health responses.
The ASEAN health landscape is heterogeneous, which complicates the development of region-wide evidence. Indonesia, Malaysia, Thailand, the Philippines, and Vietnam span broad differences in economic development, health system structure, and NCD epidemiology [6,7]. Thailand and Malaysia have achieved relatively high universal health coverage (UHC) indices and have invested substantially in NCD prevention infrastructure, whereas Indonesia and the Philippines continue to have fragmented health financing systems and large uninsured populations [6]. This variation provides a natural cross-national gradient for examining whether health system strength modifies the association between physical inactivity and population-level NCD outcomes. If UHC expansion attenuates inactivity-related NCD burden, this finding would have direct implications for sequencing public health investments in resource-constrained settings by positioning health system strengthening as a complementary—and potentially prerequisite—intervention alongside physical activity promotion [8–10].
This study used a multi-method longitudinal framework to extend prior cross-sectional and individual-level cohort research [2,3]. Fixed-effects panel regression, partial least squares structural equation modeling (PLS-SEM) [11–13], and interpretable machine learning using XGBoost with Shapley additive explanations (SHAP) were combined to examine multilevel relationships and generate exploratory projections of NCD burden in 2025 [14–17]. The conceptual framework evaluated two related pathways linking behavioral risk and health system capacity. First, UHC was examined as an observable moderator that may attenuate the association between population-level physical inactivity and NCD burden. Second, health system resilience (HSR), a broader latent construct incorporating UHC, physician density, and health expenditure, was evaluated as a mediator. In this framework, widespread physical inactivity may increase physiological risk and generate sustained demand for chronic care, thereby straining health system capacity over time. Reduced health system capacity may then limit effective NCD management and contribute to higher population-level disease burden. Fixed-effects regression was used to estimate the moderating effect of the UHC index, whereas PLS-SEM was used to assess the mediating role of the broader HSR construct. Together, these methods allowed the study to evaluate both immediate buffering effects and longer-term system-level pathways.
This study had three objectives: (1) to quantify longitudinal associations between physical inactivity and multiple NCD outcomes across five ASEAN countries from 2010 to 2021, accounting for key socioeconomic and health system confounders; (2) to examine whether health system coverage moderated the physical inactivity–NCD relationship using interaction analysis and PLS-SEM; and (3) to generate country-specific 2025 NCD burden projections under an exploratory business-as-usual scenario using an XGBoost machine learning model. The findings are intended to inform health system investment priorities and physical activity policy decisions across ASEAN member states. Because NCD deaths in ASEAN are projected to outpace health system capacity by 2030, identifying modifiable system-level factors that may buffer inactivity-related disease burden is important for public health planning. To translate the longitudinal analyses into practical policy signals, this study also used an interpretable machine learning model to generate near-term projections of NCD burden in 2025. These projections may help ASEAN policymakers prioritize resources and identify high-risk settings before projected disease burden exceeds health system capacity.
This study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) checklist for longitudinal ecological panel studies and the Transparent Reporting of a Multivariable Prediction Model for Individual Prognosis or Diagnosis statement for the machine learning prediction model.
Study Design and Data Sources
A longitudinal ecological panel design was used, with country-year as the unit of analysis. Data were obtained from four open-access international databases: (1) World Bank Open Data for economic and demographic indicators; (2) the World Health Organization (WHO) Global Health Observatory for physical inactivity prevalence, the UHC index, physician density, and health expenditure; (3) the Global Burden of Disease Study 2021, conducted by the Institute for Health Metrics and Evaluation, for NCD-related disability-adjusted life years (DALYs) and cause-specific mortality; and (4) the United Nations Sustainable Development Goals Database for Human Development Index and Sustainable Development Goal progress indicators. Data were extracted for 2010–2021.
Study Population and Setting
The panel included five ASEAN member states—Indonesia, Malaysia, Thailand, the Philippines, and Vietnam—selected because complete time-series data for the primary exposures and outcomes were available across the full study period. Together, these countries represent approximately 76% of the ASEAN population and cover a wide range of economic development (2010 gross domestic product [GDP] per capita range: US$4210–22 180) and health system capacity. The final analytical dataset included 60 country-year observations (5 countries×12 years).
Exposure, Outcomes, and Covariates
The primary exposure was physical inactivity. For the fixed-effects panel regression, this was operationalized as the observed age-standardized prevalence of insufficient physical activity (%), defined according to WHO criteria as the failure to meet recommended thresholds of 150 minutes of moderate-intensity or 75 minutes of vigorous-intensity physical activity per week. However, for the PLS-SEM analysis, physical inactivity was conceptualized as a broader reflective latent construct to capture the multidimensional nature of the behavior. This latent construct was indicated by two observed variables sourced from the WHO Global Health Observatory: (1) the prevalence of insufficient physical activity, and (2) the prevalence of sedentary behavior. Both indicators were coded such that higher values consistently denoted greater physical inactivity. The primary outcome was NCD-related DALYs per 100 000 population. Secondary outcomes were age-standardized cardiovascular mortality per 100 000 population and type 2 diabetes prevalence (%). Candidate covariates included GDP per capita (purchasing power parity [PPP], constant 2017 US$), urbanization rate (%), health expenditure per capita (PPP US$), physician density per 10 000 population, and the UHC Service Coverage Index (0–100 scale). In the primary fixed-effects outcome models, GDP per capita, urbanization rate, and health expenditure per capita were included as confounders, along with country and year fixed effects. Physician density was not included as a separate covariate in the primary fixed-effects models because it was incorporated as an indicator of the latent HSR construct in the PLS-SEM framework. The UHC index was also examined as a moderator in interaction analyses. For interpretation of the interaction term, the UHC index was rescaled to 10-point units; therefore, the interaction coefficient represents the change in the physical inactivity slope associated with each 10-point increase in the UHC index.
