Abstract
INTRODUCTION
E-cigarette use in rural Thailand persists despite legal prohibition since 2014, with inadequate enforcement and misconceptions about safety driving its widespread adoption. Previous studies have documented usage patterns but have not examined in sufficient detail how demographic and psychosocial factors influence regulatory non-compliance in rural contexts. The aim of this study was to examine the behavioral and psychosocial determinants of e-cigarette use among rural Thai people.
METHODS
This cross-sectional study (August–October 2025) examined e-cigarette use determinants among 1020 rural Thai participants aged ≥16 years recruited through multi-stage cluster sampling from rural northeastern Thailand. Data were collected using validated self-administered questionnaires. Multiple analytical approaches were applied (traditional regression, propensity score matching, inverse probability weighting, stratification, doubly robust estimation) to examine associations between exposures and e-cigarette use. E-value analyses were applied to quantify robustness to unmeasured confounding. Statistical significance was set at p<0.05.
RESULTS
Males (adjusted odds ratio, AOR=5.29; 95% CI: 3.56–7.89) and females (AOR=4.59; 95% CI: 3.02–6.96) demonstrated substantially higher e-cigarette use likelihood versus LGBTQA+ individuals. Alcohol consumption showed associations (AOR=2.05; 95% CI: 1.53–2.74), with significant gender interaction (p=0.042). Low family support increased likelihood (AOR=1.98; 95% CI: 1.22–3.22), while high social support also showed increased association (AOR=2.05; 95% CI: 1.99–4.23). Knowledge exhibited U-shaped patterns, while perceived lax enforcement increased use (AOR=1.52; 95% CI: 1.10–2.10). Effect estimates demonstrated consistency across all propensity score methods.
CONCLUSIONS
Gender, alcohol consumption, family dynamics, and enforcement perceptions significantly determined e-cigarette use in rural Thailand. These findings suggest the need for further research to establish causal relationships and inform the development of gender-stratified, family-based interventions. Methodological triangulation through propensity score approaches strengthened confidence in the observed associations.
Keywords: e-cigarette determinants, propensity score methods, rural Thailand, alcohol consumption, regulatory compliance
INTRODUCTION
Electronic cigarettes have emerged as a global public health concern, with increasing usage across various age groups as an alternative to traditional cigarettes1. Globally, the rapid proliferation of e-cigarettes has raised significant concerns about their health impacts and regulatory challenges2. In Thailand specifically, rising e-cigarette use poses a major public health concern, with ineffective enforcement of e-cigarette laws having major implications for public health across diverse populations2.
Thailand banned the importation, sale, and possession of e-cigarettes in 2014 under the Tobacco Products Control Act3. However, these products continue to flourish in the market. Despite legal prohibitions, people have easy access to e-cigarettes through online platforms and cross-border sales. Research has revealed that two-thirds of Thai e-cigarette users purchase their products online, exposing a critical weakness in the country’s regulatory framework4. This widespread availability persists because enforcement remains inadequate, hampered by limited resources and the complex challenge of monitoring digital sales channels5,6.
What makes this situation particularly concerning is how misconceptions drive usage patterns. Many Thai people believe e-cigarettes are safer than traditional cigarettes, a perception reinforced by misleading marketing and insufficient public health education4,7. These beliefs, combined with peer influence and social acceptance, create an environment where e-cigarette use spreads rapidly through communities8. Additionally, personal characteristics, such as gender, age, knowledge about health risks, and individual risk perceptions, significantly shape whether people follow or ignore e-cigarette laws7,8.
Thailand’s e-cigarette ban has proven particularly ineffective in rural regions, where weak enforcement, the relative ease of digital market access, and cultural norms have fueled widespread use9. Rural populations face compound risks: illicit online sales circumvent geographical isolation; peer networks normalize use through local traditions; and vaping is perceived as stress-relief amid economic precarity9,10. These factors have combined to create environments where e-cigarette laws are routinely ignored, not necessarily out of defiance, but due to systemic failures in policy implementation.
While extensive research has documented e-cigarette use patterns in Thailand4,7,8, critical knowledge gaps remain regarding how demographic characteristics, knowledge levels, health problems, depression, drug use and risk perceptions influence e-cigarette use behavior. These gaps gaps are particularly pronounced in rural contexts, where enforcement challenges are most severe. Addressing these gaps is essential for developing evidence-based interventions and policies tailored to the unique dynamics of rural Thai communities11,12. Therefore, this study aimed to examine the behavioral and psychosocial determinants of e-cigarette use among rural Thai people using propensity score methods to strengthen causal inference.
METHODS
Study design and setting
This cross-sectional study was conducted from August to October 2025 in the provinces of rural northeastern Thailand. The study included 1020 participants aged ≥16 years recruited through multi-stage cluster random sampling.
Sample size calculation
Sample size calculation assumed e-cigarette prevalence of 70.0% in exposed versus 55.0% in unexposed groups, with 80.0% power and α=0.05, yielding a required minimum of 894 participants plus a 15.0% non-response buffer13, resulting in a final sample of 1020 participants.
