Introduction

Adolescents and young adults aged 15‐24 years, defined as “youth” by the United Nations [-], are disproportionately affected by HIV around the globe. This age range is widely used in international health research and reporting, which allows comparability across studies and alignment with global HIV surveillance data. Alarmingly, in 2023, youth in this age group comprised nearly one-third of the 3600 daily new HIV infections recorded worldwide. Youth are especially vulnerable to HIV due to high rates of unprotected sex, inconsistent condom use, and co-occurring risk behaviors such as alcohol and drug use [,]. Still, a significant proportion of youth around the world lack access to accurate and age-appropriate information on sexual and reproductive health, rendering them susceptible to misinformation, psychological distress, and engagement in high-risk sexual behaviors []. To address this public crisis, scalable evidence-based health interventions targeting safer sex practices should be prioritized in this vulnerable population [].

Digital health interventions (DHIs) have emerged as a promising strategy for health promotion in recent years []. Digital health, conceptualized as an umbrella term by the World Health Organization [], refers to the use of digital and wireless platforms to facilitate health care delivery or health interventions, including but not limited to electronic health, mobile health, telehealth, and artificial intelligence-based applications. On the other hand, with the growing accessibility of smartphones and internet services among youth, digital technologies have become a dominant force to shape their sexual behaviors [,]. It is more convenient for young people to meet sexual partners, including casual, one-night, and anonymous partners, through web-based platforms, dating apps, and social networking sites [,], which further increases the likelihood of having unprotected sex frequency []. While digital technologies have facilitated riskier sexual behaviors among youth, they also create opportunities for DHIs that leverage young people’s existing online engagement patterns and preferences [-].

Accumulating evidence suggests that DHIs can improve HIV-related knowledge, risk perception, prevention intentions, and behavioral outcomes among youth [], with the types of DHIs including mobile apps, text messaging, online videos, social media platforms, and interactive websites [-]. A stage-based computer-delivered intervention, for example, targeting heterosexual young men demonstrated significant improvements in condom use intention and subsequent condom use behavior []. Similarly, a study evaluating a social media-based intervention via Facebook reported a 23% increase in condom use and a 54% reduction in chlamydia incidence among adolescents []. In contrast, a large (randomized controlled trial (RCT) delivering sexual health promotion via SMS and email enhanced sexually transmitted infection (STI) knowledge and testing uptake, particularly among women, but showed no significant impact on condom use []. Another study reported that intervention based on a peer-led safer-sex Facebook group for Chinese college students found no significant change in contraceptive use intention or frequency []. Similarly, a social media-based crowdsourced HIV testing intervention among youth did not increase facility-based HIV testing, condom use, or syphilis testing []. Therefore, different types of DHIs may differentially affect sexual health outcomes, yet existing trials rarely distinguish the relative effectiveness of each DHI modality. Clarifying which intervention types are most effective for specific behavioral and biological outcomes is essential for optimizing digital HIV prevention strategies among youth [,].

In addition, several systematic reviews (SRs) have synthesized the evidence on DHIs for HIV prevention, but important limitations remain. Some SRs are purely descriptive, lacking quantitative synthesis [,-]. Other reviews have focused narrowly on specific DHI types (eg, social media or telehealth) [,], or have failed to examine key behavioral outcomes like condom use [,]. In addition, traditional meta-analyses are constrained to pairwise comparisons [], leaving uncertainty about which types of DHIs are most effective in head-to-head comparisons []. To address these knowledge gaps, we conducted a SR and network meta-analysis (NMA) to evaluate and compare the effectiveness of different DHIs in promoting safer sex behaviors among youth. The study aimed to: (1) identify the most effective types of DHIs in promoting safe sex among youth; (2) construct a network-based ranking of intervention effectiveness; and (3) inform the design of scalable, evidence-based digital health programs for HIV prevention among youth.

MethodsOverview

This SR and NMA follow the Preferred Reporting Items for Systematic reviews and Meta-Analyses (PRISMA-NMA) guidelines []. The completed PRISMA-NMA checklist is provided in . The protocol for this study has been registered with PROSPERO (CRD42024527317).

Eligibility CriteriaTypes of Population

Studies were eligible for inclusion if they involved participants aged 15‐24 years or if at least half of the participants were within this age range. Those that did not report, or for which data could not be extracted, for this specific age group were excluded.

