Data sources and sample selection

This study is a secondary analysis of the Korea Youth Risk Behavior Web-based Survey (KYRBS), an anonymous online survey conducted annually by the Korea Disease Control and Prevention Agency (KDCA) since 2005. The KYRBS provides national data on risk behaviors among Korean adolescents and uses a two-stage stratified cluster sampling design. The KYRBS data are available for free download on the official KYRBS website (http://yhs.cdc.go.kr). The sample is stratified into 117 strata based on 39 regional clusters and three levels of education (middle schools, general high schools, and specialized high schools). Participants are equally distributed across all grade levels in 800 middle and high schools. The study samples were derived from the KYRBS conducted in 2018 (n=60040), 2019 (n=57303), 2020 (n=54948), 2021 (n=54848), and 2022 (n=51850). Data from KYRBS 2018 to KYRBS 2022 were compiled into a mixed cross-sectional dataset. This study used Stata 17.0 software for the analysis. All statistical tests were two-tailed and the significance levels are as follows: (*p<0.1, **p<0.05, ***p<0.01). In studies examining early sexual behavior among Korean minors, 207964 samples were included in the regression analysis after excluding missing values. Furthermore, since information on the timing of first sexual behavior was only collected in KYRBS 2018 and 2021, we selected data from these two years for further analysis, with a final sample size of 3151 after excluding missing values (Figure 1).

Figure 1

Flowchart of participants selection, South Korea (2018–2022)

Outcome variable

This study’s dependent variable is the sexual behavior of Korean adolescents, focusing on both the occurrence of sexual activity and the timing of first sexual intercourse. The sexual behavior of adolescents was evaluated through the following questions: ‘Have you had sexual relations?’ and ‘When was the first time you had sex?’. For the first question, responses were recoded such that affirmative answers were coded as 1, and negative answers were coded as 0. Participants who gave an affirmative response to the first question proceeded to answer the second question. The responses to the second question were treated as a continuous variable, with values ranging from 1 to 13, corresponding to ‘before entering school’ to ‘the third year of high school’. Higher values indicate a later timing of initial sexual intercourse.

Predictive variables

Smoking or not

The independent variable in this study is adolescent smoking behavior, specifically focusing on three types of tobacco products: general cigarettes (cigarettes), liquid e-cigarettes with nicotine and cigarette-type electronic cigarettes (such as iCos, Glo, Lil, etc.). Three binary variables were created based on the following questionnaire items: ‘Have you ever used conventional cigarettes (tobacco cigarettes)?’, ‘Have you ever used liquid e-cigarettes with nicotine so far?’, and ‘Have you ever used cigarette-style electronic cigarettes (heated tobacco products, such as iCos, Glo, Lil)?’. Responses were categorized as ‘yes’ or ‘no’, with ‘yes’ assigned a value of 1 and ‘no’ a value of 0.

The time of initial cigarette use

To analyze the timing of adolescents’ first cigarette use, we excluded respondents who indicated they had never smoked. For the remaining participants, their reported age at first cigarette use was transformed into a continuous variable ranging from 1 to 13, representing different stages or ages of smoking initiation.

Controlled variables

We have introduced a set of covariates that may influence adolescent sexual behavior. These include gender, age, family socioeconomic status (rated on a 1–5 scale, with higher scores indicating better socioeconomic status), academic performance (scored on a scale of 1–5, with higher scores indicating better performance), grade level, fathers’ education level, mothers’ education level, educational stage, school type, school level and city size.