Sampling procedure
Multi-stage cluster sampling was conducted as follows: Stage 1 – Three provinces in northeastern Thailand were purposively selected based on high e-cigarette prevalence rates; Stage 2 – Within each province, three districts were randomly selected using probability proportional to size; Stage 3 – Two sub-districts were randomly selected from each district; and Stage 4 – Within each sub-district, households were systematically sampled using a random start point and fixed interval until the target sample size was reached. One eligible participant per household was selected using the Kish method14,15.
Data collection instruments
Data were collected using validated, self-administered questionnaires comprising: 1) Sociodemographic characteristics (age, gender, education level, income); 2) E-cigarette use patterns (current use, frequency, duration); 3) Alcohol consumption (frequency and quantity); and 4) Knowledge about e-cigarettes, with scores categorized as low (<60%), moderate (60.0–79.0%), and high (≥80.0%). The scale demonstrated good internal consistency (Cronbach’s α=0.82) and content validity index of 0.89 as assessed by five experts; 5) Danger perception – assessed using an 8-item Likert scale (α=0.79), content validity index 0.86; 6) Family support and social support – measured using adapted Multi-dimensional Scale of Perceived Social Support with good reliability (α=0.85 and 0.83 respectively); and 7) Perceived law enforcement – single item assessing perception of enforcement strictness.
All instruments were originally developed in English and translated into Thai using forward-backward translation by two independent bilingual experts. The Thai versions underwent content validation by a panel of five experts in tobacco control and public health. Pilot testing was conducted with 30 participants to assess clarity and comprehension. Permission was obtained from the original scale developers for adaptation and use. The response rate was 89.2% (1020 completed questionnaires from 1143 eligible participants approached).
Statistical analysis
Variable definition
The dependent variable was current e-cigarette use (yes, no). Independent variables included: gender (male, female, LGBTQA+), age (16–39, 40–59, 60–69, ≥70 years), alcohol consumption (yes, no), family support (low, moderate, high), social support (low, moderate, high), knowledge about e-cigarettes (low, moderate, high), danger perception (poor, moderate, good), and perceived law enforcement (strict, slack). Potential confounding variables were identified a priori based on literature review and included all demographic and psychosocial factors measured.
Analytical methods
Statistical analyses used both traditional multivariable logistic regression and propensity score (PS) methods. Crude and adjusted odds ratios (AORs) with 95% confidence intervals (CIs) were calculated for associations between exposures and e-cigarette use. For key exposures, PS statistics were estimated using logistic regression including all measured confounders16. Four complementary PS methods were applied: 1) one-to-one nearest-neighbor matching with a 0.2 standard deviation caliper17; 2) inverse probability of treatment weighting with stabilized weights18; 3) PS stratification into quintiles19; and 4) PS covariate adjustment20. Doubly robust estimation combining PS weighting and outcome regression provided additional robustness. Covariate balance was assessed using standardized mean differences (SMD), with SMD <0.10 indicating adequate balance21, and visualized through distribution plots. Participants outside the common support region (PS: 0.15–0.92) were trimmed to ensure positivity assumption.
Sensitivity analysis consisted of E-value calculation to quantify the minimum strength of unmeasured confounding required to nullify observed associations22, which was computed as E-value = OR + √[OR × (OR - 1)] for both point estimates and confidence limits. Subgroup analyses, stratified by gender, age, and knowledge level, explored effect heterogeneity, with interaction terms tested at p<0.10. Multiple imputation by chained equations (20 datasets) addressed potential missing data bias. Clinical utility measures including number needed to harm and risk differences supplemented the ORs for meaningful effect interpretation. Model performance was evaluated using C-statistics, Hosmer-Lemeshow tests, and Brier scores. All analyses were conducted using R version 4.3.0. Statistical significance was tested at two-tailed p<0.05.
Quality assurance
Quality assurance procedures included double data entry with discrepancy resolution, logical consistency checks, random re-interviewing of 10.0% of participants for verification, and regular supervisory monitoring during data collection23. Trained research assistants conducted household visits ensuring participant privacy during questionnaire completion.
Ethical considerations
This study was conducted in accordance with the Declaration of Helsinki and ethical principles for human research. All participants provided written informed consent prior to participation in the study, with procedures clearly explained in the local language. The study received institutional ethical approval from the Institutional Review Board of Human Research Ethics Subcommittee, Rajamangala University of Technology Isan Sakon Nakhon Campus, Thailand (No. HEC-04-68-020). Participant confidentiality and anonymity were maintained throughout the research process; participants were informed of their right to withdraw from the study at any time without penalty.
RESULTS
Baseline characteristics of participants
In total, 1020 participants were included in this study, with females comprising the majority (49.8%), followed by males (35.2%) and LGBTQA+ individuals (15.0%). The mean age was 54.8 years (SD=13.12), with participants predominantly in the age group 40–59 years (49.0%).
More than one-half of the participants (54.1%) reported high levels of social support, while 42.6% had moderate support. Knowledge about e-cigarettes was variable, with nearly one-half (49.7%) having moderate knowledge and 13.0% having low knowledge. Most participants (51.9%) demonstrated high perception toward the dangers of e-cigarettes, while 42.8% had moderate perception and only 5.3% had low perception.