Missing Data
Missing values ranged from 2.1% to 6.4% across indicators. Little’s test for missing completely at random was applied before imputation (p=0.38). Multiple imputation by chained equations was then performed using predictive mean matching for continuous variables, with 20 imputations and 10 iterations per imputation. Pooled estimates across imputations are reported throughout. Complete-case analyses (n=54) were conducted as a sensitivity check.
Statistical Analysis
Fixed-effects panel regression. Hausman specification tests were used to determine the appropriate model type for each outcome. Fixed-effects models were appropriate for all primary analyses (p<0.05 for all). To ensure that the estimated associations reflected exposure–outcome relationships rather than common secular trends affecting all countries simultaneously, a two-way fixed-effects specification was used. This model included country fixed effects to control for unobserved time-invariant country-level heterogeneity and year fixed effects to account for common temporal shocks, global economic shifts, and macro-level time trends during 2010–2021. Driscoll–Kraay standard errors were applied to account for heteroskedasticity and cross-sectional dependence. Interaction terms between physical inactivity and the UHC index were included to examine moderation of the primary exposure–outcome relationship. Two-tailed statistical significance was set at α=0.05.
PLS-SEM was conducted in SmartPLS 4.0 using 5000 bootstrap resamples and bias-corrected confidence intervals. HSR was modeled as a reflective latent construct indicated by three macro-level indicators: the UHC Service Coverage Index, physician density, and per capita health expenditure. Although the standalone UHC index served as an observable moderator in the panel regression, its inclusion as an HSR indicator in the PLS-SEM framework allowed broader systemic capacity to be assessed. The structural mediation pathway (physical inactivity →HSR→NCD burden) was justified conceptually as a systemic feedback loop in which rising inactivity-related NCD prevalence increases ambulatory and chronic care demand, strains health system resources, and may erode resilience over time. Reduced resilience may subsequently limit the health system’s capacity to deliver effective secondary prevention and NCD management, thereby further increasing population-level disease burden. Measurement model validity was assessed using average variance extracted (>0.50), composite reliability (>0.70), Cronbach’s α, and heterotrait–monotrait ratios (<0.85). Model fit was evaluated using standardized root mean square residual, normed fit index, root mean square error of approximation, goodness-of-fit, and Stone–Geisser Q2 statistics. Indirect effects were tested using bias-corrected bootstrapping, and the proportion mediated was calculated as the ratio of the indirect effect to the total effect.
To ensure that the latent construct represented physical inactivity rather than physical activity, all indicator coding was re-examined and standardized before model estimation. The primary indicator—population-level prevalence of insufficient physical activity, defined as failure to meet WHO-recommended thresholds—was coded so that higher values denoted greater physical inactivity. All factor loadings were positive and statistically significant, confirming that higher latent construct scores represented greater physical inactivity. Accordingly, structural path coefficients were interpreted directionally: a positive path coefficient from physical inactivity to NCD burden indicated that higher population-level inactivity was associated with greater NCD burden. To address temporal precedence in the structural model, the PLS-SEM analysis used a lagged temporal specification. Physical inactivity was measured at time T (2010–2019), HSR was measured at time T+1 (2011–2020), and NCD burden was measured at time T+2 (2012–2021). After temporal lagging, the PLS-SEM analytical sample included 50 country-year observations aligned across T, T+1, and T+2.
The dataset had a hierarchical structure, with 60 country-year observations nested within five countries. Treating these observations as independent would violate the assumption of independence and increase the risk of pseudoreplication. To address this limitation, two methodological adjustments were implemented. First, all time-varying variables were group-mean centered within countries, isolating within-country temporal dynamics from between-country baseline structural differences, such as differences in economic development or health infrastructure between Thailand and Indonesia. Second, country-clustered bootstrapping was performed using 5000 resamples clustered at the country level rather than the observation level. This approach accounted for the non-independence of repeated measures within countries, generated more robust standard errors, and reduced the risk of inflated type I error rates. In the PLS-SEM framework, physical inactivity was modeled as a reflective latent construct indicated by two observed variables sourced from the WHO Global Health Observatory: the prevalence of insufficient physical activity and the prevalence of sedentary behavior. This approach allowed the structural model to capture the broader behavioral phenotype of inactivity beyond a single metric. To ensure that the latent construct represented physical inactivity (rather than physical activity), all indicator coding was standardized prior to model estimation so that higher values denoted greater inactivity. Consequently, all factor loadings were positive and statistically significant, and structural path coefficients were interpreted directionally: a positive path coefficient from the physical inactivity construct to NCD burden indicated that greater population-level inactivity was associated with a higher NCD burden. HSR was modeled as a reflective latent construct indicated by the UHC Service Coverage Index, physician density, and per capita health expenditure. NCD burden was modeled as a reflective latent construct indicated by NCD-related DALYs per 100 000 population and age-standardized NCD mortality per 100 000 population. Socioeconomic status, represented by GDP per capita, and urbanization were included as exogenous contextual variables in the structural model where applicable.

