Mechanisms of action for empirically supported interventions to reduce adolescent sexual risk behavior: A randomized controlled trial

Abstract

Purpose

Adolescents are at high risk for sexually transmitted infections including HIV. Interventions to reduce adolescent sexual risk often have modest outcomes. Understanding of the mechanisms of program effectiveness is needed in order to develop stronger interventions. We utilized a randomized controlled trial (RCT) to examine mechanisms of response to two empirically supported interventions: motivational interviewing (MI) versus behavioral skills training (BST).

Methods

262 adolescents (M age = 15.89, SD = 1.24, 34% female, 61% Latinx) were recruited from juvenile justice programs in the United States; all youth were randomized to MI or BST. Primary outcomes included 1) theory-based mechanisms: condom use attitudes, norms, self-efficacy, and intentions measured before and immediately after the interventions and 2) risky sexual behavior (frequency of unprotected sex) and condom use measured three months post-intervention.

Results

Both interventions significantly increased positive attitudes towards condom use, self-efficacy for condom use, and intentions to use condoms posttest, with no differences by condition. Neither intervention impacted norms for condom use. Both interventions significantly decreased risky sexual behavior and increased condom use at the 3-month follow-up. The pattern of associations of TPB constructs to behavior were of similar magnitude in the two groups.

Conclusions

Results indicate common mechanisms of action for these two interventions and support use of transdiagnostic mechanisms of treatment impact for sexual risk reduction.

Despite rates of human immunodeficiency virus (HIV) and sexual transmitted infections (STI) decreasing among many groups, youth (ages 15–24) continue to account for more than 1 in 5 new HIV diagnoses in the United States [1], including half of the newly diagnosed cases of chlamydia, gonorrhea, and syphilis [2]. Low rates of condom use is a primary challenge to HIV/STI prevention in this age group [3]. While theory-based interventions show some success in adolescent HIV risk reduction [4], the mechanisms by which they achieve their success remains under examined; these data are critical to inform how to improve and promote adolescent health [5].

The Theory of Planned Behavior (TPB) states that health protective behavior (e.g., condom use), is the proximal result of intentions to engage in that behavior. Intentions are in turn the result of attitudes about the behavior, norms supporting the behavior, and self-efficacy or confidence in one’s ability to perform the behavior [6]. Recent meta-analyses support these assertions across a range of health behaviors [7]. The question asked in this research is which empirically-supported intervention best targets TPB constructs to increase condom use among adolescents.

Motivational interviewing (MI) [8] is an empirically-supported intervention which has been viewed as synergistic with TPB constructs in prior studies [9, 10]. This synergy has also been supported by a recent randomized controlled trial with adolescents, which showed that an MI intervention which successfully increased condom use operated via TPB-constructs [11, 12].

Another common intervention for sexual risk reduction is Behavioral Skills Training (BST) [13, 14]. However, we could find no published research on theoretical mechanisms of BST treatment outcomes in the context of adolescent sexual risk reduction, though TPB constructs may also serve as the driving factors in BST. For example, BST may increase self-efficacy, because it helps adolescents learn, practice, and gain proficiency in basic behavioral skills (such as putting on a condom) [15]. In addition, the rewards-based nature of the intervention may result in more positive attitudes toward condom use.

Compellingly, these two empirically-supported interventions, and their underlying mechanisms of action, have never been directly compared in the context of adolescent sexual behavior. Rates of risky sexual behavior and its health sequelae are disproportionately pronounced for adolescents involved with the juvenile justice system compared to the general population [16, 17]. Given that there are over 30 million adolescents in the US juvenile justice system, their disproportionately high levels of sexual risk, and the justice system as a point of contact for health intervention, youth involved with the justice system represent an especially critical population in which to conduct and evaluate sexual risk reduction interventions in order to improve their efficacy.

