Effectiveness of text messaging interventions on prevention, detection, treatment, and knowledge outcomes for sexually transmitted infections (STIs)/HIV: a systematic review and meta-analysis

Background Rates of STIs continue to rise worldwide, and novel evidence-based interventions such as text messaging aimed at improving client services are needed. We conducted a meta-analysis to evaluate text messaging to support STI/HIV prevention and treatment interventions. Methods We included articles that reported findings from randomized controlled trials (RTCs) involving adults and youth who were at risk of acquiring (or who currently had) a STI and/or HIV, a text message and comparator intervention, and reported provided outcome data on adherence to STI/HIV treatments. Articles were excluded if they were not published in English. We only included studies that have full-text publications so certainty and risk of bias assessments could be performed. Eight databases were searched to retrieve articles published between 1996 and March 2017. The Cochrane risk of bias tool was used and certainty of the evidence was assessed using GRADE. Effect estimates were pooled using a random effects model. Results A total of 35 RCTs were found, 6 of which were considered at low risk of bias. Eight studies found an increased association using text messaging in appointments attended compared to standard care (OR 1.64, 95% CI 1.28 to 2.10). Participants receiving text messages had an increase in HIV testing compared to standard care (n = 6; OR 1.73, 95% CI 1.39 to 2.15). Ten text messaging RCTs measuring adherence using micro-electro-mechanical systems (MEMS) pill counts has a non-significant association (OR 1.17, 95% CI 0.95–1.45) while five studies measuring adherence by self-report was found to be significant (OR 1.64, 95% CI 1.28–2.11). Conclusions The effectiveness of text message interventions is equivocal. While text messaging has the potential to enhance the delivery of STI/HIV interventions, program planners are encouraged to evaluate any SMS intervention to ensure it is achieving the desired result. Systematic review registration PROSPERO CRD42013006503 Electronic supplementary material The online version of this article (10.1186/s13643-018-0921-4) contains supplementary material, which is available to authorized users.

A large body of literature has emerged demonstrating the effectiveness of SMS on STI/HIV control [17][18][19][20]. However, individual study results are markedly disparate. This points to the need for a comprehensive systematic review and meta-analysis. Several systematic reviews have been published examining SMS interventions in relation to broad questions such as attendance in clinics [21], sexual health in adolescents [22][23][24], reminder interventions to retest [25], and countless others. Three previous meta-analyses [26][27][28] and one network metaanalysis [29] have been conducted that pooled the results of randomized controlled trials (RCTs) on the effectiveness of text messaging interventions on HIV adherence and viral load. These programs take on varied purposes for engaging individuals through the prevent ion-to-treatment spectrum; however, the evidence of the effectiveness of text messaging interventions must be clear before such programs could be transitioned to policy and recommended as a tool for STI prevention and control globally.
We are improving previous research by including and reviewing more current studies, including only RCTs, assessing the risk of bias of the included RCTs, and grading the overall certainty of evidence [30]. We conducted a systematic review and meta-analysis of RCTs aimed at evaluating the impact of SMS on STI/HIV prevention, detection, and control.

Methods
This systematic review was registered with PROSPERO (CRD42013006503), the protocol has been published [31], and a PRISMA statement checklist has been completed (Additional file 1).