Machine learning projections

An XGBoost gradient boosting model was trained on 2010–2018 country-year data (n=45) and validated on 2019–2021 data (n=15). Hyperparameters were tuned using five-fold cross-validation. Although standard linear forecasting methods are typically preferred for small samples, XGBoost was selected because it can capture complex non-linear interactions between macroeconomic and health system variables that may be missed by linear models. To reduce the risk of overfitting given the limited sample size, strict regularization parameters were applied during hyperparameter tuning: maximum tree depth was constrained to 3, and L1 (alpha=0.1) and L2 (lambda=1.0) regularization penalties were applied to support model parsimony. Model performance was assessed using mean absolute error and R2 in the held-out validation set. SHAP values were computed for each prediction to identify dominant predictive features. Given the limited sample size (n=45 for training and n=15 for validation), the machine learning analyses were treated as exploratory rather than definitive predictive modeling. Country-specific 2025 NCD DALY projections were generated under an exploratory business-as-usual scenario by extrapolating observed 2010–2021 trends, with 95% confidence intervals derived from 1000 bootstrap resamples of the validation data. We note that although 2025 has now passed at the time of publication, updated empirical surveillance data to confirm or evaluate these projections are not yet available; therefore, they are presented strictly as modelled estimates based on historical trends.

Robustness checks

To address potential small-sample bias (n=60), bias-corrected bootstrapping with 5000 resamples was used for PLS-SEM, least absolute shrinkage and selection operator (LASSO) penalized regression was used to support predictor parsimony, and complete-case sensitivity analysis (n=54) was performed to assess coefficient stability (Δβ<0.04). Spatial autocorrelation was assessed using Moran’s I. Detailed sensitivity analysis outputs are provided in Supplemental Material 1.
Ethics Statement
This study utilized publicly available, de-identified secondary data aggregated at the country-year level from international repositories (e.g., WHO Global Health Observatory, World Bank Open Data). Because the analysis did not involve direct interaction with human participants and contained no identifiable individual-level information, formal ethical review and informed consent were waived by the Institutional Review Board (Ethics Committee) of Universitas Negeri Malang, in strict accordance with national regulations and institutional guidelines for ecological analyses of public, aggregated statistics.
Study Population and Data Characteristics
The final analytical dataset included 60 country-year observations from five ASEAN member states during 2010–2021. Data completeness was high, with missing values ranging from 2.1% to 6.4% across indicators. Table 1 summarizes demographic, socioeconomic, and health system characteristics for the five countries. From 2010 to 2021, all countries experienced GDP growth, ranging from 29.7% in Malaysia to 126.8% in Vietnam, along with increases in urbanization. Health system development varied substantially across countries: by 2021, Thailand and Malaysia had UHC Service Coverage Index values of 80 and 76, respectively, whereas Indonesia and the Philippines remained below 60.
Temporal Trends in Physical Inactivity and Non-communicable Diseases Burden
Physical inactivity prevalence increased in all five countries between 2010 and 2021, although the magnitude of change differed considerably by country. The Philippines had the largest absolute increase in physical inactivity prevalence, rising by 6.5 percentage points from 32.5% (95% CI, 31.1 to 33.9) in 2010 to 39.0% (95% CI, 37.4 to 40.6) in 2021, which also represented the largest relative increase (+20.0%). Malaysia had the highest overall prevalence in 2021, rising from 37.2% (95% CI, 35.8 to 38.6) to 43.3% (95% CI, 41.7 to 44.9), an absolute increase of 6.1 percentage points. Thailand had the smallest increase (+6.1%). Vietnam had the lowest physical inactivity prevalence throughout the study period, reaching 17.4% in 2021. Figure 1 shows country-specific time-series trajectories for physical inactivity and NCD-related DALYs, and Supplemental Material 2 summarizes trends in physical inactivity and NCD burden across all five countries.
NCD-related DALYs per 100 000 population also increased in all countries during the study period. The Philippines had the largest relative increase in NCD burden (+21.9%), followed by Indonesia (+17.8%). Countries with higher UHC indices, including Thailand and Malaysia, had smaller relative increases (+8.8 and +9.4%, respectively).
Fixed-effects Panel Regression Analysis
Hausman tests supported fixed-effects specifications for all primary outcomes (p<0.05 for all). After adjustment for GDP per capita, urbanization rate, health expenditure per capita, and country and year fixed effects, physical inactivity was positively and significantly associated with all NCD outcomes (Table 2). The largest coefficient was observed for the primary outcome, NCD-related DALYs per 100 000 population (β=0.421; 95% CI, 0.251 to 0.591; p<0.001; adjusted R2=0.68), followed by NCD mortality per 100 000 population (β=0.389; p<0.001), cardiovascular mortality per 100 000 population (β=0.318; p<0.001), and diabetes prevalence (β=0.274; 95% CI, 0.160 to 0.388; p=0.002). The interaction between physical inactivity and the UHC index was negative and statistically significant (β=−0.183; 95% CI, −0.358 to −0.008; p=0.041), suggesting that higher health system coverage attenuated the population-level association between physical inactivity and NCD burden. Specifically, because the UHC index was rescaled to 10-point units, each 10-point increase in the UHC index was associated with a 0.183-unit decrease in the physical inactivity slope (β=−0.183; 95% CI, −0.358 to −0.008; p=0.041). This indicates that higher health system coverage significantly attenuated the population-level association between physical inactivity and NCD burden in absolute terms, acting as a structural buffer against inactivity-related disease accumulation.
Measurement Model Validation (Partial Least Squares Structural Equation Modeling)
The measurement model demonstrated adequate convergent validity and confirmed the directional coding of the physical inactivity latent construct. Both observed indicators loaded positively and significantly onto the physical inactivity construct. The insufficient physical activity prevalence indicator showed a strong standardized factor loading (λ=0.89; p<0.001), and the sedentary behavior prevalence indicator also loaded significantly (λ=0.82; p<0.001). The average variance extracted (AVE) for the physical inactivity construct was 0.74, and composite reliability was 0.89, exceeding the recommended thresholds of 0.50 and 0.70, respectively. This confirmed that higher latent construct scores consistently represented greater population-level physical inactivity. Detailed factor loadings, cross-loadings, and discriminant validity metrics, including heterotrait–monotrait ratios <0.85, are provided in Supplemental Material 3. Physical inactivity had a significant direct association with NCD burden (β=0.538; 95% CI, 0.361 to 0.714; p<0.001). The indirect association through HSR was β=0.133 (95% CI, 0.048 to 0.241; p=0.007), accounting for 19.8% of the total association (Figure 2).
Structural Model and Hypothesis Testing
Using the lagged temporal specification, the structural model estimated the longitudinal indirect association between physical inactivity and NCD burden through HSR. The indirect association was statistically significant (β=0.133; 95% CI, 0.048 to 0.241; p=0.007), with 19.8% of the total association mediated through this pathway. Given the observational panel design and exploratory purpose of the structural pathway, this finding should be interpreted as an exploratory indirect association rather than as evidence of causal or dynamic mediation. In the lagged model, higher physical inactivity at time T was associated with lower HSR at time T+1, which in turn was associated with higher NCD burden at time T+2. This temporal ordering supports the proposed system-level pathway but does not establish causality.
Supplemental Material 4 reports the explanatory power and predictive relevance of the endogenous constructs. The model explained 61.4% of the variance in HSR (R2=0.614) and 78.2% of the variance in NCD burden (R2=0.782). Predictive relevance was also high for NCD burden (Q2=0.448).