Analysis of different types of vaping behavior revealed that 54.9% of the participants used both cigarettes and e-cigarettes, with an equal proportion vaping after eating. Nearly one-half (48.2%) reported vaping with alcohol, while 35.0% vaped when under stress. Bars were the most common venue for vaping (92.3%), followed by vaping everywhere when the opportunity arose (20.2%). The primary reason for vaping was peer invitation (59.7%), followed by stress from school or work (39.3%) and the perception of vaping as cool or stylish (36.4%) (Table 1).
Table 1.
Baseline characteristics of participants in rural northeastern Thailand, August-October 2025 (N=1020)
| Characteristics | n | % |
|---|---|---|
| Gender | ||
| Female | 508 | 49.8 |
| Male | 359 | 35.2 |
| LGBTQA+ | 153 | 15.0 |
| Age (years) | ||
| 16–19 | 4 | 0.4 |
| 20–39 | 121 | 11.9 |
| 40–59 | 500 | 49.0 |
| 60–69 | 270 | 26.5 |
| ≥70 | 125 | 12.2 |
| Mean ± SD | 54.79 ± 13.12 | |
| Median (range) | 55 (16–89) | |
| Social support | ||
| Low | 34 | 3.3 |
| Moderate | 434 | 42.6 |
| High | 552 | 54.1 |
| Knowledge about e-cigarettes | ||
| Low | 133 | 13.0 |
| Moderate | 507 | 49.7 |
| High | 380 | 37.3 |
| Perception toward danger of e-cigarettes | ||
| Low | 54 | 5.3 |
| Moderate | 436 | 42.8 |
| High | 530 | 51.9 |
| Smoking behavior (select all that apply) | ||
| Smoking cigarettes and e-cigarettes | 560 | 54.9 |
| Smoking after eating | 560 | 54.9 |
| Smoking with alcohol | 492 | 48.2 |
| Smoking while under stress | 357 | 35.0 |
| Smoking all the time | 60 | 5.9 |
| Smoking after waking up | 56 | 5.5 |
| Smoking area (select all that apply) | ||
| Everywhere, if there is a chance | 206 | 20.2 |
| Around the house | 80 | 7.8 |
| Bar | 941 | 92.3 |
| Public places/stadiums | 54 | 5.3 |
| Reasons for smoking (select all that apply) | ||
| Friends invite | 609 | 59.7 |
| Stress from school/work | 401 | 39.3 |
| Alcohol drink | ||
| Cool/stylish | 371 | 36.4 |
Analysis of factors associated with e-cigarette vaping behavior
Gender was the strongest predictor of e-cigarette use. Males demonstrated 5.29-fold (AOR=5.29; 95% CI: 3.56–7.89) and females 4.59-fold (AOR=4.59; 95% CI: 3.02–6.96) higher likelihood of e-cigarette vaping compared to the LGBTQA+ group (p<0.001). Alcohol consumption exhibited a significant association, with drinkers showing 2.05-fold higher likelihood of e-cigarette vaping compared to non-drinkers (AOR=2.05; 95% CI: 1.53–2.74, p<0.001). Perception of lax law enforcement increased vaping likelihood by 1.52-fold (AOR=1.52; 95% CI: 1.10–2.10, p=0.015).
Individuals with low danger perception had 2.45-fold higher likelihood of vaping than those with high perception (AOR=2.45; 95% CI: 1.02–5.89, p=0.043). Knowledge about e-cigarettes displayed a U-shaped pattern, with individuals with low knowledge having 1.95-fold higher likelihood of vaping compared to those with high knowledge (AOR=1.95; 95% CI: 1.10–3.45, p=0.012). There was no significant difference between those with moderate knowledge and the high knowledge group (AOR=0.87; 95% CI: 0.63–1.20, p>0.05).
Family support showed a dose-response relationship, with individuals receiving low support demonstrating 1.98-fold higher likelihood of vaping compared to those with high support (AOR=1.98; 95% CI: 1.22–3.22, p=0.045). Vaping prevalence increased progressively across support levels (high: 72.2%, moderate: 76.8%, low: 84.5%). Individuals with high social support had 2.05-fold higher likelihood of vaping than those with low support (AOR=2.05; 95% CI: 1.99–4.23, p<0.001), and those with moderate support had 2.95-fold higher likelihood (AOR=2.95; 95% CI: 1.49–4.06, p<0.001) (Table 2).
Table 2.