Search strategy
This systematic review and meta-analysis was performed according to the Preferred Reporting Items for Systematic review and Meta-Analysis (PRISMA) guidelines [32]. The search was conducted in two phases in order to provide an up-to-date review. The initial search (phase 1) was conducted in the Cochrane Database of Systematic Reviews, MEDLINE, ACP Journal Club, Database of Abstracts of Reviews, EMBASE, EBM Reviews, and included articles published from 1996 to August 31, 2013. We chose to limit the search starting at 1996 as it was thought that this was when SMS first became widely used. In addition, the table of contents for the following journals was reviewed: Sexually Transmitted Diseases, Sexually Transmitted Infections, and AIDS Patient Care and STDs. Grey literature was searched using Google and Grey Literature Report (www.greylit.org) to identify studies that may have been missed in the abovementioned searches. Medical subject headings (MeSH) and keywords included mobile health, mHealth, cell phone, mobile phone short message services, text messaging, texting, SMS, MMS, communication technologies, patient monitoring devices, wireless technologies, STI testing, sexually transmitted diseases, sexually transmitted infections, HIV, chlamydia, gonorrhea, herpes, Trichomonas vaginalis, and syphilis. The search strategies are included in Additional file 2. The second search (phase 2) was conducted in MEDLINE and mhealthevidence.org (Additional file 3) using the same search terms. Limiting our search to MEDLINE and mhealthevidence.org was done in this phase due to a lack of resources. This included articles published between September 1, 2013, and March 30, 2017. We also examined the reference lists of any relevant systematic reviews retrieved.

Eligibility criteria
RCTs examining the effect of SMS on STI/HIV prevention and control outcomes among adults and youth published in English between January 1, 1996, and March 30, 2017, were included. For phase 2, we only included studies that were identified in MEDLINE and mhealthevidence.org between 2013 and 2017. For the purposes of this review, SMS was defined as a text message that is delivered to a mobile phone either manually or by an automated system. We included only RCTs published in English as we did not have the financial and human resources to translate non-English studies. A detailed description of the inclusion/exclusion criteria using a PICO (Population, Intervention, Comparison, Outcome) format can be found in Additional file 4.

Selection of studies
Titles and abstracts were independently examined for eligibility by two reviewers (PL, OW). Disagreements were resolved by a third reviewer (DT). Studies that were deemed eligible based on title and abstract were then reviewed by two independent researchers to confirm eligibility. Forward searching was done using the included references. Our updated search involved screening by one reviewer (CL) with a second reviewer checking the full text of included studies (DT). No test of agreement between reviewers was conducted for either of the search phases. All discrepancies were discussed until consensus was achieved.

Data extraction
Data were extracted independently by two reviewers (PL, OW) using a standardized pre-tested data extraction form. Data extracted included (1) purpose of intervention, (2) duration of intervention, (3) frequency of text messages, (4) setting, (5) intervention, (6) comparison, (7) one way or two-way SMS, (8) type of participants, (9) follow-up period, (10) outcome results, and (11) reported effect measure (risk ratio, odds ratio, etc.) from the study. The studies included in our updated search were extracted by one reviewer (CL) and 100% checked by a second reviewer (DT). No test of agreement between reviewers was conducted for either of the search phases. All discrepancies were discussed until consensus was achieved.

Risk of bias assessment
Risk of bias was assessed using Cochrane's risk of bias (ROB) tool [33]. In Cochrane's "other" ROB domain, we included "incomplete outcome data (intention to treat [ITT] analysis)." This was defined as randomized participants having been analyzed according to their allocated treatment, irrespective of whether they were eligible, received the allocated treatment, received another treatment, or received no treatment. Study certainty was rated independently by PL and OW, and any discrepancies were discussed until consensus was reached. The studies included from our updated search involved assessment by one reviewer (CL) and checking by a second reviewer (DT). RCTs were considered at low risk of bias if they were rated at low risk for the following domains: random sequence generation, allocation concealment, incomplete outcome data (attrition bias and intention-to-treat [ITT] analysis), and selective reporting bias.

Unit of analysis issues
As recommended in the Cochrane Handbook [33], we combined multiple intervention arms to overcome unit-of-analysis error and to create a single pair-wise comparison.

