Measurements

The prevalence of baseline sexual experience and sexual initiation during 12-month follow-up were the main study outcomes. Baseline sexual experience was assessed with one yes-or-no question (“Have you ever had sexual intercourse?”). Sexual initiation during follow-up was assessed with a yes-or-no question in the 12-month questionnaire (“During the past twelve months, have you had sexual intercourse?”). Subjects giving an affirmative answer were asked about partners’ number (“During the past twelve months, with how many people have you had sexual intercourse?”) and condom use (“When you have had sexual intercourse during the past twelve months, how often did you or your partner use a condom?”). Those items were recoded as dichotomous variables (1 versus multiple, partners; rarely/never versus always/almost all the time, used a condom). Depending on the report of any of those two factors, a secondary outcome variable with three possible categories was computed: no sexual initiation; sexual initiation, no risk factors reported; risky sexual initiation, one or both risk factors reported.

Several exposure variables were assessed by SEYLE’s baseline questionnaire. Baseline risk behaviours were investigated through the Global School-based Student Health Survey (GSHS) [24]. The following behaviours were defined by dichotomized variables, in agreement with a previous work [1]: excessive alcohol use (≥ twice/week), illegal drug use (≥ three times/lifetime), heavy smoking (> five cigarettes/day), reduced sleep (≤ six hours/night), sedentary behavior (physical activity < once/week), high media use (≥ five hours/day of school/work unrelated use of Internet, television and video-games). Reduced sleep, sedentary behaviour and high media use are referred to as “invisible” risk behaviours throughout the paper, because they may not be perceived as harmful as overt behaviours (e.g. substance abuse) by adult observers [1].

The individual psychological domain was investigated at baseline through the Strengths and Difficulties Questionnaire (SDQ), an internationally validated screening tool which relates well to childhood mental health [25]. The SDQ includes 4 difficulty scales (emotional, peer, conduct, hyperactivity) and a prosocial behaviour scale. In this study, we adopted the broader sub-scale division of SDQ’s difficulty items into internalizing and externalizing scales, which has been recommended for general population samples [26]. The internalizing scale is obtained by combining the emotional and peer items whereas the externalizing scale combines the conduct and hyperactivity items. In analogy with previous research [27], both scales were dichotomized at the 90th percentile, considering as abnormal the measurements above the cut-off. Abnormal externalizing was defined by scores ≥10 for both boys and girls, while abnormal internalizing was defined by scores ≥7 for boys and ≥10 for girls.

Context vulnerabilities were investigated at baseline through measures of problematic interaction with peers, school and family. Trauma exposure was evaluated with SEYLE-specific items [28]. Several bullying manifestations were investigated with yes-or-no questions (“In the past 12 months have others often: spread rumors about you; teased you; deliberately left you out of activities; taken money, property, or food from you; called you names; made fun of how you look or talk”), and a dichotomous variable was created by categorizing any positive answer as being bullied. Physical abuse was investigated with a yes-or-no question about attacks in the previous 12 months. Involvement in fighting was assessed with one question on the number of physical fights in the previous 12 months, and a dichotomized variable was obtained (0 versus ≥1). Truancy was defined as having missed class or school without permission during the previous two weeks. Parental involvement was assessed by asking how often parents/guardians helped making important decisions; took time to talk about things that happened; came to see in case of special activity; paid attention to opinion. These questions had three possible answers (never or almost never; sometimes; often), and were recoded into a new variable where a low parental involvement was defined by two or more “never or almost never” answers.

Age and sex were self-reported at baseline. Age was dichotomized as <16 versus ≥16 years old, because literature shows that, in Europe, sexual debut is infrequently reported by 15 years old [29], and an age of 16 and below is perceived as too young to have sex [30].

