The literature search was conducted in the PubMed, Embase, and Cochrane Library databases for relevant studies without time period restriction, and only studies written in English were included. Medical Subject Headings (MeSH) terms were used as follows: “smartphone apps,” “type 1 diabetes mellitus,” and “randomized controlled trials.” Meanwhile, manual searching was performed to find further relevant studies. The final electronic database search was performed on 18 February 2023.

We regarded studies as eligible for inclusion if they were randomized clinical trials performed among children, adolescents, or young adults (ages ranging from 0 to 24 years old) with T1D, studies including compared treatment involving smartphone apps and usual care to treatment with usual diabetes care only, had at least 4 weeks’ duration of intervention, and reported changes in HbA1c. Exclusion criteria were as follows: observational and retrospective studies, studies with less than 4 weeks duration of intervention, studies that did not assess smartphone apps, studies with participants aged over 24 years old, studies with participants diagnosed with other types of diabetes, and studies with incomplete data.

We extracted the following data from each selected study: research designs, population demographics, trial duration, sample size, study duration, a summary of interventions, main outcomes, and findings. Two independent investigators (LP and XS) performed the literature search, study selection, data extraction, and quality assessment. Disagreements were resolved by a third investigator (ZZ).

The primary outcomes were the mean change in HbA1c from baseline to study end (described as weighted mean difference, WMD) and the person-year incidence of mild and severe hypoglycemia between the smartphone app-usual care combination treatment group (intervention group) and the usual care-only group (control group). WMD was calculated using the inverse variance random-effects model. The overall effect size with 95% confidence intervals (CIs) was used to evaluate smartphone apps for glycemic control, and a value of p of <0.05 was considered statistically significant.

We assessed the possibility of publication bias by constructing a funnel plot of each trial’s effect size against the standard error. The “one study removed” approach was performed to evaluate the influence of each study on the overall pooled estimate. We used the Cochrane Q-test to assess heterogeneity between studies. We also performed I² testing to assess the magnitude of the heterogeneity between studies, with values greater than 50% regarded as being indicative of moderate-to-high heterogeneity. Subgroup analysis was conducted to explore the source of heterogeneity, including smartphone app style and the duration of trials. We used RevMan (version 5.3) for all statistical analyses.

A total of 1,190 studies were initially identified from PubMed, Embase, and the Cochrane Library. Then, 332 studies were removed because of duplications. After further screening based on title and abstract, 858 studies were excluded due to the lack of relevance to our topic or other reasons. A total of 43 studies were eligible for subsequent full-text review, and nine studies were ultimately included in the meta-analysis. The detailed process of the literature selection is illustrated in Figure 1.

All nine included studies were randomized controlled trials published between 2014 and 2021 (most in the last 5 years except for one in 2014) (Table 1). Sample sizes ranged from 32 to 168 among studies. Study durations varied from 4 weeks to 12 months (five trials with a duration ≥6 months). Participants were located in Canada, the United States, Germany, Denmark, Greece, Switzerland, Italy, and China. The mean age ranged from 12.65 to 17.8 years, and the maximum age was 24 years.

The smartphone apps included in our analysis were complex in nature. They were mainly divided into two categories: auxiliary style and recording style. The former, which applied to three trials, was designed to assist in calculating carbohydrate content or insulin bolus; the latter, which applied to the other six trials, was designed to have one or more functions as follows: collecting biodata, tracking patterns or trends in diabetes management, self-monitoring, diabetes education, and social support in Table 2.

The risk of bias was assessed by two independent reviewers with consultation from a third reviewer if necessary and is presented in Figure 2. All nine included studies showed overall fair levels of randomization. However, in three studies, there was no report of random sequence generation, and in four studies, allocation concealment was not reported.

We defined the intervention and control groups as receiving smartphone apps and usual care combination treatment (n = 390) and only usual diabetes care (n = 358), respectively. The pooled analysis of nine included studies showed a non-significant reduction in HbA1c (WMD = −0.26, 95% CI: −0.56 to 0.05; p = 0.10) in the intervention group compared to the control group. Meanwhile, there was a moderate level of heterogeneity in the overall pooled effect (I² = 69%) (Figure 3).

Subgroup analysis was conducted to explore whether smartphone app style could affect glycemic control; auxiliary-style apps and recording-style apps were described more fully in previous studies. Pooled analysis of three studies showed that auxiliary-style apps yielded a significant reduction in HbA1c (WMD = −0.83, 95% CI: −1.10 to −0.56; P < 0.001, I² = 0%) in the intervention group compared to the control group. The pooled analysis of six studies showed that recording-style apps exerted a similar effect on the reduction in HbA1c between the intervention group and the control group (WMD = −0.03, 95% CI: −0.23 to 0.16; p = 0.73, I² = 0%) (Figure 4).

Another subgroup analysis was conducted to explore whether smartphone app intervention duration could affect glycemic control. The pooled analysis of five or more studies with intervention duration >6 months and four studies with intervention duration <6 months showed a non-significant reduction in HbA1c compared to their corresponding control groups (WMD = −0.27, 95% CI: −0.57 to 0.03, p = 0.08, I² = 11%) and (WMD = −0.24, 95% CI: −0.77 to 0.29, p = 0.37, I² = 86%), respectively (Figure 5).

The heterogeneity assessment indicated a moderate level of heterogeneity in the overall pooled effect (I² = 69%). The “one study removed” approach was used to assess the heterogeneity, and the removal of the Chatzakis et al. study (22) led to a reduced HbA1c change from baseline to study end between the intervention group and control group. However, it did not alter the final result. Meanwhile, the value of I² became 9% (Figure 6).