To assess whether the structural pathways were robust to the country-year data structure, the model was re-evaluated using within-country centered variables and country-clustered bootstrapping. The longitudinal indirect association remained statistically significant under this multilevel specification (β=0.133; 95% CI, 0.048 to 0.241; p=0.007, clustered). This result suggests that the observed pathway was not solely an artifact of pooling countries with different baseline characteristics, although the small number of countries still warrants cautious interpretation.
Machine Learning Projections and Feature Importance
The XGBoost model trained on 2010–2018 data achieved a mean absolute error of 1830 DALYs per 100 000 population and an R2 of 0.847 in the 2019–2021 validation set, indicating acceptable apparent performance within the constraints of the small sample. Because the number of observations was limited, these outputs should be interpreted strictly as exploratory scenario-generating results rather than as robust or generalizable predictions. SHAP analysis identified physical inactivity prevalence as the dominant predictive feature (mean |SHAP=1.847), followed by GDP growth rate (1.423) and the UHC index (1.218). Table 3 presents country-specific 2025 NCD DALY projections under the exploratory business-as-usual scenario. The Philippines and Indonesia were projected to have the largest relative increases in NCD burden (+20.9 and +17.3%, respectively) and were classified as exploratory high-priority intervention targets. Thailand, which had the highest UHC index, was projected to have the smallest increase (+7.8%).
Robustness and Sensitivity Analyses
Complete-case analysis and LASSO penalized regression supported the stability and parsimony of the primary estimates. No significant spatial autocorrelation was detected in panel residuals (Moran’s I=0.11; p=0.14). Full robustness-check outputs are provided in Supplemental Material 5.
The accelerating epidemiological transition in Southeast Asia presents a major public health challenge: NCDs now account for more than 70% of regional mortality, but population-level evidence linking behavioral risk factors to systemic disease burden remains limited. In this longitudinal analysis of five ASEAN countries, physical inactivity was independently associated with NCD-related DALYs, cardiovascular mortality, and diabetes prevalence after adjustment for macroeconomic and health system confounders. Extensive epidemiological evidence has established physical inactivity as a major modifiable risk factor for NCDs [2,4]; this study extends that evidence by quantifying its macro-level population association over a 12-year period. These findings are consistent with regional and global burden analyses showing that inactivity-related mortality is increasingly important in rapidly urbanizing and middle-income economies (e.g., ecological and national survey evidence from Chile, Ghana, and broader multi-country analyses) [5,18,19] and support a shift from individual-focused behavioral messaging alone toward structural, population-wide interventions [20–22].
A key finding was the statistically significant attenuating interaction between physical inactivity and the UHC index. Because the UHC index was modeled in 10-point units, each 10-point increase in the UHC index was associated with a 0.183-unit reduction in the physical inactivity regression coefficient (β=−0.183). This indicates that more comprehensive health systems act as downstream buffers, significantly weakening the absolute magnitude of the population-level association between physical inactivity and NCD burden. Mechanistically, this interpretation is consistent with the idea that resilient health systems can identify and manage preclinical risk before it progresses to irreversible morbidity at the population level, acting as downstream buffers by improving primary care access and supporting sustained chronic disease management [23–28]. However, although the fixed-effects models accounted for time-invariant country characteristics, residual confounding may remain because major time-varying NCD risk factors, including changes in smoking prevalence, dietary patterns, and ambient air pollution, were not included. These unmeasured temporal shifts may have led to overestimation of the independent association between physical inactivity and NCD burden. Nevertheless, the consistency of the UHC interaction across sensitivity analyses and analytic approaches supports the interpretation of UHC expansion as a complementary strategy alongside physical activity promotion, rather than as a competing public health priority.
To examine whether the results were sensitive to the structural disruption caused by the coronavirus disease 2019 (COV-ID-19) pandemic, we conducted a sensitivity analysis exclud-ing observations from 2020–2021, as detailed in Supplemental Material 6. Excluding these two years produced only modest changes in the primary coefficients. The fixed-effects estimate for the association between physical inactivity and NCD-related DALYs shifted from β=0.421 in the full sample to β=0.398 in the restricted sample (95% CI, 0.224 to 0.572; p<0.001), corresponding to a 5.5% attenuation. The interaction between physical inactivity and the UHC index remained directionally consistent but was attenuated and did not reach conventional statistical significance in the restricted sample (β=−0.167; p=0.053). This pattern suggests that the pandemic years (2020–2021), characterized by simultaneous mobility restrictions and emergency health system mobilization during the late-pandemic period, may have amplified the observed buffering association of UHC infrastructure. In the structural model, the variance explained in NCD burden decreased only slightly, from R2=0.782 to R2=0.761, and the indirect association through HSR remained significant (β=0.121; 95% CI, 0.039 to 0.228; p=0.012), compared with β=0.133 in the full sample. These findings have methodological implications for interpreting the main analyses. The directional consistency and relative stability of the estimates across the full and restricted samples suggest that the observed physical inactivity–NCD relationship was not solely driven by pandemic-period anomalies in health care utilization or mobility surveillance. However, the attenuation of the UHC interaction after exclusion of 2020–2021 indicates that the full-sample moderation estimate may reflect both steady-state and crisis-period health system dynamics. This distinction is important when interpreting the UHC moderation result and when applying the 2025 forecasting parameters, which incorporate pre-pandemic trajectories, pandemic disruption, and the early sustained pandemic period.
The structural equation modeling identified a significant longitudinal indirect association from physical inactivity through HSR to NCD burden, accounting for nearly 20% of the total association. By using a lagged temporal design, the analysis quantified a system-level pathway in which widespread physical inactivity was associated with subsequent changes in health system capacity, plausibly through increased demand for ambulatory and chronic NCD care. Lower HSR was then associated with higher NCD burden in subsequent years, suggesting that constrained health systems may have less capacity to manage NCD complications effectively. Because the data are observational and ecological, this pathway should not be interpreted as definitive causal mediation. Nevertheless, the temporal ordering and consistency of the association support its interpretation as an exploratory longitudinal indirect pathway. The partial nature of this association is both theoretically coherent and practically important: resilient systems may absorb some inactivity-related demand, but they are unlikely to offset it fully without concurrent upstream prevention.
Socioeconomic development showed a dual relationship with NCD burden. Higher-income settings often have stronger health systems, but rapid urbanization can also reduce traditional movement environments and contribute to physical inactivity. This pattern is consistent with the epidemiological transition framework and suggests that urban planning and transportation policies should be integrated with health promotion through a Health in All Policies approach. Without structural interventions that support active mobility, economic growth may accelerate sedentary behavior and offset some gains from improved health care access.