Types of Interventions and Comparison

The interventions included in this review were DHIs, which were defined in accordance with the World Health Organization’s broad definition of digital health technologies []. For the purpose of this review, the included DHIs were further classified into four mutually exclusive categories based on their delivery modes and characteristics: (1) mobile app-based interventions (MAIs), (2) telecommunication-based interventions (TCI), (3) static web-based interventions (SWIs), and (4) interactive online-based interventions. This operational classification was developed to reflect the interventions identified in the included studies and to avoid overlap across categories. A detailed description of common DHI subcategories was provided in ; the subcategories of the DHIs were based on a previous SR []. The control group received nondigital interventions (NDI), referring to traditional approaches without digital technology, such as face-to-face counseling, printed materials, or group sessions.

Table 1. Subcategories of digital health interventions (DHIs) and abbreviations used to classify interventions evaluated.Subcategories of DHIsAbbreviationDescriptionMobile app–based interventionsMAIPrograms delivered primarily via dedicated software apps installed on smartphones or tablets, leveraging device features (eg, notifications, sensors, data storage) to provide interactive content, personalized feedback, tracking, and behavior change support.Telecommunication-based interventionTCIInterventions using traditional telecommunication methods such as SMS text messages or telephone calls. These interventions typically involve sending reminders, educational messages, or conducting counseling via phone communication without the need for internet-based platforms.Static web-based interventionsSWIInterventions provided through websites that offer static, noninteractive content. This may include informational pages, downloadable resources, or educational materials without features for user engagement or real-time feedback.Interactive online-based interventionsIOIInterventions delivered via web-based platforms or websites that enable user interaction, such as quizzes, tailored feedback, chatbots, or real-time communication with health professionals. These platforms actively engage users to enhance learning and behavior change.

Any appropriate comparator group was included, such as usual care, placebo, no intervention, waitlist, attention control, or different DHIs. To reduce inconsistency among trials, we excluded studies that combined non-DHIs with DHIs, unless the distinction between the intervention and control groups lay solely in the DHIs. In multi-arm trials, intervention arms representing the same modality without meaningful differences in content, intensity, or delivery were considered a single treatment node for eligibility purposes and later combined analytically to avoid double-counting.

Outcomes

The primary outcomes were specific condom use behaviors, defined as follows:

Condom Use Rate in the Last Sexual Contact: The percentage of individuals reporting condom use during their most recent penetrative sexual act [].Consistent Condom Use Rate: The percentage of individuals reporting consistent condom use during all their penetrative sexual acts over the recall period specified in each study [].Proportion of Condom Use: The overall proportion of sexual acts in which a condom was used, calculated as the total number of times a condom was used divided by the total number of sexual acts. Unlike the consistent condom use rate, which measures whether individuals always use a condom, this indicator captures the frequency of condom use across all reported sexual encounters, allowing for partial or occasional use [].

The secondary outcomes included (1) self-efficacy for condom use, measured by the overall mean score on a validated condom use self-efficacy scale, such as Brafford and Beck’s [] condom use self-efficacy scale, Lawrance et al’s [] self-efficacy for HIV prevention scale and others, (2) number of sexual partners, and (3) the incidence rate of STIs (including HIV). Because follow-up length varied substantially across trials, we included studies reporting at least one postintervention follow-up outcome and extracted the longest follow-up time point for synthesis to enhance comparability [,].

Types of Studies

Only RCTs were included, including crossover trials and cluster-randomized trials. Studies using nonrandomized, quasi-experimental, observational, or qualitative designs were excluded. Only peer-reviewed articles published in English were eligible, as non-English or non-peer-reviewed sources lack sufficient methodological detail for reliable data extraction and risk-of-bias assessment.

Search Strategy

The search strategy was developed and reported in accordance with the PRISMA-S guideline []. Searches for RCTs were conducted in PubMed (including MEDLINE), EMBASE, Web of Science, and the Cochrane Library. Searches were performed through the native interfaces of each database (PubMed via NCBI, EMBASE via Elsevier, Web of Science via Clarivate, and Cochrane Library via Wiley). In addition, the reference lists of relevant SRs were checked to ensure that no eligible trials were missed.

