The methods reported in this section were first published in the trial protocol [39] and the statistical analysis plan [40].This was a parallel group, individually randomized superiority trial with a 1:1 allocation ratio that evaluated the effect of the intervention delivered by CIES’s app. Women were eligible to take part if they were aged 16 to 24 years, owned a personal Android mobile phone, lived in La Paz or El Alto, reported an unmet need for contraception (ie, are sexually active, not using effective contraception, and want to avoid a pregnancy), and could read Spanish. Participants must have been willing to download the app and receive messages about contraception on their mobile phone. The trial was promoted through CIES’s service delivery points in La Paz and El Alto, the CIES website, flyers distributed through CIES’s youth network, and social media sites. Participants provided informed consent through the secure web-based trial database and randomization system.

The trial was granted ethical approval by LSHTM Interventions Research Ethics Committee on May 16, 2016, and by La Comisión de Ética de la Investigación del Comité Nacional de Bioética on September 20, 2016.

The intervention was developed with young Bolivian people in 2015 to 2016, guided by a systematic protocol for the development of behavior change interventions [36]. The development process involved the following steps: (1) needs assessment (activities included establishing a project planning group, a literature search, focus group discussions, and interviews with the target group and interviews with local service providers), (2) specifying behavioral change to result from the intervention, (3) designing the intervention components by selecting behavior change methods, and (4) producing and refining the intervention content. The needs assessment revealed that young people in El Alto and La Paz were eager to receive information about contraception on their phone, lacked comprehensive knowledge about contraception, and expressed a range of negative beliefs about effective methods. The intervention messages were tested with young people and refined after each consultation in an iterative process until the context was acceptable to them.

The intervention provided accurate information about contraception, targeted the beliefs identified in the development phase that influence contraceptive use (eg, specific misconceptions about the side effects and health risks of contraception), and aimed to support young women in believing that they can influence their reproductive health. The messages contained 10 behavior change methods [41], adapted for delivery by mobile phone: belief selection, facilitation, anticipated regret, guided practice, verbal persuasion, tailoring, cultural similarity, arguments, shifting perspective, and goal setting. Participants allocated to the intervention group received 0 to 3 messages per day (a total of 183 messages) for 120 days. Please see the protocol [39] and the intervention development publication [36] for a more detailed description of the intervention.

The Tú decides app itself contained standard family planning information and no behavior change methods. Participants allocated to the intervention arm had access to the app and the intervention instant messages. Participants allocated to the control arm had access to the Tú decides app and 7 control instant messages about the importance of their participation and reminding them to contact the project coordinator if they change their number (which intervention participants also received). All participants who received usual care were free to seek any other support, whether existing or new.

After providing informed consent, participants completed the baseline questionnaire through the database and randomization system. The allocation sequence was generated by the remote computer-based randomization software. Randomization occurred immediately after baseline data were submitted. All participants downloaded the app immediately after they submitted their baseline data. The messages commenced within 24 hours after participants downloaded the app.

Owing to the nature of the intervention, participants would have been aware of the allocation soon after they started receiving the messages. Local research staff collecting outcome data were masked to allocation unless the participant revealed it to them. Researchers who analyzed the data were masked to treatment allocation.

Data were collected at baseline and at 4 months post randomization using questionnaires. At baseline, we collected personal and demographic data and the coprimary outcome acceptability. At follow-up, we collected all outcomes and the following: if participants report using an effective method, where they obtained it; current pregnancy intention; whether they knew someone else who had also participated in the study and, if so, if they read each other’s messages; and to assess potential adverse outcomes, we asked participants if they have experienced physical violence since being in the study and if anything good or bad happened as a result of receiving the messages. An instant message that included a link to the database to complete the follow-up questionnaire was sent to all participants through the app 4 months after downloading the app. If participants did not complete the follow-up questionnaire themselves, local research staff unaware of participants’ allocation contacted them by telephone to collect their data.

The trial was powered to detect a 10% absolute difference in the use of effective contraception between the intervention and control groups at 4 months. The 10% difference was based on a trial evaluating a postabortion mobile phone intervention, which found that 18% more women in the intervention arm than those in the control arm were using effective contraception at 4 months (64% vs 46%) [27]. We assumed that our trial would observe a smaller increase in contraceptive uptake, as it does not specifically involve women who had just had an abortion, who had already accessed services, and who may also have a greater intention to use contraception compared with the women in our trial. Therefore, we powered our trial to detect a smaller absolute difference of 10% uptake in effective contraception at 4 months.