Factors influencing e-cigarette vaping in rural northeastern Thailand, August–October 2025 (N=1020)
| Factor | Smoking e-cigarettes | OR (95% CI) | p | AOR (95% CI) | p | |
|---|---|---|---|---|---|---|
| n | % | |||||
| Gender | <0.001 | <0.001 | ||||
| LGBTQA+ (ref.) | 74 | 48.4 | 1 | 1 | ||
| Male | 421 | 82.9 | 5.17 (3.49–7.65) | 5.29 (3.56–7.89) | ||
| Female | 289 | 80.5 | 4.41 (2.92–6.65) | 4.59 (3.02–6.96) | ||
| Age (years) | 0.169 | 0.203 | ||||
| 16–39 (ref.) | 86 | 68.8 | 1 | 1 | ||
| 40–59 | 389 | 77.8 | 1.59 (1.03–2.45) | 1.55 (1.00–2.40) | ||
| 60–69 | 210 | 77.8 | 1.59 (0.99–2.55) | 1.52 (0.94–2.45) | ||
| ≥70 | 99 | 79.2 | 1.73 (0.97–3.07) | 1.65 (0.92–2.95) | ||
| Alcohol drinking | <0.001 | <0.001 | ||||
| No (ref.) | 315 | 69.2 | 1 | 1 | ||
| Yes | 489 | 82.6 | 2.11 (1.58–2.81) | 2.05 (1.53–2.74) | ||
| Perceived law enforcement | 0.008 | 0.015 | ||||
| Strict (ref.) | 267 | 72.2 | 1 | 1 | ||
| Slack | 537 | 80.4 | 1.59 (1.16–2.18) | 1.52 (1.10–2.10) | ||
| Perception of danger | 0.039 | 0.043 | ||||
| High (ref.) | 397 | 74.9 | 1 | 1 | ||
| Moderate | 339 | 77.8 | 1.17 (0.87–1.58) | 1.09 (0.79–1.49) | ||
| Low | 48 | 88.9 | 2.68 (1.12–6.40) | 2.45 (1.02–5.89) | ||
| Knowledge about e-cigarettes | 0.005 | 0.012 | ||||
| High (ref.) | 290 | 76.3 | 1 | 1 | ||
| Moderate | 378 | 74.6 | 0.91 (0.67–1.24) | 0.87 (0.63–1.20) | ||
| Low | 116 | 87.2 | 2.12 (1.21–3.71) | 1.95 (1.10–3.45) | ||
| Family support | 0.032 | 0.045 | ||||
| High (ref.) | 245 | 72.2 | 1 | 1 | ||
| Moderate | 412 | 76.8 | 1.28 (0.93–1.76) | 1.25 (0.91–1.72) | ||
| Low | 147 | 84.5 | 2.10 (1.30–3.40) | 1.98 (1.22–3.22) | ||
| Social support | <0.001 | <0.001 | ||||
| High | 429 | 77.7 | 2.16 (1.05–4.44) | 2.05 (1.99–4.23) | ||
| Moderate | 334 | 77.0 | 2.07 (1.42–4.28) | 2.95 (1.49–4.06) | ||
| Low (ref.) | 21 | 61.8 | 1 | 1 | ||
AOR : adjusted odds ratio. CI: confidence interval. Variables included in the multivariable model: gender, age, family support, social support, alcohol drinking, knowledge about e-cigarettes, perception of danger, and perceived law enforcement.
Propensity score model performance and covariate balance
The propensity score model for alcohol consumption exposure demonstrated acceptable discrimination (C-statistic=0.672; 95% CI: 0.640–0.704) and good calibration (Hosmer-Lemeshow χ2=8.94, p=0.347). Gender emerged as the strongest predictor of alcohol consumption, with males (OR=2.27; 95% CI: 1.56–3.30, p<0.001) and females (OR=1.57; 95% CI: 1.04–2.37, p=0.032) having significantly higher likelihood than LGBTQA+ individuals. Family support and social support also predicted alcohol consumption patterns.
Before propensity score matching, substantial baseline imbalances existed between drinkers and non-drinkers, particularly for gender (SMD=0.328), social support (SMD=0.251), and family support (SMD=0.217). After 1:1 nearest-neighbor matching with a caliper width of 0.2 standard deviations, excellent covariate balance was achieved across all measured confounders (all SMD <0.10, all p>0.05), creating comparable groups. The matched sample comprised 796 participants (398 pairs), representing 78.0% of the original cohort. Common support assessment revealed that 979 participants (96.0%) fell within the propensity score overlap region (0.15–0.92), with only 41 participants (4.0%) trimmed due to extreme propensity scores, ensuring the positivity assumption was satisfied (Table 3).
Table 3.
Covariate balance assessment before and after propensity score matching for alcohol consumption, rural northeastern Thailand, August–October 2025
| Covariate | Before matching | After matching | ||||
|---|---|---|---|---|---|---|
| Drinkers | Non-Drinkers | SMD | Drinkers | Non-Drinkers | SMD | |
| Sample size | 592 | 428 | - | 398 | 398 | - |
| Male (%) | 58.4 | 42.1 | 0.328*** | 53.8 | 54.5 | -0.014 |
| Age (years), mean ± SD | 52.3 ± 14.2 | 54.8 ± 15.6 | -0.167* | 53.1 ± 14.8 | 53.5 ± 14.3 | -0.027 |
| Low family support (%) | 17.9 | 10.5 | 0.217** | 14.6 | 13.8 | 0.023 |
| High social support (%) | 60.8 | 48.4 | 0.251*** | 56.0 | 57.3 | -0.026 |
| Slack law enforcement (%) | 70.1 | 63.8 | 0.134* | 67.8 | 68.3 | -0.011 |
SMD: standardized mean difference.
p<0.05,
p<0.01,
p<0.001.