Statistical analysis

First, descriptive analyses were conducted to summarize the characteristics of all variables. Concurrently, the annual proportions of adolescents using different types of cigarettes was calculated. These types were categorized into four groups: non-users, single-type users, dual-type users, and triple-type users. Second, since sexual activity is a dummy variable and the use of an Ordinary Least Squares (OLS) model could lead to heteroskedasticity, we employed a logit model for regression analysis in this study and provided adjusted odds ratios (AORs) and 95% confidence intervals (95% CIs). Specifically, Models 1–3 investigate the association between tobacco use and sexual behavior, whereas Model 4 incorporates all three tobacco types into the regression analysis. The OLS model was used in further analysis to examine the association between the timing of initial cigarette use and sexual initiation. Additionally, the moderating effect of gender was introduced. In the moderation effect analysis, an interaction term between gender and the timing of first tobacco use was included. In the baseline regression analysis, control variables included gender, age, economic situation, academic achievement, grade, parental education level, educational stage, school type, and city scale. In the further moderation effect analysis, control variables included economic situation, academic achievement, grade, educational stage, school type, and city scale. All analyses control for time and city fixed effects Third, Kaplan-Meier (KM) survival estimates, which combine event outcomes (sexual activity occurrence) with time to event (timing of sexual activity), provide more reliable visual results. We shall present a graph where the x-axis represents survival time (adolescent age), while the y-axis represents the survival rate (proportion of individuals who have engaged in sexual activity in the total sample). A value of 1 on the y-axis indicates that no individuals in the sample have engaged in sexual activity. The line segments in the graph represent the proportion of individuals who have engaged in sexual activity within each age group. In comparison to the Kaplan-Meier (KM) survival curve, the Cox regression analysis estimates the risk ratio for event time data, considering the effects of grade, academic achievement, educational stage, school type, and city scale.

Descriptive statistical analysis

Table 1 provides descriptive statistics for the key variables. The proportion of male students was 48.4%, while female students accounted for 52.6%, resulting in a nearly equal gender ratio. The average age of the adolescents was approximately 14.9 years. Most adolescents perceived their family’s economic status as moderate. Students who maintained upper middle academic performance constituted the largest proportion, with 86.5% achieving grades at or above the intermediate level. The majority of parents had completed college education. In terms of school types, mixed-gender schools were the most prevalent, followed by girls’ schools and boys’ schools, which account for 17% and 15%, respectively. Regarding city size, students from small- and medium-sized cities represented the largest group, followed by those from large cities, while those from county areas are the least numerous. The mean average prevalence (proportion within the sample) of sexual behavior among Korean adolescents was 0.044. The rate of adolescent sexual activity increased from 4.69% in 2018 to 4.93% in 2022. Among the three types of tobacco products, conventional cigarettes had the highest usage rate, averaging 10% of adolescents. Additionally, Figure 2 reveals a declining trend in adolescents who use only one type of cigarette, with the majority having experience with multiple types of tobacco products.

Figure 2

The trend in the number of types of tobacco used by adolescents, South Korea (2018–2022) (N=207964)

Regression results

Table 2 presents an analysis of the association between the use of different types of cigarettes and sexual behavior, with adjustments made for potential confounding factors. The findings of Models 1–3 indicate that, under constant other conditions, smoking, whether conventional cigarettes or e-cigarettes, was significantly associated with the occurrence of sexual behavior among minors. Specifically, there was a positive association between smoking behavior and sexual activity, exhibiting a convergent trend. Individuals who used ordinary cigarettes were more likely to engage in sexual behavior than those who do not (AOR=7.85; 95% CI: 7.49–8.23). Adolescents who use liquid e-cigarettes with nicotine were more likely to engage in sexual behavior than those who do not (AOR=8.21; 95% CI: 7.80–8.64). When controlling for all other variables, adolescents who use combustible electronic cigarettes were more likely to engage in sexual behavior compared to those who do not (AOR=9.86; 95% CI: 9.26–10.49). In Model 4, the regression analysis included all three types of cigarettes, and the results remained statistically significant. However, the coefficients were smaller compared to those from individual analyses; specifically, ordinary cigarette (AOR=4.20; 95% CI: 3.94–4.48), liquid e-cigarettes with nicotine (AOR=2.02; 95% CI: 1.87–2.18), and cigarette type e-cigarette (AOR=2.17; 95% CI: 2.01–2.34).

