Sexual Risk Outcomes of an Emotion Regulation Intervention for At-Risk Early Adolescents

Additional details regarding study methods have been previously published.^18,19 With institutional review board approval, students from 5 urban Rhode Island public schools were enrolled between September 2009 and February 2012. Inclusion criteria included being in the seventh grade, 12 to 14 years old, English-speaking, and exhibiting at least 1 symptom of emotional or behavioral problems or suspected sexual or substance use behavior. School professionals identified at-risk students by using a form listing mental health symptoms (eg, withdrawing, hyperactivity, nervousness). Exclusion criteria included a history of sexual aggression, HIV infection, current pregnancy, developmental delays, or a participating sibling.

Approximately 27% of students attending seventh grade during the 3 study years were referred to the program, consistent with national prevalence data for adolescent mental health disorders.²⁰ Consent and assent were obtained for 40% of referred students (n = 420; Fig 1). Sample size was determined by power analysis by using the following primary study outcome: time to first vaginal sex. Using well-validated psychiatric screening instruments (adolescent-reported Youth Inventory-4 and parent-reported Adolescent Symptom Inventory),²¹ we found that 65% of the sample had a T-score of ≥65 for at least 1 of 7 subscales (attention-deficit/hyperactivity disorder [any type] was 46%, conduct disorder 21%, oppositional defiant disorder 20%, generalized anxiety disorder 16%, major depressive disorder 31%, dysthymic disorder 37%, and bipolar disorder was 32%).

The sample was composed of 53% boys, and the average age was 12.94 years (SD = 0.53) at baseline. Racial backgrounds included 32% white, 28% African American, 3% native Hawaiian or other Pacific Islander, 2% American Indian, 1% Asian American, and 18% multiracial (16% did not endorse a racial category). Thirty-eight percent of participants reported Hispanic ethnicity. Thirty percent of parents reported an annual family income <$20 000 (18% did not report family income). Vaginal or anal sex before baseline assessment was reported by 11% of participants. There were no statistically significant baseline differences by intervention condition on age, sex, race, ethnicity, household income, previous sexual activity, or presence of clinically significant symptoms on mental health screening. There was no difference between conditions on attendance at the 12 after-school sessions ([9.1 vs 9.0]; t [418] = 0.20; P = .84).

Adolescents participated in 1 of 2 after-school interventions in which structured curricula (ER or HP) were used. Each consisted of 12 twice-weekly, hour-long sessions, run in single-sex groups of 4 to 8 adolescents. Two booster sessions were provided to both conditions at 6- and 12-month follow-ups. Groups were led by male-female pairs that included a mental health clinician (or clinician in training) and a research assistant. In both interventions, the same techniques were used to convey and personalize information, including interactive games, videos, group discussions, and workbook activities. ER sessions contained education in ER (eg, recognizing feelings in self, strategies for reducing momentary emotional arousal) and sexual health (eg, sexual anatomy, STIs). Addressed in HP sessions were health topics including sexual risk, substance use, and nutrition but not emotion education.

To minimize contamination effects that might reduce the treatment effect and to avoid nesting treatment condition within schools, a cluster-randomized crossover design²² was used. Schools were randomly assigned by coin flip to an order of conditions over 2 school years (eg, ER in year 1 and HP in year 2). Thus, the 2 conditions were offered in each school over different school years to different groups of seventh-graders.

The ER intervention was aimed to enhance ER skills to reduce poor decision-making associated with risk behaviors. The program was used to present the relationship between emotions and behaviors and approaches to recognizing and labeling emotions and recognizing their sources (eg, feeling pressured to have sex). It was used to present developmentally appropriate strategies for regulating emotions (both positive and negative) during moments of decision-making by using the following 3 strategies: (1) “Get Out,” getting away (physically or cognitively) from triggers for strong emotions, (2) “Let It Out,” releasing emotional energy in healthy ways (verbally or physically), and (3) “Think It Out,” changing cognitions and appraisals about emotional triggers. Games and role plays were used to apply these strategies to various risk situations. Sexual health education was integrated, including information about sexual development, STIs, and disease and/or pregnancy prevention, which included discussions of abstinence, nonpenetrative sexual behaviors, and condom use.