Statistical analysis

Baseline prevalences of risk behaviours, psychological difficulties, trauma, low parental involvement and sexual experience were calculated. The association of age, sex and exposure variables with baseline sexual experience or initiation during follow-up was evaluated through chi-squared tests and simple logistic regression. Age and sex stratified multivariable logistic regression measured the strength of association between potential baseline predictors and the dependent variable sexual initiation (reference category “no sexual initiation”) during the 12-month follow-up. Multiple multinomial logistic regression analyses measured the independent association between predictors and the outcome sexual initiation (reference category “no sexual initiation”), which was subcategorized depending on the absence (Sexual initiation) or presence (Risky sexual initiation) of reproductive risk factors (multiple partners or infrequent condom use). All predictor variables showing a significative association with the outcome at unadjusted analyses were initially included, and only predictors showing significant association in at least one stratum were kept in the final multivariable analyses.

Odds ratios (OR) were calculated together with 95% confidence intervals (CI). Statistical significance of differences was defined by p < 0.05, two-tailed. The analyses were performed with IBM® SPSS® Statistics ver. 23 for Mac OsX.

Ethics statement

SEYLE was approved by the European Commission, as a precondition for funding, and by the ethics committees of each national centre [21]:

1. Austria: Ethikkomission der Medizinischen Universität Innsbruck
2. Estonia: Tallinna Meditsiiniuuringute Eetikakomitee
3. France: Comité de Protection des Personnes Sud-Méditerranée II
4. Germany: Ethikkommission Medizinische Fakultät Heidelberg
5. Hungary: Egészségügyi Tudományos Tanács Titkárság, Pályázati Iroda, Tudományos És Kutatásetikai Bizottság
6. Ireland: Clinical Research Ethics Committee of the Cork Teaching Hospitals
7. Italy: Comitato Bioetico di Ateneo, Università Degli Studi Del Molise
8. Romania: Comisia De Eticã, A Universitãtii De Medicinã Si Farmacie, Cluj Napoca
9. Slovenia: Komisija Republike Slovenije Za Medicinsko Etiko
10. Spain: Comité Ètico de Investigación Clinica, regional del Principado de Asturias

An external advisor from the University of Basel, Switzerland, provided independent ethical assessment and supervision of the project [23]. Participation to the project was voluntary and recruitment occurred after obtaining written informed consent from parents/caregivers as well as assent from pupils, in agreement with the guidelines of the involved local ethic committees [23].

Clinical trial registration

SEYLE was registered in the German Clinical Trials Register (DRKS00000214).

Key findings

We have identified correlates of adolescents sexual initiation among non-sexual risk behaviours, psychological attributes and social factors, in a large multinational cohort of European adolescents. Baseline sexual experience was reported significantly more often by adolescents ≥ 16 years old and by those exposed to smoking, alcohol and illegal drugs consumption, poor sleep, sedentariness and high media use. Baseline consumption of alcohol, illegal drugs, and tobacco, were strong predictors of sexual initiation during a 12-month follow-up period. Sexual debut was also predicted by baseline reduced sleep, but not sedentary behaviour or high media use (Table 2). The association between risk behaviours and sexual initiation was particularly strong in case of younger females and, among girls, when the outcome was associated with multiple partners and/or infrequent condom use. Significant associations were also observed between sexual initiation and childhood trauma (being bullied, involved in fighting or physically attacked), truancy, and low parental involvement. Externalizing symptoms independently predicted sexual debut, particularly for older girls. Internalizing symptoms were negatively associated with early and risky sexual debut among boys.

Strengths and limitations

This study included a large sample of European adolescents recruited from randomly selected schools in a large geographic area and ten different countries. Each national sample was population-based and fairly nationally representative. Selection criteria, measurements and procedures were strict and homogenous. The comprehensive sample characteristics, systematically evaluated in SEYLE, allowed to study several exposure variables in the same cohort, including more or less overt risk behaviours, psychological difficulties and problematic interactions with peers, school or family. Another major strength is the longitudinal design, aiming at identifying risk factors which are not simply associated with sexual initiation but also precede it. Additionally, through stratification we could measure disparities between groups of adolescents in relation to their age and sex.

