The evaluation protocol was prospectively registered in PROSPERO (International Prospective Register of Systematic Reviews; CRD42024523025). We conducted a systematic review and meta-analysis following the guidelines outlined in the Cochrane Handbook [18] and adhered to the 2020 PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) guidelines for systematic review reporting [19].

We conducted a comprehensive systematic search across 5 databases: PubMed, Web of Science, CINAHL, Embase, and Cochrane. Our search included all forms of electronic communication, such as virtual reality, augmented reality, artificial intelligence, smartphone apps, TM, and SMS text messages. The search covered all records from the inception of each database to September 26, 2024. The complete search strategy for each database is detailed in Multimedia Appendices 1-5. Additionally, we manually searched the reference lists of retrieved articles.

The inclusion criteria, defined using the PICOS (Patients, Implementation, Comparison, Outcomes, Study) format, were as follows: (1) patients diagnosed with type 1 diabetes; (2) implementation of the CC method via TM; (3) studies categorizing participants into experimental and control groups, including both usual and standard care (comparison); (4) HbA_1c levels (outcomes); and (5) RCTs (study).

The exclusion criteria were as follows: (1) studies that were incomplete, including research protocols and ongoing studies; (2) reports lacking sufficient details on patient outcome measures; and (3) studies with insufficient statistical data for quantitative outcomes, such as mean, SD, and median with range.

First, the literature search results from all retrieved databases were exported into EndNote X9 (Clarivate Analytics) for management and duplicate removal. Two researchers (YL and YY) independently screened titles and abstracts, followed by a full-text review to confirm eligibility. A third researcher (FL) resolved any disagreements or discrepancies through deliberation until a consensus was reached.

Two researchers (YL and YY) independently extracted data from the 10 articles, and another researcher (FL) verified the extracted information. Discrepancies were resolved through discussion. If data were unavailable, the authors were contacted.

We extracted trial characteristics (study name, author, year, country, number of centers, design, duration, and sample size) and patient characteristics (age, sex, and duration of diabetes) from the selected studies. Additionally, we extracted details of the TM intervention (type, duration, and other specifics) and results (mean, SD, SE, 95% CIs, statistical significance at follow-up time points, primary and secondary outcomes, and validated measurement instruments).

The primary outcome was the HbA_1c level. Secondary outcomes included treatment satisfaction (measured using a validated tool), total daily insulin dose (TDD), and time in range (TIR).

Two reviewers independently assessed the risk of bias in RCTs using the revised Cochrane Risk of Bias (RoB) 2 tool and calculated the weighted Cohen κ coefficient to measure agreement [20]. In case of disagreement, a third reviewer (FL) facilitated discussions to reach a consensus. The RoB 2 tool evaluates 5 key areas of potential bias in RCTs: the randomization process, deviations from intended interventions, missing outcome data, outcome measurement, and selective reporting of outcomes. Each area includes a series of questions with 3 response options—“low,” “some concern,” and “high”—which classify the level of bias risk. Based on these classifications, studies are categorized as having “low,” “some concern,” or “high” overall risk of bias. The assessments in each area contribute to the overall judgment of bias risk in the study results. If no bias is identified in any area, the overall risk is deemed low. If at least one area raises some concern, the overall risk is categorized as “some concern.” If a high risk of bias is found in any area, the study is considered to have a high overall risk of bias [20].

We assessed publication bias using the Egger test [21] and examined contour-enhanced funnel plots [22]. This is particularly important in meta-analyses, as studies with positive and significant results are more likely to be published in high-impact journals compared with those with negative findings [23].

Statistical analysis was performed using the metafor package in R 4.4.1 software (R Foundation) [24]. For HbA_1c and TIR, the mean difference (MD) and 95% CI were calculated. Given that treatment satisfaction was measured using different validation scales, and TDD used other measurement methods and units, we calculated the standardized mean difference (SMD) and 95% CIs. When studies reported baseline and follow-up values but not change-from-baseline SDs, the missing SDs were calculated based on the baseline and follow-up SDs, and the average correlation coefficient (r) was estimated from other identified studies using the following formula: If the SE was reported instead of the SD, it was converted to SD using the following formula: where n is the sample size. By contrast, if a 95% CI was reported instead of SD or SE, the SD can be calculated according to the Cochrane Manual [25]. Studies that did not report SDs, SEs, or 95% CIs were excluded from meta-analysis [25].