The best estimate at the time of designing the trial was that the proportion of women aged 16 to 24 years in a partnership living in La Paz or El Alto using effective contraception was around 44% [44]. In total, 1048 participants provided 90% power to detect a 10% absolute difference, at the 5% significance level, assuming 44% use in the control group versus 54% in the intervention group (corresponding to an odds ratio [OR] of 1.49). Allowing for 20% loss to follow-up, the sample size was 1310.

The trial protocol was accepted for publication on November 3, 2017 [39], and the detailed statistical analysis plan was publicly released on November 7, 2017 [40]. Analyses were according to the randomized arm, and only participants with complete outcome data were included in the principal analysis. All statistical tests were two sided and were considered significant at the 5% level. The analysis was conducted using Stata 15 (StataCorp). Unmasking occurred on February 6, 2018, after the analyses outlined within the analysis plan were complete on masked data.

We used a chi-square test to investigate the evidence for whether losses to follow-up differed by trial arm. We used logistic regression to compare the baseline characteristics of participants who completed follow-up with those who did not. We investigated whether predictors of loss to follow-up differed by trial arm by testing for an interaction.

Between March 1, 2017, and July 29, 2017, there were 645 randomizations by the system. Follow-up ended on February 8, 2018. During the trial follow-up, we discovered that 3 participants enrolled and were randomized twice. For the 1 participant who was allocated to the same arm on both randomizations, we kept this participant in the analysis using the baseline data from their first record. For the 2 participants who were allocated to different arms, we excluded them from the analysis. This resulted in 640 participants included in the trial.

Of the 640 participants, 321 were allocated to the intervention arm and 319 participants were allocated to the control arm (Figure 1).

The baseline characteristics of trial participants are reported in Table 1. Mean age was 20 years. In addition, 90.4% (579/640) of the participants did not have children, and only 8% (26/640) of the participants found at least one method of effective contraception acceptable. Acceptability was very low for the individual methods (1%-3%). Characteristics were similar between the 2 groups.

Overall, 67.0% (429/640) participants completed the trial follow-up for the coprimary outcome use (control, n=215 and intervention, n=214; Figure 1), and 63.4% (406/640) participants completed the follow-up for the coprimary outcome acceptability (control, n=203 and intervention, n=203). Retention did not differ between the arms (215/319, 67.4% in the control arm and 214/321, 66.7% in the intervention arm; P=.84). Among the participants who completed the use coprimary outcome, the strongest predictor of retention was being aged 20 to 24 years (OR 1.33, 95% CI 0.96 to 1.86; P=.09). There was some evidence that the effect of this predictor differed by arm (interaction test P=.09). Detailed characteristics of participants who completed follow-up and those who did not are reported in Multimedia Appendix 1.

In the intervention arm, 37.4% (80/214) of participants reported the use of effective contraception compared with 33.5% (72/215) of participants in the control arm (Table 2). There was no evidence of a difference in use between the groups (crude OR 1.19, 95% CI 0.80 to 1.76; P=.40 and adjusted OR 1.19, 95% CI 0.80 to 1.77; P=.40).

There were no significant differences in any of the secondary outcomes between the groups (Table 3).

In the intervention arm, 71.9% (146/203) of the participants reported that at least one method of contraception was acceptable compared with 62.6% (127/203) of participants in the control arm (Table 2). There was borderline evidence of a difference in acceptability between the groups (crude OR 1.53, 95% CI 1.01 to 2.33; P=.05 and adjusted OR 1.49, 95% CI 0.98 to 2.28; P=.06).

When we considered that participants who were lost to follow-up did not use an effective method or find an effective method acceptable, the effects observed in the principal analysis were reduced (use: OR 1.14, 95% CI 0.79-1.64; P=.48 and acceptability: OR 1.26, 95% CI 0.92-1.74; P=.15).

The strongest predictor of retention was being aged 20 to 24 years. Age was a baseline covariate, so the model in the second sensitivity analysis (adjusting for the main baseline predictors of missingness) is the same as the primary analysis model.

There was no evidence that the effect of the intervention differed within the different levels of the subgroups (Figures 2 and 3).

The observed difference in contraceptive use between the groups (absolute difference=3.89%) was smaller than expected. Although the use was higher in the intervention group, the difference was not statistically significant. There was a borderline significant effect regarding the acceptability coprimary outcome, which favored the intervention group. No statistically significant differences between the groups in any of the secondary or process outcomes were observed. The intervention dose received was low, based on participants’ report.

The main limitations of this trial were that we did not recruit to target and achieved less than 80% follow-up completion. Although effect estimates or both primary outcomes favored sending the intervention messages with the Tú decides app, the differences between the groups were smaller than what we expected. The recruitment and follow-up challenges meant that the trial was underpowered and therefore unable to produce unequivocal estimates regarding the effect of sending the intervention messages in addition to the app (with 429 participants, the trial had 54% power to detect a 10% absolute difference in use of effective contraception between the groups). Another limitation is in relation to the self-reported outcome measures. As they are self-reported, they are more likely to be biased than if the outcomes were objective, such as clinic-verified use of contraception.