Although the fixed-effects design controlled for time-invariant country-level heterogeneity and common secular trends, the findings should be interpreted within the broader context of co-occurring NCD risk factors. Physical inactivity does not operate in isolation; it is linked to concurrent, time-varying epidemiological changes, including shifts in smoking prevalence, dietary patterns characterized by greater ultra-processed food consumption, and rising ambient air pollution. These factors often share common socioeconomic drivers, including rapid urbanization and economic development, creating an environment in which multiple behavioral and environmental risk factors may combine to increase population-level disease burden.
The omission of these dynamic time-varying confounders introduces residual uncertainty when interpreting the magnitude of the estimates. The independent association between physical inactivity and NCD-related DALYs may be modestly overestimated if the model attributes variance from unmeasured co-occurring risk factors to physical inactivity alone. This limitation does not eliminate the public health relevance of the findings, but it does require caution in interpreting the effect size. Physical inactivity remained a statistically significant predictor of NCD burden despite these limitations, supporting its importance as a population-health target. In addition, the observed buffering associations of UHC and HSR may extend beyond inactivity-related demand and reflect broader health system capacity to respond to multiple intersecting behavioral and environmental risk factors.
The exploratory XGBoost projections and SHAP interpretability analyses, although constrained by the small sample, provide scenario-based signals for resource planning. These analyses identified physical inactivity prevalence and health system capacity as important predictors of projected NCD burden. Under current trajectories, countries with lower UHC indices had the steepest projected increases in DALYs, a cross-national gradient broadly consistent with global burden forecasts. These projections should not be interpreted as deterministic outcomes or as strong evidence for policy prioritization; We acknowledge that although 2025 has now elapsed, empirical validation of these modelled trajectories is pending the release of updated national surveillance data; readers should therefore interpret these figures as scenario-based estimates rather than observed outcomes. Rather, they can help identify settings where weak primary care infrastructure coincides with rapid lifestyle transition. The convergence of findings across regression, structural modeling, and algorithmic forecasting strengthens the overall pattern of evidence, but it does not establish causality. Within these limits, the results suggest that directing prevention resources toward high-burden, low-resilience settings may offer meaningful near-term opportunities for NCD prevention.
Limitations and Methodological Considerations
Despite its use of multiple analytic approaches, this study has several limitations. First, the ecological design precludes individual-level causal inference, even though the fixed-effects panel approach controlled for time-invariant country characteristics and common secular trends. In addition, although major macroeconomic and health system covariates were included, specific time-varying NCD risk factors, including longitudinal shifts in smoking rates, nutritional transitions, and ambient air pollution, were not modeled and may have introduced residual confounding. As noted above, this limitation may have led to modest overestimation of the independent association between physical inactivity and NCD burden and highlights the need for future studies that incorporate these dynamic covariates. Second, the limited panel size constrained model complexity and required the exclusion of Cambodia, Laos, and Myanmar because of incomplete data, which may limit regional generalizability and requires caution when extrapolating to lower-income contexts. Third, physical inactivity estimates were based on self-reported surveillance instruments, which are susceptible to recall bias and cross-country measurement heterogeneity; future studies should incorporate accelerometer-derived metrics where feasible to improve comparability. Finally, although the analysis accounted for the hierarchical data structure through within-country centering and country-clustered bootstrapping, the inclusion of only five countries limits the generalizability of the findings. A sup-plementary sensitivity analysis excluding the COVID-19 pan-demic years (2020–2021), reported in Supplemental Material 6, showed that the primary physical inactivity–DALY association remained directionally consistent and statistically significant, with attenuation of no more than 5.5%. The UHC interaction also remained directionally stable but was attenuated and borderline significant in the restricted sample (p=0.053), indicating that the moderation estimate should be interpreted cautiously. These findings nevertheless support the stability of the 2025 forecasting parameters. Future research should use quasi-experimental designs, such as instrumental-variable or difference-in-differences approaches, to strengthen causal identification in post-crisis longitudinal analyses.
The convergent findings from fixed-effects regression, structural equation modeling, and machine learning projections demonstrate that physical inactivity and health system capacity are interrelated determinants of near-term NCD trajectories in Southeast Asia. Physical inactivity was independently associated with NCD-related DALYs, cardiovascular mortality, and diabetes prevalence across the five ASEAN member states over the 12-year study period, while UHC expansion significantly attenuated this association. To translate these findings into policy action, we propose a three-level strategy: nationally, integrate physical activity screening and brief behavioral counseling into UHC primary care benefit packages; regionally, harmonize surveillance protocols and establish shared data platforms for real-time policy monitoring across ASEAN; and strategically, direct prevention funding toward countries where low system capacity coincides with the steepest projected increases in NCD burden, such as Indonesia and the Philippines. If policymakers prioritize one intervention area based on this evidence, expansion of primary care capacity may be particularly important because it can support both inactivity prevention and chronic disease management. Ultimately, physical activity promotion and UHC expansion appear to be complementary public health investment priorities. Sustainable progress toward Sustainable Development Goal 3 will likely require coordinated behavior-change strategies and health system reform rather than either approach alone, although extrapolation to the broader ASEAN region should remain cautious until more complete regional data are available.
In conclusion, the convergent findings from fixed-effects regression, structural equation modeling, and machine learning projections demonstrate that physical inactivity and health system capacity are interrelated determinants of near-term NCD trajectories in Southeast Asia. Physical inactivity was independently associated with NCD-related DALYs, cardiovascular mortality, and diabetes prevalence across the five ASEAN member states over the 12-year study period, while UHC expansion significantly attenuated this association, and the indirect pathway from physical inactivity through health system resilience to NCD burden supported a system-level reinforcing dynamic. Under the exploratory business-as-usual scenario, Indonesia and the Philippines had the highest projected increases in NCD burden by 2025 and may warrant priority attention. Physical activity promotion and UHC expansion appear to be complementary public health investment priorities for the five ASEAN countries studied. Sustainable progress toward Sustainable Development Goal 3 will likely require coordinated behavior-change strategies and health system reform rather than either approach alone, although extrapolation to the broader ASEAN region should remain cautious until more complete regional data are available.
All data were obtained from publicly accessible repositories, including the WHO Global Health Observatory, World Bank Open Data, the Global Burden of Disease Study 2021, and the United Nations Sustainable Development Goals Database. Analytical code, including R scripts for panel regression and XGBoost/SHAP interpretation and the SmartPLS 4.0 project file, is available from the corresponding author upon reasonable request.
Supplemental materials are available at https://doi.org/10.3961/jpmph.26.543.