The comparison condition was loosely modeled on another intervention²³ that was used to encourage healthy decision-making through information, similar to many public school health curricula. Topics, which were matched for time and delivered using similar activities, included internet safety, substance use, violence, nutrition, exercise, sleep, and cigarette smoking. The sexual health content of the ER intervention was included, with modifications to eliminate discussion of emotions related to decision-making.

Facilitators were trained in the structured curriculum through mock groups, practiced for each session. Annual trainings were used to review session goals, role plays, behavior management, and protocol adherence. Facilitator-completed ratings revealed excellent (98%) manual adherence, as did senior project staff observations (observed for 15% of sessions; 97% adherence).

Participants completed audio computer assisted self-interviews of sexual behavior at baseline and every 6 months postbaseline for 30 months (6 assessments). Data from an abbreviated assessment given immediately postintervention were used only for providing information for time to sexual initiation; the effects of the intervention immediately after program completion and at 1-year follow-up have been published elsewhere.^18,19 Parents completed baseline demographics and adolescent mental health symptom questionnaires on laptops in the language of their choice (English, Spanish, or Portuguese).

Participants were asked at each assessment whether they had ever engaged in vaginal sex (the primary outcome). Adolescents responding yes were asked additional questions about the last 6 months, including whether this was the first time they had vaginal sex and, if so, in which month it had occurred.

Adolescents were also asked whether they had engaged in anal sex in the last 6 months, although not whether this was the first time. Adolescents who endorsed vaginal or anal sex in the previous 6 months were asked about the number of vaginal and/or anal sexual partners, number of vaginal and/or anal sex acts, number of times condoms were used, how often (on a 5-point Likert scale) they used drugs or alcohol in the 4 hours before sex, and whether they used substances before their last sexual encounter. Behavioral definitions were provided to ensure understanding; definitions did not reference partner gender to allow for both opposite- and same-sex encounters.

Reported vaginal and anal sexual acts were summed across the 30-month follow-up period. The same procedure was repeated for condomless sexual acts. At each assessment, the number of partners in the last 6 months was asked; however, the identity of partners was not recorded, meaning that the cumulative number of partners could not be determined (ie, having 1 partner at each of 4 assessment periods could reflect 1 partner for 2 years or 4 different partners). Instead, with our analyses, we used a dichotomous variable (any versus none) examining the presence of any 6-month period during follow-up in which participants endorsed ˃1 partner.

A sexual risk composite was calculated such that those who reported ˃1 partner during any 6-month period, any condomless vaginal or anal sex, or any substance use before sex were coded as “1.” Paricipants reporting none of these behaviors or no sexual activity during the 30-month follow-up were coded as “0.”

To account for participant nesting within the 5 schools, school was included in all statistical models as a categorical covariate, entered as a set of 4 dummy-coded contrasts.

Unless otherwise specified, analyses included all 420 participants. Seventy-two percent of participants completed all 5 follow-up assessments with 12% completing 4, 6% completing 3, 4% completing 2, 3% completing 1, and 4% completing no follow-up assessments. Fifty multiple imputations, generated by using chained equations,²⁴ were used for all analyses to address potential bias because of dropout.

In accordance with current recommendations, inverse probability of treatment weights were used to adjust for potential imbalance between conditions on baseline characteristics.^25,26 Weights were generated by using boosted decision trees, a flexible, semiparametric ensemble modeling approach found to balance baseline characteristics among treatment conditions.²⁷ Weights were stabilized by dividing the baseline probability of being assigned a treatment by the estimated probability from the boosted regression. Stabilized weights have been shown to improve performance in smaller samples.²⁸ 

To assess the delay of sexual initiation, we used a Cox-proportional hazard model to test differences between conditions. Time in months from beginning the study to first vaginal event was calculated for each participant. Those who reported vaginal sex before beginning the study (HP = 12%, n = 24; ER = 9%, n = 20; χ²(1) = 1.09; P = .30) were not included. Multiple imputations were used to protect against violations of the independent censoring assumption of the proportional hazard model.²⁹ 

To assess the impact of the intervention on sexual risk behaviors, we used negative binomial models to evaluate behavioral counts of vaginal or anal sex acts through the 30-month follow-up. The sexual risk composite was evaluated by using logistic regression. Baseline values were not included in the models given the low rate of risk behavior at baseline.