Notably, six (21–23, 25–27) of nine included studies reported the risk of hypoglycemia during the intervention period. All six studies demonstrated that using smartphone apps did not increase the risk of hypoglycemia. Pooled data from two studies (n = 132 participants) showed no increased frequency of mild hypoglycemia of 1.87 (95% CI: −1.52 to 5.27, p = 0.28, I² = 0%) (Figure 7A) or severe hypoglycemia of −0.04 (95% CI: −0.35 to 0.27, p = 0.80, I² = 0%) (Figure 7B) in the intervention group compared to the control group.

We assessed the possibility of publication bias by constructing a funnel plot from the analyses of the effect on glycemic control. As shown in Figure 8, more than half of the included studies were distributed around the top of the funnel plot, and it was relatively symmetric overall. Therefore, there was evidently no publication bias.

A total of nine randomized controlled studies of fair quality were included in our study. The results showed that, compared with the control group, using smartphone apps can yield a trend in HbA1c from baseline to study reduction without an increased frequency of hypoglycemia, although not statistically significant. Furthermore, in the subgroup analysis to assess the effect of app style, compared with recording-style apps, auxiliary-style apps yielded a significant reduction in HbA1c. Meanwhile, in the subgroup analysis to assess the effect of intervention duration, it was found that compared to an intervention duration of less than 6 months, an intervention duration of more than 6 months was associated with a similar reduction in HbA1c.

Achieving glycemic control plays a vital role in avoiding late longstanding complications, which is also a great challenge in T1D management. Patients with T1D are prone to require lifelong comprehensive management, which includes insulin therapy, metabolic monitoring, nutritional guidance, diabetes education, lifestyle management, and psychosocial care (28). Good adherence to the diabetes regimen would help to meet the target of metabolic control (29). Only 14% of youths with T1D, however, had met the glycemic control recommendation from the American Diabetes Association (29, 30), which might be attributed to the special period of maturation—the transition from childhood to adulthood—that involves a more important role of self-management and reduced engagement with caregivers (31, 32). In the context of more than 2.7 billion individuals in the world using smartphones (8) and approximately 0.5 billion people already using mobile apps for diet (9), physical activity (10), and chronic disease management (11–14), smartphone apps, a newly emerging technology, have demonstrated enormous potential to provide an effective tool aiding self-management of T1D. Different types of smartphone apps have been developed to assist self-management of T1D among youth, such as “iSpy” (19) and “Euglyca” (22). These are intended to improve glycemic control by encouraging glucose monitoring, assisting data collection, coaching people with diabetes, guiding healthy nutrition and medication dosing, and maintaining lifestyle modifications (33). Some small-scale trials of smartphone apps targeting glucose medication adherence, weight loss, and quality of life have shown promising results (34, 35).

A previous study indicated a consistent reduction in HbA1c in patients with type 2 diabetes mellitus, while in patients with T1D, the result was inconclusive (15). Patients with T1D receive advertisements regarding many kinds of apps to improve their treatment. It is important to know whether they are also useful in improving glycemic balance. The results of the present study suggest that apps with functional modularity containing carbohydrate counting or insulin dosage adjustment provide a generally positive improvement in HbA1c among young people with T1D. Nearly all individuals with T1D receive insulin, and insulin therapy is essential for optimal glucose control in patients with T1D (36, 37). The other pillar of T1D management is maintaining a healthy diet. Accurate counting of carbohydrates is paramount to the appropriate dosage of insulin, and regulating food intake and counting carbohydrates have a positive effect on metabolic control in children and adolescents with T1D (38). It is plausible that diabetes apps with functional modularity containing carbohydrate counting or insulin dosage are more effective. These characteristics deserve more consideration in future smartphone app design. While some smartphone apps have recording functions, the results regarding such functionality were inconsistent; further studies are needed to explore the effectiveness of this kind of smartphone app.

In addition, our results demonstrated no association between follow-up duration and reductions in HbA1c. This finding resonates with previous literature (39, 40). Lifelong self-management is needed for all age groups of patients with T1D; however, the longest follow-up duration in our study was 12 months, and the long-term effects of smartphone apps are still largely unknown.

Safety is one of the largest issues that limit the implementation of diabetes apps. Two studies compared the incidence of hypoglycemia between the smartphone app group and the control group, and the other studies reported the safety of the apps with descriptive words. Overall, smartphone apps were safe and associated with no increase in the number of hypoglycemia episodes.

The strengths of the analysis include a sensitive search to identify eligible trials and independent study identification, selection, data extraction, and adjudication of risk of bias by two reviewers. The limitations are as follows: (1) the sample size of individual studies, ranging from 16 to 86 in the intervention group, was relatively small, although all included trials were randomized controlled trials; (2) the optimal intervention duration was unknown, as lifelong self-management is needed for all age groups of patients with T1D; (3) as the included studies were mainly conducted in Europe (6/9), North America (2/9), and Asia (1/9), geographical or ethnic disparities should be considered for underlying bias; and (4) the overall heterogeneity was high, with an I² of 69%. When we removed the study of Chatzakis et al. (22), the value of I² decreased to 9%, which indicated that the high heterogeneity is mainly caused by this article with a great contribution. Meanwhile, a reduced HbA1c change from baseline to study end was found between the intervention and control groups, but it did not alter the final result.