One-Year Follow-up After a Pregnancy Prevention Intervention for LGB+ Teens: An RCT

Advarra Institutional Review Board, an Office of Human Research Protections-approved IRB, reviewed and approved the research protocol. A waiver of parental permission was granted to avoid putting girls in potentially unsafe situations needing to disclose their sexual identity to their parents to obtain permission to participate in the study.²⁶ 

Baseline surveys were completed between January 23, 2017 and January 12, 2018. Twelve-month follow up surveys were completed between March 2018 and June 2019.

Girls were recruited across the United States. Eligible participants: (1) were 14 to 18 years old; (2) identified as cisgender (ie, assigned female sex at birth and endorsed a female gender identity); (3) identified as sexual minority (eg, lesbian, bisexual); (4) had not graduated high school, including those who did not finish or dropped out; (5) were English speaking; (6) had a cell phone and did not share it with anyone else; (7) were on an unlimited text messaging plan; (8) planned to have the same cell number for the next 12 months; (9) had used text messaging for at least 6 months; and (10) provided informed assent (for those under 18 years of age) or consent (for those 18 years old), including demonstrating a capacity to consent and passing the self-safety assessment.²⁷  Exclusion criteria included knowing another girl enrolled in the RCT or having participated in an earlier intervention development activity (eg, focus groups).²⁸ 

Participants were recruited nationally over social media, primarily through Facebook and Instagram. Profiles that indicated the user was female, between 14 to 18 years of age, and “interested in females” or “interested in males and females” were targeted. Youth who clicked on the study advertisement were sent to the online screener form.

Youth who appeared to be eligible based upon their screener responses were contacted for telephone enrollment. Youth were contacted sequentially while also considering preset demographic targets to promote sample diversity. For example, once the targeted number of non-Hispanic white girls was enrolled, we did not reach out to additional girls with the same racial or ethnic identities. After providing verbal assent or consent, girls were emailed a link to the baseline survey. Participants were randomized after they completed the survey.

Participants were randomized at a 1-to-1 ratio to the Girl2Girl intervention (n = 473) or the attention-matched control group (n = 475) by our software program. Randomization was stratified by (1) sexual experience (ever or never) and (2) sexual identity (lesbian or gay versus all others). Participants, but not researchers, were blind to arm allocation.

Youth did not receive incentives for completing the baseline survey. Participants received between $5 and $25 for completing the survey at intervention end. The incentive for the 3-month survey was $10 to $35; 6-month, $30 to $35; 9-month, $10 to $40; and 12-month, $10 to $45. Amounts varied based upon whether they completed a brief text survey or longer online survey.

Baseline surveys were collected online. Some follow-up surveys were planned as online and others as text messaging-based. When funding uncertainties arose, the data collection mode shifted accordingly. As such, outcomes within wave (eg, 3-month follow-up) were sometimes measured both online and through text messaging.

Program participation length for intervention participants depended on their interaction with the program. For example, if they did not respond to a level up question, they received a reminder in subsequent days before intervention messages resumed. Those who responded right away had messages resume right away. Thus, some girls’ intervention experience was longer than 20 weeks.

As described previously,²⁸  Girl2Girl is a 20-week, text messaging-based healthy sexuality program aimed at reducing teen pregnancy risk. Participants receive between 4 and 12 messages per day for 7 weeks. Then, they receive about 1 to 2 messages per week for 12 weeks to have time to integrate new behaviors into their daily lives. The intervention ends with a review week of messages that reiterate the main concepts discussed in the first 7 weeks of the program.

The Information-Motivation-Behavioral Skills model guided the content.^29,30  Messages discuss information about how pregnancy happens and can be prevented, motivations to avoid pregnancy (eg, reasons to use condoms), and behavioral skills (eg, how to negotiate condoms with partners). Additional content describes topics and scenarios that are relevant to sexual decision-making for sexual minority girls (eg, sexual identity development).^31,32  Content is tailored based on participants’ self-reported sexual experience (ever versus never had sex) and sexual identity (lesbian or gay versus all others except for girls who were queer. These girls were assigned a content “path” based upon their sexual attractions).