Limitations of the study should be acknowledged. SEYLE only surveyed pupils, and self-report of sensitive data may be biased by social or legal pressure. Nevertheless, participants were thoroughly informed about the strict confidentiality of all procedures, and response rates were high. Another limitation relates to the absence of a definition of the main outcome variable, as respondents may have a different understanding of what “sexual intercourse” is, and adolescents are known to engage in alternative sexual behaviours, such as oral sex. A related limitation is the absence of data on sexual orientation. A normative interpretation of sexual intercourse as “vaginal-penile sex” can overlook sexual minorities, among whom health risks may be higher [31,32]. The questionnaire also lacked a marker for frequency of sex after initiation, which arguably contributes to the health outcomes of adolescent sexuality, as repetitive small life events do [8].

Finally, SEYLE did not collect individual data on socioeconomics, pubertal development, motivations and intentions, all of them being potential correlates of adolescent sexuality [12,33,34].

Interpretation and implications

Gaining knowledge about adolescents’ risk-taking is challenging because it is common or almost normative, and multiple behaviours often cluster in the same subjects [1,9]. The case of sexual initiation is particularly complex because, apart from being a normative life event, it is also a physiological component of adolescent development. Nevertheless, early sexual activity is associated with reproductive health risk, whose most overt manifestations are unwanted pregnancies and sexually transmitted disease [2–5]. Dangerous health outcomes are commonly mediated by practices which can obviously be correlates of sexuality at any age, although they are particularly prevalent among adolescents [2,5]. In our study, more than one third of adolescents having their sexual debut during 12-month follow-up also reported multiple partners or infrequent condom use.

The association of early sexual initiation with reproductive risk may be interpreted through a gateway model identifying sexuality as an occasion for unsafe sexual behaviour, thus leading to negative health outcomes. As a result, adolescent sexuality is often addressed by downstream individual risk policy responses [3,9] attempting to avoid or delay sexual initiation, or to minimize its potential consequences, through interventions in the reproductive and sexual health domain.

A relevant obstacle to the management of adolescents risk behaviours is that those have been traditionally approached as if belonging to isolated domains of adolescents health [9]. Therefore, our interpretation of interactions between multiple behaviours, as well as its potential to inform policy interventions, is limited. However, data from established surveillance programmes show cross-nationally consistent associations between early sexual activity and overt risk behaviours [10,11]. This is confirmed by our study, as adolescents exposed to alcohol, illegal drugs and smoking reported the highest sexual initiation rates (respectively 52.7%, 63.4%, 53.7%). Gateway interpretations, according to which substance abusers have increased opportunities for sexual risk behaviours [9], are supported by our findings that overt risk behaviours predict earlier and riskier sexual initiation among adolescents, independently of other individual and contextual vulnerabilities.

The gateway-effect interpretation of the association between “invisible” risk behaviors [1], which adult observers may not consider harmful, and sexuality appears more challenging. The association with media consumption may depend from increased exposure to sexual content or opportunity to meet sexual partners [14]. Studies on physical activity have given ambiguous results, as both sedentary behaviour and participation in sports might be associated with early sexual activity [14,35]. However, the only “invisible” risk behavior predicting sexual initiation in our analysis was reduced sleep, a known correlate of adolescents’ development, risk-taking, sedentariness, and high media use [1,36–38].