Subgroup analyses were conducted to determine whether the intervention had an impact on sexual risk behaviors among those who were sexually active. Sexual behavior was summed for the period from vaginal or anal sexual initiation (for some, this was baseline) to the end of the study for each participant. The rate of sexual behaviors across this period was evaluated by using negative binomial models. The risk composite was evaluated by using logistic regression.

Kaplan-Meier survival curves representing time to sexual initiation are presented in Fig 2. Of those who had not had sex at baseline, 68 participants in the HP condition (39%) and 63 participants in the ER condition (31%) reported having vaginal sex for the first time between baseline and the 30-month assessment. Controlling for school, we found that the adjusted hazard ratio (AHR) for the difference between conditions indicated a delay in the transition to vaginal sex over 30 months for participants in ER compared with participants in HP (AHR = 0.61; 95% confidence interval [0.42 to 0.89]).

Unadjusted imputed means for cumulative (over 30 months) behavioral counts and probabilities for cumulative dichotomous outcomes along with adjusted odds ratios (aORs) and adjusted rate ratios (ARRs) of sexual risk behaviors are listed in Table 1. Effect sizes are depicted in Fig 3. ER participants were less likely than HP participants to endorse any behavior in the sexual risk composite during the 30-month follow-up (aOR = 0.52; 95% confidence interval [0.32 to 0.84]). ER participants were less likely than HP participants to endorse multiple partners in any 6-month period (aOR = 0.54; 95% confidence interval [0.30 to 0.99]) and less likely to endorse substance use before sex (aOR = 0.42; 95% confidence interval [0.23 to 0.75]). The groups exhibited less difference with respect to whether they had any condomless sex (aOR = 0.81; 95% confidence interval [0.45 to 1.45]). However, examination of the number of condomless sex acts by group indicated that ER participants reported fewer condomless vaginal or anal sex acts during follow-up than those in the HP condition (ARR = 0.36; 95% confidence interval [0.14 to 0.90]). They also reported fewer total vaginal or anal sex acts (ARR = 0.39; 95% confidence interval [0.20 to 0.77]).

Among adolescents who endorsed any vaginal or anal sexual experience during the study (ER: 43%; 95% confidence interval [36% to 50%]; HP: 50%; 95% confidence interval [43% to 57%]; aOR = 0.66; 95% confidence interval [0.42 to 1.02]), those in the ER condition were less likely to endorse any of the sexual risk composite behaviors (aOR = 0.66; 95% confidence interval [0.39 to 1.13]), although this did not reach statistical significance, nor did the group difference between those who endorsed multiple sexual partners in any 6-month period (aOR = 0.70; 95% confidence interval [0.38 to 1.30]). Sexually active ER participants were less likely than their HP counterparts to endorse using substances before sex (aOR = 0.52; 95% confidence interval [0.28 to 0.95]). ER adolescents also reported less frequent vaginal or anal sexual activity after sexual debut than HP adolescents (ARR = 0.57; 95% confidence interval [0.32 to 0.99]). Condomless vaginal or anal sex after debut did not reach statistical significance (ARR = 0.53; 95% confidence interval [0.24 to 1.17]), and groups were similar with regards to any occurrence of condomless sex (aOR = 1.08; 95% confidence interval [0.59 to 1.97]).

To explore the impact of mental health symptoms on treatment effect, we examined the endorsement of clinically significant symptoms (T-score ≥65) on any of the Youth Inventory-4 or Adolescent Symptom Inventory subscales as a potential moderator of the primary outcome (sexual initiation) and the presence of any sexual risk among those who were ever sexually active. Neither moderating effect was statistically significant, although the study was not powered for subgroup analysis. The effect sizes suggest relatively similar treatment effects on sexual initiation for participants with and without clinically significant symptoms (AHR_{no diagnosis} = 0.60; 95% confidence interval [0.31 to 1.17]; AHR_diagnosis = 0.65; 95% confidence interval [0.42 to 0.99]). In contrast, regarding the presence of any sexual risk among participants who were ever sexually active, participants with clinically significant symptoms showed a stronger treatment effect (aOR_diagnosis = 0.55; 95% confidence interval [0.30 to 1.04]) than those with lower levels of symptoms (aOR_{no diagnosis} = 1.01; 95% confidence interval [0.37 to 2.72]).

