Adolescent Sexual Behavior Patterns, Mental Health, and Early Life Adversities in a British Birth Cohort

ABSTRACT

This study tested adolescent sexual behavior patterns at age 14, their association with mental health at age 17 (psychological well-being, substance use, and self-harm attempts), and the influence of early life adversities upon this association. A British birth cohort (5,593 boys and 5,724 girls from the Millennium Cohort Study) was used. Latent class analysis suggested five subgroups of adolescent sexual behaviors: a “no sexual behavior” (50.74%), a “kisser” (39.92%), a “touching under clothes” (4.71%), a “genital touching” (2.64%), and an “all sexual activities” class (1.99%). Adolescents from the “kisser,” “touching under clothes,” “genital touching,” and “all sexual activities” classes reported significantly more substance use and self-harm attempts compared to adolescents from the “no sexual behavior” group. The associations became weaker after controlling for early life adversities (reducing around 4.38% to 37.35% for boys, and 9.29% to 52.56% for girls), and reduced to a smaller degree after further controlling for mental health variables at 14. The associations between sexual behaviors and psychological well-being became non-significant after controlling for early life adversities. Adolescents who have engaged in low-intensity sexual activities at early age may have poorer reported mental health, a pattern that is stronger for girls and early life adversity may partially explain this association.

Introduction

Early sexual debut may be a risk factor for subsequent poorer mental health (Ramrakha et al., 2000). However, prior studies have focused mostly on early sexual intercourse and ignored the diversity of sexual activities that adolescents may engage in. Different sexual behavior patterns may be associated with different health outcomes. Studying adolescents who have engaged in a range of sexual activities at an early age (outside of simply sexual intercourse) could help identify adolescents with greater risk of negative health outcomes. This information could then inform the development of interventions targeted toward these groups.

There are few studies on a range of early sexual activities and mental health but those that exist have produced mixed results (Espinosa-Hernández & Vasilenko, 2015; Hershenberg & Davila, 2010; Hipwell et al., 2011). In general, this body of work suggests that sexual activities were associated with poorer mental health, and these associations were worse for girls than boys (Meier, 2007), possibly due to more negative appraisals (Donald et al., 1995) and greater social stigma attached to having earlier sexual experiences among girls compared to boys (Kreager et al., 2016). Prior studies also have important methodological limitations. Most studies are cross-sectional, likely to suffer from recall bias, and have difficulty in determining whether early sexual behaviors precede or are followed by mental health outcomes (Espinosa-Hernández & Vasilenko, 2015). Others have included only one sex (Hipwell et al., 2011), have small sample sizes (Hershenberg & Davila, 2010), or have included a restricted range of sexual activities (Wesche et al., 2017). Studies which use latent class or latent profile analysis include risky sexual behavior outcomes (e.g., condom use) as indicators of latent class membership of adolescent sexual behaviors. These bias the estimates of any association between sexual behaviors and health-related outcomes (Vasilenko et al., 2015). Some studies also use complete cases to deal with missing information, and a two-step approach for latent class analysis (LCA) with distal outcomes (Hipwell et al., 2011). Finally, almost all prior research has not controlled for early life adversities as potential confounders associated with adolescent sexual behaviors and health-related outcomes (Hershenberg & Davila, 2010).

Early life adversity, which encompasses a range of negative exposures from infancy through adolescence (e.g., abuse, neglect, parental separation, parental psychopathology, domestic violence, familial socioeconomic inequality), is a prominent risk for all classes of mental health disorders in adolescence (Hughes et al., 2017). Prolonged exposures to adverse early life adversities could cause depletion of somatic and psychological resources without sufficient recovery, dysfunction of hypothalamic–pituitary–adrenal axis, deficits in emotional regulation, or increased attention to negative stimuli which increases risks for mental health problems (Pechtel & Pizzagalli, 2011).

Several versions of life history theory also propose that early life adversity may be associated with early sexual activities because it constrains physiological and behavioral development, dysregulates stress response systems, or acts as a forecast for possible negative future conditions which results in youth engaging in earlier sexual behavior (Del Giudice et al., 2015). The biosocial model suggests that early life adversity may be associated with earlier physical maturation and pubertal timing (Ellis, 2004), which is associated with early sexual activities due to the increases in pubertal hormones and sensation-seeking or impulsivity (Halpern et al., 1997; Suleiman et al., 2017). For example, one study reported that the increase in testosterone during puberty was associated with earlier sexual intercourse among girls (Halpern et al., 1997). Other types of early life adversity, such as childhood maltreatment, may be associated with greater impulsivity (Liu, 2019), which is in turn associated with early sexual debut (Khurana et al., 2012). Problem behavior theory also suggests that behaviors that depart from social or age-graded norms (e.g., early sexual debut) may tend to cluster together due to the common biological factors, perceived environment (e.g., lack of parental control or parental absence), and personality (e.g., poor impulse control) (Madkour et al., 2010).

Meta-analyses have found that some markers of early life adversity are associated with increased risk of poorer mental health (Hughes et al., 2017) and, in a separate meta-analysis, with early sexual debut (Xu et al., 2018). Although prior research is limited to studying sexual intercourse rather than exploring the relationship to the diverse range of sexual behaviors, they do suggest that the relationship between adolescent sexual behaviors and mental health outcomes may become either weaker or even reduce to non-significance when adverse early life adversities are controlled for. Thus, early life adversity may be a common cause confounder between sexual behavior patterns and mental health.