Supplemental Material 1.

Trends in Physical Inactivity and NCD- Related DALYs Across Five ASEAN Countries, 2010–2021
jpmph-26-543-Supplementary-Material-1.pdf

Supplemental Material 2.

Measurement Model Indicators, Factor Loadings, and Reliability Metrics
jpmph-26-543-Supplementary-Material-2.pdf

Supplemental Material 3.

Factor Loadings and Variance Inflation Factors for Latent Construct Indicators
jpmph-26-543-Supplementary-Material-3.pdf

Supplemental Material 4.

PLS-SEM Model Fit Indices
jpmph-26-543-Supplementary-Material-4.pdf

Supplemental Material 5.

(A) Results of PLS-SEM Hypothesis Testing (Bootstrap Resamplses)
jpmph-26-543-Supplementary-Material-5.pdf

Supplemental Material 6.

SHAP (Shapley Additive Explanations) summary plot displaying the top 10 predictive features for NCD-related DALYs per 100,000 population projection in 2025 (XGBoost).
jpmph-26-543-Supplementary-Material-6.pdf

Conflict of Interest

The authors have no conflicts of interest associated with the material presented in this paper.

Funding

None.

Acknowledgements

None.

Author Contributions

Conceptualization: Yunus M, Azam M. Formal analysis: Aditya RS, Azam M. Funding acquisition: None. Methodology: Yunus M, Wahjuni ES, Aditya RS. Visualization: Azam M, Aditya RS. Writing – original draft: Yunus M, Supriyadi S. Writing – review & editing: Yunus M, Azam M, Wahjuni ES, Aditya RS, AlMutairi RL, Supriyadi S.

Figure 1
Temporal trends in (A) physical inactivity prevalence and (B) non-communicable disease (NCD)-related disability-adjusted life years (DALYs) across five Association of Southeast Asian Nations countries, 2010–2021. Lines represent locally estimated scatterplot smoothing trajectories, and shaded ribbons represent 95% confidence intervals. The vertical dashed line indicates the onset of the coronavirus disease 2019 (COVID-19) pandemic in 2020. Data sources: World Health Organization Global Health Observatory and Global Burden of Disease Study 2021.
jpmph-26-543f1.jpg
Figure 2
Partial least squares structural equation modeling path diagram showing relationships among physical inactivity (PI), health system resilience (HSR), socioeconomic status (SES), urbanization (URB), and non-communicable disease burden (NCDB). Standardized β coefficients were estimated using SmartPLS 4.0 with 5000 bootstrap resamples. Solid arrows indicate significant direct associations, and dashed arrows indicate the exploratory longitudinal indirect association pathway. R2 values are shown for endogenous constructs. Model fit indices were as follows: standardized root mean square residual=0.061, normed fit index=0.873, root mean square error of approximation=0.068, goodness-of-fit=0.624, and Stone–Geisser Q2=0.448. *p<0.05, **p<0.01, ***p<0.001.
jpmph-26-543f2.jpg
jpmph-26-543f3.jpg
Table 1
Demographic, socioeconomic, and health system characteristics of five ASEAN countries, 2010–20211
Characteristics Indonesia Malaysia Thailand Philippines Vietnam
Population (million, 2021) 273.5 32.4 69.8 109.6 97.3
Urbanization rate (%)
 2010 50.2 70.8 46.4 47.3 30.5
 2021 57.3 77.2 52.2 47.8 38.4
 Change (%) +14.1 +9.0 +12.5 +1.1 +25.9
GDP per capita (PPP, 2017 USD)
 2010 8920 22 180 13 540 6340 4210
 2021 12 450 28 760 18 920 8670 9540
 Change (%) +39.6 +29.7 +39.7 +36.8 +126.8
Health expenditure per capita (PPP, USD)
 2010 287 891 542 198 142
 2021 412 1247 789 287 398
Physician density (per 10 000)
 2010 2.8 14.2 7.9 5.3 7.1
 2021 4.3 16.8 9.4 6.7 9.8
UHC Service Coverage Index (0–100)
 2010 48 68 73 47 64
 2021 57 76 80 56 74
Life expectancy at birth (year)
 2010 69.2 74.1 74.8 68.9 75.2
 2021 69.8 75.0 77.2 69.4 75.4