Example program messages can be found elsewhere.¹³  Most program messages are unidirectional (ie, were sent but did not ask for a reply). Bidirectional, interactive components include: (1) links to online videos that demonstrate and reinforce behavioral skills, (2) girls also have access to G2Genie, an on-demand automated advice channel^33,34 , (3) “leveling up” by answering a program message about the week’s content, and (4) being awarded “badges” for achieving behavioral skills that are tied to intervention outcomes of interest (eg, using condoms).

Each Girl2Girl participant is also paired with a “Text Buddy,” who is another participant in the intervention.^33,34  Messages encourage girls to talk about program content and provide support to each other. Youth are required to accept a Buddy Code of Conduct, which outlines acceptable (eg, positive support) and unacceptable (eg, bullying) behavior. Buddy messages are routed through the study server to protect participant phone numbers. All buddy conversations are monitored by research staff from 6 am to 10 pm Pacific Standard Time.

The control arm receives a similar number of messages for a similar number of days as the intervention arm. Content addresses nonsexual health topics relevant to adolescents (eg, exercise, bullying). To help blind this arm, girls in the control group receive 2 days of pregnancy prevention content so that they read messages aligned with the stated intended outcome (eg, pregnancy prevention). Interactive components (eg, Text Buddy) were not available to this group.

Given the lack of published data in this area, we relied upon unpublished pilot data to inform the power analysis. Assuming 80% power and statistical significance at P = .05, a sample size of 420 girls in each arm (840 total) would be sufficiently powered to detect differences in condom use. Recruitment was more successful than anticipated; 948 participants were recruited and randomized.

Main outcomes were assessed at 12 months; they also were collected at 3-, 6-, and 9-months. Outcome measures were adapted from those recommended by the Office of Adolescent Health³⁵  and included: (1) condom-protected penile-vaginal sex in the past 3 months (a count of number of events), (2) current contraception use, other than condoms (binary indicator of use), (3) not having penile-vaginal sex in the past 3 months (binary indicator of abstinence), and (4) pregnancy since program enrollment (binary indicator of pregnancy). Because of the 160-character limitation in text messages, questions in the text messaging-based intervention end survey varied from the online version. That said, rates of behavioral outcomes were similar whether girls completed the survey online or via text messaging.

The CONSORT table can be found in Figure 1. As a preliminary step, between-group differences in baseline socio-demographics, sexual history and contraceptive use were examined using t tests (for continuous variables), χ² tests (for categorical variables) and nonparametrics as appropriate. Socio-demographic and related variables that differed significantly between treatment and control groups were considered confounders in subsequent models if they were significantly associated with the outcome variable at a modest P < .10 level.

Using a series of longitudinal mixed effects models, we examined treatment group effects on primary outcomes. Specifically, outcomes at each follow-up were simultaneously regressed on group, follow-up time, group*follow-up time, as well as potential confounders of the group effect, including baseline variables that were statistically significant despite randomization. Models included a subject-specific intercept to account for repeated measurements within participants over time. As sexual experience may vary over time, it was included as a time-varying covariate in the model. Mixed models have the advantage of allowing for a wide range of outcomes distributions (including normal, negative binomial, and Bernoulli), as well as specifications of the time effect (linear, nonlinear). Standard errors are adjusted for the repeated measures within participant.

Analyses were intention-to-treat (N = 948). Models used a likelihood-based approach to estimation, making use of all available data without directly imputing missing data. This allows for regression estimates that are consistent, even with missing data.

All analyses were run in SAS 9.3 and significance level was set at 0.05 a priori. We were interested in estimating both group effects over time and measures of effect size (in this case, f²). A medium effect size would correspond to f² = 0.15.³⁶  Per CONSORT guidelines, we note that short-term outcomes¹³  and subgroup analyses of the 12-month outcomes are reported elsewhere (manuscript under review).