The search terms were formulated according to the PICOs framework, including participants or populations, interventions, outcomes, and types of research design. Both Medical Subject Headings and free-text terms were included as appropriate. Boolean operators (“AND,” “OR”) were used to combine search terms, and database-specific search techniques such as truncation, phrase marks, and wildcards were applied. The complete search terms and algorithm were provided in , and search strategies for the other databases were adapted accordingly. The search was designed and executed by 2 reviewers (YZ and WP), who were trained in SR methodology, and the strategy was cross-checked for completeness and accuracy.

The literature search was initially conducted on June 13, 2024 and was last updated on November 15, 2025. All retrieved records were imported into EndNote X9 for citation management, and duplicates were removed using both automated and manual deduplication. Additional search methods included checking the reference lists of the included studies or relevant SRs. Gray literature was also searched via Google Scholar, OpenGrey, and ProQuest Dissertations. The search was limited to studies published in English due to resource constraints for translation.

Selection Procedure and Data Extraction

Two reviewers (YZ and DH) independently screened the titles and abstracts against predefined protocol criteria. Full texts were retrieved for all potentially eligible studies. When multiple articles were identified from the same randomized controlled trial, the most recent or most comprehensive publication was retained for data extraction. Earlier reports were used to supplement missing information on study design, intervention details, or outcomes when necessary. Any discrepancies between the 2 reviewers were resolved by discussion. If disagreements persisted, a third reviewer (WP) was invited for adjudication. At the title and abstract screening stage, we excluded 14,788 records that clearly did not meet the eligibility criteria, most commonly because of wrong study design (eg, cross-sectional surveys, qualitative studies, reviews, protocols), wrong population (non-youth samples), ineligible intervention or comparator (ie, the difference between study arms did not lie in the use of a DHI), or an unrelated topic.

Data were extracted using a standardized and piloted form. Extracted variables included: first author, publication year, recruitment region, participant characteristics (mean age, SD, gender distribution), type of intervention and comparator, sample size per arm, intervention duration, and reported endpoints. Detailed characteristics of the DHIs were also extracted to facilitate subcategorization (). Further, outcomes and corresponding measurement methods were recorded, such as self-reported condom use and validated self-efficacy scales.

When outcome data were incomplete or unclear, study authors were contacted by email for clarification; trials with essential missing data were excluded from the quantitative synthesis and documented in . All eligible studies were included in the SR, and only studies with usable and connected outcome data were included in the NMA.

Risk of Bias Assessment

The methodological quality of the included studies was independently assessed by 2 reviewers (YZ and WP), with disagreements resolved by a third reviewer (CZ). Risk of bias was evaluated using version 2 of the Cochrane risk of bias 2 tool (RoB 2) for randomized trials []. For each domain, studies were rated as having “low risk,” “some concerns,” or “high risk” of bias according to the Cochrane Handbook (version 6.5) []. Domain-level risk of bias judgments for each trial are summarized in , with extended graphs and contribution matrices provided in . The reference list of included studies is provided in [,-].

‎Figure 1. Risk of bias assessment of included randomized controlled trials using Cochrane Risk of Bias 2 tool [,-].

Because blinding was generally not feasible for these nonpharmacological interventions, many trials were judged at high risk of bias in the domain of deviations from intended interventions [,]. As this limitation was expected and unlikely to influence objectively measured outcomes, we did not consider this domain when grading the certainty of evidence. The overall certainty of evidence was determined using the CINeMA (Confidence in NMA) web application, which is based on the GRADE framework [,]. In addition, we constructed GRADE “Summary of Findings” tables using the official template provided by the GRADE Working Group to summarize the key relative and absolute effects and certainty ratings for the primary outcomes ().

RoB 2 assessments were incorporated into the interpretation of NMA findings and into the GRADE/CINeMA evaluation of certainty, but they were not used to weight studies in the statistical synthesis.

Statistical AnalysisGeometry of the Evidence Network

We examined the geometry of the treatment network by mapping each trial arm to one of the predefined intervention nodes and summarizing the pattern of direct comparisons. A network plot was generated to visually depict the evidence base, with node size proportional to the number of randomized participants and edge thickness reflecting the number of trials informing each comparison. We further assessed potential network-related biases by identifying sparse nodes, single-study comparisons, and imbalance in the distribution of direct evidence.