ASEAN, Association of Southeast Asian Nations; GDP, gross domestic product; USD, US dollar; PPP, purchasing power parity; UHC, universal health coverage.

1 Missing data (2.1–6.4%) were handled via multiple imputation by chained equations (m=20).

Sources from: World Bank Open Data (2023), United Nations Sustainable Development Goals Database (2023), and World Health Organization Global Health Observatory (2023).

Table 2
Fixed-effects panel regression of physical inactivity on NCD outcomes (n=60 country-years)1
Variables β coefficient (95% CI) SE p-value Adjusted R2
Primary outcome
 NCD DALYs per 100 000 0.421 (0.251, 0.591) 0.087 <0.001 0.68
Secondary outcomes
 Cardiovascular mortality per 100 000 0.318 (0.193, 0.443) 0.064 <0.001 0.61
 Diabetes prevalence (%) 0.274 (0.160, 0.388) 0.058 0.002 0.57
 NCD mortality per 100 000 0.389 (0.221, 0.557) 0.085 <0.001 0.64
Interaction analysis for NCD DALYs
 Physical inactivity×UHC index −0.183 (−0.358, −0.008) 0.089 0.041 0.71

NCD, non-communicable disease; DALY, disability-adjusted life year; CI, confidence interval; SE, standard error; UHC, universal health coverage; GDP, gross domestic product; PPP, purchasing power parity.

1 All models used two-way fixed effects (country and year) and Driscoll–Kraay standard errors to account for heteroskedasticity and cross-sectional dependence; The primary and secondary outcome models were adjusted for GDP per capita (PPP), urbanization rate (%), and health expenditure per capita; The interaction model additionally included the UHC index and the physical inactivity×UHC index term; The UHC index was rescaled to 10-point units in the interaction model; For the main effects, β coefficients represent the estimated change in the outcome associated with a one-percentage-point increase in physical inactivity prevalence; For the interaction term, the β coefficient represents the change in the association between physical inactivity and NCD burden for each 10-point increase in the UHC index, rather than the effect of a one-percentage-point increase in physical inactivity itself; For the interaction term, because the UHC index was rescaled to 10-point units, the coefficient (β=−0.183) represents the absolute change in the physical inactivity slope (in DALYs per 100 000 population) for each 10-point increase in the UHC index, rather than a relative percentage reduction.

Table 3
Projected NCD-related DALYs in 2025 under an exploratory business-as-usual scenario using XGBoost1
Country DALYs 2021 (actual) Projected DALYs 2025 Δ (%) 95% CI for Δ (%) Priority level
LL UL
Indonesia 21 712 25 480 +17.3 13.8 20.8 High
Malaysia 14 105 15 630 +10.8 7.9 13.7 Moderate
Thailand 14 726 15 870 +7.8 5.1 10.5 Low
Philippines 20 570 24 890 +20.9 17.2 24.6 High
Vietnam 15 420 16 840 +9.2 6.3 12.1 Moderate

NCD, non-communicable disease; DALYs, disability-adjusted life years per 100 000 population; CI, confidence interval; LL, lower limit; UL, upper limit; MAE, mean absolute error; SHAP, Shapley additive explanations; GDP, gross domestic product.

1 XGBoost model: training 2010–2018 (n=45), validation 2019–2021 (n=15); Model validation: MAE=1830 DALYs per 100 000 population; R2=0.847; Top predictors by SHAP: physical inactivity prevalence (1.847), GDP growth rate (1.423), Universal health coverage index (1.218); 95% CIs represent the lower and upper limits of the projected percentage change (Δ) derived from 1000 bootstrap resamples; Priority level classification is exploratory and was defined post hoc based on the projected percentage change (Δ) in NCD-related DALYs: high (≥15.0%), moderate (8.0 to <15.0%), and low (<8.0%); These thresholds are intended solely for exploratory scenario planning and do not represent prespecified, validated clinical or epidemiological cutoffs.