Certainty of the evidence assessment
We graded the overall certainty of evidence using the Grading of Recommendations Assessment, Development and Evaluation (GRADE) approach [30]. GRADE assessments were conducted by a single reviewer. GRADEpro software [34] was used to create Summary of Findings tables for the following primary outcomes: appointment adherence, adherence to ART by MEMS (medication event monitoring system) pill count, and adherence to ART by self-report. The certainty of the evidence was categorized as high certainty, moderate certainty, low certainty, or very low certainty based on a judgment of the confidence in the effect estimate, whether the true effect is likely to be substantially different from the estimate of effect. The certainty of the evidence is downgraded from high, moderate, low to very low for the following reasons: risk of bias, inconsistency (unexplained heterogeneity, inconsistency of results), indirectness of evidence (indirect population, intervention, control, outcomes), imprecision of results (wide confidence intervals), and risk of publication bias.

Analysis
Clinical heterogeneity was assessed within each pairwise comparison by comparing characteristics [33]. In order to minimize heterogeneity, outcomes were categorized as follows: (1) SMS used for prevention of STIs/HIV, (2) SMS used to impact adherence to HIV medication adherence, and (3) SMS used to impact HIV treatment outcomes. Sub-analyses were conducted based on these different categories. As the characteristics of the 35 included studies were variable, we chose to use the Mantel-Haenszel random effects model. Statistical heterogeneity was examined using the I 2 statistic and interpreted as an I 2 estimate of 0% = no heterogeneity, 25% = low, 50% = moderate, and 75% = high heterogeneity of effect sizes [33]. We summarized dichotomous outcomes using odds ratios (OR), with 95% confidence intervals (CI). We used RevMan 5.3 to conduct the statistical analysis and graphs.

Publication bias
If more than ten studies reported the same outcome, publication bias was explored with the Egger test and a funnel plot using STATA 13 software with the metan, metafunnel, metabias, and metatrim packages [35][36][37][38][39]. A p value less than 0.05 was considered statistically significant. Stata's trim and fill method was used to estimate the number of missing null studies from the meta-analysis. Stata's metatrim command performs the Duval and Tweedie nonparametric "trim and fill" method of accounting for publication bias in meta-analysis. The method, a rank-based data augmentation technique, formalizes the use of funnel plots, estimates the number and outcomes of missing studies, and adjusts the meta-analysis to incorporate the theoretical missing studies.
Sensitivity analyses were conducted with and without statistical outliers to assess their effect on the overall findings. In addition, a sensitivity analysis was conducted to assess the impact of RCTs judged to have low methodological certainty on the summary effect. If three or more studies exist in each group, we subgrouped adherence as "self-reported" and "objectively-measured" adherence (MEMS pill count).

