This systematic review and meta-analysis was conducted in accordance with the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) 2020 guidelines (Checklist 1). It was registered in PROSPERO with the number CRD420251055972. A protocol for this systematic review was not prepared.

A comprehensive literature search was performed by 3 independent reviewers (TTYC, CYF, and AKCW) across 6 electronic databases: PubMed, MEDLINE, CINAHL, Embase, Web of Science, and the Cochrane Library. Studies published between April 1, 2005, and March 31, 2025, were included. The reference lists of included studies were also hand-searched independently by the 3 reviewers (TTYC, CYF, and AKCW) to identify additional eligible studies. Any disagreements among the reviewers were resolved through discussion.

Only peer-reviewed articles were considered; gray literature and review articles were excluded to ensure the inclusion of high-quality evidence. Search strategies were developed using key terms including “prediabetes” and “text messaging” and were expanded using Medical Subject Headings (MeSH) terms where appropriate. Detailed search strategies for each database are provided in Multimedia Appendix 1.

Studies were eligible for inclusion if they met the following criteria: randomized controlled trials (RCTs), participants aged ≥18 years diagnosed with prediabetes, interventions consisting of lifestyle modification programs (ie, diet control, exercise, or behavioral therapy), and use of text messages—defined as SMS, multimedia messaging service, or instant messaging via mobile apps—as the primary intervention delivery channel.

The exclusion criteria were as follows: participants diagnosed with type 2 diabetes mellitus (DM) or gestational diabetes, participants taking medications known to affect glucose tolerance, and studies without complete analytical outcome data.

The primary outcome of interest was change in BMI. Secondary outcomes included changes in body weight, waist circumference, HbA_1c, total cholesterol, and incidence of DM.

The risk of bias for the included studies was assessed using the Risk of Bias tool (Cochrane Collaboration). This tool evaluates the following five domains: (1) bias arising from the randomization process, (2) bias due to deviations from intended interventions, (3) bias due to missing outcome data, (4) bias in the measurement of outcomes, and (5) bias in the selection of reported results. Any disagreements among the 3 reviewers were resolved through discussion. As there were fewer than 10 included studies, a funnel plot to assess publication bias was not constructed.

Beyond study-level risk of bias, we appraised the outcome-level certainty of evidence using the GRADE (Grading of Recommendations Assessment, Development and Evaluation) approach for each meta-analyzed outcome (BMI, body weight, waist circumference, HbA_1c, total cholesterol, and diabetes incidence). Because all included studies were RCTs, each outcome began at high certainty and could be rated down for risk of bias, inconsistency, indirectness, imprecision, and publication bias. The domains were assessed as follows:

We present GRADE ratings and explicit downgrading rationales in a summary of findings table (Table S1 in Multimedia Appendix 2) and detailed footnotes (Multimedia Appendix 2). No outcomes were upgraded.

Data from the included studies were retrieved and independently entered into a Microsoft Excel spreadsheet by 3 independent reviewers (TTYC, CYF, and AKCW). Additional variables—including author, study location, study population, study duration, intervention and control details, data collection time points, outcome measures, and results—were also extracted from the selected studies (Multimedia Appendix 3).

Meta-analyses were conducted using Review Manager (version 5.4.1; Cochrane Collaboration). Outcomes were analyzed at the longest reported follow-up time point and illustrated by forest plots. For continuous variables, mean differences (MDs) with 95% CIs were calculated under a random-effects model. The meta-analysis was conducted using the postintervention values reported by the individual studies. For dichotomous variables, odds ratios and 95% CIs were computed using the inverse variance method.

Statistical heterogeneity was assessed using the chi-square test, the I² statistic, and associated P values (P<.05 was considered statistically significant). I² values were interpreted as follows: unimportant (0%‐40%), moderate (30%‐60%), substantial (50%‐90%), or considerable (75%‐100%) heterogeneity. Given the limited number of included trials (<10), subgroup analyses, meta-regression, and sensitivity analyses were not conducted.

A total of 19,940 publications were identified after duplicate removal. Title and abstract screening excluded 19,850 (99.5%) publications. Of the remaining 90 publications, 75 (83.3%) were available for full-text review and assessed for eligibility. In total, 68 (90.7%) publications were excluded, primarily due to inappropriate intervention delivery modes, ineligible populations, or study designs other than RCTs. Only 7 (9.4%) publications met the eligibility criteria and were included in this review [16,23-28] (Figure 1).

