This analysis employed a repeated cross-sectional study, using national data from the 2015 and 2021 STEPS surveys conducted in Viet Nam. Both surveys followed the standardized WHO STEPwise approach to NCD risk factor surveillance [16] and included individuals aged 15 years and older. A two-stage stratified cluster sampling design was applied, using Enumeration Areas as the primary sampling units, to ensure national representativeness across all 63 provinces and cities. For STEPS 2 and 3, anthropometric and biochemical indicators were collected only from participants aged 18 years and older. Accordingly, this study focused on respondents aged 18 and above who completed all three components of the survey (interview, physical measurements, and blood test), yielding analytic samples of 3,074 in 2015 and 3,712 in 2021. For latent class analysis (LCA), we restricted the sample to STEPS 2021 participants with complete data on all behavioral risk factors and required blood tests, yielding a final analytic sample of n = 3,306. LCA was applied to the 2021 dataset only because it provides the most recent cross-sectional snapshot of population risk profiles and ensures consistency in measurement across all indicators.

Four NCD behavioral risk factors were examined: current smoking, current drinking, insufficient physical activity, and low consumption of fruits and vegetables. Current smoking was defined based on the question “Do you currently smoke tobacco on a daily basis, less than daily, or not at all?”; participants who reported daily or less-than-daily use were classified as current smokers. Current drinking was defined as having consumed at least one standard alcoholic drink within the past 30 days. Physical activity was measured according to the WHO recommendation of at least 600 MET-minutes per week across work, transport, and leisure domains; participants with a total score below this threshold were considered not to meet the recommendation. Finally, fruit and vegetable intake was assessed as the average number of servings per day; fewer than 5 servings per day were classified as insufficient. To standardize data collection, interviewers used visual showcards (for tobacco, alcohol, fruits, vegetables, and physical activity) to help participants report quantities and frequencies in standard units. Data were collected using electronic data capture systems in both survey rounds; the 2021 survey used a REDCap mobile application with integrated questionnaires and showcards, while the 2015 survey used an earlier electronic platform.

Four metabolic abnormalities were assessed: hypertension, elevated fasting blood glucose, overweight/obesity, and elevated total cholesterol. Hypertension was defined as systolic blood pressure ≥130 mmHg, diastolic blood pressure ≥85 mmHg, or current use of antihypertensive medication [17]. Elevated fasting blood glucose (indicative of diabetes or prediabetes) was defined as ≥5.6 mmol/L or current use of diabetes medication. Overweight/obesity (BMI ≥23 kg/m²) was classified according to cutoffs commonly used for Asian populations. Elevated total cholesterol was defined as a concentration of ≥5.2 mmol/L [13, 14]. Physical measurements were obtained using standardized WHO-recommended equipment, including digital blood pressure monitors (BOSO), electronic weighing scales (Seca), and stadiometers and measuring tapes for anthropometry. Biochemical measurements, including fasting blood glucose and total cholesterol, were conducted using point-of-care devices (CardioChek Plus) following standardized STEPS protocols, with consistent procedures applied across both survey rounds.

Demographic characteristics were assessed across several domains. Age was categorized into five groups: 18–24 years, 25–34 years, 35–44 years, 45–54 years, and ≥55 years. Education level was classified as primary school or less, secondary school, high school, university/college, or no information. Marital status was categorized into three groups: not married, married, and divorced/widowed. Occupation referred to the main activity over the past 12 months and included working for a government or non-government agency/organization, being self-employed, being a student, being a homemaker, being retired, or being in another category. Sex was recorded as male or female, place of residence as urban or rural, and ethnicity as Kinh or other. Socioeconomic status was measured using the household wealth index, constructed through principal component analysis of household assets and housing characteristics. This index is standardized in many large household surveys in low- and middle-income countries and has been validated in Viet Nam [18]. The index was divided into quintiles, ranging from the poorest (Q1) to the richest (Q5).

We used the svy procedures in Stata 18 to estimate the survey-weighted prevalence and 95% confidence intervals (CIs) for all NCD risk factors and the clustering of metabolic abnormalities for both 2015 and 2021. Survey weights were applied in descriptive prevalence analyses to account for the complex sampling design. For clustering patterns, we present: (i) the proportion of individuals with zero, one, two, three, or four metabolic abnormalities and (ii) the most common two- and three-factor combinations.