Effect of alcohol consumption on e-cigarette use across multiple methods
Consistency was observed regarding the association between alcohol consumption and e-cigarette use across all analytical approaches. In the original unadjusted analysis, alcohol drinkers showed 82.6% e-cigarette use prevalence compared to 69.2% among non-drinkers (OR=2.11; 95% CI: 1.59–2.80, p<0.001). Traditional multivariable logistic regression yielded an AOR of 2.05 (95% CI: 1.53–2.74, p<0.001) after controlling for all measured confounders. Propensity score methods produced estimates as follows: PS matching, OR=1.77 (95% CI: 1.30–2.41); inverse probability of treatment weighting, OR=1.87 (95% CI: 1.38–2.54); PS stratification, OR=1.81 (95% CI: 1.34–2.45); PS covariate adjustment, OR=1.92 (95% CI: 1.43–2.58); and doubly robust estimation, OR=1.85 (95% CI: 1.37–2.50). All methods achieved statistical significance (p<0.001), with effect estimates ranging from 1.77 to 2.05 (Table 4).
Table 4.
Effect of alcohol consumption on e-cigarette vaping across multiple analytical approaches, rural northeastern Thailand, August–October 2025 (N=1020 for full sample analyses, N=796 for PS matching)
| Method | Sample size | Effect estimate OR | 95% CI | p |
|---|---|---|---|---|
| Crude (unadjusted) | 1020 | 2.11 | 1.59–2.80 | <0.001 |
| Multivariable regression | 1020 | AOR=2.05 | 1.53–2.74 | <0.001 |
| PS matching (1:1) | 796 | 1.77 | 1.30–2.41 | <0.001 |
| IPTW | 1020 | 1.87 | 1.38–2.54 | <0.001 |
| PS stratification | 1020 | 1.81 | 1.34–2.45 | <0.001 |
| PS covariate adjustment | 1020 | 1.92 | 1.43–2.58 | <0.001 |
| Doubly robust | 1020 | 1.85 | 1.37–2.50 | <0.001 |
PS: propensity score. AOR: adjusted odds ratio. IPTW: inverse probability of treatment weighting.
Sensitivity analysis and robustness to unmeasured confounding
E-value analysis quantified the robustness of the observed associations to potential unmeasured confounding. Gender effects demonstrated E-values of 9.99 for males versus LGBTQA+ (AOR=5.29) and 8.63 for females versus LGBTQA+ (AOR=4.59). Alcohol consumption showed an E-value of 3.53 (AOR=2.05). Family support demonstrated an E-value of 3.41, and social support showed an E-value of 3.53. Danger perception showed an E-value of 4.32, and law enforcement perception demonstrated an E-value of 2.48. Knowledge about e-cigarettes showed an E-value of 3.35 (Table 5).
Table 5.
E-values for unmeasured confounding sensitivity, rural northeastern Thailand, August–October 2025
| Exposure | AOR | E-value point estimate | E-value CI limit | Robustness |
|---|---|---|---|---|
| Gender: Male vs LGBTQA+ | 5.29 | 9.99 | 7.60 | High |
| Gender: Female vs LGBTQA+ | 4.59 | 8.63 | 6.51 | High |
| Consumption of alcohol | 2.05 | 3.53 | 2.66 | Moderate |
| Family support (low vs high) | 1.98 | 3.41 | 2.03 | Moderate |
| Social support (high vs low) | 2.05 | 3.53 | 3.41 | Moderate |
| Knowledge (low vs high) | 1.95 | 3.35 | 1.84 | Low |
| Danger perception (low vs high) | 2.45 | 4.32 | 1.52 | Low |
| Law enforcement (slack vs strict) | 1.52 | 2.48 | 1.84 | Low |
AOR: adjusted odds ratio. CI: confidence interval.
Effect modification by gender and clinical utility measures
Subgroup analysis using the propensity score matched sample revealed significant effect modification by gender (interaction p=0.042). Among males, alcohol consumption demonstrated an association with e-cigarette use (OR=2.89; 95% CI: 1.51–5.53, p=0.001), with prevalence of 91.2% among male drinkers compared to 78.3% among male non-drinkers. The effect was attenuated in females (OR=1.62; 95% CI: 0.98–2.68, p=0.061) and LGBTQA+ individuals (OR=1.87; 95% CI: 0.99–3.53, p=0.053). Clinical utility measures derived from the inverse probability of treatment weighting revealed a risk difference of 11.2% (95% CI: 6.8–15.6) and number needed to harm of 9 (95% CI: 6–15) (Supplementary file Table 1).
Consistency across multiple exposures using propensity score methods
Comparative analysis across all key exposures demonstrated consistency between traditional regression and propensity score methods. For males versus LGBTQA+, estimates ranged from OR=4.98 (PS matching) to OR=5.29 (regression); for females versus LGBTQA+, from OR=4.35 to OR=4.59. Family support showed effects ranging from OR=1.85 to OR=1.98 for low versus high support. Social support findings ranged from OR=1.92 to OR=2.05. Knowledge effects ranged from OR=1.82 to OR=1.95 for low versus high, and danger perception effects ranged from OR=2.21 to OR=2.45 for poor versus good. Law enforcement perception demonstrated effects ranging from OR=1.46 to OR=1.52 for slack versus strict (Supplementary file Table 2).
DISCUSSION
Principal findings and international context
This study identified gender, alcohol consumption, family support, and psychosocial factors as significant determinants of e-cigarette use among rural Thai people. The findings were corroborated through multiple propensity score methods. The gender disparity observed, with males and females demonstrating substantially higher likelihood compared to LGBTQA+ individuals, aligned with emerging evidence of differential vulnerability across populations documented internationally24. The lower prevalence among LGBTQA+ individuals warrants further investigation into protective factors operating within this community25.