Model Specification and Synthesis Methods

Model convergence was assessed through Markov Chain Monte Carlo diagnostics, including the Gelman-Rubin potential scale reduction factor and inspection of leverage plots. The number of adaptation iterations, burn-in period, and total iterations were set to ensure adequate mixing and convergence. Effect estimates were expressed as pooled odds ratios (ORs) and 95% credible intervals (CrIs), which served as the primary summary measure for all dichotomous outcomes.

Bayesian NMAs were conducted using the R package BUGSnet to compare the effectiveness of 4 subcategories of DHIs and control groups. Binomial likelihood models with a logit link function were specified, and both fixed-effect and random-effects consistency models were fitted. Given anticipated clinical heterogeneity, random-effects models were treated as primary, with fixed-effect models used in sensitivity analyses. Noninformative priors were assigned to treatment effects and heterogeneity parameters to minimize prior influence. Model fit and parsimony were evaluated using the deviance information criterion (DIC), with lower values indicating better fit.

To evaluate the transitivity assumption, we compared mean age, sex distribution, intervention intensity, and follow-up duration across treatment comparisons. We further restricted inclusion to trials in which ≥50% of participants were aged 15‐24 years, excluded trials in which nondigital components were offered only to one arm, and extracted outcomes at the longest reported follow-up to harmonize follow-up time. These design and population characteristics showed broadly overlapping ranges across interventions and no systematic differences between comparisons, so transitivity was judged plausible. Forest plots of posterior ORs with 95% CrIs from the Bayesian consistency model were generated to summarize the magnitude and uncertainty of estimated treatment effects.

To complement these Bayesian estimates and quantify uncertainty in effects that might be observed in new settings, we also performed frequentist random-effects NMAs using the netmeta package in R, specifying NDI as the reference group []. For each outcome, we estimated ORs and 95% CIs for each intervention versus NDI and derived 95% prediction intervals (PIs) by combining the average treatment effect with between-study heterogeneity, in line with recent recommendations that NMAs should routinely report PIs when heterogeneity is present [].

Assessment of Inconsistency and Heterogeneity

Consistency between direct and indirect evidence was assessed by comparing the DIC between consistency and inconsistency models. A substantially lower DIC in the consistency model indicated acceptable agreement between sources of evidence. Due to the limited number of included studies, we did not formally investigate small-study effects (eg, using funnel plots or Egger’s regression test), which are typically used to explore potential publication bias as one of several possible explanations for such effects. Selective outcome reporting could not be formally assessed due to insufficient reporting in the included trials; however, the potential for selective reporting was considered when interpreting the cumulative evidence. Because the number of studies informing most comparisons was limited, local inconsistency (eg, node-splitting) could not be reliably assessed; in the presence of any potential inconsistency, we planned to explore differences in study characteristics and reassess the plausibility of the transitivity assumption.

Handling of Multi-Arm Trials and Node Merging

For multi-arm trials, if 2 or more arms delivered essentially the same intervention category (eg, different versions of the same SWI content without meaningful variation in delivery or timing), we merged these arms by summing the number of events and participants. This ensured that each intervention was represented by a single node in the network and avoided duplicate contributions from the same trial [].

Ranking of Interventions

Ranking probabilities and surface under the cumulative ranking curve (SUCRA) were computed to summarize the relative effectiveness of each intervention across the posterior distribution. Rankograms and cumulative ranking plots were used to visualize intervention hierarchies, and league tables and heatmaps were generated to present pairwise comparisons and their relative effect estimates. No additional analyses, such as sensitivity analyses, subgroup analyses, or meta-regression, were conducted because the limited number of studies and the sparse network geometry did not permit reliable implementation of these methods. All statistical analyses were conducted using R (version 4.3.2; R Foundation for Statistical Computing) with the gemtc, BUGSnet, and netmeta packages.