Although the interest in evaluating the coexistence of risk behaviours is justified by gateway hypotheses and the possible incremental physical health outcomes, adolescent sexual behaviours also have potential effects on mental and social health domains of wellbeing, which are arguably mediated by the evaluation that young people make of their own behaviours [8]. A distinctive feature of the present study is the evaluation of several individual and contextual factors which may influence adolescents’ engagement in risk behaviours as well as their perception of the same behaviours. For instance, age and sex related disparities should not be overlooked. The significantly low rate of sexual initiation among subjects younger than 16 years old (14.2% versus 41% in those ≥16) was expected [10,39]. European adolescents’ median age at sexual debut is 17 years, while less than 15% become sexually active before the age of 15 [29]. In addition, despite legal ages of consent variating between 14 and 16, European teenagers and parents consider people under 17 years old to be “too young to have sexual intercourse”, and higher age norms among parents are negatively associated with early debut [30]. It is not even surprising that less girls than boys reported sexual activity at baseline, as previously described at European level [12,29], although the gap is inconsistent across countries and becomes narrower with age, as shown by our longitudinal data. Interestingly, overt risk behaviours such as substance abuse, which are in general more common among boys [1], were strongly predictive among younger girls. Additionally, a strong association between non-sexual risk behaviours and risky sexual initiation was suggested among females. These findings are consistent with previous studies [30,40] and may partially be explained by sexual norms, stigma and double standards, whose consequences on sexual behaviours and their perception-mediated outcomes are often worse among women [20]. It therefore seems plausible that differences between girls and boys are not limited to their timing or level of engagement in specific behaviours but extend to the trajectories between risk factors and outcomes, which are arguably modulated by individual and contextual vulnerabilities [41,42].

For instance, adolescent psychological difficulties are related to risk-taking patterns [1,17]. Abnormal internalizing and externalizing symptoms are more prevalent among adolescents engaging in multiple risk behaviours [1], and both were significantly associated with sexual initiation in our study. Externalizing is common in undercontrolling adolescents who engage in overt risk [1,41] and we found that it predicts sexual debut in girls and boys, independently of other risk behaviours. On the contrary, internalizing, which is particularly frequent among adolescents with sedentary behaviour and high media consumption [1], was not predictive for girls whereas it was negatively associated with sexual initiation during follow-up among boys. The positive baseline association may therefore be age-specific or even depend on an inverse directionality between sexual activity and internalizing symptoms.

Our study also evaluated the interaction between adolescents and their social context, including peers, school, and family, which previous research has highlighted as source of both risk and protective factors [4,16,43–45]. Bullying victimization is associated with self-injurious behaviour though peer and parental support have a protective role [45]. Childhood trauma and violent behaviors, whether as victims or perpetrators, are related to adolescent mental health and sexual activity [15,29,39]. It should also be reminded that forced sexual initiation, despite being relatively infrequent in developed countries, is significantly more prevalent among individuals reporting an earlier debut and females [46,47]. Our study could not specifically address exposure to non-consensual sex or sex-related trauma though it shows a significant association between fighting, being bullied, or low parental involvement, and earlier sexual initiation. The protective potential of an inclusive, vigilant and caring social context is further supported by the findings that risky sexual behaviour was predicted by low parental involvement among boys and by truancy among girls.

In conclusion, the practice and research implications of our findings share reservations about single-domain perspectives on adolescent risk behaviour and health. Responses which are limited to the reproductive health domain overlook cross-domain interactions that could guide the early identification of risk subjects and the development of preventive initiatives. Further studies of the relationship between adolescent sexual and non-sexual risk behaviour, also in look of possible reverse gateway effects, may result in useful information to downstream policies. However, it seems simplistic to reduce any association between risk behaviours and early sexual initiation to a gateway effect, which overlooks the role of individual and contextual factors in determining behaviours and modulating their effect on physical as well as mental and social health [8,9,17,18]. Refuting simplified gateway hypotheses in favour of a more comprehensive model would therefore be particularly valuable, in order to achieve a deeper understanding of sexual behaviours and their potential health outcomes, and to promote the positive effects of sexuality on adolescent wellbeing [8]. Resulting research and policies would respectively study and implement tailored interventions targeting adolescent health as a whole instead of focusing on limited health outcomes. Special consideration may, for instance, be needed for younger girls because the strong association between non-sexual and sexual behaviours identifies them as a particularly vulnerable group, in view of double standards and distinctive reproductive and gynaecological health risks.