Results
A total of 14,850 articles were initially identified through database and manual searches conducted between 1996 and March 2017 (both phases) (Fig. 1). Four thousand six hundred eighty duplicates were removed, and 9949 were not included as they were not relevant to our topic. Of the remaining 221 full text articles, 186 were excluded because they were case studies (n = 2), duplicate records of the same study (n = 9), included no data and authors were unreachable (n = 15), had no relevant outcomes (n = 40), did not involve texting (n = 22), both arms received text messages (n = 3), was not a full article (n = 1), and not a RCT (n = 94). The remaining 35 RCTs  were included. Table 1 displays the characteristics of the 35 included studies. A description of the content of the SMS messages can be found in Additional file 5.
Of the 35 studies included in this systematic review, the majority investigated either the effectiveness of text messaging to support STI clinic appointment attendance (8 studies), or the effectiveness of text messaging to support HIV medication adherence (15 studies). These studies were conducted in 14 countries, the majority being in the USA (13 studies), Kenya (5 studies), and South Africa (3 studies) and were conducted in a variety of settings including clinics (19 studies), hospitals (6 studies), universities (4 studies), and others such as an emergency room (1 study) and a music festival (1 study). The age of participants ranged from 16 to 84 years and the length of follow-up ranged from 7 days to 24 months. Text messages were either sent daily, weekly, monthly, or as one-off reminders. The majority of studies (83%) sent a one-way text with the remaining sending two-way texting.
In most studies, medical staff were masked to the knowledge of the allocated interventions; however, in the majority of cases, blinding of participants to intervention allocation was not done. Only two studies successfully blinded participants to the intervention [57,67]. Participants in the Morris et al. [57] study were from a registry so they were blinded to their randomization group, and individuals in Rand et al.'s [67] control group received the same initial message as the intervention group, followed by a control message about a different general adolescent health topic each time reminders were sent to the intervention group; therefore, they were unlikely to ascertain their allocation.
High attrition bias was encountered in three studies [40,56,73] and was unclear in three others [45,46,70] as they had not recorded the number of participants lost to follow-up. Thirteen studies were at high risk of incomplete outcome bias (intention-to-treat [ITT] analysis) as they had not conducted the ITT analysis correctly or at all [40,42,44,45,48,49,53,56,57,61,62,69,73]. An intention-to-treat (ITT) analysis is recommended as the least biased way to estimate intervention effects in randomized trials [33].
Selective reporting bias was at high risk in three studies [43,53,73] and at unclear risk in eight studies [42,44,45,47,48,54,71,74] because of pre-protocol or registry information being unavailable. Six studies were considered at low risk of bias as they were rated at low risk for the following domains: random sequence generation, allocation concealment, incomplete outcome data (attrition bias and ITT analysis bias), and selective reporting bias [41,50,52,63,66,68]. Only three of these measured the same outcome (HIV adherence by pill count) and were therefore available for sensitivity analysis [50,63,66]. Risk of bias for each study is included in the forest plots provided (Figs. 2, 3, and 4). In addition, a summary graph of the risk of bias for all included studies and a detailed graph of the risk of bias of individual studies can be found in Additional file 6.

Outcomes
Outcomes included in our analysis were divided into three categories: prevention interventions, HIV drug adherence interventions, and HIV treatment outcome interventions. A table organizing studies for subgroup analysis can be found in Additional file 7.

Prevention interventions (Fig. 2)
Appointment adherence Eight studies [41,43,49,58,59,61,65,70] with a total of 5027 participants found an associated increase in appointment attendance among the SMS group compared to standard care (OR 1.64, 95% CI 1.28 to 2.10). There was low heterogeneity among the pooled studies (I 2 45%). Two studies that recruited youth (age 16-30 years) had a pooled odds ratio of 2.03 with a wide confidence interval of 1.39 to 2.97 indicating imprecision [58,70]. In the case of Rutland et al.'s study [70], imprecision may have been a result of a small sample size.
Condom use Two studies [53,73] with 1046 participants (aged between 16 and 29) examining the   [73] (n = 52 females; 30% attrition) that reported a positive effect of the SMS intervention at 3 months follow-up. There was low statistical heterogeneity (I 2 13%) among the two studies; however, the two studies differed in the type of participants recruited and outcome definition. Lim et al. [53] recruited males and females at a music festival and defined condom use as always using condoms in the past 3 months with risky partners. In contrast, Suffoletto et al. [73] recruited females with hazardous drinking behavior and recent risky sexual encounters and defined condom use as always using condoms in the past month.
Uptake of STI/HIV testing Pooled results from six studies (n = 4753; 58% attrition) revealed a positive association between SMS and uptake of STI/HIV testing (OR 1.73, 95% CI 1.39 to 2.15; I 2 0%). Two studies [44,58] evaluated the effectiveness of SMS on the uptake of HIV testing compared to standard care, one study [53] examined the effectiveness of SMS on the uptake of STI testing in the past 6 months compared to standard care, one study [74] examined those undergoing chlamydia test of cure, and two others examined CD4 testing before 26 weeks gestation [45] and viral load testing in infants by 8 weeks postpartum [62].  Fig. 2). Two studies with populations aged between 18 and 26 were subgrouped and produced a pooled odds