The sample consisted of 948 participants randomized to intervention (n = 473) or control (n = 475) at baseline. Participants were 16 years of age on average (SD = 1.2, Range: 14–18). Most (99%) identified as cisgender girls. About 1 in 25 (4%) had ever been pregnant. One in 3 (33%) had ever had vaginal-penile sex. At the end of the intervention, 69% had not had vaginal-penile sex in the past 90 days; at 12-month follow-up, 70% had not. Over half (56%) identified as white and 14% as Black race. A full description of the study sample, including sexual experience at baseline, is presented in Table 1. No adverse events were reported.

Assessment of between-group differences at baseline indicate that age was borderline significant, with intervention participants slightly older than control participants (16.1 years vs 16.0 years; P = .051). There were no other significant group differences at baseline.

Longitudinal models of protected sex events had a negative binomial distribution (over-dispersed count outcome) and a log link function. Results, shown in Table 2, indicate that Girl2Girl participants had higher rates of protected penile-vaginal sex events over time (f² = 0.11) compared with controls. Specifically, at 6 months postintervention, the incidence rate of protected sex events was 89% higher for intervention versus control (incident rate ratio [IRR] = 1.89, 95% confidence interval [CI]:1.07–1.3.34), adjusting for baseline rate of condom use, age, and a time-varying indicator of sexual experience. Similarly, at 9 months postintervention, the incidence rate of protected sex events was 16% higher for intervention versus control (IRR = 1.16, 95% CI:1.06–1.76), adjusting for baseline rate of condom use, age, and a time-varying indicator of sexual experience. At 12 months postintervention, this difference between groups increased to a 39% difference (IRR = 1.39, 95% CI: 1.09–2.11).

Longitudinal models examining use of birth control (other than condoms) assumed a Bernoulli distribution of the outcome variable and a logit link function. Results indicate that Girl2Girl participants were more likely than control participants to report use of birth control other than condoms (f² = 0.13). Specifically, results indicate an increase of 35% in the odds of birth control use other than condoms for Intervention versus Control at 6 months postintervention (odds ratio [OR] = 1.35, 95% CI:1.22–1.39), 14% at 9 months postintervention (OR = 1.14, 95% CI:1.02–1.34) and a difference of 7% at 12 months postintervention (OR = 1.07, 95% CI:1.006–1.26). All models adjusted for baseline rate of birth control use other than condoms, age, and a time-varying indicator of sexual experience.

Finally, models of abstinence and pregnancy rates did not suggest statistically significant group differences across time. However, effect sizes were in the small to medium range and point estimates favored Girl2Girl versus Control in both cases (f² = 0.09 and f² = 0.07 respectively).

To date, Girl2Girl is the only teen pregnancy prevention intervention of which we are aware targeted toward cisgender sexual minority girls that has been evaluated among ethnically diverse adolescents within a national RCT design. Results suggest that Girl2Girl has long-term impact on pregnancy preventive behaviors through 1 year. This is important not only because it is 1 of the few adolescent pregnancy programs to report results through 12 months postintervention, but also because it was delivered over text messaging. Although meta-analyses consistently find that mobile interventions are able to affect change,¹⁹  few report sustained changed – particularly to 1 year. That said, it is unclear if the dearth of similar literature is because of publication bias against null findings or reflective of funding and research designs that do not afford a longer-term observation period or adequate number of cases to ensure powered analyses. Both the field of teenage pregnancy prevention and mobile health interventions would benefit from more research on longer-term outcomes to facilitate a discussion of program designs that promote lasting behavior change.