Risk of bias among the 7 included RCTs, assessed using the Cochrane Risk of Bias tool, ranged from “some concerns” to “high risk.” All included studies demonstrated low risk in the domain “bias in selection of the reported result.” However, 3 of 7 included studies were rated as “some concerns” for the domains of bias arising from the randomization process, missing outcome data, and measurement of outcomes. All included studies were rated as “some concerns” for bias due to deviations from intended interventions, likely reflecting the nonblinded nature of the interventions. A summary of the risk of bias assessments is presented in Table 1.

Outcome-level certainty ratings and rationales are reported in Table S1 (summary of findings) in Multimedia Appendix 2. In brief, certainty was most commonly rated down for inconsistency (between-study heterogeneity) and imprecision (wide CIs and limited sample size per outcome). Publication bias could not be formally assessed due to fewer than 10 studies, and no outcomes were upgraded.

The 7 included RCTs enrolled a total of 4632 participants diagnosed with prediabetes. All participants owned smartphones and were free of major comorbidities. The mean age of participants ranged from 43.5 to 54.2 (SD 5.6) years. Of 4632 participants, 2843 (61.4%) were male.

Among the participants, 2056 (44.4%) individuals received lifestyle modification programs incorporating SMS text messaging, while 2576 (55.6%) individuals received comparable programs without SMS text messaging support. Intervention durations ranged from 12 to 48 months, with 24 months (n=2) and 48 months (n=2) being the most common durations. Delivery formats varied: 4 studies delivered regular text messages promoting lifestyle modification [16,23-25], 2 studies combined mHealth apps with SMS text messaging support [26,27], and 1 study integrated face-to-face sessions with text message follow-ups [28]. For the control group, usual care included related education (n=4), paper-based health information materials (n=2), and physical activity management (n=1). Characteristics of all included studies are summarized in Multimedia Appendix 3.

The forest plots of both primary and secondary outcomes are shown in Figures 2-7.

All 7 studies assessed BMI (4102/4634, 88.5% participants). The pooled MD in BMI change was not statistically significant between intervention and control groups (MD –0.17, 95% CI –0.85 to 0.25; P=.43), with low heterogeneity observed (χ²₆=8.8; I²=32%; P=.18).

This systematic review and meta-analysis synthesized evidence on SMS text message–delivered lifestyle programs for adults with prediabetes, focusing on changes in BMI, weight, waist circumference, HbA_1c, total cholesterol, and diabetes incidence. Overall, the pooled estimates did not demonstrate statistically significant improvements in these metabolic outcomes compared with standard care, although several individual trials showed favorable trends.

mHealth technologies can extend the reach of prevention programs by lowering barriers related to access, workforce capacity, and cost. SMS text messaging is especially attractive because it is simple, low-cost, and highly scalable [18,29]. For adults with prediabetes—a group often characterized by low disease awareness and variable motivation—SMS text messaging may provide timely prompts for early lifestyle change. However, in contrast to prior meta-analyses reporting benefits of SMS text messaging for smoking cessation, weight loss, cardiovascular risk reduction, or glycemic control in type 2 diabetes [30-37], our results showed null effects on anthropometry and cardiometabolic markers in prediabetes. These differences likely reflect population context and mechanism: approaches successful in patients with established chronic disease may not translate directly to prevention in largely asymptomatic prediabetes.

Several design features may explain the limited impact observed. First, interventions varied widely in duration, message frequency, delivery channel, and content design. Frequencies ranged from multiple messages per week to biweekly contact; the literature is mixed on whether a higher cadence improves outcomes or induces message fatigue, suggesting an outcome- and population-specific optimum. Second, most programs used standardized, nontailored content. Evidence from digital behavior change research indicates that dynamic tailoring—linking message content, timing, and cadence to a person’s current behavior, preferences, and psychosocial state—improves engagement and maintenance [38]. Third, the predominant communication style was unidirectional. Bidirectional designs that solicit brief replies, troubleshoot barriers, and escalate support after nonresponse may enhance accountability and adherence but were rarely implemented. Finally, several trials combined SMS text messaging with additional components (eg, app access, in-person education, or follow-up calls), which improved ecological validity but reduced the ability to attribute effects specifically to SMS text messaging alone and added heterogeneity in behavioral dose and support intensity [39-41].