Knowledge of STIs, ART dosage, and instructions
Two studies [53,54]  Circumcision One study [40] that examined the effectiveness of sending promotional texts to HIV-negative and uncircumcised men after HIV clinic testing to encourage circumcision compared to standard clinic referral found a positive association (OR 1.59, 95% CI 1.18 to 2.14). However, another study of men who underwent circumcision [61] reported no association between men Fig. 4 Forest plot of the effectiveness of text messaging on HIV treatment outcomes who resumed sex before the waiting period that was recommended by their doctor in the SMS group and the control group (RR 1.09, 95% CI 0.99 to 1.19).
The five RCTs measuring adherence by self-report were homogenous. Heterogeneity in the ten studies measuring adherence by pill count (I 2 56%) was moderate when all studies were combined. However, when the outlier study by Sabin et al. [71] (OR 7.45, 95% CI 2.90 to 19.15) was omitted, the RCTs became homogenous (I 2 0%), and the effect measure weakened to show no association (OR 1.17, 95% CI 0.95 to 1.45).
HIV treatment outcomes (Fig. 4) Uptake of CD4 testing One study by Dryden-Peterson et al. (n = 366; 11% attrition) [45] examined the effectiveness of text messaging on the uptake of CD4 cell count testing revealing no significant association (OR 1.41, 95% CI 0.91 to 2.19).

Viral load suppression
Five studies (n = 1607; 2% attrition) examined the effectiveness of SMS to support HIV viral suppression [52,56,63,71,72]. In pooled analysis, the five RCTs showed no association in virological suppression below the level of detection between intervention and control groups with low heterogeneity (OR 1.08, 95% CI 0.81 to 1.44; I 2 22%).

Presence of a new opportunistic infection
One study [75] (n = 200; 0% attrition) reported the presence of new opportunistic infections in HIV positive participants at 3 months post-intervention. This study showed that SMS was not associated with the incidence of new opportunistic infections (OR 1.56, 95% CI 0.85 to 2.85).
CD4 cell count increase Two studies (n = 208; 15% attrition) [56,71] examining the proportion of patients that had an increase in CD4 cell count when a text message was received found no association (OR 1.13, 95% CI 0.62 to 2.06; I 2 0%). In addition to these two studies, Maduka (n = 104) [54] measured pre-and post-intervention median CD4 cell counts among intervention and control groups and found a greater increase in CD4 cell counts in the text message group than the control group (SMS group increased from 193 cells/ml to 575.0 cells/ml compared to the control group that saw an increase from 131.0 cells/ml to 361.5 cells/ml; p = 0.007).
Uptake of CD4 testing One study (n = 366; 11% attrition) [45] examined the uptake of CD4 testing when SMS were sent to the intervention group compared to standard care found no association (OR 1.41. 95% CI 0.91, 2.19).
ART initiation before 30 weeks gestation One study (n = 366; 11% attrition) [45] examining text messaging to support ART initiation before 30 weeks gestation showed no association between the intervention and standard care groups (OR 1.59, 95% CI 0.98 to 2.60).

Test of publication bias
Medication adherence was the only outcome that had sufficient studies to conduct a test of publication bias. The funnel plot with the trim and fill method (Additional file 7) for the 15 medication adherence studies indicated asymmetry. The Egger test of small-study effects revealed an estimated bias coefficient of 1.53 (standard error 0.41; p = 0.002). The trim and fill method predicted that two RCTs were missing from this analysis.