In this study, we randomly assigned justice-involved adolescents to receive time-matched sessions of MI vs. BST. We measured TPB-based mechanisms of change (condom use attitudes, norms, self-efficacy, and intentions) immediately before randomization and post-treatment to determine degree of proximal change. Further, we examined whether changes in TPB mechanisms were associated with risky sexual behavior for all youth (defined as frequency of unprotected sex) and condom use among sexually active youth, three months post-intervention. Based on past research [19, 20], we hypothesized that the MI intervention would change condom use attitudes, norms, and self-efficacy, which would in turn relate to intentions, and subsequently, three-month risky sex outcomes. We expected that the BST intervention would change condom use attitudes and self-efficacy, but not norms as there is no content directly targeting norms in BST. In addition, rather than working indirectly through intentions to influence future behavior, we instead predicted a direct path from attitudes and self-efficacy to three-month risky sex outcomes for the BST group. We expected both interventions to improve sexual behavior outcomes of lower frequency of unprotected sex and increased condom use.

Methods

Participants.

This study included N = 262 adolescents (M age = 15.85, SD = 1.24, 34% female) invited to participate in a larger-scale translational study aimed at understanding neurocognitive mechanisms associated with improving adolescent health. The primary aim of this longitudinal study is to examine neural moderators of two STI/HIV behavioral risk interventions (#NCT02141646; 1R01NR013332).Most participants (64%) reported their ethnicity as Latinx. Approximately 61% of participants reported that they were still in school, and on average, the last grade completed was 9^th grade (M = 9.11, SD = 1.25). Most (67.5%) reported that they were eligible for free or reduced-priced school lunch.

Procedures and Informed Consent.

Enrollment, intervention, and follow-ups were completed between January 2012 and December 2015. Research staff introduced the project at several day treatment juvenile justice programs (i.e., community diversion and alternative to incarceration programs) housed in one center in the southwest United States. Youth were informed that participation was voluntary and separate from their justice involvement. In addition, youth were informed that they could leave the study at any time without penalty. Youth completed informed written assent and parent/guardian consent was obtained via telephone [20]. All study procedures were conducted with University Institutional Committee on Human Subjects approval, a federal Certificate of Confidentiality, and a waiver of the Health Insurance Portability and Accountability Act (1996).

Eligible participants were between the ages of 14–18 years and currently involved with a juvenile justice program. Exclusion criteria included requirements for the magnetic resonance imaging (MRI) component of the study (not reported herein) and included current MRI contraindications (e.g., current pregnancy, non-removable metal in the body), and loss of consciousness (>5 min during past 6 months). All measures were completed on a laptop via audio computer assisted self-interview [21], which facilitates participation for youth with less school exposure and lower literacy. Participants could skip any question without penalty. Participants received $20 in return for completing baseline measures, $20 for each intervention session and completing posttest measures, and $40 for completing the follow-up assessment.

All baseline assessments and interventions were delivered in one of three confidential rooms dedicated to the study at the partnering juvenile justice center. Justice center staff were not involved in recruitment, consenting, or treatment delivery. Participants completed follow-up measures at a location convenient to them (e.g., juvenile justice center, partnering university). For more information regarding follow-up procedures, see [22].

Design.

This study was a between-subjects randomized controlled trial with two conditions, comparing changes in mechanisms of sexual risk behavior change pre- to post-intervention, and behavior change three months after receiving MI vs. BST. Sample size was selected to permit analysis of the primary research questions at two-tailed alpha of .05 and power level of .80. Estimates of effect size follow Cohen and were conducted in G*Power 3 [23]. The larger trial from which these data come utilized three behavioral measurement points, and it is these three time points that were used in the power analysis. Using a small effect size (f = .10), two groups, three assessment points, and an average correlation of ~ .50 between assessment points, we estimated a minimum of 156 participants (78 per group) would be required. Our sample size thus exceeds minimum requirements for power.

Randomization and Masking.

Study staff generated a random assignment table for each gender using a random number generator. Upon completion of consent and assent, a research assistant enrolled and block-randomized participants by gender to either the MI or BST intervention. Given the nature of behavioral trials, it was not possible for participants to be fully blind to condition. However, they were blind to the nature of the other intervention condition as well as the study hypotheses.

Measures.

Participants completed measures querying demographics and putative mediators at baseline and post-intervention. Baseline values for both groups on these assessments appear in Table 1.