Among the control variables, compared to females, male adolescents exhibit a significant positive correlation with the occurrence of sexual behavior (p<0.01). Adolescent males were significantly more likely to engage in sexual behavior than females (AOR=1.17; 95% CI: 1.10–1.24). The probability of adolescents engaging in sexual behavior increased with age. For each unit increase in age, the likelihood of engaging in sexual behavior increased by 0.14% (AOR=1.14; 95% CI: 1.09–1.19). A better family economic status was associated with a higher probability of sexual behavior (AOR=1.07; 95% CI: 1.04–1.10). Higher grade level was associated with an increased probability of sexual behavior (AOR=1.19; 95% CI: 1.13–1.25). Academic performance (AOR=0.941; 95% CI: 0.923–0.960), father’s education level (AOR=0.938; 95% CI: 0.902–0.976), and mother’s education level (AOR=0.928; 95% CI: 0.892–0.966) were protective factors against engaging in sexual behavior. Students in co-educational and boys’ schools were more likely to engage in sexual behavior than those in girls’ schools. School type with a higher proportion of males was associated with a higher probability of sexual behavior (AOR=1.07; 95% CI: 1.02–1.12). Larger city size was associated with a higher probability of engaging in sexual behavior (AOR=1.12; 95% CI: 1.03–1.23) (Tables 2 and 3).

Further analysis examines the relationship between smoking and sexual behavior in the time dimension. Table 4 Model 1 shows that: under constant conditions, for every unit increase in the time of first use of tobacco, the time to first sexual intercourse increases by 0.526 indicating that earlier tobacco use is possibly associated with earlier sexual behavior. The visualization results in Figure 3 further validate the findings.

Figure 3

The visualization of the regression results for first-time smoking and first-time sexual activity, South Korea (2018–2021) (N=3151)

The analysis of moderating effects

This analysis explores gender differences in the relationship between smoking and the timing of first sexual behavior. Figure 4A uses survival analysis, revealing that males initiate sexual behavior earlier than females. The log-rank test was significant [χ²(1)=430.09, Pr >χ²=0.0000]. Figure 4B further validates this result through the use of Cox proportional hazards regression. Table 4, Model 2 introduces gender as a moderating variable and creates an interaction term between gender and the timing of first tobacco use. The results show a positive correlation between the timing of first tobacco use and first sexual behavior (p<0.01), with a coefficient of 0.591. Gender was also positively associated with the timing of first sexual behavior. However, the coefficient of the interaction term is -0.100, suggesting that males reduce the positive association between the timing of first tobacco use and first sexual behavior. The visualization in Figure 5 clearly illustrates that gender significantly inhibits this relationship, with the effect being stronger for males.

Figure 4A

Kaplan-Meier (KM) survival estimates of initial sexual behavior by gender, South Korea (2018–2021)

Figure 4B

Cox proportional hazards regression of initial sexual behavior by gender, South Korea (2018–2021)

Figure 5

Moderating effect of gender, South Korea (2018–2021) (N=3151)

Limitations

Although this study controls for time and city fixed effects, it relies on mixed cross-sectional data, which cannot fully capture unobserved individual-level variations over time or cannot establish causal relationships. Second, self-reported assessments may introduce reporting biases. Third, residual confounding factors may remain because not all potential confounders (such as peer influences and major family events) were controlled for in the analysis. Unmeasured or inadequately controlled variables, such as mental health status or previous exposure to risk behaviors, may still affect the observed relationships. Fourth, the generalizability of the results may be limited. The study’s target population consists of Korean adolescents. Additionally, tobacco policies, cultural norms, and sexual attitudes vary across countries, which could influence the study’s applicability to other settings. Therefore, the specific context of this study may not be applicable to other settings or populations, particularly those with different cultural or policy environments.