In the present study, we investigated, (1) whether there are subgroups of adolescent sexual behaviors at age 14 years via a LCA; (2) whether these sexual behavior patterns are associated with mental health outcomes at age 17 years via probability-weighted regressions; and (3) whether these associations remain after controlling for early life adversity (measured prospectively) and mental health at age 14. These research objectives were tested in a large, well-characterized British prospective birth cohort, the Millennium Cohort Study (MCS). The first 7 years postnatally are characterized by important processes, such as greater cognitive development from 5 to 7 years, and the sensitivity of brain development to early life events (Danese & McEwen, 2012; Ellis, 2004). The sex differences in somatic physiology, aggression, and social behavior expand during middle childhood (6–11 years; Del Giudice, 2014). Thus, several developmental stages may be important sensitive periods for the influences of early life adversity upon early sexual behaviors and mental health outcomes. Accordingly, we classified the early life conditions into two relevant developmental periods, including postnatal before 7 years of age and postnatal after 7 years of age. Given the exploratory nature of the LCA, we do not have specific hypotheses regarding the predicted number of subgroups. However, based on the extant literature reviewed, we hypothesized that there would be associations between clusters of adolescent sexual behaviors at 14 and mental health outcomes at 17; that these would be stronger for girls than boys; and that these associations would become either statistically weaker or non-significant when measures of early life adversity were controlled for.

Method

Participants

Participants were part of the MCS, an ongoing longitudinal study of children born in the UK between September 2000 and January 2002, which has been described in detail elsewhere (Connelly & Platt, 2014). To ensure adequate representation of ethnic minority and children living in disadvantaged areas, a random sample was collected from the eligible population using a disproportionate stratified cluster sampling design. The initial sample recruited 18,827 children from 18,552 families, and 1,389 new families who had moved into the sampling wards after their initial registration were recruited on the second occasion. Seven waves of the MCS cohort have been carried out so far. Attrition is a problem common to all longitudinal studies. Among 19,941 families, the proportion of productive cases were 93.03%, 78.18%, 76.46%, 69.49%, 66.63%, 58.80%, and 53.28% for wave 1 to wave 7, respectively. The drop out was mainly due to refusal. This study was performed in accordance with the Helsinki Declaration of 1975, as revised in 2008, and informed parental consent was obtained at each wave. No ethical approval was needed since secondary data were used which were completely anonymous and unidentifiable.

The current study analyzed MCS data reported by parents and children across different waves. Adolescents who reported a valid response of at least one item on the Adolescent Sexual Activities Index (ASAI) at 14 years old were included, N = 11,317 (5,593 boys and 5,724 girls). The majority of the adolescents in the sample were ethnically White (78.89%), and the median equivalized weekly family income (divided the family weekly income by number of household members according to their weights on the modified Organization for Economic Co-operation and Development’s equivalized income scale) was £394.75 when adolescents were 14 years old.

Measures

Sexual Behavior

At 14 years old, adolescents reported whether they had engaged in any of ten sexual activities on the ASAI with another young person in the last 12 months (Hansen et al., 1999). For the assessment of ASAI in MCS, those sexual activities were presented in order from low (e.g., kissing) to high (e.g., have oral sex) intensity and were categorized into four stages based on the intensity: holding hands, kissing, and cuddling together were in the first stage; letting others touch them under clothes and touching others under clothes in the second stage; touching others’ genitals and letting others touch their genitals in the third stage; and oral sex and sexual intercourse questions in the fourth stage. Adolescents who reported having not engaged in a particular sexual activity received a score of 0 on that activity and those who reported having engaged in the activity received a score of 1 on that activity. Adolescents did not progress to the next stage if they reported having not engaged in all sexual activities in one stage. The ASAI is a validated self-report measure and has high internal consistency (Cronbach’s α = .93) (Hansen et al., 1999).

Mental Health Outcome

The mental health outcomes at 14 and 17 years old were chosen based on availability in the MCS and robust associations with early sexual behaviors reported by prior research.

Mental Well-being

When adolescents were 17 years old, the Short Warwick-Edinburgh Mental Well-being Scale (SWEMWBS) was used to measure their mental well-being in the past 2 weeks. The SWEMWBS is a validated self-report inventory with seven items (McKay & Andretta, 2017). The SWEMWBS has acceptable reliability and validity (Cronbach’s α = .85; Stewart-Brown et al., 2009). Each item was measured on a 5-point scale ranging from 1 = none of the time to 5 = all of the time. An example item is “I have been feeling useful.” The total score of the seven items was transformed into metric scores using the Rasch model and was used in the analysis, with a higher score indicating higher positive mental wellbeing (Stewart-Brown et al., 2009).

Psychological Distress

When adolescents were 17 years old, the Kessler Psychological Distress Scale (K6) was used to measure their psychological distress in the past 30 days. K6 is a validated self-report inventory with six items on a 5-point scale ranging from 1 = all of the time to 5 = none of the time (Kessler et al., 2002). K6 also has good reliability and validity (Cronbach’s α = .84; Mewton et al., 2016). An example item is “During the last 30 days, about how often did you feel hopeless?” The total score of the six items was used in the analysis, with a higher score indicating greater psychological distress.

Emotional and Behavioral Difficulties

When adolescents were 14 and 17 years old, the Strength and Difficulties Questionnaire (SDQ) was used to measure their psychological adjustment in the past 6 months (Goodman, 2001). The SDQ is a validated self-report inventory with 25 items divided into 5 subscales (emotional symptoms, conduct problems, hyperactivity, peer relationship problems, and prosocial behavior). The SDQ has acceptable reliability and validity (Cronbach’s α = .82; Goodman, 2001). Each subscale consists of five items rated by the main carers (usually mothers) on a 3-point scale ranging from 0 = not true to 2 = certainly true. An example of an item is “Child’s name is restless, overactive, cannot stay still for long.” The total difficulties score (the sum of all subscales except the prosocial behavior subscale since it is independent of the difficulties measured by the other subscales) was used in the analysis, with a higher score indicating greater emotional and behavioral difficulties. The total difficulties score was chosen over five separate subscales scores in our sample since it has been validated in prior research (Achenbach et al., 2008). In contrast, subscale scores were suggested to be only justified when studying high-risk children (those with mental disorders or with higher scores on SDQ subscales) (Goodman et al., 2010).