Table 1.

Baseline Demographic Characteristics and Pretest Equivalence Tests.

	Overall Sample (n=222)	MI Group (n=115)	BST Group (n=107)	Equivalence test p-value
Age (years)	15.85 (1.22)	15.77 (1.24)	15.93 (1.22)	p = .33
Gender (% female)	33.8%	28.7%	39.3%	p = .13
Ethnicity (% Latinx)	64.25%	60.5%	68.22%	p = .29
Last grade completed	9.11 (1.25)	9.15 (1.22)	9.07 (1.28)	p = .62
Baseline Risky Sex1	8.62 (8.38)	7.80 (7.67)	9.51 (9.05)	p = 17
Baseline Condom Use2	2.92 (1.54)	2.89 (1.50)	2.95 (1.59)	p = 79
Baseline Attitudes3	5.42 (1.27)	5.34 (1.36)	5.52 (1.16)	p = .30
Baseline Norms	2.63 (0.86)	2.64 (0.87)	2.61 (0.84)	p = 75
Baseline Self-Efficacy4	3.38 (0.42)	3.36(0.41)	3.41 (0.44)	p = 35
Baseline Intentions4	2.87 (0.86)	2.83 (0.90)	2.91 (0.82)	p = .52

Note.

^1.Past 3-month risky sex index (range 1–30);

^2.Past 3-month condom use scored on a Likert scale, (1=Never, 5=Always);

^3.7-point semantic differential scale;

^4.4-point Likert Scales

Theoretical Mediators:

Condom Use Attitudes.

Condom use attitudes were measured using a TPB semantic differential attitudes scale [5, 6]. This 7-item measure included items such as “On a scale from 1–7, where 1 means harmful and 7 means beneficial, using a condom every time I have sex would be…” (α = .89).

Condom Use Self-Efficacy.

The condom use self-efficacy measure was based on the Condom Use Self-Efficacy Scale [24]. The 19-item measure included items such as “I am confident in my ability to use a condom correctly”; responses were on a 1–4 Likert scale (1=Disagree at lot, 4=Agree a lot) (α = .78).

Condom Use Norms.

Condom use norms were measured on a previously validated 8-item scale on the same 1–4 Likert scale as self-efficacy and included items such as “Most of my friends use condoms when they have sex” (α = .91) [12].

Condom Use Intentions.

Intentions to use condoms in the future were measured on a 4-item scale (1= Will NOT happen, 4= Will DEFINITELY happen) previously utilized in Schmiege et al. [12] that included items such as “How likely is it that you will use a condom every time you have sexual intercourse in the next three months?” (α = .88).

Condom Use and Risky Sexual Behavior.

In line with prior work [12], a risky sex index was used to evaluate three-month sexual risk behavior. Component variables included condom use frequency, “In the past three months, how much of the time did you use condoms when you had sexual intercourse?” (5-point Likert scale ranging from “Never” to “Always) and frequency of sex, “On average, in the past three months only, how often have you had sexual intercourse?” (6-point Likert scale ranging from “Never” to “Almost Every Day”). The risky sex index was calculated by multiplying frequency of sex by the reverse-scored condom use measure resulting in a range of 1 to 30 where higher numbers indicate a higher frequency of unprotected sex. Participants who said that they had not had sex in the past three months received a score of 0 on this index, reflecting no sexual risk. Thus the risky sex index included all participants. For adolescents who had had sexual intercourse at least once in the prior three months, we also examined outcomes on the condom use frequency variable alone.

Intervention Content.

All youth received two, one-hour sessions of either MI or BST spaced one week apart to provide an opportunity to practice newly acquired skills in the intervening weekend. All youth in this project (100%) attended both treatment sessions. Participants in both conditions received information about HIV/STI transmission and prevention. All interventions were conducted by trained research staff, matched by gender to the participant. Therapists were affiliated with the investigators’ institution and were not known to participants prior to the intervention. To ensure the accuracy, integrity and consistency of the MI and BST interventions, both interventions were operationalized into manuals (blinded for review). These interventions are based on our prior work along with seminal publications in each respective field [8, 25]. All therapists were trained to criterion by one of the MPIs of this study. With participant permission, all sessions were audio-recorded for the purpose of evaluating therapist fidelity and preventing therapist drift. Intervention integrity and fidelity were monitored and maintained by one of the MPIs, who reviewed random audio-recorded sessions each week during therapist supervisory meetings.