To identify latent risk profiles, we conducted a latent class analysis (LCA) using 2021 data, incorporating all 8 binary indicators (4 behavioral and 4 metabolic). These indicators were selected a priori based on their established role as core behavioral and metabolic risk factors within the WHO STEPS framework and their availability across survey components [16]. The LCA models were estimated without directly incorporating survey weights or complex survey design variables because of methodological limitations in latent class estimation with gsem and lclass () in Stata. We evaluated candidate models with 2–7 classes using model fit criteria, including the Bayesian Information Criterion (BIC), Akaike Information Criterion (AIC), sample-size adjusted BIC, entropy, and clinical interpretability. Model fit statistics showed that the three-class model had the lowest Bayesian Information Criterion (BIC), indicating the best balance between model fit and parsimony. Although AIC continued to decrease with additional classes, improvements beyond three classes were modest. Entropy values were similar across models, suggesting no clear gain in classification quality with more complex solutions. Therefore, the three-class model was selected based on BIC, parsimony, interpretability, and conceptual coherence (see Supplementary Material 1). We report class-specific item-response probabilities and estimated population proportions (with 95% CIs). Classes were labeled based on dominant response patterns as follows: “Health-Conscious Group” (Class 1), “Metabolic-Risk Aware” (Class 2), and “High Behavioral Risk with Moderate Metabolic Comorbidity” (Class 3).

Participants were assigned to their most likely latent class using modal posterior probabilities. We then used multinomial logistic regression to examine associations between class membership and sociodemographic characteristics, treating Class 1 as the reference category. Covariates included age group, sex, education level, ethnicity, urban/rural residence, and household socioeconomic status (measured in quintiles), selected a priori based on epidemiological relevance and data availability. Results are presented as Odds Ratios (ORs) with 95% CIs and p-values, using a significance threshold of p = 0.05. A sensitivity analysis restricted to individuals with high classification certainty (posterior probability >0.80) yielded similar findings in both direction and magnitude.

There was no funding source for this study.

This study analyzed de-identified secondary data from the 2015 and 2021 WHO STEPwise (STEPS) surveys in Viet Nam. The national STEPS surveys received ethics approval from the Ministry of Health and the Hanoi University of Public Health, and written informed consent was obtained from all participants before data collection. Since the datasets were fully de-identified, this secondary analysis did not require additional approval from the institutional review board.

The analytic sample comprised 3,074 participants in 2015 and 3,712 in 2021, aged 18 years and older. In both survey years, the largest age groups were 35–44 and 45–54 years. Most participants were married and identified as members of the Kinh ethnic group. Educational attainment ranged from primary to tertiary, with a substantial proportion having completed high school or higher. Self-employment was the most common occupation. Both men and women were included, with a near-equal sex distribution in 2021. Urban and rural residents were well-represented across the survey years. Household socioeconomic status was relatively evenly distributed across quintiles in both survey years. However, the proportion in the middle quintile (Q3) increased from 15.7% in 2015 to 21.1% in 2021, while the proportions in the lower quintiles declined modestly. Detailed sociodemographic characteristics are presented in Table 1.

Table 2 presents the prevalence of risk factors in 2015 and 2021 among individuals who completed all three rounds of the STEPS survey. For behavioral risk factors, current smoking showed a slight decline, from 25.54% (95% CI: 23.58–27.50) to 22.56% (20.62–24.49), with overlapping confidence intervals suggesting no clear evidence of change; current drinking declined from 44.17% to 38.35%; low fruit and vegetable intake also decreased from 27.47% to 22.51%; while insufficient physical activity remained high with little change (56.87% in 2015% vs. 59.47% in 2021).