GRADE assessment
We assessed all outcomes using the GRADE approach. Results for three of the most important outcomes are in Table 2. The remaining assessments are available in Additional file 8. The eight RCTs contributing data to the appointment adherence outcome showed serious risk of bias [41,43,49,58,59,61,70]. The majority of studies were single-blinded; however, the outcome assessment appeared free from bias because the investigators blinded outcome assessors. There was a lack of blinding of patients in studies, and only one study clearly used intention to treat analysis. Inconsistency and indirectness were assessed as serious because of moderate heterogeneity of effect estimates, and populations varied significantly. Imprecision was considered serious because of wide confidence intervals. Publication bias was strongly suspected. Appointment adherence was therefore assessed as very low-certainty evidence. The ten studies contributing data to the adherence to ART by pill count outcome showed serious risk of bias [42, 47-50, 56, 63, 66, 71, 72]. Seven out of ten RCTs were assessed at high risk of bias due to unclear randomization and allocation concealment. There was unexplained inconsistency, with moderate heterogeneity, and one study's confidence intervals did not overlap [71]. Indirectness was assessed as serious because populations studied varied, as well as the frequency of text reminders sent. Adherence to ART by pill count was therefore assessed as very low-certainty evidence.
The five RCTs contributing data to the adherence to ART by self-report outcome showed serious risk of bias [46,52,54,55,59] due to the unclear risk of allocation concealment (one study) and inadequate blinding of staff to the study arms (four out of five studies). Inconsistency was not serious as studies showed statistical homogeneity. Indirectness was serious because populations varied across studies, as did the frequency of text reminders sent. There was also serious imprecision because of wide confidence intervals. The adherence to ART by self-report outcome was therefore assessed as very low-certainty evidence.

Deviations from planned protocol
In our published protocol [31], we stated we would include randomized and nonrandomized controlled trials, pre-and post-test designs, observational (cross-sectional, case-series, case studies, and qualitative studies) that examined the effectiveness of SMS interventions on STIs. However, we only include RCTs in order to manage the large amount of data collected and because these represent the highest quality research. Similarly, we only reported on the 14 outcomes that were most dominant in the literature aimed at assessing the effectiveness of interventions (not qualitative outcomes such as acceptability). We also made our eligibility criteria broader than our originally planned criteria to allow for all types of comparisons (e.g., phone, emails) to be inclusive of other means of communicating health messages.
The initial search was conducted in 2013 in multiple databases (MEDLINE, ACP Journal Club, Database of Abstracts of Reviews, EMBASE, EBM Reviews, and Cochrane Library); however, the update search was conducted only in MEDLINE and mhealthevidence.org to contain the search results due to restricted resources available (staff and funding). We feel however that searching mhealthevidence.org in addition to MEDLINE would have retrieved the majority of randomized trials on this topic as it is a current database repository for all mHealth evidence implemented by Johns Hopkins Center for Communication Programs. Furthermore, in our more comprehensive search from 1996 to 2013 of five databases, we retrieved only 14 RCTs as compared to the additional 21 we retrieved from August 2013 to end of March 2017 (with consideration of the probable increase in SMS studies published) using the MEDLINE and mhealthevidence.org databases.

Summary of main results
The purpose of this systematic review was to determine the effectiveness of SMS on STI/HIV prevention and control outcomes. Overall, the results of our systematic review highlight the equivocal nature of the evidence surrounding the effectiveness of SMS interventions of STI/HIV outcomes. SMS interventions vary greatly in different populations and when used for different purposes. SMS may be effective for some prevention interventions such as those aimed at improving appointment adherence (eight studies) and those aimed at increasing the uptake of circumcision (to prevent HIV) (two studies). They may also be effective in increasing adherence to HIV medications(15 studies) although this effect was primarily found in studies depending on self-report suggesting that social desirability bias may have come into play. Conversely, pooled results from seven studies indicate that there is no statistically significant association between providing SMS interventions and improving HIV treatment outcomes.
It should be noted that the certainty of the evidence was found to be low due to inconsistency, indirectness, and imprecision. This may be explained by the markedly variant in terms of SMS methods (one-way versus two-way, frequency of text messages sent, their content, and the populations targeted) among the studies we examined. This explanation is supported by others who have experienced disparate effects of SMS using identical study designs, geographic location, and SMS mode of delivery but among a slightly different population group [76]. Moreover, uncertainty among our results may be due to the high percentage of studies (75%) having high risk of selection bias and performance bias due to inability to conceal randomization allocation and inability to blind study participants and study personnel.