Self-esteem

When adolescents were 14 and 17 years old, a shortened version of Rosenberg’s Self-Esteem Scale was used to assess self-esteem (Rosenberg, 2015). Five items including self-satisfaction, having a number of good qualities, being able to do things similar to others, a person of value, and feeling good about oneself were used. Adolescents reported how much they agreed with the above statements on a 4-point scale ranging from 1 = strongly agree to 4 = strongly disagree. Cronbach’s alpha for the scale in our sample was .91. Exploratory factor analysis yielded one factor (eigenvalue = 3.71) accounting for 74.16% of the variance in our sample. The total score of the five items was used in the analysis, with a higher score indicating lower self-esteem.

Reported Self-harm Attempts

Adolescents reported whether they had hurt themselves on purpose in an attempt to end life in the last 12 months at 17 years old (yes or no).

Substance Use

Adolescents reported the frequency of smoking, binge drinking (having five or more alcoholic drinks at a time), and drug use at 14 and 17 years old. Two items were used to assess the behavior: frequency of smoking cigarettes on a 6-point scale ranging from 1 = never to 6 = more than six cigarettes a week and frequency of smoking electronic cigarettes on a 4-point scale ranging from 1 = never to 4 = every day at 14 years old or a 6-point scale ranging from 1 = never to 6 = more than six times a week at 17 years old. One item was used to assess the frequency of binge drinking in the last 12 months on a 5-point scale ranging from 1 = never to 5 = 10 or more times. One item was used to assess the frequency of cannabis use in the last 12 months on a 5-point scale ranging from 0 = never to 4 = more than ten times.

Early Life Adversities

The early life adversities were chosen based on availability in the MCS and known associations with both early sexual behaviors and mental health outcomes reported by prior research. These included reported victimization experiences, domestic violence, parental psychosocial distress, parental absence, parental-child relationship quality, family socioeconomic position, and parental substance use. Details on the measurement of parental absence (a derived variable based on household members’ reports of relationships to adolescents and their sexes at each wave), parental psychosocial distress (K6) measured at each wave and reported by adolescents’ parents, family socioeconomic position (a summary score incorporating parents’ education and occupation, and household income at each wave), and parental-child relationship quality (Child Parent Relationship Scale) when adolescents were 3 years and reported by their parents have been described elsewhere (Xu et al., 2020).

Victimization Experiences

When adolescents were 14 years old, four binary items were used to assess the experience of victimization (verbal, sexual, and physical) in the past 12 months. Verbal abuse was measured via one item: “whether someone has insulted you, called you names, threatened or shouted at you in a public place, at school or anywhere else.” Sexual harassment or assault was measured via one item: “whether someone has made an unwelcome sexual approach to you or assaulted you sexually.” Reported sexual harassment or assault and verbal abuse was coded as present if they reported “yes.” Physical abuse was measured via two items: “whether someone has been physically violent towards you, e.g., pushed, shoved, hit, slapped or punched you” and “whether someone has hit you with or used a weapon against you.” Physical abuse was coded as present if they reported “yes” to either of the two items.

Domestic Violence

At each wave, adolescents’ parents reported whether their partners had ever used force on them for any reason. We recoded this into two variables: domestic violence before 7 years of age and domestic violence since 7 years of age. To do this, domestic violence before 7 was coded as present if either adolescents’ mothers or fathers reported that their partners had used force on them when adolescents were 9 months, 3 years old, and 5 years old. We did the same for domestic violence since 7 years of age.

Parental Substance Use

When adolescents were 3 years old, the CAGE was used to measure parents’ problem drinking (Mayfield et al., 1974). The CAGE is a validated self-report inventory with four items assessing whether respondents had ever experienced four alcohol-related symptoms in their lifetime on a binary scale: 1 = yes and 0 = no (Mayfield et al., 1974). The scale also has good reliability and validity (Dhalla & Kopec, 2007). Cronbach’s alpha for the scale in our sample was .60. An example item is “Have you ever felt you should cut down on your drinking?”. Mothers’ and fathers’ problem drinking were coded as present if they reported one or more alcohol-related symptoms. When adolescents were 3 and 5 years old, adolescents’ parents reported their frequency of any illicit drug use (e.g., cannabis, cocaine or ecstasy) in the last 12 months on a 3-point scale: 1 = never, 2 = occasionally, and 3 = Regularly. Mothers’ and fathers’ drug use was coded as present if they reported regular drug use at any time point and absent if they reported never or occasional drug use at all time points.

Statistical Analysis

Missing Data

The 10 sexual activities items had 0.40%–94.61% missing information, mainly due to the non-responses to high-intensity sexual activities (about 44% and 90% of adolescents did not progress to the second and third stage of sexual activities, respectively). Research has found that most adolescents follow a progressive sexual trajectory from kissing and touching over the clothes to touching under the clothes and then having sexual intercourse (Shtarkshall et al., 2009). Thus, adolescents who did not progress to answer these sexual activities were recoded as not having engaged in that sexual activity. As a result, the 10 sexual activities items had 0.04%–0.51% missing information in the final analysis sample, mainly due to the non-responses to the first stage of sexual activities. The mental health outcomes had 19.66%–20.80% missing information. For LCA and multigroup confirmatory factory analysis (CFA), this missing data were handled via full information maximum likelihood estimation with robust standard errors (MLR) in Mplus 7.4.

The mental health factor scores, self-harm attempts, and early life adversity variables had 0.02%–57.86% missing information (Supplemental Tables 1 and 2). This was handled using multiple imputation stratified by sex using Stata 16.0, which could help to reduce biases due to selective attrition (Twisk & de Vente, 2002). For the imputation model, all variables in the analysis model were included, a chained equations algorithm model was used, a predictive mean matching approach with 10 nearest-neighbor donors was used for continuous variables, and 58 imputations (should be at least as large as the percentage of missing data) were created (White et al., 2011). Trace plots and other diagnostics provided no cause for concern regarding the imputed values. Sensitivity analysis found no substantial difference in estimates for mental health outcomes associated with latent class membership of sexual behaviors between univariate regression using complete-case and analysis using imputed data. Due to the relatively high proportion of missing data in the sample, no sensitivity analysis comparing multivariate regressions based complete-case and imputed data, controlling for all early life adversities, was conducted. But our sample size allowed us to conduct complete-case analysis that included some early life adversities. When early life adversities variables with missing data less than 15%, including parental absence, victimization experiences, and parental-child relationship quality, were included in the multivariate regressions, sensitivity analyses found no substantial difference in estimates between multivariate regression using complete-case and analysis using imputed data.