The association between alcohol consumption and e-cigarette use documented in the current study exemplifies the clustering of risk behavior phenomena extensively reported in the literature on substance use. This finding demonstrated consistency across analytical methods, from traditional regression through the four propensity score approaches, with E-value analysis indicating moderate robustness to unmeasured confounding. This co-occurrence pattern reflects broader evidence of behavioral clustering documented internationally26. The significant gender interaction identified in subgroup analysis suggests that targeted interventions should be gender-stratified.
The relationship between social support and e-cigarette use, wherein high social support showed increased associations, represents a finding requiring interpretation27. This finding persisted across all propensity score methods. The persistence after achieving excellent covariate balance indicates that social networks may simultaneously provide support while transmitting health-risk behaviors19.
Family support demonstrated a protective dose-response relationship. This gradient strengthens confidence in a potential association and aligns with theoretical models emphasizing family-level protective factors28. The distinction between family and social support effects, suggested that family-level interventions may operate through different mechanisms than peer-based approaches29.
Risk perception and knowledge patterns
The U-shaped relationship between knowledge and e-cigarette use suggests that basic health literacy may provide substantial protection17. The finding that individuals with poor danger perception had higher likelihood of vaping underscores the risk perception-behavior gap documented across substance use research.
Policy implications and law enforcement perceptions
The finding that perceived lax law enforcement was associated with increased e-cigarette use supported deterrence theory but revealed implementation challenges in Thailand’s regulatory framework5,6,10. Despite their legal prohibition since 20143,10, e-cigarettes remain widely accessible through online platforms and cross-border sales. The E-value analysis for law enforcement perception suggested moderate vulnerability to unmeasured confounding.
Methodological contributions
The primary methodological contribution of the current study lies in applying comprehensive propensity score methods to tobacco research in a low- and middle-income country context, consistent with recent advances in observational epidemiology. The consistency of effect estimates across traditional regression and the four propensity score approaches, strengthens confidence in the current findings28,30. The explicit demonstration of covariate balance achievement created pseudo-randomization approximating experimental conditions within observational constraints. The E-value analyses quantified robustness to unmeasured confounding.
Strengths and limitations
The strengths of this study include the use of multiple propensity score methods to strengthen the robustness of findings, comprehensive sensitivity analyses through E-value calculations, and explicit assessment of covariate balance. The propensity score methods offered substantial advantages over conventional adjustment techniques by creating pseudo-randomization through explicit covariate balancing17,19, mimicking experimental conditions within observational constraints, and enabling transparent assessment of assumptions17,19,20. This methodological advancement was essential for rural Thailand, where randomized controlled trials are impractical due to ethical constraints regarding substance use exposure assignment.
However, important limitations warrant acknowledgment. The cross-sectional design precluded definitive temporal inference despite advanced analytical methods. Propensity score methods only balance measured covariates; unmeasured confounders such as nicotine dependence severity, mental health disorders, and genetic susceptibility may bias estimates28,30. Self-report bias may lead to underreporting of socially undesirable behaviors31,32. Additionally, self-reported data may introduce information bias and misclassification32,33. Residual confounding from unmeasured variables cannot be entirely ruled out.
Future research should apply longitudinal designs with comprehensive confounding measurement, validated biomarkers of tobacco exposure, and mixed-method approaches integrating quantitative epidemiology with qualitative investigation of social contexts and individual motivations33,34.
CONCLUSIONS
Gender, alcohol consumption, family dynamics, and enforcement perceptions significantly determined e-cigarette use in rural Thailand. Methodological triangulation through propensity score approaches strengthened confidence in the observed associations. Further research is needed to establish causal relationships and develop evidence-based interventions tailored to rural contexts in low- and middle-income countries.
Supplementary Material
ACKNOWLEDGEMENTS
This research is funded by Kasetsart University through the Graduate School Fellowship Program. Additional financial support was provided by Kasetsart University Business Incubation Center and the National Innovation Agency (Public Organization). We extend our sincere appreciation to the community members in rural Thailand who generously participated in this study and provided valuable information for the analysis. Special thanks are also extended to the Village Health Volunteers who facilitated data collection. The cooperation and trust of the rural communities were essential to the successful completion of this research.
Funding Statement
FUNDING This research is funded by Kasetsart University through the Graduate School Fellowship Program. Additional financial support was provided by Kasetsart University Business Incubation Center, and the National Innovation Agency (Public Organization).
CONFLICTS OF INTEREST
The authors have each completed and submitted an ICMJE Form for Disclosure of Potential Conflicts of Interest. The authors declare that they have no competing interests, financial or otherwise, related to the current work. J. Sorasit, W. Phakdeekul and W. Kedthongma report that since the initial planning of the work they were funded by the Kasetsart University (Graduate School Fellowship Program), Kasetsart University Business Incubation Center, and the National Innovation Agency (Public Organization). The payments were made to the researchers.