Figure & Data

References

    Citations

    Citations to this article as recorded by  

      Figure
      • 0
      • 1
      • 2
      Physical Inactivity and Non-communicable Disease Burden in Five ASEAN Countries: A 12-Year Panel Analysis Using Structural Equation Modeling and Machine Learning Projections
      Image Image Image
      Figure 1 Temporal trends in (A) physical inactivity prevalence and (B) non-communicable disease (NCD)-related disability-adjusted life years (DALYs) across five Association of Southeast Asian Nations countries, 2010–2021. Lines represent locally estimated scatterplot smoothing trajectories, and shaded ribbons represent 95% confidence intervals. The vertical dashed line indicates the onset of the coronavirus disease 2019 (COVID-19) pandemic in 2020. Data sources: World Health Organization Global Health Observatory and Global Burden of Disease Study 2021.
      Figure 2 Partial least squares structural equation modeling path diagram showing relationships among physical inactivity (PI), health system resilience (HSR), socioeconomic status (SES), urbanization (URB), and non-communicable disease burden (NCDB). Standardized β coefficients were estimated using SmartPLS 4.0 with 5000 bootstrap resamples. Solid arrows indicate significant direct associations, and dashed arrows indicate the exploratory longitudinal indirect association pathway. R2 values are shown for endogenous constructs. Model fit indices were as follows: standardized root mean square residual=0.061, normed fit index=0.873, root mean square error of approximation=0.068, goodness-of-fit=0.624, and Stone–Geisser Q2=0.448. *p<0.05, **p<0.01, ***p<0.001.
      Graphical abstract
      Physical Inactivity and Non-communicable Disease Burden in Five ASEAN Countries: A 12-Year Panel Analysis Using Structural Equation Modeling and Machine Learning Projections
      Characteristics Indonesia Malaysia Thailand Philippines Vietnam
      Population (million, 2021) 273.5 32.4 69.8 109.6 97.3
      Urbanization rate (%)
       2010 50.2 70.8 46.4 47.3 30.5
       2021 57.3 77.2 52.2 47.8 38.4
       Change (%) +14.1 +9.0 +12.5 +1.1 +25.9
      GDP per capita (PPP, 2017 USD)
       2010 8920 22 180 13 540 6340 4210
       2021 12 450 28 760 18 920 8670 9540
       Change (%) +39.6 +29.7 +39.7 +36.8 +126.8
      Health expenditure per capita (PPP, USD)
       2010 287 891 542 198 142
       2021 412 1247 789 287 398
      Physician density (per 10 000)
       2010 2.8 14.2 7.9 5.3 7.1
       2021 4.3 16.8 9.4 6.7 9.8
      UHC Service Coverage Index (0–100)
       2010 48 68 73 47 64
       2021 57 76 80 56 74
      Life expectancy at birth (year)
       2010 69.2 74.1 74.8 68.9 75.2
       2021 69.8 75.0 77.2 69.4 75.4
      Variables β coefficient (95% CI) SE p-value Adjusted R2
      Primary outcome
       NCD DALYs per 100 000 0.421 (0.251, 0.591) 0.087 <0.001 0.68
      Secondary outcomes
       Cardiovascular mortality per 100 000 0.318 (0.193, 0.443) 0.064 <0.001 0.61
       Diabetes prevalence (%) 0.274 (0.160, 0.388) 0.058 0.002 0.57
       NCD mortality per 100 000 0.389 (0.221, 0.557) 0.085 <0.001 0.64
      Interaction analysis for NCD DALYs
       Physical inactivity×UHC index −0.183 (−0.358, −0.008) 0.089 0.041 0.71
      Country DALYs 2021 (actual) Projected DALYs 2025 Δ (%) 95% CI for Δ (%) Priority level
      LL UL
      Indonesia 21 712 25 480 +17.3 13.8 20.8 High
      Malaysia 14 105 15 630 +10.8 7.9 13.7 Moderate
      Thailand 14 726 15 870 +7.8 5.1 10.5 Low
      Philippines 20 570 24 890 +20.9 17.2 24.6 High
      Vietnam 15 420 16 840 +9.2 6.3 12.1 Moderate
      Table 1 Demographic, socioeconomic, and health system characteristics of five ASEAN countries, 2010–20211

      ASEAN, Association of Southeast Asian Nations; GDP, gross domestic product; USD, US dollar; PPP, purchasing power parity; UHC, universal health coverage.

      Missing data (2.1–6.4%) were handled via multiple imputation by chained equations (m=20).

      Sources from: World Bank Open Data (2023), United Nations Sustainable Development Goals Database (2023), and World Health Organization Global Health Observatory (2023).

      Table 2 Fixed-effects panel regression of physical inactivity on NCD outcomes (n=60 country-years)1

      NCD, non-communicable disease; DALY, disability-adjusted life year; CI, confidence interval; SE, standard error; UHC, universal health coverage; GDP, gross domestic product; PPP, purchasing power parity.

      All models used two-way fixed effects (country and year) and Driscoll–Kraay standard errors to account for heteroskedasticity and cross-sectional dependence; The primary and secondary outcome models were adjusted for GDP per capita (PPP), urbanization rate (%), and health expenditure per capita; The interaction model additionally included the UHC index and the physical inactivity×UHC index term; The UHC index was rescaled to 10-point units in the interaction model; For the main effects, β coefficients represent the estimated change in the outcome associated with a one-percentage-point increase in physical inactivity prevalence; For the interaction term, the β coefficient represents the change in the association between physical inactivity and NCD burden for each 10-point increase in the UHC index, rather than the effect of a one-percentage-point increase in physical inactivity itself; For the interaction term, because the UHC index was rescaled to 10-point units, the coefficient (β=−0.183) represents the absolute change in the physical inactivity slope (in DALYs per 100 000 population) for each 10-point increase in the UHC index, rather than a relative percentage reduction.

      Table 3 Projected NCD-related DALYs in 2025 under an exploratory business-as-usual scenario using XGBoost1

      NCD, non-communicable disease; DALYs, disability-adjusted life years per 100 000 population; CI, confidence interval; LL, lower limit; UL, upper limit; MAE, mean absolute error; SHAP, Shapley additive explanations; GDP, gross domestic product.

      XGBoost model: training 2010–2018 (n=45), validation 2019–2021 (n=15); Model validation: MAE=1830 DALYs per 100 000 population; R2=0.847; Top predictors by SHAP: physical inactivity prevalence (1.847), GDP growth rate (1.423), Universal health coverage index (1.218); 95% CIs represent the lower and upper limits of the projected percentage change (Δ) derived from 1000 bootstrap resamples; Priority level classification is exploratory and was defined post hoc based on the projected percentage change (Δ) in NCD-related DALYs: high (≥15.0%), moderate (8.0 to <15.0%), and low (<8.0%); These thresholds are intended solely for exploratory scenario planning and do not represent prespecified, validated clinical or epidemiological cutoffs.


      JPMPH : Journal of Preventive Medicine and Public Health
      TOP