BMI is an imperfect proxy for metabolic risk because it conflates fat and lean mass and cannot distinguish visceral adipose tissue (VAT) from subcutaneous fat or capture ectopic depots (hepatic or intramuscular fat). Small but meaningful reductions in VAT—or preservation or increases in skeletal muscle mass with modest fat loss—can yield little or no net change in BMI while still improving insulin sensitivity, lipid handling, and inflammatory tone. Early behavior change can also redistribute fat (from visceral to subcutaneous compartments) or prevent sarcopenic weight loss, effects that BMI cannot detect.

In our synthesis, waist circumference similarly showed a precise null effect with relatively narrow CIs; however, centimeter-level changes over short follow-up periods may be below detection thresholds, and meaningful body composition adaptations often require ≥6 to 12 months to manifest. Moreover, adherence decay and metabolic adaptation (reductions in energy expenditure during weight loss) can flatten BMI trajectories when interventions are light touch and largely unidirectional, as seen in most included programs. Therefore, future trials should include body composition outcomes (eg, dual-energy X-ray absorptiometry–derived percent fat and lean mass, VAT area by imaging, and multifrequency bioimpedance) and waist-to-height ratio alongside the BMI.

Regarding publication bias, the small number of studies per outcome precluded formal small-study tests. Notably, several pooled estimates centered closely on the null value with relatively narrow CIs (eg, total cholesterol), suggesting that, at the tested dose and design—generic, 1-way prompts over limited durations—SMS text messaging alone is unlikely to generate the sustained energy deficit and resistance-type stimulus required to materially shift BMI or central adiposity in prediabetes. Thus, the persistence of null results appears plausible even if publication bias might ordinarily favor positive findings.

SMS text messaging appears best positioned as a reach-and-reinforcement layer rather than a stand-alone metabolic intervention. In clinical pathways, brief messaging can reinforce clinic-initiated goals, prompt self-monitoring (weight, steps, and dietary logs), and support attendance at higher-intensity services (group classes, dietitian visits, or structured exercise). At the service level, programs should incorporate closed-loop features (eg, automated outreach after nonresponse, brief telecoaching, or referral triggers when weight plateaus) and integrate device-captured data to enable tailoring.

Commissioners planning population strategies should pair SMS text messaging with an effective dose (≥12‐24 weeks), link to evidence-based nutrition and progressive resistance training supports, and evaluate outcomes beyond BMI, including waist-to-height ratio, body fat percentage, VAT surrogates, and glycemic variability. Reporting of engagement analytics (delivery, open, reply, and attrition rates) and behavioral mediators (diet quality, step counts, and resistance training adherence) is essential to clarify mechanisms and optimize deployment.

Several limitations of this study should be acknowledged. Moderate to high heterogeneity was identified in outcomes such as HbA_1c levels and diabetes incidence, primarily due to the limited number of available studies (n=3‐4) for these outcomes. Variability in study designs—including differences in intervention duration, SMS text messaging content, and follow-up periods—also posed challenges in synthesizing and comparing results across studies. Due to the small number of studies for certain outcomes, conducting meta-regression analyses and subgroup analyses to adjust for these sources of heterogeneity was not feasible, potentially impacting the reliability of pooled estimates. Assessment of funnel plot asymmetry for publication bias was also precluded due to the limited number of studies.

Furthermore, this meta-analysis did not exclude studies that integrated additional mHealth components, such as mobile apps, alongside SMS text messaging. The inclusion of multicomponent interventions may have introduced bias, limiting the ability to isolate the independent effects of SMS text messaging on clinical outcomes in prediabetes. This review is also limited by its focus on BMI, which may require a longer follow-up duration to detect meaningful changes. Other measures of body composition, such as fat distribution or muscle mass, were not assessed.

In summary, this systematic review found that lifestyle modification programs using SMS text messaging did not produce significant improvements in BMI among adults with prediabetes compared with standard care. The limited effectiveness observed may reflect variations in intervention design, including content, frequency, and delivery mode, highlighting the need for more optimized, theory-driven messaging strategies.

Future research should focus on developing standardized, tailored, and interactive SMS text messaging interventions with larger sample sizes and longer follow-up periods. Isolating SMS text messaging as a primary intervention component will also be critical to better determine its role in prediabetes management and diabetes prevention efforts.