Despite the limitations, this study had several strengths. Our trial database and randomization system generated and concealed the allocated and achieved well-balanced groups. There was no evidence that the intervention was associated with an increase in self-reported violence, a potential adverse outcome related to others viewing the messages, unwanted by the participant. We considered this a potential adverse outcome, given the stigma associated with sex before marriage in Bolivia [36]. However, we cannot determine the effect of the app on partner violence because both groups had access to it. Despite this, it is reassuring that the self-reported prevalence in this trial was low (2%).

Trials evaluating contraceptive behavioral interventions delivered by mobile phone have had mixed results [25-35], with some showing an improvement in contraceptive use [25-27] and knowledge [25,28-30]. The results of this trial are not inconsistent with our trials of similar interventions among young people in Tajikistan [50] and Palestine [51]. In the Palestine trial, participants who received the intervention were more than twice as likely to find at least one method of contraception acceptable (OR 2.34, 95% CI 1.48-3.68; P<.001). There were also improvements in knowledge, acceptability of individual methods, perceived norms about friends using contraception, and intention to use contraception compared with the control group. In the Tajikistan trial, there was contamination between the intervention and control groups, and no differences were found between the groups. As with this Bolivian trial, the Tajik and Palestinian trials also did not suggest that the intervention was associated with an increase in violence.

The uncertainty regarding the efficacy of the intervention means that we cannot strongly recommend implementation in Bolivia. The low dose of the intervention is likely to have reduced the effect estimates. Only 25.7% (53/206) of the participants reported that they read all or most of the intervention messages, with 28.6% (59/206) reporting that they read none of the messages. This could have contributed to the smaller than anticipated observed differences. As we did not collect data on why these people did not read most of the messages, we do not know the reasons for this. A total of 47% (97/206) of intervention participants who answered the question “Did anything good or bad happen during the study?” (97/321, 30.2% of all intervention participants and 97/214, 45.3% of those who completed follow-up) reported something positive about the messages. Although participants in the development work were positive about the message content, sending the messages to the target group as they are intended to be sent (eg, over the entire 4 months) and then interviewing participants about their experience may have identified barriers to successful intervention receipt. We have done this in previous studies, but our timeline and resources did not allow for this in this trial.

Another possibility for the smaller than anticipated differences could be that the intervention is only moderately more effective than offering standard family planning information on a mobile app. It may be that offering good-quality family planning information on mobile app pages in this context is sufficient enough to improve the use of and attitudes toward contraception. However, the results suggest that the intervention messages could increase the acceptability of effective contraception if they were offered alongside the download of the Tú decides app and would not cause harm if done so.

It is difficult to determine exactly why we were unable to recruit to target. Some factors that likely contributed to the under-recruitment are (1) the trial promotion was not targeted to eligible people as much as it could have been, (2) young women were reticent to admit to being sexually active, (3) they did not want to have the app on their phone (although the development work indicated that this would not be a problem for them), and (4) they were not interested in taking part in a trial. Although the target sample size was much lower in the Tajik and Palestine trials (n=570 for each), this Bolivian trial actually recruited more participants and had a narrower inclusion criterion (women aged 16-24 years with an unmet need for contraception). Potential ways to improve recruitment in future trials of contraceptive behavioral interventions delivered by a mobile phone could be to promote the trial in settings where recruiters have a very good chance of accessing eligible people. In addition, only assessing potential participants on their own may improve the chance that they would admit to being sexually active.

In deciding how to analyze the scales, we thought it would be better to avoid false positives. For participants to score acceptable for a method, they must have chosen agree or strongly agree to the positively worded stems and disagree or strongly disagree to the negatively worded stems. To illustrate how stringent this definition of acceptability is, a participant could respond strongly agree to the statement that the method has unwanted side effects but could still use the method because the benefit of using it (avoiding a pregnancy) outweighs the risk [51]. This does not have implications on the effect of the intervention relative to the control. Nevertheless, the very low baseline acceptability of all methods highlights the need for interventions such as the one evaluated in this trial—interventions that provide accurate, nonjudgmental information about contraception and that address negative beliefs and misconceptions.

This trial was unable to determine unequivocally if the intervention was effective at increasing the use and acceptability of effective contraception among young women with an unmet need in Bolivia. The intervention messages when delivered in addition to an app providing standard family planning information may moderately improve acceptability. Future research could first identify why around one-third of participants did not read the intervention messages and then evaluate the effect of the intervention on the use of contraception in different contexts.