Significance of this review
Our study points out that the evidence related to SMS interventions remains equivocal and individual studies should be interpreted with caution. Randomized controlled trials are considered the gold standard for determining the effectiveness of an intervention. However, when bias is introduced by lack of blinding, small sample sizes, high attrition rates, and unclear allocation concealment, confidence in the evidence is reduced. Our study highlights the challenges that are encountered when attempting to pool results from RCTs that are prone to bias. In addition, there is a need for standardized reporting, using common definitions for outcomes for RCTs aimed at evaluating SMS to reduce unexplained inconsistencies. Finally, we recommend authors of RCT studies to include statements about registration of study protocols in their published work to minimize concerns of reporting bias and include intention-to-treat analyses in their research to avoid bias resulting from crossover.

Agreements and disagreements with other studies or reviews
Our results are consistent with the findings of a 2013 Cochrane review (n = 8) that showed low to moderate certainty related to SMS improving attendance rates at any healthcare appointments [77]. Our findings also agree with a 2017 network meta-analysis [29] reporting results of a pairwise direct comparison of HIV treatment adherence using a SMS intervention versus standard care in four RCTs (OR of 1.70) (95% CI 1.16, 2.49, certainty of the evidence, moderate). Another 2012 Cochrane review (n = 2) evaluated SMS for promoting adherence to antiretroviral therapy in patients with HIV conducted in Kenya only [27]. Our analysis included these two high-certainty studies and included an additional 13 RCTs.
Two other reviews pooled the results of RCTs and quasi-experimental controlled trials on the effectiveness of SMS interventions on HIV adherence and viral load [26,28]. One systematic review with meta-analysis by Mayer and Fontelo examined (1) appointment non-attendance and (2) HIV adherence when an SMS reminder was sent to participants in intervention studies compared to a control group or pre-intervention group [28]. They included both pre-post studies and randomized controlled trials but did not include a subgroup analysis by study design. The pooling of the Mayer and Fontelo data on non-attendance is somewhat counterintuitive as clinicians, researchers, and funders want to know if the addition of SMS reminders will increase appointment attendance at STI/HIV clinics. Further, Mayer and Fontelo's pooled analysis of the effectiveness of SMS reminders to improve medication adherence was done using a standardized mean difference. The majority of studies in our analysis (67%, n = 10/15) reported data as proportion of patients 90-95% adherent compared to control; therefore, we chose to analyze the HIV adherence data on a dichotomous scale. This decision, as well as the decision to pool only RCTs, resulted in a lower level of heterogeneity (I 2 44% over the 15 RCTs) in our pooled analysis compared to Mayer and Fontelo's [28] high heterogeneity (I 2 99%). When performing sensitivity analyses, we omitted the one study that differed in its definition of adherence (patients were categorized as optimally and suboptimally adherent) which resulted in a homogenous analysis (I 2 0%) and an OR of 1.35 with a narrow confidence interval (95% CI 1.15-1.59).
Two other previous reviews are available. Lim et al. [78] provided a review of SMS interventions. However, this review is outdated (2008) and was limited to describing SMS programs. The authors acknowledge that the programs the report on were not rigorously evaluated. We are providing more current evidence that has been rigorously produced. Zou et al. [79] provided a review that is 4 years old and had a narrow scope. They evaluated the effect of SMS on screening rates for bacterial STIs among men who have sex with men. Our study provides a more comprehensive review of all STIs for all populations. The reference list for this review was utilized during our search strategy.
Systematic reviews by Chavez et al. [80], L'Engle et al. [24], and Zou et al. [81] present important findings with respect to text messaging and sexual health outcomes; however, they do not cover the effectiveness of SMS interventions for a broad spectrum of STI prevention and control strategies. Our review fills an important gap by including more RCTs. In addition, the scope of our review is wider because it provides evidence from developed and developing countries.