Given the expected heterogeneity in study populations and procedures, a random-effects model was used to combine the effect sizes and SDs of the studies. Heterogeneity was assessed by examining the forest plot and calculating the degree of inconsistency (I²) among studies. I² represents the proportion of variability in study results attributable to heterogeneity rather than chance [26]. An I² value of 0%-40% indicated low heterogeneity, 30%-60% indicated moderate heterogeneity, 50%-90% indicated substantial heterogeneity, and 75%-100% indicated considerable heterogeneity. To assess the influence of individual studies, we conducted a leave-one-out analysis, systematically excluding each study and reanalyzing the data set to determine whether any single study disproportionately affected the results. Sensitivity analyses were performed to evaluate the robustness of the findings. Additionally, subgroup analyses based on the type of intervention were conducted to explore potential sources of heterogeneity. Finally, meta-regression was used to examine the impact of demographic and intervention characteristics on variations in HbA_1c levels, aiming to identify underlying sources of heterogeneity.

The initial database search yielded 3612 records, with an additional 10 identified through reference list screening. After removing 185 duplicates, 3437 abstracts were screened, of which 3317 were excluded for not meeting the selection criteria, and 2 lacked full-text access. Subsequently, 118 full-text articles were assessed, with exclusions detailed in Figure 1. Ultimately, 19 articles met the inclusion criteria and were included in the final analysis. One of these articles reported 2 distinct intervention groups within a 3-arm trial, comparing both intervention groups against a control, resulting in the inclusion of 20 trials in the meta-analysis [27-45]. The interrater agreement between evaluators (YL and YY) was strong, with a κ value of 0.91.

Table 1 summarizes the characteristics of the 20 included studies, encompassing 1627 participants from 14 regions. The publication years ranged from 2004 to 2023, with over 50% (12/20, 60%) published after 2018. Geographically, the studies were conducted in Europe (n=12), America (n=6), and Asia (n=2). All studies were RCTs, with 17 applying a parallel-group design and 3 utilizing a crossover design. Regarding the study population, 7 focused on children and adolescents with T1DM, 11 on adults with T1DM, and 2 included participants of all ages with T1DM.

CC interventions using different TM models exhibited considerable variability in glycemic control outcomes among patients with T1DM (Table 2). The interventions were categorized into 3 groups based on their content: smartphone apps (n=13), web-based systems (n=3), and connected or wearable glucometers (n=4).

Currently, smartphone apps are the most widely used, encompassing various types. These include the iSpy mobile app, OneTouch Reveal, Young with Diabetes mHealth app, Webdia mHealth app, GLIC APP, MySugr, Euglia, Social Diabetes, and the “artificial pancreas.” The web-based category comprises integrated network systems such as the ELKa system, cloud-based platforms, and computer programs accessible via the internet. Internet-connected glucometers, such as the Accu-Chek Connect by Roche, integrate self-monitoring of BG results with additional features, including an insulin calculator and a food diary. These tools assist users in calculating carbohydrate intake and managing BG levels more effectively while enabling clinicians to review accurate BG patterns for treatment adjustments [46].

Figure 2 presents the risk of biased outcomes for the 19 included articles, highlighting variability in study quality. Of these, 7 articles (37%) exhibited a low risk of bias, 11 (58%) demonstrated some risk of bias, and 1 (5%) had a high risk of bias. The high risk of bias was primarily attributed to inadequate descriptions of randomization methods, the absence of participant and researcher blinding during intervention allocation, and deviations from the planned intervention. Given the impracticality of blinding participants in TM interventions, all trials were conducted using an open-label design, resulting in at least one identified risk of bias per study. Our assessment revealed that 15 out of 19 articles (79%) adequately reported and described appropriate randomization methods, while 13 (68%) effectively communicated assignment concealment procedures. Additionally, 11 studies (58%) adhered to the intention-to-treat principle. As shown in Table 3, interrater agreement for risk of bias assessments was high, with Cohen κ values ranging from 0.642 to 1.00 across domains. A risk of bias summary is presented in Figure 3.

Egger test revealed significant asymmetry in study distribution (P=.003). This asymmetry was also evident in the contour-enhanced funnel plot of HbA_1c (Figure 4; see also [27-45]), suggesting a degree of publication bias consistent with the Egger test results. In the profile-enhanced funnel plot after applying the trim-and-fill method (Figure 5; see also [27-45]), 9 additional studies (represented by modest circles) were required to correct the observed asymmetry. While some studies fell within the nonsignificant region (white areas), indicating that no significant studies were left unpublished, others appeared in the statistically significant region (gray area), suggesting that certain significant studies may not have been published. This implies that factors beyond publication bias may have contributed to the observed funnel plot asymmetry for HbA_1c.

Three studies [28,33,39] did not mention hypoglycemia, and inconsistencies in its definition and reporting prevented a meta-analysis. Eight studies [27,32,34,40-44] defined hypoglycemia based on objective BG values, with cut-off thresholds ranging from <3.9 mmol/L to <2.5 mmol/L. Meanwhile, 7 studies [29-31,35-38] defined hypoglycemic events as those requiring assistance from another person.