With recent “stay at home” orders and closure of physical spaces, such as schools, because of the global coronavirus disease 2019 pandemic, the benefit of sexual health programming that goes to where youth are, literally in the palm of their hands, is even clearer. Additionally, this targeted approach may help fill existing gaps and improve sexual health equity for LGB+ young women. Likewise, an intervention that can be implemented confidentially, without requiring a stigmatized population to attend programming in person or publicly disclose their identity to participate, has the potential to engage marginalized young people who might otherwise feel unsafe or reluctant, for whatever reason, to participate. Given the positive results reported in this investigation, coupled with the reach and relatively low cost to deliver mobile health interventions, efforts to disseminate other health programming via text messaging to adolescents appears warranted.

The attenuation of intervention effect on the use of birth control over time aligns with the fact that youth were not receiving ongoing reinforcement for this behavior once the intervention ended. Although differences between the intervention and control group were still detectable at 12 months, the patterning suggests that a second booster at 6- or 9-months postintervention end might sustain the effects longer. In contrast, the rate of condom-protected sex acts for the intervention versus control was the highest at 12 months. This long-term increase in protected sex is a positive sign for supporting both unintended pregnancy and sexually transmitted infection prevention. The sustained behavior change suggests that the strategies used in the intervention, including messaging on where to find and purchase condoms and dental dams, as well as animated demonstrations of how to use barriers, and encouragement for Text Buddy conversations about using barriers, may together have provided the knowledge, skills, and reinforcement to establish the behavior and sustain it over time.

It is notable that 2 in 3 girls had not had penile-vaginal sex when they started the Girl2Girl RCT and yet, behavioral changes related to condom use and contraception were still detected. However, Girl2Girl did not demonstrate significant effects on abstinence compared with participants in the control condition, an expected finding given the sex-positive framing of messages. As noted above, the aim was not to delay the onset of sexual behaviors per se, but rather to encourage youth to wait to have sex until they felt ready to make that decision, and when they did so, to be sure that they used barrier methods. Pregnancy rates were also not significantly different between intervention and control groups over time, although the effect sizes at every time point were in favor of the intervention group. It is possible that this was an issue of power given that about 3% of girls in the study reported a pregnancy experience. Future studies that include more girls who are having penile-vaginal sex may provide more informative data in this regard.

Although the sample is not nationally representative, both the reach (national) and the size of the sample are unusual and provide greater generalizability than would a program tested in 1 specific area (eg, Los Angeles). Moreover, the inclusion of sexually inexperienced and experienced, and lesbian, bisexual, and queer girls of diverse racial and ethnic backgrounds, provides applicability to a wider range of sexual minority girls than would a more homogeneous sample. However, the current analyses did not examine whether the intervention was equally effective for all diverse groups. Moreover, rates of behaviors should not be interpreted as population-level prevalence estimates. Nonetheless, there is little reason to believe that selection bias would affect the intervention mechanism; the relative rates of behavior in intervention versus control groups are affected more by internal rather than external validity. Future work that examines how the intervention affects girls in a more “real world” setting that includes self-enrollment and no incentives, might further inform the intervention’s potential public health impact. The intervention was developed for girls who have a cell phone and unlimited text messaging plan. Although this reflects most youth today, it nonetheless excludes those who do not have a phone or lack privacy in their technology use to safely engage with intervention messages. It may also have skewed the sample to youth living in higher socioeconomic status environments. The intervention included only cisgender sexual minority girls, yet transgender and nonbinary adolescents assigned female at birth also can get pregnant. Future research should consider adaptations to include gender minority youth. More generally, future research could additionally include whether usage of intervention features (eg, text buddy) was related to behavior change and whether the update of specific types of birth control were more easily affected and sustained than other types.

The current study builds upon previously reported short-term outcomes and suggests that Girl2Girl was associated with long-term sexual behavioral change among sexual minority, cisgender adolescent girls across the United States. With so few interventions targeted to the reproductive and sexual health needs of sexual minority youth, Girl2Girl offers a promising option for addressing this gap. Future studies might consider for whom and under what circumstances the intervention has greater impact and consider adapting Girl2Girl for both gender minority youth and youth living in other countries.

LGB+

lesbian, gay, bisexual, and other sexual minority girls