All four metabolic indicators showed increasing trends over time: hypertension rose from 30.31% (28.35–32.27) to 41.13% (38.86–43.40); elevated fasting blood glucose increased sharply from 11.99% (10.63–13.36) to 37.67% (35.25–40.09), marking the largest absolute change; overweight/obesity (BMI ≥23 kg/m²) increased from 33.87% (31.80–35.94) to 38.68% (36.37–41.00); and elevated total cholesterol increased from 23.54% (21.73–25.35) to 35.26% (32.98–37.53).

Table 3 presents the clustering of metabolic risk factors across four abnormalities: overweight/obesity, hypertension, elevated fasting blood glucose, and elevated total cholesterol. The proportion of adults with no abnormality declined substantially from 40.2% (95% CI: 37.9–42.5) in 2015 to 23.3% (21.1–25.7) in 2021. Conversely, the prevalence of individuals with two or more metabolic abnormalities increased markedly from 27.6% to 46.6%. The share with exactly two abnormalities rose from 18.9% to 26.5%, while those with three or more abnormalities more than doubled, from 8.7% (three: 6.9%; four: 1.8%) to 20.1% (three: 15.7%; four: 4.4%).

Combinations involving elevated fasting blood glucose increased over time: the prevalence of hypertension and elevated fasting blood glucose rose from 1.4% to 4.4%, overweight/obesity and elevated fasting blood glucose from 1.2% to 3.8%, and elevated fasting blood glucose and elevated total cholesterol from 1.1% to 3.3%. Triple clusters including glucose (e.g., hypertension, elevated fasting blood glucose, and overweight/obesity) increased from 1.4% to 4.9%, and the share with all four abnormalities more than doubled, from 1.8% to 4.4%.

In 2021, LCA identified three distinct classes of behavioral and metabolic risk among participants. Class 1, “Health-Conscious” (38.1%), showed low probabilities across all risk factors, including smoking, alcohol use, physical inactivity, low fruit and vegetable intake, and metabolic indicators. Class 2, “Metabolic Risk-Aware” (34.6%), had low behavioral risk but higher levels of metabolic abnormalities, including overweight/obesity, hypertension, elevated fasting blood glucose, and elevated total cholesterol. Class 3, “High Behavioral Risk with Moderate Metabolic Comorbidity” (27.3%), was characterized by high probabilities of smoking and alcohol use together with moderately elevated metabolic indicators. Class-specific item-response probabilities are presented in Table 4.

Older age was strongly associated with membership in both higher-risk classes. Compared with adults aged 18–24 years, individuals aged ≥55 years had substantially higher odds of membership in the “Metabolic Risk-Aware” class (OR: 25.97, 95% CI: 13.70–50.33, p < 0.001) and the “High Behavioral Risk with Moderate Metabolic Comorbidity” class (OR: 8.08, 95% CI: 4.90–13.72, p < 0.001).

Higher levels of education were associated with lower odds of membership in the higher-risk classes, particularly the “High Behavioral Risk with Moderate Metabolic Comorbidity” class. Individuals with university or college education had 57% lower odds of membership in this class (OR: 0.43, 95% CI: 0.30–0.63, p < 0.001) compared to those with primary education or less.

Women had significantly lower odds than men of membership in either higher-risk class, most notably for the “High Behavioral Risk with Moderate Metabolic Comorbidity class”, where the OR was 0.01 (95% CI: 0.01–0.02, p < 0.001).

Rural residence was associated with 30% lower odds of membership in the “Metabolic-Risk Aware” class (OR: 0.70, 95% CI: 0.58–0.84, p < 0.001), but no significant association was observed for the “High Behavioral Risk with Moderate Metabolic Comorbidity” class (p = 0.574).

Regarding household socioeconomic status, individuals in the richest quintile (Q5) had lower odds of membership in the “High Behavioral Risk with Moderate Metabolic Comorbidity” class than those in the poorest quintile (OR: 0.59, 95% CI: 0.39–0.90, p = 0.014). No significant socioeconomic gradient was observed for the “Metabolic-Risk Aware” group.

A sensitivity analysis restricted to individuals with high class assignment certainty (posterior probability >0.80) yielded similar results. While point estimates were modestly attenuated, particularly for age and SES, the overall direction and interpretation of the main associations (age, education, sex, and rurality) remained unchanged. Detailed multinomial regression results are presented in Table 5.