LCA

We evaluated whether there were meaningful subgroups of adolescents who showed similar early sexual behavior patterns at 14 years old through LCA in Mplus 7.4, with 10 binary items of ASAI as the manifest indicators of latent class memberships. We started with a single-class model and fitted successive models with an increasing number of latent classes (one at a time) until the results were no longer interpretable. The best-fitting model was determined by the interpretability of the model, sample size of each latent class, and model fit statistics. Smaller values of Akaike Information Criterion (AIC), Bayesian Information Criterion (BIC), and adjusted BIC, and a significant p value for the Lo-Mendell-Rubin likelihood ratio test (LMR LR), adjusted LMR LR test, and bootstrapped likelihood ratio test (BLRT) indicate that the k class model fit the data better than the k-1 class model (Asparouhov & Muthén, 2012). Models were estimated multiple times using 1000 random sets of starting values to ensure the model converged at its global maxima. Once the best-fitting model was determined, adolescents were assigned their most likely latent class membership representing their sexual behavior profile.

Multigroup CFA

Based on theory, correlation matrix (Supplemental Table 3), and exploratory factor analysis, factor loadings from two-factor CFA were used as item weightings to generate summary scores incorporating mental health outcomes at age 14 and 17, respectively: psychological well-being and substance use, with a higher score indicating more negative psychological well-being and more substance use (Figure 1). Self-harm attempts were not included in the CFA since it presents a different construct and the model fit statistics were not affected when self-harm attempts were excluded. Likert scales were treated as continuous and MLR was used to handle the missing data and potential multivariate normality violation in Mplus 7.4. MLR assuming continuous scaling was used because the latent mental health outcomes measured here are more likely to be continuous constructs, and simulation studies show that MLR performed well even with skewed responses when sample sizes were large (Sass et al., 2014). Sensitivity analysis indicated that there were no substantial differences in model fit statistics between analysis using MLR and analysis using robust weighted least squares, which treats Likert scored items as ordered categorical variables (Supplemental Table 4). We also followed the guideline to test whether measurement invariance at age 17 is supported across different latent class membership of sexual behaviors (Putnick & Bornstein, 2016). Since likelihood ratio test is sensitive to large sample sizes, measurement invariance is considered to be supported when constraints lead to a change in the comparative fit index ≤0.01 among model comparison (Putnick & Bornstein, 2016).

Figure 1. Models tested in this study

Regressions

All regressions stratified by sex were performed in Stata 16.0. To test whether early life adversities could predict latent class membership of adolescent sexual behaviors, a multivariate probability-weighted ordered logistic regression was estimated with latent class membership regressed on all early life adversities. The maximum posterior probability of class membership for each observation was included as a sampling weight to account for measurement error from the determination of most likely latent class (Kamata et al., 2018). The Brant test was used to test the proportional odds assumption and the assumption was relaxed for variables where the test was significant at the 5% level (use gologit2 function).

A three-step hierarchical multivariate probability-weighted regression was used to test the association between latent class membership of sexual behaviors and mental health outcomes at age 17 (linear regression for psychological well-being and substance use, and logistic regression for self-harm attempts). First, a univariate regression was estimated including only latent class membership as a predictor (Model 1). In the second step, all early life adversities were entered (Model 2). In the final step, mental health outcomes at age 14 were entered (Model 3). Substance use at age 17 was log-transformed (adding 0.40 to convert negative values to positive values before transformation) since its distribution was skewed. In all models, the maximum posterior probability of class membership for each observation was included as a sampling weight to account for measurement error from the determination of most likely latent class (Kamata et al., 2018). There was no indication of multicollinearity problems based on the variance inflation factor (all <2.5 except family SEP before and since 7 with a maximum value of 5.69). For linear regressions, model residual checks were performed on each imputed dataset (White et al., 2011). Residuals were normally distributed based on QQ plot and homoscedasticity was not violated.

As a potential check against possible mischievous responding between adolescents who had not engaged in any sexual activities and those who had, we compared both groups on several screening items that could potentially capture such responding (Robinson-Cimpian, 2014). These include items on “have computer of their own” and “have seen a dentist in the past 12 months.” These items were selected because they were unrelated to sexual behavior development. We found no evidence of potential mischievous responding (Supplemental Table 6).

Results

LCA of Adolescent Sexual Behavior

The lowest BIC was for the six class model, while the lowest AIC and ABIC were for the seven class model (Supplemental Table 7). The results of BLRT tests also suggested that the seven class model was optimal. However, we selected the five class model over the six and seven class models as providing the most parsimonious explanation of the adolescent sexual behavior patterns because all classes were qualitatively distinct from one another based on item-response probabilities, and thus more interpretable. For example, compared to the five class model, classes 4 and 5 from the six class model (n = 338), and classes 4, 5, and 6 from the seven class model (n = 371) were all marked by a high probability of having genital touching (see Table 1, Supplemental Tables 8 and 9 for the comparison among five, six, and seven class models). The distinction among those classes seems to be whether adolescents have engaged in touching others under clothes or genitals, versus having engaged in being touched under the clothes or their genitals. Sensitivity analyses found no significant differences in demographic characteristics (ethnicity, family income, and parental absence) and mental health outcomes at 17 between Classes 4 and 5 from the six class model, and among classes 4, 5, and 6 from the seven class model, which further support that all classes from the six and seven class models were not qualitatively distinct from one another and it was reasonable to choose the five class model over the six and seven class models. Sensitivity analyses also found no substantial differences in estimates for mental health outcomes associated with latent class membership of sexual behaviors between analysis using the five class model and analysis using the six or seven class models.