ResultsDescription of Included Studies

From a total of 25,659 records initially retrieved, 24 RCTs published between 2004 and 2024 were included in the final analysis (). These studies were conducted across 8 diverse countries, predominantly in the United States (n=14), with the remainder from China (n=2), the United Kingdom (n=2), Uganda (n=2), Singapore (n=1), the Netherlands (n=1), Australia (n=1), and Spain (n=1). The trials collectively enrolled 20,134 participants (range 50‐6248; mean 838.9, SD 1358.2), with a mean age of 19.5 years. Overall, 10,228 participants (53.4%) were male, although sex composition varied substantially—some studies enrolled only males [-], only females [,], or mixed populations. Of the 24 included studies, 21 studies were two-arm, and 3 studies were multi-arm. Across the included studies, 6 trials evaluated TCI, 8 assessed interactive online-based intervention (IOI), 6 examined MAI, and 8 investigated SWI. The total number of intervention approaches (n=28) exceeded the number of included studies (n=24) because several trials directly compared 2 or more active interventions (eg, IOI vs SWI) without including a conventional control group. Intervention durations ranged from brief sessions lasting 10‐20 minutes up to 12 months. Follow-up periods were heterogeneous, spanning from immediate postintervention assessments to 24 months. Most studies reported outcomes at 3‐6 months, while only a few provided longer-term follow-up beyond 12 months. Five studies performed analyses for more than one time point. To enhance consistency and reduce potential bias associated with short-term variability, we extracted outcomes at the longest follow-up time point, thereby facilitating a more comprehensive evaluation of the intervention’s long-term effectiveness [,].

‎Figure 2. The flow diagram of the literatureease clarify th selection process for randomized controlled trials of digital health interventions included in this review.

For condom use at last sexual contact, 4 out of 7 studies reported ORs greater than 1, suggesting a possible beneficial effect of the interventions; however, only one trial [] showed a clear statistical significance (OR 1.13, 95% CI 1.01‐1.25). Regarding consistent condom use, 6 of 11 studies showed ORs above 1, with substantial heterogeneity; one study reported a very large effect (OR 4.55, 95% CI 1.15‐17.95) []. For the proportion of condom use, 5 out of 6 trials reported ORs above 1, suggesting a tendency towards higher condom use in the intervention groups; however, CIs were wide and often included the null (overall OR range 0.48‐2.43), indicating that the evidence for this outcome is imprecise. Finally, for STIs incidence, including HIV, effects varied substantially across 7 trials, with ORs ranging from 0.53 to 2.10. Four of the 7 trials had point estimates below 1 [,,,], and 3 trials had 95% CI that excluded 1 [,,], indicating heterogeneous and partly conflicting evidence for this outcome. These results summarized the observed effects across trials, highlighting that effect estimates varied considerably across outcomes and studies. The characteristics and effect estimates of the included studies were summarized in .