MI Condition.

All MI interventions were conducted in an MI-consistent manner, that is, open, strength-based, affirming, non-judgmental, empathic, with a goal of reducing resistance and highlighting ambivalence around current sexual risk behavior in order to foster and support participants’ intrinsic motivation for behavior change.

BST Condition.

The goal of this behavioral intervention was on education/instruction, modeling, and rehearsal. For instance, trained therapists demonstrated the steps for successful condom negotiation with a sexual partner, which adolescents practiced in a role-play discussion. To reinforce short-term behavioral goals, youth were informed that successful goal achievement would be rewarded with immediate positive reinforcers (tokens = $1), which were awarded in cash to the youth at the close of each session. If participants failed to demonstrate aptitude in a task, they received feedback and continued practice until they met the goal criterion prior to advancing to the next topic; all youth successfully completed all required session content.

In both conditions, therapists demonstrated the steps for correct condom use on a wooden penis model. Youth then practiced and performed the steps on a wooden model.

Data Analysis.

To test successful random assignment to condition, we evaluated baseline demographic variables and outcome measures of interest using independent samples t-tests. Next, we examined mechanisms of interest (TPB constructs of attitudes, norms, and self-efficacy) by regressing the difference score for the posttest measure minus the pretest measure of the TPB construct, or the difference score for the three month measure of behavior minus the baseline measure of behavior on intervention group (contrast coded; MI condition = −0.5, BST condition = +0.5). Note that participants were given the option to skip questions (given the answer choice “I do not feel comfortable answering this question”), and thus, degrees of freedom vary slightly across each variable. In additional analyses, we examined the above models including covariates and moderators; specifically, we examined whether controlling for age or gender changed the relationship between condition and pretest/posttest change, and also examined whether three-month intervention response differed based on these factors. Neither controlling for these variables nor including them as moderators of intervention response changed the pattern of results; we thus did not include them in presented results. Finally, we estimated a cross-group path analytic model to test whether the relationship between posttest values on the mediators and follow-up condom use differed by condition [26]. Based on prior theorizing [27, 28], we estimated both direct and indirect paths from condom-use attitudes and self-efficacy to sexual risk behavior.

Results

Descriptive Statistics and Attrition

75.8% of participants reported that they had had sexual intercourse in the past three months at baseline. Of those, 71.7% reported having sex at least once per week, and only 21.7% said that they “always” used condoms. At baseline, participants in the BST vs. MI conditions did not significantly differ by age, gender, ethnicity, last grade completed (all p’s > .13), or proposed mediators (all p’s > .17; see Table 1). We did not observe differential attrition by condition from baseline to posttest (χ² = .70, p = .40), nor did we observe differential attrition by condition from posttest to follow-up (χ² = .95, p = .33). Figure 1 shows the flow of participants through these phases of the study.

Figure 1.

Subject flow

Degree of Change in Proposed Theoretical Mechanisms (Baseline/Posttest)

Table 2 shows mean level of change in the proposed mechanisms by condition. Condom use attitudes significantly increased from baseline to posttest; this increase did not differ by condition. Condom use norms did not significantly change over time; this did not differ by condition. Condom use self-efficacy significantly increased over time; this increase did not differ by condition. Condom use intentions significantly increased over time; this increase did not differ by condition.

Table 2.

Pretest and Posttest Means for Theoretical Mediators and Sexual Behavior Outcomes and Tests for Intervention Effects.