The inclusion of ER training in a small-group behavioral intervention reduced sexual risk behaviors among seventh-graders with suspected mental health symptoms over a 2.5-year follow-up beyond that achieved with more traditional health education. This was true across a range of behaviors, such as engaging in fewer condomless sex acts, being less likely to have multiple partners, and being less likely to use substances before sex. The intervention conditions were similar in design (ie, they used the same group format, core sexual health information, and interactive teaching techniques) but differed in content. ER skills appear relevant to early adolescent sexual decision-making and amenable to change in this developmental period.

The ER intervention appears to have reduced risk in the following 2 ways: delaying sexual activity and reducing sexual risk behaviors among sexually active youth. First, ER participants delayed onset of vaginal intercourse relative to peers. Teenagers in the HP group were 1.6 times more likely than teenagers in the ER group to initiate vaginal sex. The longer teenagers wait to have sex, the longer they postpone exposure to risk. Delaying the onset of sex beyond early adolescence has been associated with fewer sexual risk behaviors throughout adolescence and early adulthood.^1,–4 

Second, after participants became sexually active, those who learned ER strategies engaged in vaginal or anal sex less frequently and used substances before sex less often, which also reduces risk of STIs and pregnancy. In addition, although not statistically significant in the smaller sexually active subsample, the treatment effect for the rate of condomless sex acts was sizable (1.89 times higher for HP compared with ER), as were the effects for the sexual risk composite (1.52 times higher) and multiple partners (1.43 times higher). In exploratory analyses, it was also suggested that the treatment effect on sexual risk behaviors after debut may have been stronger among participants with clinically significant mental health symptoms.

With these findings, we suggest that building ER skills among vulnerable adolescents helps to delay sexual debut and reduce the number of sexual risk behaviors after debut. Because ER is a skill that could influence other adolescent risk behaviors such as substance use, violence, and truancy, addressing ER during this sensitive period in adolescent development promises significant public health benefits. The challenge is the scale-up and dissemination of ER interventions. Increasing the reach of programs in which health education is enhanced with emotion education may be an important step toward improving the lives of adolescents, because they are prone to beginning risk behavior. With this study, we selected identified at-risk youth, but additional approaches may be needed for dissemination. Integrating ER concepts into standard classroom-based health education could reach more youth, as well as innovate how young people learn to make health decisions. Other strategies might include mobile health (eg, via electronic games) or pairing ER with other interventions, such as parenting programs that might be used to teach ER skills to parents, who might then teach their children.

This study has limitations. Like many sexual risk studies, the data rely on self-reports. However, audio computer assisted self-interview was used to minimize biases. The sample was selected because of emotional and behavioral health symptoms and may not be generalizable to all early adolescents, although they are a population in need of effective intervention. Randomization was not at the individual level, allowing for the possibility of treatment confounders. Confounders were minimized by conditioning on school and baseline characteristics, but cohort effects may have existed. The sample may not be generalizable to other geographic regions. The measurement of sexual partners did not include identifiers, thus limiting the ability to examine the total number of partners across assessment windows. Finally, the study was powered to examine the delay of vaginal sex. As a result, a portion of the sample was not yet sexually experienced, providing less power for comparisons of sexual behaviors among the sexually active. Nonetheless, with the observed effect sizes, it is suggested that these, too, were influenced by the intervention in expected directions.

With this study, we reduced risk by intervening during the critical period of early adolescence (before the onset of most risk behavior) by using developmentally tailored strategies that may be generalizable to other health behaviors and continue to develop as adolescents gain greater cognitive and emotional maturity. Furthermore, the intervention was effective in teaching ER to adolescents challenged by these skills because of mental health difficulties. Enhancing ER during early adolescence represents a promising direction that deserves continued attention.

AHR

adjusted hazard ratio

aOR

adjusted odds ratio

ARR

adjusted rate ratio

ER

emotion regulation

HP

health promotion

STI

sexually transmitted infection