Table 2. Study characteristics and effect estimates of the included trials.Study (author, year)Region / countryStudy designAge (years)Males, n (%)Intervention nameTheoretical frameworkDelivery mode or digital toolsIntervention / comparatorSample size (intervention / comparator)Intervention dosage / Intervention durationFollow-upEndpointsEffect estimates (OR, 95% CI)Mean (SD)MedianHu et al []ChinaRCT16.09 (0.84)—1691 (53.67)“You and Me—Online education sessions; Internet-based educational platform; cartoon videos; PowerPoint slidesIOI / NDI1760/139145 min sessions/8 wkEnd of intervention / 12 wk①, ④0.72 (0.31‐1.69) ①
1.42 (1.16‐1.74) ④
Tan et al []SingaporeRCT23.90 (2.98)—300 (100.00)People Like Us (PLU) web drama video seriesPredetermined theory of behavior changeWeb-basedSWI / NDI150/1506 videos, each about 10 min in length/1 wk6 mo②, ④1.22 (0.59‐2.51) ②
0.82 (0.40‐1.70) ④
Nuwamanya et al []UgandaRCT21.00 (2.00)—407 (36.60)MPA-SRH (mobile phone apps-sexual reproductive health）—Mobile appMAI / NDI556/556Within 6 mo, open for use to the intervention group.End of intervention①Wilson et al []The United KingdomRCT23.00 (3.55)—846 (41.01)SH:24 website—Website, online serviceSWI / NDI1031/1032All participants were free to use any other sexual health services or interventions during the trial period/Median =28.8 d (participant-dependent duration)6 wk④Lau et al []ChinaRCT——396 (100.00)Online intervention based on STD-related cognitions involving videos (SC)/online intervention based on both STD-related cognitions and emotions (eg, fear) involving fear-arousing imagery and videos (SCFI)—Online videos (Sent Intervention package email）SWI / NDI261/13510‐20 min3 mo③Bannink et al []NetherlandsRCT15.81 (0.68)—446 (54.00)The E-health4Uth Intervention—Web-based tailored messagesIOI / NDI392/4341 mo4 mo②Lim et al []AustraliaRCT—19417 (41.99)Email and SMS to a group of young people (intervention gro—SMS, emailTCI / NDI507/486SMS messages were sent every 3‐4 wk (a total of 14) emails were sent less than monthly (a total of 8)/12 moend of intervention (6 mo) / end of intervention (12 mo)②Rotheram-Borus et al []The United StatesRCT—23136 (77.71)Telephone intervention—Telephone sessionsTCI / NDI59/1162 h/6 wk15 mo②Ballester-Arnal et al []SpainRCT20.90 (1.90)—60 (25.00)Fear induction group/website groupInformation-Motivation-Behavioral Skills (IMB) ModelVideo, music/website; computer basedSWI / NDIEstimated 34‐35 participants per group (total n=239; balanced allocation across 7 groups)1 h1 months / 4 mo③Free et al []The United KingdomParallel group RCT20.35 (2.1)—2162 (34.60)text messaging intervention (safetxt)COM-B (capability, opportunity, motivation, and behavior) modelDelivered by text messages to improve safer sex behaviorsTCI / NDI3123/3125Day 1‐3: 4 items per day Day 4‐28: 1‐2 items per day Second month: 2‐3 items per week Month 3‐12: 2‐5 items per month/12 mo4 weeks / 12 mo①, ④1.13 (1.01‐1.25) ①
1.11 (0.98‐1.25) ④
Santa Maria et al []The United StatesPilot RCT21.20 (2.10)—56 (52.34)MY-RID (Motivating Youth to Reduce Infection and Disconnection)Information-Motivation-Behavioral Skills (IMB) ModelSmartphone-based Just-in-Time Adaptive Intervention (JITAI) Android smartphones with unlimited data were provided to participantsMAI / NDI48/49Participants will receive a customized message for each EMA assessment completed.• Weeks 1‐2: 3 times per day• Weeks 3‐4: 2 times per day• Weeks 5‐6: 1 time per day• Each EMA took 1‐5 min to complete/6 wkend of intervention②Whiteley et al []The United StatesPilot RCT18.60 (2.30)—37 (61.67)Online HIV/STI prevention interventionInformation-Motivation-Behavioral Skills (IMB) ModelPublicly available websites and YouTube videos related to HIV/STI prevention Content accessible via computer, smartphone, or tabletSWI / NDI31/29Twice per week for 4 wk, each contained 2‐3 website or video links Total exposure: 8 emails with up to 19 different resources/4 wk3 mo②Mustanski et al []The United StatesRCT23.82—901 (100.00)“Keep It Up!”Information-Motivation-Behavioral Skills (IMB) modelFully online (eHealth intervention), delivered via computers and tablets (not available on mobile phones).IOI / SWI445/456Each session lasted about 1 h./≥3 d for core intervention (3 sessions ≥24 h apart); booster sessions at 3 and 6 mo12 mo③, ④1.34 (1.00‐1.80) ③
0.56 (0.35‐0.89) ④
Peipert et al []The United StatesRCT—220 (0)Project PROTECTTranstheoretical (TTM) model of behavior changeComputer-based multimedia programIOI / SWI272/270Three computer-based sessions /80 d24 mo②, ④0.99 (0.70‐1.38) ②
0.96 (0.61‐1.53) ④
Bull et al []The United StatesRCT14.94 (1.08)—415 (48.71)Teen Outreach Program (TOP)+ text message program: Youth All Engaged (YAE!)Integrated Theory of mHealthText messages based on social mediaTCI / NDI436/4165 and 7 messages weekly/25 wkend of intervention③Ybarra et al []UgandaRCT16.10 (1.40)—307 (83.88)CyberSengaInformation-Motivation-Behavioral Skills (IMB) modelWebsite based; computer-basedIOI / NDI183/183One module per week, a total of 5 modules need to be completed./5 wk3 mo②Bauermeister et al []The United StatesPilot RCT21.67 (1.81)—123 (100.00)myDEx(My Desires & Expectations)The dual processing cognitive-emotional decision-making frameworkOnline-delivered HIV prevention interventions; AppMAI / SWI95/28myDEx intervention includes 6 personalized online courses, completed within 3 mo, with participants logging in an average of about 5 times, and a total conversation volume of about 7 times/3 moend of intervention③Rinehart et al []The United StatesPilot RCT15.90 (1.60)—0 (0)Texts for Sexual Health Education and Empowerment (t4she)Health belief modelText messageTCI / NDI122/12258 automated messages sent over 12 wk/12 wk3 months / 6 mo①Miller et al []The United StatesPilot RCT16.90 (1.00)—26 (28.57)SexHealthTheory of Planned Behavior to inform intervention content and the Social Ecological ModelA tablet-based, interactive intervention: The educator used a tablet to deliver the intervention, intermittently sharing the screen with the participant.IOI / NDI44/4725 min6 mo①Cordova et al []The United StatesPilot RCT18.82 (2.1)—4 (8.00)Storytelling 4 Empowerment (S4E)An ecodevelopment and empowerment frameworkMultilevel Mobile Health AppMAI / NDI25/25Complete 3 interactive modules at once /30‐45 min1 mo③Shafii et al []The United StatesPilot RCT21—176 (64.71)e-KISSInformation-Motivation-Behavioral Skills (IMB) modelAn interactive computer-based intervention; videoIOI / NDI130/14215‐20 min, a single, one-time interactive /15‐20 min2 mo④Chernick et al []The United StatesPilot RCT17.74 (1.27)—0 (0)Dr. Erica (Emergency Room Interventions to improve the Care of Adolescents)Intervention mapping, a program-planning framework; the Social Cognitive Theory and Motivational InterviewingMultimedia text messaging; video; mobile-basedIOI / NDI72/74A minimum of 56 and maximum of 121 texts, with additional texts sent based on keywords./3 mo3 mo①, ②0.95 (0.41‐2.21) ①
1.53 (0.55‐4.25) ②
Yarger et al []The United StatesA Cluster Randomized Trial15.7—340 (40.72)In the Know——an in-person, group-based sexual health education program integrating digital technologies,—Technology-based; mobile-based; appMAI / NDI348/4871.5 h/4 wk3 mo②Suffoletto et al []The United StatesPilot RCT21.44 (2.04)—0 (0)SMS programThe Health Belief Model; the Information Motivation Behavior modelText messageTCI / NDI23/29Once a week, send a text message at noon every Sunday/12 wk3 mo①, ②1.86 (0.48‐7.12) ①
1.60 (0.37‐6.96) ②