Table 1. Item-response probabilities for the five classes model of adolescent sexual behavior at 14

	Latent class membership of early sexual behaviors at 14 years
Sexual behaviors at 14 years, yes	No sexual behavior (N = 5,742)	Kisser (N = 4,518)	Touching under clothes (N = 533)	Genital touching (N = 299)	All sexual activities (N = 225)
Hold hands	.20	.95	.96	.96	.98
Kiss	.03	.66	.93	.97	1.00
Cuddle together	.14	.89	.98	.98	1.00
Let others touch under clothes	.00	.01	.78	.97	.99
Touch others under clothes	.00	.00	.70	.85	.99
Touch others genitals	.00	.00	.08	.80	1.00
Let others touch their genitals	.00	.00	.09	.93	.99
Perform oral sex on others	.00	.00	.00	.09	.91
Let others perform oral sex on them	.00	.00	.00	.17	.90
Sexual intercourse	.00	.00	.00	.17	.62

Based on item-response probabilities (Table 1), the five class model was interpreted as: “no sexual behavior” (class 1), “kisser” (class 2); “touching under clothes” (class 3); “genital touching” (class 4); and “all sexual activities” (class 5). The “no sexual behavior” class contained 50.74% (n = 5,742) of participants and was characterized by a high probability of having not engaged in any sexual activities. The “kisser” class contained 39.92% (n = 4,518) of participants and was marked by a high probability of having engaged in kissing. The “touching under clothes” class contained 4.71% (n = 533) of participants and was characterized by a high probability of having engaged in kissing and touching under clothes. The “genital touching” class contained 2.64% (n = 299) of participants and was characterized by a high probability of having engaged in kissing, touching under clothes, and genital touching. The “all sexual activities” class contained 1.99% (n = 225) of participants who had a high probability of having engaged in all sexual activities. The relatively small sample sizes of class 3, 4, and 5 may be due to the measurement of sexual activities at the early stage of adolescence.

Early Life Adversity and Adolescent Sexual Behaviors

Compared to boys from the “no sexual behavior” class, boys with reported verbal abuse experiences, physical abuse experiences, father’s problem drinking, and parental absence were more likely to engage in early sexual activities as captured by the remaining four latent classes among boys (odds ratios ranging from 1.21 to 2.65, all ps < .05; Table 2). Compared to girls from the “no sexual behavior” class, girls with reported verbal abuse experiences, physical abuse experiences, sexual harassment or assault experiences, and parental absence were significantly more likely to have engaged in early sexual activities, as captured by the remaining four latent classes (odds ratios ranging from 1.34 to 6.26, all ps < .01; Table 3).

Table 2. Multivariate ordered logistic regressions for latent class membership of sexual behavior at 14 years old among boys

	Multivariate ordered logistic regression
Variable	OR	OR^a	OR^b	OR^c	OR^d
Maternal psychological distress before 7	0.98 (0.96, 1.01)
Maternal psychological distress since 7	1.00 (0.97, 1.02)
Paternal psychological distress before 7		0.99 (0.95, 1.02)	1.02 (0.97, 1.07)	1.03 (0.97, 1.09)	0.99 (0.91, 1.08)
Paternal psychological distress since 7	0.98 (0.95, 1.02)
Family socioeconomic position before 7	0.96 (0.90, 1.03)
Family socioeconomic position since 7	1.02 (0.95, 1.08)
Parent-child relationship		1.01 (1.00, 1.02)	1.00 (0.98, 1.01)	0.99 (0.97, 1.01)	0.99 (0.96, 1.02)
Verbal abuse (Ref = no)	1.30***(1.15, 1.47)
Physical abuse (Ref = no)		1.56*** (1.36, 1.79)	2.23*** (1.84, 2.70)	2.29*** (1.76, 2.97)	2.65*** (1.77, 3.97)
Sexual harassment or assault (Ref = no)	1.56 (0.96, 2.54)
Domestic violence before 7 (Ref = no)	1.09 (0.89, 1.33)
Domestic violence since 7 (Ref = no)	1.20 (0.96, 1.49)
Mother’s problem drinking (Ref = no)		1.09 (0.89, 1.33)	1.33 (0.97, 1.82)	1.76**(1.20, 2.60)	1.22 (0.65, 2.27)
Father’s problem drinking (Ref = no)	1.21*(1.03, 1.43)
Mother’s drug use (Ref = no)	1.23 (0.75, 2.02)
Father’s drug use (Ref = no)	1.06 (0.70, 1.61)
Parental absence (Ref = parents presence)
Either parent absence before 7	1.38**(1.12, 1.69)
Either parent absence since 7	1.37***(1.18, 1.60)

Brant test was applied to test the proportional odds assumption. If the proportional odds assumption was not violated, one odds ratio in the column OR is reported; if it was violated, generalized ordered logit model (gologit2) was used and four odds ratios in the columns OR^a, OR^b, OR^c, and OR^d are reported. The five classes were interpreted as: “no sexual behavior” (class 1), “kisser” (class 2); “touching under clothes” (class 3); “genital touching” (class 4); and “all sexual activities” (class 5).

^aClass 1 versus class 2, 3, 4, and 5.

^bClass 1 and 2 versus class 3, 4, and 5.

^cClass 1, 2, and 3 versus class 4 and 5.

^dClass 1, 2, 3, and 4 versus class 5.

*p < .05. **p < .01. ***p < .001.