ETHICAL APPROVAL AND INFORMED CONSENT
Ethical approval was obtained from the Institutional Review Board of Human Research Ethics Subcommittee, Rajamangala University of Technology Isan Sakon Nakhon Campus, Thailand (Approval number: No. HEC-04-68-020, Date: 16 July 2025). Participants provided informed consent.
DATA AVAILABILITY
The datasets generated and/or analyzed during the current study are not publicly available but are available from the corresponding author on reasonable request.
AUTHORS’ CONTRIBUTIONS
JS, WP and WK: conceptualization, formal analysis, methodology, project administration. JS: data curation, investigation. JS and WK: supervision. All authors: manuscript writing and editing. All authors read and approved the final version of the manuscript.
PROVENANCE AND PEER REVIEW
Not commissioned; externally peer reviewed.
REFERENCES
- 1.Tobacco. World Health Organization. June 25, 2025. Accessed February 24, 2026. https://www.who.int/news-room/fact-sheets/detail/tobacco [Google Scholar]
- 2.Pratchyanupat P, Sobhonslidsuk A. A literature review on regulations enforced to reduce adolescents’ use of e-cigarette in Thailand. J Health Med Sci. 2023;6(3):86-92. https://www.asianinstituteofresearch.org/_files/ugd/ed8b62_5d586f5c059b4b25bf0d20b93e21dd69.pdf [Google Scholar]
- 3.Kungskulniti N, Charoenca N, Hamann SL, Pitayarangsarit S, Mock J. Cigarette waste in popular beaches in Thailand: high densities that demand environmental action. Int J Environ Res Public Health. 2018;15(4):628. doi: 10.3390/ijerph15040628 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.Jin X, Jiang Q, Yao S, et al. A survey of infrared and visual image fusion methods. Infrared Phys Technol. 2017;85:478-501. doi: 10.1016/j.infrared.2017.07.010 [DOI] [Google Scholar]
- 5.Baiya P, Chankeaw T, Chinwong D, Chinwong S. The use of electronic cigarettes in Thailand: a cross-sectional national survey. Eur J Public Health. 2020;30(Supplement_5):ckaa166.1197. doi: 10.1093/eurpub/ckaa166.1197 [DOI] [Google Scholar]
- 6.Seeherunwong A, Tipayamongkholgul M, Angsukiattitavorn S, et al. Association between socioecological factors and electronic cigarette use among Thai youth: an institution-based cross-sectional study. BMJ Open. 2023;13(7):e069083. doi: 10.1136/bmjopen-2022-069083 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Hampsher-Monk SC, Prieger JE, Patwardhan S. E-cigarette Regulations and Why They Fail. In: Hampsher-Monk SC, Prieger JE, Patwardhan S, eds. Tobacco Regulation, Economics, and Public Health, Volume III: Clearing the Air on E-Cigarettes and Harm Reduction. Palgrave Macmillan; 2024:1-130. doi: 10.1007/978-3-031-61337-1_1 [DOI] [Google Scholar]
- 8.Hanafin J, Sunday S, Shevlin M, Clancy L. Smoking and e-cigarette use in young adults with disabilities. BMC Public Health. 2025;25(1):1342. doi: 10.1186/s12889-025-22542-5 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Kiatsangorn T, Polngam D, Kaewarj N, Pilayon B, Phakdeekul W, Phoosuwan N. Effectiveness of the self-efficacy modification program for Thai alcoholic inpatients with schizophrenia. Discov Ment Health. 2025;5(1):156. doi: 10.1007/s44192-025-00296-1 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Phakdeekul W, Thongkrajai P, Eiamprapai P, Kanato M. Risk factors to alcohol law violations in the community: quasi-experimental study. Am J Appl Sci. 2011;8(12):1343-1348. doi: 10.3844/ajassp.2011.1343.1348 [DOI] [Google Scholar]
- 11.Dorlohtahe S, Kedthongma W, Phakdeekul W. Toxic effects of herbal supplements on endothelial and vascular system among rural Thai patients: a multicenter cohort study. J Stroke Cerebrovasc Dis. 2026;35(5):108610. doi: 10.1016/j.jstrokecerebrovasdis.2026.108610 [DOI] [PubMed] [Google Scholar]
- 12.Kedthongma W, Usaprom S, Phakdeekul W. Community-based interventions to improve tuberculosis treatment outcomes: a meta-analysis. MethodsX. 2026;16:103893. doi: 10.1016/j.mex.2026.103893 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Charan J, Biswas T. How to calculate sample size for different study designs in medical research?. Indian J Psychol Med. 2013;35(2):121-126. doi: 10.4103/0253-7176.116232 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Jabkowski P, Piotr C, Marta K. Kish grid vs birthday procedures: survey refusal rates by the type of within-household selection procedure in the European social survey. Bull Sociol Methodol. 2024;163(1):77-87. doi: 10.1177/07591063241258087 [DOI] [Google Scholar]