aRCT: randomized controlled trial.

bNot available.

cIOI: interactive online-based intervention.

dNDI: nondigital intervention.

eSWI: static web-based intervention.

fTCI: telecommunication-based intervention.

gMAI: mobile app-based intervention.

hSTI: sexually transmitted infection.

iThe included studies did not report SD values for these mean estimates, and the SDs cannot be derived from the available information.

Risk-of-bias assessments using the RoB 2 tool are summarized in . Overall, most trials were judged to be at low risk of bias for the randomization process, outcome measurement, and selection of the reported result. However, a substantial minority of studies had some concerns or high risk of bias in at least one domain, most frequently for deviations from the intended interventions and missing outcome data. Consequently, several trials were rated as having some concerns or a high overall risk of bias.

Although condom use self-efficacy and number of sexual partners were prespecified as secondary outcomes, too few studies reported these measures to allow meta-analysis. Only one trial evaluated condom use self-efficacy. In Rinehart et al [], this construct was assessed using 3 items developed within the Health Belief Model (range 0‐12; Cronbach α=0.72). At the 3rd month, the intervention group reported significantly higher self-efficacy scores than the control group (7.38 vs 6.68; P=.04), but this difference was no longer significant at the 6th month (7.39 vs 6.99; P=.20). Two trials reported on the number of sexual partners. In Shafii et al [], participants in the intervention arm reported a 29% reduction in the number of sexual partners at follow-up, although the effect did not reach statistical significance (IRR 0.71, 95% CI 0.50‐1.03, P=.07). Changes in the control group were not reported. By contrast, Free et al [] examined the proportion of participants reporting 2 or more sexual partners over 12 months. At one year, this outcome was reported by 56.9% of intervention participants compared with 54.8% of controls (OR 1.11, 95% CI 1.00‐1.24, P=.06). Overall, the evidence on the impact of digital interventions on the number of sexual partners remains limited and inconsistent.