Table 3. Multivariate ordered logistic regressions for latent class membership of sexual behavior at 14 years among girls

	Multivariate ordered logistic regression
Variable	OR	OR^a	OR^b	OR^c	OR^d
Maternal psychological distress before 7		0.98 (0.96, 1.01)	0.96 (0.92, 1.00)	1.00 (0.95, 1.05)	0.99 (0.92, 1.06)
Maternal psychological distress since 7	1.02 (0.99, 1.04)
Paternal psychological distress before 7	0.97 (0.93, 1.01)
Paternal psychological distress since 7		1.00 (0.96, 1.04)	1.01 (0.95, 1.07)	1.00 (0.92, 1.09)	1.09 (0.98, 1.21)
Family socioeconomic position before 7	0.95 (0.89, 1.02)
Family socioeconomic position since 7	1.04 (0.98, 1.11)
Parent-child relationship	1.00 (0.99, 1.01)
Verbal abuse (Ref = no)	2.25***(2.01, 2.53)
Physical abuse (Ref = no)	1.68***(1.43, 1.98)
Sexual harassment or assault (Ref = no)	6.26***(4.69, 8.36)
Domestic violence before 7 (Ref = no)	1.21 (0.98, 1.50)
Domestic violence since 7 (Ref = no)	0.92 (0.70, 1.21)
Mother’s problem drinking (Ref = no)	0.96 (0.79, 1.18)
Father’s problem drinking (Ref = no)	1.15 (0.96, 1.38)
Mother’s drug use (Ref = no)	1.59 (0.77, 3.28)
Father’s drug use (Ref = no)	1.22 (0.80, 1.88)
Parental absence (Ref = parents presence)
Either parent absence before 7	1.34**(1.09, 1.65)
Either parent absence since 7	1.62***(1.38, 1.90)

Brant test was applied to test the proportional odds assumption. If the proportional odds assumption was not violated, one odds ratio in the column OR is reported; if it was violated, gologit2 was used and four odds ratios in the columns OR^a, OR^b, OR^c, and OR^d are reported. The five classes were interpreted as: “no sexual behavior” (class 1), “kisser” (class 2); “touching under clothes” (class 3); “genital touching” (class 4); and “all sexual activities” (class 5).

^aClass 1 versus class 2, 3, 4, and 5.

^bClass 1 and 2 versus class 3, 4, and 5.

^cClass 1, 2, and 3 versus class 4 and 5.

^dClass 1, 2, 3, and 4 versus class 5.

*p < .05. **p < .01. ***p < .001.

Sexual Behavior Classes and Mental Health-Related Outcomes

The two-factor CFA had acceptable model fit statistics (Supplemental Table 4). Scalar invariance for CFA at age 17 across latent class membership of sexual behaviors was supported (Supplemental Table 5). Compared to adolescents from the “no sexual behavior” class, boys and girls from the “kisser,” “touching under clothes,” “genital touching,” and “all sexual activities” classes reported significantly more substance use at age 17 in univariate regressions (Tables 4 and 5). These associations reduced to a smaller degree but remained statistically significant after controlling for early life adversities (Model 2), and mental health outcomes at 14 (Model 3). Compared to adolescents from the “no sexual behavior” class, boys and girls from the “kisser,” “touching under clothes,” “genital touching,” and “all sexual activities” classes were also at greater risk of self-harm attempts at age 17 in univariate regressions (Tables 4 and 5). These associations (except for boys from the “genital touching” class) reduced to a smaller degree but remained statistically significant after controlling for early life adversities (Model 2). For boys, these associations remained statistically significant after further controlling for mental health outcomes at 14. However, for girls, only the increased risk of self-harm attempts at age 17 among girls from “all sexual activities” compared with girls from the “no sexual behavior” class remained statistically significant after further controlling for mental health outcomes at 14. Early life adversities could explain 6.33% and 34.28% of the increased risk of substance use and self-harm attempts associated with “all sexual activities” compared to boys from “no sexual behavior” class, respectively. For girls, the corresponding figures were 11.62% and 52.56%, respectively.

Table 4. Univariate and multivariate regressions for mental health outcomes for boys

Mental health	Latent class membership of early sexual behaviors	Model 1	Model 2	Model 3
Psychological Well-being	Class 2	−0.00 (−0.23, 0.22)	−0.24* (−0.46, −0.02)	−0.15 (−0.37, 0.07)
	Class 3	0.80** (0.30, 1.31)	0.21 (−0.30, 0.72)	0.22 (−0.27, 0.72)
	Class 4	0.97** (0.30, 1.64)	0.27 (−0.38, 0.92)	0.16 (−0.49, 0.80)
	Class 5	1.37** (0.51, 2.23)	0.55 (−0.29, 1.40)	0.11 (−0.77, 0.98)
Substance use	Latent class membership of early sexual behaviors
	Class 2	1.37*** (1.29, 1.45)	1.31*** (1.24, 1.39)	1.17*** (1.10, 1.25)
	Class 3	1.86*** (1.74, 1.97)	1.73*** (1.62, 1.85)	1.35*** (1.24, 1.47)
	Class 4	2.07*** (1.94, 2.21)	1.94*** (1.80, 2.07)	1.41*** (1.28, 1.55)
	Class 5	2.37*** (2.23, 2.51)	2.22*** (2.08, 2.37)	1.48*** (1.31, 1.64)
Self-harm attempts	Latent class membership of early sexual behaviors
	Class 2	1.97*** (1.41, 2.77)	1.73** (1.21, 2.46)	1.77** (1.24, 2.55)
	Class 3	3.12*** (1.79, 5.44)	2.32** (1.28, 4.22)	2.23* (1.19, 4.18)
	Class 4	3.24** (1.57, 6.65)	2.03 (0.92, 4.45)	1.60 (0.68, 3.74)
	Class 5	6.68*** (3.24, 13.77)	4.39*** (2.05, 9.38)	2.86* (1.20, 6.82)

Substance use at 17 was log-transformed. For psychological well-being and substance use at 17, the mean differences and 95% confidence interval are reported here, with higher scores indicating more negative psychological well-being and more substance use. For self-harm attempts at 17, the odds ratios and 95% confidence interval are reported here. The five classes were interpreted as: “no sexual behavior” (class 1), “kisser” (class 2); “touching under clothes” (class 3); “genital touching” (class 4); and “all sexual activities” (class 5). Boys from the “no sexual behavior” class were the reference group. Model 1 is the univariate regression. Model 2 controlled for early life adversities. Model 3 is the same as Model 2 with mental health outcomes at 14 being further controlled for.

*p < .05. **p < .01. ***p < .001.