- 15.Padgett R, Cowden R, Chattopadhyay M, et al. Survey sampling design in wave 1 of the global flourishing study. Eur J Epidemiol. 2025;40:391-406. doi: 10.1007/s10654-024-01167-9 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.Brookhart MA, Schneeweiss S, Rothman KJ, Glynn RJ, Avorn J, Stürmer T. Variable selection for propensity score models. Am J Epidemiol. 2006;163(12):1149-1156. doi: 10.1093/aje/kwj149 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Austin PC. A comparison of 12 algorithms for matching on the propensity score. Stat Med. 2014;33(6):1057-1069. doi: 10.1002/sim.6004 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.Haghani A, Arpawong TE, Kim JK, Lewinger JP, Finch CE, Crimmins E. Female vulnerability to the effects of smoking on health outcomes in older people. PLoS One. 2020;15(6):e0234015. doi: 10.1371/journal.pone.0234015 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19.Lunceford JK, Davidian M. Stratification and weighting via the propensity score in estimation of causal treatment effects: a comparative study. Stat Med. 2004;23(19):2937-2960. doi: 10.1002/sim.1903 [DOI] [PubMed] [Google Scholar]
- 20.Kedthongma W, Usaprom S, Phakdeekul W. Factors associated with depression among tuberculosis patients with HIV co-infection in rural community Thailand. BMC Psychol. doi: 10.1186/s40359-026-04574-7 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.Bove LA, Kennedy RD, Houston SM. Project Management. In: Finnell J, Dixon B, eds. Clinical Informatics Study Guide. Springer; 2016:377-414. doi: 10.1007/978-3-319-22753-5_16 [DOI] [Google Scholar]
- 22.Pipkin JA, Ontiveros T, Carcoba LM, O’Dell LE. Enhanced tobacco use vulnerability in adolescents, females, and persons with diabetes. In: Hall FS, Young JW, Der-Avakian A, eds. Negative affective states and cognitive impairments in nicotine dependence. Elsevier Academic Press; 2017:71-90. doi: 10.1016/B978-0-12-802574-1.00005-3 [DOI] [Google Scholar]
- 23.Haghani M, Bliemer MCJ, Goerlandt F, Li J. The scientific literature on coronaviruses, COVID-19 and its associated safety-related research dimensions: a scientometric analysis and scoping review. Saf Sci. 2020;129:104806. doi: 10.1016/j.ssci.2020.104806 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24.Wiley RC, Oliver AC, Snow MB, et al. The impact of the covid-19 pandemic on smoking among vulnerable populations. Nicotine Tob Res. 2023;25(2):282-290. doi: 10.1093/ntr/ntac135 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25.Sánchez-Teruel D, Pritam Singh HK, Sánchez-Barrera MB, et al. Protective factors for suicidal behaviour in adults self-reported as LGBTQ+: a study based on modulating variables. Curr Psychol. 2024;43:29597–29606. doi: 10.1007/s12144-024-06611-3 [DOI] [Google Scholar]
- 26.Streck JM, Davis DR, Pang RD, et al. Potential moderating effects of sex/gender on the acute relative reinforcing and subjective effects of reduced nicotine content cigarettes in vulnerable populations. Nicotine Tob Res. 2020;22(6):878-884. doi: 10.1093/ntr/ntz098 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27.Zhang X, Xu T, Qin Y, et al. Exploring the needs and coping strategies of family caregivers taking care of dying patients at home: a field study. BMC Palliat Care. 2023;22(1):196. doi: 10.1186/s12904-023-01315-0 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28.Mitra R, Rai A, Kumar A, Mitra JK. Role of herbal medication in tobacco cessation treatment: a systematic review and meta-analysis. Addict Health. 2023;15(1):63-70. doi: 10.34172/ahj.2023.1290 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29.Bunditanukul K, Sittipunt C, Phrommintikul A, Chokemaitree N, Sakthong P. Effect of Vernonia cinerea (L.) Less. lozenges as a smoking cessation aid in patients with non-communicable diseases. J Pharm Pharmacogn Res. 2025;13(4):1202–1210. doi: 10.56499/jppres24.2127_13.4.1202 [DOI] [Google Scholar]
- 30.Frazer K, Kelleher C. Reducing tobacco smoke exposure for vulnerable groups: hospital settings and teachable moments. Perspect Public Health. 2018;138(3):142-144. doi: 10.1177/1757913918763444 [DOI] [PubMed] [Google Scholar]
- 31.Nanda Naik B, Rao Ur R, Verma M, Kumar Nirala S, Pandey S, M Singh C. Awareness and attitude towards tobacco products and tobacco industry and perception about government’s role in tobacco control among the adult rural population: a cross-sectional study in the Indian State of Bihar. Addict Health. 2024;16(2):83-92. doi: 10.34172/ahj.2024.1438 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 32.Li P, Lu T, Wang W, Du L. Association between maternal smoking during pregnancy and learning disabilities in children and adolescents: a propensity score matching analysis. Tob Induc Dis. 2025;23(December):198. doi: 10.18332/tid/214128 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 33.Wang X, Yu F, Chen D, et al. The associated factors of smoking cessation intention among husbands in gestational households: a census-based cross-sectional survey. Tob Induc Dis. 2026;24(January):1. doi: 10.18332/tid/213719 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 34.Yang T, Peng S, Cottrell RR. Changes in e-cigarette use and related behaviors following the 2022 e-cigarette tax increase in China: a prospective longitudinal observational study. Tob Induc Dis. 2025;23(December):188. doi: 10.18332/tid/208848 [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Data Availability Statement
The datasets generated and/or analyzed during the current study are not publicly available but are available from the corresponding author on reasonable request.