	Cell means by condition			Effect of time						Time x Condition (MI vs. BST) Interaction
Construct	MI (n =115)	BST(n =107)		β		95% C.I.	t(df)	Partial n2		β	95% C.I.	t(df)	Partial n2
Attitudes				.47	***	[.311, .636]	t(218) = 5.72	.13		−.21	[−.54, .109]	t(218)=−1.29	.01
Pretest	5.34 (1.36)	5.52 (1.16)
Posttest	5.92 (1.26)	5.88 (1.17)
Norms				.00		[−.107, .116]	t(209) = .08	.00		.05	[−.172, .273]	t(209) = .45	.00
Pretest	2.64 (0.87)	2.61 (0.84)
Posttest	2.64 (0.80)	2.62 (0.75)
Self-Efficacy				.12	***	[.071,.172]	t(214) = 4.72	.09		−.03	[−.129, .074]	t(214) = −.53	.00
Pretest	3.36 (0.41)	3.41 (0.44)
Posttest	3.49 (0.41)	3.52 (0.44)
Intentions				.31	***	[.201, .431]	t(210)=5.42	.12		−.1	[−33,.13]	t(210)=−.85	.00
Pretest	2.83 (0.90)	2.91 (0.82)
Posttest	3.21 (0.81)	3.19 (0.82)
Risky Sex				−2.81	***	[−3.90, −1.73]	t(157) = 5.12	.14		−1.45	[−3.62, .720]	t(157) = −1.32	.01
Pretest	7.79 (7.67)	9.51 (9.06)
3 Month Follow-up	4.94 (7.07)	5.42 (6.77)
Condom Use				.37	**	[.104, .64]	t(92) = 2.76	.08		−.03	[−.564, .512]	t(92) = −.10	.00
Pretest	2.89 (1.50)	2.95 (1.60)
3 Month Follow-up	3.37 (1.58)	3.28 (1.63)

Note. t tests for main effect of time and time by condition interaction. MI = Motivational interviewing, BST = Behavioral Skills training.

^***p<.001, two-tailed.

^**p<.01,

^*p <.05 two-tailed.

Intervention Effects on Risky Sexual Behavior and Condom use at Three Month Follow-Up

Risky sexual behavior significantly decreased from baseline to follow-up; this decrease did not differ by condition. Past 3-month condom use, a component variable of the risky sex index, significantly increased from baseline to follow-up across groups; this increase did not differ by condition.

Cross Groups TPB Model

The TPB model was estimated simultaneously in a cross-groups model, with structural paths free to vary across groups. We utilized the variable of condom use as the outcome, as the TPB items were articulated to condom use rather than frequency of sex. Based on prior work and TPB theorizing, we included direct paths from self-efficacy to behavior and attitudes to behavior (see Figure 2). Model fit was adequate; χ² = 1.35, (df = 2, n = 242, p = .5), CFI=1.000, RMSEA=.000, 90% C.I. [.000, .161].

Figure 2.

Cross-group path models Note. Path models demonstrating relationships between posttest theoretical constructs and follow-up condom use behavior by group. Structural paths were freely estimated across groups. Coefficients are standardized path coefficients. + indicates p < .1, * indicates p < .05, ** indicates p < .01, *** indicates p < .001.

For the BST group, posttest attitudes and self-efficacy were significantly associated with intentions at posttest, and posttest norms were marginally associated with posttest intentions. In turn, posttest intentions were significantly associated with follow-up condom use behavior. There was a significant direct path from attitudes to behavior, but not from self-efficacy to behavior. For the MI group, posttest attitudes were significantly associated with posttest intentions, but norms and self-efficacy were not associated with posttest intentions. There were no significant direct paths between attitudes and behavior or self-efficacy and behavior in the MI group.

We also estimated a fully constrained model (i.e., all structural paths and correlations were constrained to be equal across groups) and compared this to the first model where these coefficients were freely estimated across groups. This analysis indicated that the fit of the two models was not significantly different, suggesting that the magnitude of the differences in these paths was not large enough to reach statistical significance.