Strengths and limitations of the present review
A major strength of this present review is that we have included an evaluation of SMS for both STIs and HIV. This is important because public health clinicians do not separate out interventions based on these two things. Introduction of bias in our review was minimized by double independent screening and extraction, and assessing the risk of bias of each included study by two authors. Although no test of agreement was conducted between reviewers, all discrepancies were discussed and resolved.
The authors recognize that limiting our review to studies published in English may produce an unintended bias as SMS interventions conducted in non-Englishspeaking countries may have a different effect. Therefore, we caution readers from non-English speaking countries to take this into consideration when considering introducing SMS interventions in their respective jurisdictions. We did not include studies that could only be found in the grey literature (i.e., conference proceedings) as it is difficult to assess certainty and bias for these studies. This may have resulted in missing more recent studies which may have resulted in including studies with a greater effect size (due to possible increased comfort levels with text messaging). In addition, only one reviewer was involved in searching for studies and screening titles and abstracts for eligibility during phase 2. This may have resulted in some relevant studies spuriously being excluded. Moreover, GRADE assessments were conducted by a single reviewer, due to resource limitations, which may have biased the reporting in an unintended way. In addition, we did not report the results of one primary outcome-acceptability of SMS for STI interventions and one secondary outcomefeasibility of program delivery, which was reported by some RCTs in our sample. We felt that this information was better suited to a subsequent (second paper) reporting implementation results.
The initial search was conducted in 2013 in multiple databases (MEDLINE, ACP Journal Club, Database of Abstracts of Reviews, EMBASE, EBM Reviews, and Cochrane Library); however, the update search was conducted only in MEDLINE and mhealthevidence.org to contain the search results due to restricted resources available (staff and funding). We feel however that searching mhealthevidence.org in addition to MEDLINE would have retrieved the majority of randomized trials on this topic as it is a current database repository for all mHealth evidence implemented by Johns Hopkins Center for Communication Programs. Furthermore, in our more comprehensive search from 1996 to 2013 of five databases, we retrieved only 14 RCTs as compared to the additional 21 we retrieved from August 2013 to end of March 2017 (with consideration of the probable increase in SMS studies published) using the MEDLINE and mhealthevidence.org databases.
The studies we included were of varying certainty, and the certainty of the evidence on two important outcomes was considered low, thus limiting the confidence in the summary effect estimates. We were also only able to pool a sufficient number of high-certainty studies for one outcome-adherence to ART. Both certainty of evidence and study quality were considered when making conclusions, reducing the influence of lower certainty studies. In addition, rigorous searching of multiple databases was conducted; however, publication bias is suspected due to non-publication of many negative trials.

Future research
High certainty RCTs are required to adequately assess the effect of text messaging to support all STI/HIV prevention and treatment outcomes included in this review. Ongoing trials into text messaging support to STI/HIV interventions might provide more power to update of these meta-analyses [82,83]. Research into messaging interventions should also include other messaging platforms including WhatsApp, Facebook, etc. and compare results with cell phone texting. Moreover, our study did not report reported adverse events related to the SMS interventions. This should be included in a future review. Future research should also examine whether reminders alone are as effective as reminders embedded in educative messages [84,85]. Finally, RCTs can be designed as double-blinded by delivering the same number of text messages but with different content to the patient and control groups, thus making the trials less prone to biased results.

Application to clinical care
SMS is a flexible method to deliver health messages as it allows for instantaneous delivery directly to individuals at any time, place, or setting via their phones [86]. However, it is important to evaluate the effectiveness of an intervention once it has been initiated. The use of digital technology in designing interventions aimed at improving health outcomes including those related to STI and HIV has been approached with great enthusiasm. However, as others have pointed out [80], digital technology is rapidly changing and therefore research that has been conducted with SMS interventions may soon only be useful for a short period of time.