Table 5. Univariate and multivariate regressions for mental health outcomes for girls

Mental health	Latent class membership of early sexual behaviors	Model 1	Model 2	Model 3
Psychological well-being	Class 2	0.66*** (0.41, 0.90)	0.01 (−0.23, 0.25)	−0.05 (−0.27, 0.18)
	Class 3	1.59*** (0.96, 2.22)	0.56 (−0.06, 1.19)	0.33 (−0.27, 0.94)
	Class 4	1.08** (0.27, 1.90)	−0.16 (−0.96, 0.64)	−0.59 (−1.38, 0.19)
	Class 5	2.29*** (1.29, 3.30)	0.64 (−0.38, 1.66)	0.10 (−0.89, 1.09)
Substance use	Latent class membership of early sexual behaviors
	Class 2	1.40*** (1.33, 1.47)	1.27*** (1.20, 1.34)	1.08*** (1.01, 1.15)
	Class 3	1.95*** (1.84, 2.07)	1.73*** (1.60, 1.85)	1.22*** (1.09, 1.34)
	Class 4	2.06*** (1.91, 2.21)	1.78*** (1.62, 1.94)	1.11*** (0.95, 1.28)
	Class 5	2.41*** (2.29, 2.53)	2.13*** (1.99, 2.27)	1.21*** (1.06, 1.36)
Self-harm attempts	Latent class membership of early sexual behaviors
	Class 2	1.98*** (1.60, 2.44)	1.34* (1.06, 1.69)	1.24 (0.97, 1.59)
	Class 3	3.09*** (2.04, 4.67)	1.88** (1.19, 2.98)	1.51 (0.95, 2.41)
	Class 4	4.38*** (2.64, 7.28)	2.30** (1.26, 4.22)	1.66 (0.89, 3.07)
	Class 5	8.20*** (5.06, 13.29)	3.89*** (2.24, 6.73)	2.49** (1.35, 4.62)

Substance use at 17 was log-transformed. For psychological well-being and substance use at 17, the mean differences and 95% confidence interval are reported here, with higher scores indicating more negative psychological well-being and more substance use. For self-harm attempts at 17, the odds ratios and 95% confidence interval are reported here. The five classes were interpreted as: “no sexual behavior” (class 1), “kisser” (class 2); “touching under clothes” (class 3); “genital touching” (class 4); and “all sexual activities” (class 5). Girls from the “no sexual behavior” class were the reference group. Model 1 is the univariate regression. Model 2 controlled for early life adversities. Model 3 is the same as Model 2 with mental health outcomes at 14 being further controlled for.

*p < .05. **p < .01. ***p < .001.

Compared to adolescents from the “no sexual behavior” class, boys and girls from the “touching under clothes,” “genital touching,” and “all sexual activities” classes also reported significantly negative psychological well-being at age 17 in univariate regressions, unstandardized regression coefficients from 0.80 to 2.29, all ps < .01 (Tables 4 and 5). These associations became statistically non-significant after controlling for early life adversities. Compared to boys from the “no sexual behavior” class, boys from the “kisser” class reported significantly more positive psychological well-being, Cohen’s d = −0.07, 95% confidence interval = [−0.12, −0.01], p < .05 after controlling for early life adversities. However, this association became statistically non-significant after further controlling for mental health outcomes at 14, Cohen’s d = −0.04, 95% confidence interval = [−0.10, 0.01].

Discussion

This study examined adolescent sexual behavior patterns, their association with mental health, and the influence of early life adversities upon this association in a British birth cohort. We found that there was a quantitatively meaningful diversity of early adolescent sexual behaviors. The five subgroups of adolescent sexual behaviors were similar to the pattern of subgroups revealed in a different prospective birth cohort study (where they did not test the association between sexual behavior and mental health outcomes; Xu et al., 2021). However, given the exploratory nature of LCA, there may be different results if sexual activities were measured at later ages. We expect that as adolescents become older they will transition from no sexual behavior toward greater engagement in high-intensity sexual activities, resulting in different latent growth curve profiles (Xu et al., 2021). Future studies using later waves of sexual behavior are required to test this.

The findings are consistent with prior research that early life adversities may be associated with earlier onset of sexual activity (Xu et al., 2018). The strongest candidates appear to be recalled experiences of verbal and physical abuse and parental absence for both sexes. These associations likely have multifactorial causes. For example, the patterns could be due to dysregulation of stress response systems caused by experiencing early life adversity. Life history models suggest that it may also be adaptive for adolescents to allocate more resources to early sexual behaviors over body growth under adverse or resource scare conditions early in life (Del Giudice et al., 2015). Alternatively, adolescents with parental absence may be more likely to lack parental monitoring or control which increases their risk of early sexual experimentation (Wight et al., 2006). Individual differences in sexual behaviors and exposure to early life adversities may also be influenced by common genetic and/or environmental factors (Mendle et al., 2009).

Compared to adolescents who engaged in no sexual activities, those engaging in kissing, touching under clothes, genital touching, and all sexual activities were at greater risk for substance use and self-harm attempts. The risk of substance use and self-harm attempts increased in a dose-dependent manner across the subgroups in accordance with the intensity of the sexual activities. These patterns indicate that a greater range of sexual activities, rather than just sexual intercourse, should be considered in future studies on the impact of early sexual activity on health outcomes among adolescents. The poorer mental health associated with adolescents who had engaged in kissing only compared with adolescents without any sexual experiences is consistent with prior research (Kaltiala-Heino et al., 2003). Adolescents who have engaged in kissing may be likely to have earlier pubertal timing compared with adolescents without any sexual experiences (Lam et al., 2002). The incongruence between their physical maturity and cognitive and emotional resources, and the increased levels of adrenal and gonadal hormones at puberty among early maturing adolescents may increase their risk of poorer mental health (Ge & Natsuaki, 2009). Our results also suggest that the type of mental health outcome appears important. Indeed, the strength of the associations depended on the type of mental health outcome; strongest for substance use and weakest for psychological well-being.