Discussion

In this study, we examined whether two sexual risk reduction interventions, MI versus BST, affected behavior equivalently through theory-based social cognitive mechanisms. We found that the MI and BST interventions both significantly increased TPB constructs of condom use attitudes, self-efficacy, and intentions. Consistent with meta-analytic evidence that it may be most difficult to change perceived norms in this age group [8], neither intervention showed any movement on norms. Crucially, posttest measures of the proposed TPB mechanisms, most strongly attitudes and intentions, were positively associated with condom use at the three-month follow-up. For the most part, we did not find that the two interventions acted through different TPB mediators to affect post-treatment behavior, with one small exception. When structural paths were estimated freely across groups, in the BST intervention, in which adolescents were rewarded with tokens for successful condom use approximation behaviors, attitudes toward condom use showed a direct path to follow-up behavior engagement. We did not observe this path in the model for the MI group.

These findings of largely similar mechanisms of treatment efficacy are compelling, given recent attention to “common” elements that may underlie treatment outcomes across treatment modalities [5], including the recognition and movement of the field toward common transdiagnostic mechanisms [29, 30] in order to develop more effective interventions. Given that both interventions changed attitudes, self-efficacy, and intentions, and that attitudes and intentions impacted risky sexual behavior and condom use behavior in both conditions, it is possible that any intervention that successfully targets condom use attitudes and intentions may change behavior.

What our data may also support is that exposure to an empirically-supported treatment delivered by a caring clinician has significant and reliable effects on both mechanisms of change and sexual behavior outcomes. The empirically-supported treatment may be key, however, since Schmiege et al. [20] found that a group-based MI had stronger effects on TPB mediators than an information-only control intervention, despite both being delivered by the same highly trained and caring clinicians.

Contrary to hypotheses, while self-efficacy significantly improved across both the MI and the BST conditions, it did not gain support as a mechanism of behavior change of either intervention, either in bivariate correlations or in the path models. These results were unexpected, given that much of the literature demonstrates robust relationships between self-efficacy and behavior across a number of treatment approaches [7]. The lack of effects might reflect either ceiling effects for self-efficacy, suppression effects, and/or shared variance related to a relatively high correlation between the measures of attitudes and self-efficacy in this sample. Supporting ceiling effects, self-efficacy scores were near the high end of the range of the scale.

An important strength of this study is the use of a high-risk sample of adolescents as study participants. This is important due to recent research demonstrating that adult models for behavior change interventions likely do not generalize to adolescents [31]. Another strength is the testing of a relatively unexplored intervention (BST) in comparison to MI in the context of the TPB. Our examination of TPB variables in the context of these interventions represents a novel perspective for potentially clarifying mechanism(s) of action within these two interventions [32].

One limitation of the current study is the lack of a no-treatment control group. However, due to ethical concerns, it is not possible to conduct a study on a vulnerable population of juvenile justice-involved adolescents that would have involved no expected benefit. Further, we have demonstrated in prior work that a theory-based MI intervention is superior to the provision of information alone [33]. A second limitation is that while adolescents in juvenile justice are a highly relevant population for a risk behavior intervention study, we cannot be certain that the results would generalize to adolescents not involved in juvenile justice. Finally, the behavioral data reported herein are self-reported, a limitation shared with much research on sexual behavior. Taken together, results from the present study support the existence of common mechanism(s) of action for behavior change interventions among high-risk, justice-involved adolescents. Given that transdiagnostic models have gained considerable support in recent years [29, 30], it makes sense that different therapeutic approaches may leverage similar mechanisms. Our results underscore the importance of TPB variables in mediating behavior change in the context of these interventions. Further work in this area should explore whether other empirically-supported interventions for sexual risk reduction in this population (even those not explicitly designed to target these constructs) may have their effects via TPB variables. Identifying which treatments best target transdiagnostic mechanisms of change may help to maximize treatment efficacy and promote better health outcomes.[18]

Implications statement:

Rates of risky sexual behavior are disproportionately pronounced for juvenile justice-involved youth, but interventions targeting this behavior often have modest outcomes. Understanding the mechanisms through which theory-based interventions reduce adolescent sexual risk is an important step in developing effective interventions. Across two interventions, this study identifies transdiagnostic mechanisms of treatment efficacy.

Abbreviations:

RCT

Randomized Controlled Trial

MI

Motivational Interviewing

BST

Behavioral Skills Training

HIV

Human Immunodeficiency Virus

STI

Sexually Transmitted Infections

TPB

Theory of Planned Behavior