Results of Network Meta-AnalysisOverview

A total of 24 RCTs were included to evaluate the comparative effectiveness among 5 intervention types—4 DHIs (TCI, IOI, MAI, and SWI) and NDI—across the four analyzable outcomes: (1) condom use at last sexual contact, (2) consistent condom use, (3) overall proportion of condom use, and (4) incidence of STIs (including HIV). The remaining 2 outcomes of self-efficacy were excluded due to insufficient network connectivity. The network structures for each outcome were shown in , where the thickness of the lines was proportional to the number of comparisons, and the size of the nodes reflected the number of studies involving each intervention. Across outcomes, the treatment network was dominated by comparisons of each DHI category versus NDI, whereas head-to-head trials comparing different DHIs were rare. Several DHI-DHI contrasts and some STI outcomes were informed by only one or two small trials, and self-efficacy outcomes formed disconnected subnetworks. Thus, the network geometry was relatively sparse and heavily anchored on NDI, implying that several treatment rankings rely mainly on indirect evidence. The indirect comparative effectiveness of DHIs was summarized in . Forest plots of posterior ORs with 95% CrIs for each intervention versus NDI across all 4 outcomes are provided in .

‎Figure 3. Network structure diagrams for randomized controlled trials of digital health interventions among youth, by outcome: (A) Condom use rate in the last sexual contact; (B) Consistent condom use rate; (C) Proportion of condom use; (D) The incidence rate of sexually transmitted infections (including HIV). The thicknesses of the lines were proportional to the number of comparisons; the diameters of the circles were proportional to the number of treatments. IOI: interactive online-based intervention; MAI: mobile app-based intervention; NDI: nondigital intervention; SWI: static web-based intervention; TCI: telecommunication-based intervention.

In the complementary frequentist random-effects NMAs, point estimates and 95% CIs for each intervention versus NDI were broadly consistent with the Bayesian results (). Across all 4 outcomes, 95% PIs were noticeably wider than the corresponding CIs and frequently included the null value, even when the average effects suggested benefit. For example, for condom use at last sexual contact and for consistent condom use, TCI, IOI, and MAI tended to favor improved condom use versus NDI, but their PIs indicated that future trials conducted in different settings could plausibly observe smaller benefits or no clear difference from NDI. Similar patterns were observed for the proportion of condom-protected acts and for STI incidence, highlighting that between-study heterogeneity and contextual differences may lead to substantial variability in the effects realized in new populations.

Condom Use Rate in the Last Sexual Contact

Seven studies involving 4 DHIs with a total of 10,285 participants were included in the analysis of condom use at last sexual contact. The random-effects consistency model was selected based on model fit, as it showed comparable DIC and residual deviance values to the inconsistency model, indicating no substantial inconsistency. Among the interventions, only TCI showed a statistically significant improvement compared with NDI (OR 1.13, 95% CrI 1.02‐1.26). Although MAI had the highest SUCRA value (83.44%) and was most likely to rank first (65.61%), its effect was not statistically significant. The rank probabilities for all interventions were summarized in and illustrated in and , showing the descending order of MAI, TCI, NDI, and IOI. As shown in , TCI was the only intervention with its 95% CrI entirely to the right of the line of no effect, suggesting a modest but relatively certain increase in condom use at last sex compared with NDI. IOI and MAI showed point estimates on either side of 1 with wide CrIs, indicating no clear difference from NDI.

Table 3. Rank probabilities and surface under the cumulative ranking curve (SUCRA) values for digital health intervention categories in the network meta-analysis of randomized controlled trials assessing sexual health outcomes among youth.RankIOIMAINDISWITCI(A) rank probability of condom use rate in the last sexual contact15.7665.610.05—28.58212.0287.348.58—92.05320.3497.3882.46—99.824100100100—100SUCRA12.7183.4430.36—73.48(B) Rank probability of consistent condom use rate133.074.590.5461.090.7290.7710.953.3592.762.16397.7841.852.4397.7110.27499.5977.3894.6599.528.87510099.9999.99100100SUCRA80.333.6837.7487.7710.5(C) Rank probability of proportion of condom use112.669.180.80.0417.39265.1788.686.822.8236.53392.394.8724.837.7550.31499.397.8965.674.1363.115100100100100100SUCRA67.3487.6624.528.6841.84(D) Rank probability of the incidence rate of STIs (including HIV)10.38—91.780.047.8127—99.950.4392.63395.79