The present study also found sex differences in the pattern of results. Negative mental health outcomes were more common among girls who had engaged in any sexual activities compared to boys, which is consistent with prior studies (Meier, 2007). This was reflected in the larger effect sizes for the associations in the current study. This may be related to the ‘sexual double standard’ where girls and boys are evaluated differently regarding sexual behaviors (Kreager et al., 2016). Boys are socially expected to be sexually active and take sexual initiative, whereas girls are expected to be sexually reactive and submissive (Kreager et al., 2016). Accordingly, girls may be more restricted in their sexual freedom, and there are more negative appraisals (Donald et al., 1995) and greater social stigma attached to having earlier sexual experiences among girls compared to boys (Kreager et al., 2016), which increase their risk of poorer mental health. Girls who engaged in early sexual activities were more likely to have earlier pubertal timing. The greater risk of peer victimization associated with more noticeable physical characteristics of puberty exhibited by early maturing girls (Petersen & Hyde, 2009), the increased levels of adrenal and gonadal hormones at puberty (Ge & Natsuaki, 2009), and the incongruence between their physical maturity and psychological resources among early maturing girls (Ge & Natsuaki, 2009), may interact with the greater social stigma attached to having earlier sexual experiences among girls, which increase their risk of poorer mental health. This finding has both developmental and prevention implications. In term of development, sexual debut is a milestone in adolescent development. Knowledge of the influence of early sexual activities on developmental trajectories may enhance our understanding of how multiple domains of adolescent development interact. From a prevention perspective, sexual education programs for adolescents may wish to encourage students to manage expectations (including delaying) regarding early sexual activities.

Several mechanisms may help explain the link between early life adversities, early sexual behaviors and mental health outcomes reported here. Early life adversities may be one of the common causes associated with both early sexual behavior and mental health outcomes. Prolonged exposure to adverse early life conditions could trigger dysregulated stress responses which cascade into earlier sexual behavior and/or mental health problems (Pechtel & Pizzagalli, 2011). However, it should be noted that although the associations between early sexual behaviors and mental health outcomes were attenuated after controlling for early life adversities, the associations were not reduced to zero. There are a number of potential explanations for these findings. It could be that a direct effect of sexual behavior on mental health, where greater sexual behavior results in poorer mental health, exists given the indications from the existing literature. Alternatively, it may be that early life adversity does explain all of the association but measurement error attenuated the reduction in the effect observed since the assessments of adversity had limitations and were not all-encompassing. Or it could be that all of the association between sexual behavior and mental health is explained by third variables, including unmeasured confounders (e.g., genetic and/or environmental factors). Indeed, sexual behaviors and mental health outcomes have their own genetic and environmental contributions (Harden, 2014; McAdams et al., 2014). Exposure to early life adversities could also have prior causes in unmeasured genetic and/or environmental factors. Thus, there may be complex patterns of multivariate genetic and environmental associations between early adversity, sexual behavior and psychopathology. Future research using genetically sensitive longitudinal designs may help unpack these complex associations (e.g., cross-lagged twin approaches; Luo et al., 2010). Such studies will need to incorporate measures of the social context and psychosocial factors to improve our understanding of biopsychosocial causation.

The key strengths of the present study include the use of a prospective, well-characterized and representative longitudinal cohort with good statistical power. Mental health outcomes at age 17 were measured prospectively after sexual behavior at age 14, and early life adversities were also measured prospectively and often by independent ratings (e.g., caregiver). This reduces the risk of recall biases and the likelihood of reverse causation between early life factors and later sexual behavior and mental health scores. The range of early adversity factors studied closely match those reported in previous work as does the use of two theoretically relevant developmental stages. We found little evidence of differences in mischievous responding between adolescents who had no sexual experience and those with sexual experience. Our LCA also incorporated the measurement error from the determination of most likely latent class memberships of adolescent sexual behavior. The use of the ASAI provided a relatively unambiguous and validated measure of a full range of sexual activities among adolescents. The measure reduces the use of excessive introspection or deliberation regarding sexual feelings when completing the measure, again improving accuracy of recall.

Nevertheless, several limitations are important to note. Adolescents who have engaged in oral sex but not sexual intercourse may be different from those who have engaged in both sexual activities. Some adolescents may have engaged in high-intensity sexual activities without having experienced low-intensity ones. However, we are unable to detect such patterns since adolescents did not progress to the next stage of the ASAI in MCS if they reported having not engaged in all sexual activities in one stage. The ASAI may still suffer from some recall bias given the very low rates of endorsement to sexual activities in the higher-intensity range. Of course that could simply be due to the young age that the ASAI was administered. Also, within the age range examined, our statistical approach incorporated measurement error and so misclassification of class membership based on ASAI responses is less likely. The MCS study sample also suffers from additional measurement issues. Different periods were used for each mental health measure (e.g., mental well-being in the past 2 weeks and psychological distress in the past 30 days), which may affect the prevalence of poor mental health measured here. Some of the early life adversity and mental health measures were based on restricted or single-item responses (e.g., self-harm attempts), which is a key limitation. Some early life adversities, such as parental substance use, were measured only at one or two time points. Thus, we cannot be certain what the duration of such early life adversities was nor test the influence of early life adversities at different critical periods on adolescents’ mental health outcomes. As mentioned, we were also unable to control for unmeasured confounders that load simultaneously on early life adversities, early sexual behaviors, and mental health outcomes.

In sum, the present findings suggest that not only adolescents who have engaged in sexual intercourse but also adolescents who have engaged in low-intensity sexual activities at an early age may be at risk of some poor mental health outcomes. Early life adversities may help explain some of this association. Future studies should test for possible causal mechanisms and unmeasured factors that may explain these associations such as genetic and environmental confounds.

Acknowledgments

The authors are extremely grateful to all the families who took part in the Millennium Cohort Study. We are also grateful to the Centre for Longitudinal Studies, Institute of Education for the use of the data and to the UK Data Archive and Economic and Social Data Service for making them available.

Disclosure Statement

No potential conflict of interest was reported by the authors.

Data Availability Statement

The data that support the findings of this study are openly available on the UK Data Service at https://www.ukdataservice.ac.uk.

Supplementary Material

Supplemental data for this article can be accessed on the publisher’s website.

Additional information

Funding

This research received no specific grant from any funding agency, commercial or not-for-profit sectors.