This meta-analysis of RCTs evaluating TM-based CC interventions for BG control in patients with T1DM included data from 20 studies completed on September 26, 2024. TM-based CC interventions significantly reduced HbA_1c levels by –0.35% (95% CI –0.54% to –0.16%) and improved glycemic control compared with controls. Considerable heterogeneity was observed among trials (I²=81%, P<.01), which may have led to an underestimation of the intervention effect and contributed to the asymmetry in the funnel plot, potentially due to factors other than publication bias. TM-based CC interventions significantly improved TIR (9.59%, 95% CI 6.50%-12.67%). However, no significant differences were found in hypoglycemia, treatment satisfaction, or TDD, possibly due to the short duration of the included trials.

The included studies were categorized based on the type of TM intervention: smartphone apps, connected and wearable glucometers, and web-based systems. Subgroup analysis showed that all intervention types had statistically significant effects on HbA_1c levels. Meta-regression analysis indicated that the study region was significantly associated with changes in HbA_1c levels. In terms of average HbA_1c reduction, Asian participants appeared to benefit more than North American and European populations. However, this conclusion should be interpreted with caution, as only 2 studies from Asia were included, compared with 12 from Europe and 6 from North America.

This study has several strengths. First, our research strategy involved an extensive search across multiple databases, enhancing the comprehensiveness of the review. Second, we adhered to rigorous systematic review and meta-analysis standards following PRISMA (Multimedia Appendix 6) and Cochrane guidelines. Third, we conducted comprehensive sensitivity analyses to ensure the robustness of our findings. Finally, subgroup and meta-regression analyses were performed to validate our results.

This study has several limitations. The primary limitation is the considerable heterogeneity among the included studies. Despite conducting subgroup analyses and meta-regression to explore potential sources of heterogeneity, the results should be interpreted with caution due to the influence of uncontrolled or unmeasured factors. Additionally, our review excluded RCT registries, ongoing studies, and gray literature. While the inclusion of gray literature in systematic reviews remains debated [57], limiting the analysis to published studies may introduce publication bias. Furthermore, we restricted our search to English-language publications, which may affect the generalizability of our findings. Another limitation is the potential subjectivity in bias assessments using the RoB2 tool. Finally, many trials had small sample sizes, short durations, and lacked blinding. Most studies were conducted in developed countries, reflecting the limited availability and feasibility of technology-based interventions in low-income settings.

Our findings have several practical implications. First, remote medical interventions for CC demonstrate beneficial effects on glycemic control in patients with T1DM. Thus, we recommend incorporating TM-based CC interventions into long-term glucose management strategies. However, successful implementation requires consideration of factors such as patient learning ability, adaptability, acceptance of technology, and the level of medical team support. As the time needed to achieve independent glucose management varies among individuals, long-term studies are warranted to further assess these interventions [55].

Recent cross-sectional research on T1DM indicates that the likelihood of diabetic retinopathy increases with rising HbA_1c levels. The United Kingdom Prospective Diabetes Study (UKPDS) [58] found that a 1% reduction in average HbA_1c was associated with a 21% reduction in diabetes-related deaths, a 14% reduction in the risk of myocardial infarction, and a 37% reduction in microvascular complications in patients with T2DM. If remote medical care–based CC interventions were implemented universally for patients with T1DM, they could potentially reduce diabetes-related deaths by 7.4%, myocardial infarction risk by 4.9%, and microvascular complications by 13%. Such interventions may also improve glycemic control, lower the risk of macrovascular and microvascular complications, and enhance quality of life. However, other important outcomes, including hypoglycemia incidence, TIR, quality of life, and self-management behaviors, remain understudied and require further evaluation. Additional research should also focus on the needs and characteristics of special populations, such as children and older adults, to enhance acceptance and effectiveness [46].

Given these findings, our results suggest that remote medical care–based CC interventions hold significant potential for clinical application. Future efforts should focus on advancing technological innovation to develop more intelligent and user-friendly remote medical devices and apps, such as integrating artificial intelligence to provide personalized treatment recommendations. Additionally, strengthening international collaboration is essential to facilitate global implementation, particularly in low-income countries and remote areas. Encouraging patients to integrate these interventions into their daily lives and work could further enhance glycemic management in T1DM. Future research should prioritize evaluating the long-term effects and cost-effectiveness of remote medical interventions in diabetes management. Moreover, exploring their combined use with other diabetes management strategies could provide further insights. Large-scale, multicenter, and long-term follow-up clinical trials are necessary to assess their efficacy and safety in glycemic control for patients with T1DM. These findings would offer critical evidence to support policy makers in promoting and expanding the use of remote medical care in diabetes management.

Our systematic review and meta-analysis suggest that TM-based CC interventions can be effective for glycemic control in patients with T1DM in RCTs. However, a robust, large-scale trial is needed to draw definitive conclusions. These findings may inform the development of new strategies to enhance T1DM management.