A Meta-Analysis of the Effectiveness of Telemedicine in Glycemic Management among Patients with Type 2 Diabetes in Primary Care

Introduction: Telemedicine interventions are gradually being used in primary health care to help patients with type 2 diabetes receive ongoing medical guidance. The purpose of this study was to analyze the effectiveness of using telemedicine in primary health care for the management of patients with type 2 diabetes. Methods: A systematic search was conducted from database inception to August 2021 in nine databases, including PubMed, Web of Science, Cochrane Library, EMBASE, EBSCO, CNKI, Wanfang Data, VIP, and CBM. Data extraction and quality assessment were performed for studies that met the inclusion criteria. The meta-analysis was performed using Review Manager 5.4 (Cochrane) and Stata v.16.0SE (College Station, TX, USA). Results: A total of 32 articles were included in this study. Analysis showed a reduction in glycated hemoglobin, fasting glucose, and postprandial glucose after the telemedicine intervention. Systolic blood pressure and self-efficacy improved significantly, but there was no significant improvement in weight, lipid metabolism, or diabetes awareness. Subgroup analysis based on the duration of intervention showed significant improvement in glycated hemoglobin at 6 months of intervention. Conclusions: Telemedicine interventions may help patients with type 2 diabetes to effectively control blood glucose and improve self-management in primary health care. There is only moderate benefit, and the benefit may not be sustained beyond 6 months. However, the evidence for the improvement in lipid metabolism is insufficient and further studies are needed.


Introduction
The International Diabetes Federation 2021 reported that the number of people with diabetes mellitus (DM) has reached 533 million worldwide and is expected to reach 783 million worldwide by 2045. The prevalence rate is as high as 12.2% [1], of which type 2 diabetes (T2DM) [2] accounts for 90%. During the COVID-19 pandemic, the vulnerability of people with diabetes to public health emergencies became even more pronounced, with at least a two-fold increase in the risk of serious illness or mortality [3]. Telemedicine is a branch of e-health that uses communication networks to deliver healthcare services and health education to patients across regions [4]. Telemedicine is a viable alternative for patients seeking medical guidance for a chronic condition such as diabetes that requires repeated consultations with a physician, and without the risk of COVID-19 infection [5]. At the same time, telemedicine can provide patients with opportunities to connect and learn online and expand access to care through self-management and diabetes education [6]. It also allows for the involvement of both patients and the healthcare team to provide solutions to the patient's needs [7]. Globally, the vast majority of people with T2DM are treated in primary health care settings [8]. Primary care is seen as the ideal point for making

Inclusion and Exclusion Criteria
The inclusion criteria for articles were as follows: (a) patients ≥18 years of age with T2DM; (b) since telemedicine is defined as the remote collection of records and transmission of patient data via telecommunications systems to healthcare providers for analysis and decision-making [19], email, smartphone, phone, SMS and text messages, web-based platforms, and hybrid forms (video calls and text messages) were included, without any face-to-face contact by a healthcare practitioner during the intervention; (c) improving one or more areas of diabetes self-management through telehealth; (d) control group using conventional treatment; (e) randomized controlled trial.
Exclusion criteria were (a) duplicate publications, (b) inaccessibility of full texts or extracted data, (c) studies without clinical data (abstract presentations, case reports, review articles, case-control, pilot studies, or study protocols), and (d) special populations such as children and pregnant women with T2DM.

Literature Screening and Data Extraction
Two researchers screened titles and abstracts that met the requirements based on the inclusion criteria, and when disagreements arose, a third researcher was consulted. If applicable, the disagreements were recorded to ensure a final consensus among the three reviewers. The literature searches were managed using EndNote X9 (Thomson ResearchSoft; Stanford, CT, USA).
Two researchers independently extracted relevant data from the included studies using a self-designed form, including information on the author, year of publication, country, intervention method, and outcomes of both protocols in each of the clinical studies. In multiple-group studies, only telemedicine interventions and the conventional treatment groups were selected for comparison. For studies that could not be extracted or for which data were missing, attempts were made to contact the authors to obtain the data.

Methodological Quality Evaluation
The methodological quality of the included studies was independently evaluated by two investigators using the Critical Appraisal Skills Program (CASP) [20]. This list consists of 11 entries with three results: "yes", "unclear", and "no". Disagreements, if any, were resolved through discussion with a third investigator or with the entire team.

Risk Bias Evaluation
Two researchers used Risk of Bias from the Cochrane Handbook for Systematic Review of Interventions 5.1.0 (2019) [21] to independently evaluate risk bias in the included studies. Bias included random sequence generation, allocation concealment, blinding of subjects and intervention providers, blinding of outcome evaluators, incomplete outcome data, selective reporting, and other sources of bias. Disagreements, if any, were resolved by discussion with a third author or by a discussion with the entire team.

Statistical Analysis
Meta-analysis was conducted using Review Manager 5.4 (Cochrane Collaboration, Oxford, UK) and Stata v. 16.0SE (College Station, TX, USA) in this study. Continuous data were used with a mean and standard deviation as effect measures and estimates (MD) and 95% confidence intervals (CI) were given for each effect measure. We used the chisquare test (p < 0.01, statistically significant difference) and I 2 to assess the heterogeneity between studies. When I 2 < 50%, it indicated that there was no significant heterogeneity among studies, so a fixed-effect model was used; when I 2 > 50%, it indicated that there was significant heterogeneity among studies, and a random-effects model was used. Sensitivity analysis was performed using a case-by-case exclusion approach to assess the stability of the study results. The presence of publication bias was assessed by a funnel plot and Egger and Begg tests when ≥10 studies were included with a two-sided significance level (p < 0.05, difference statistically significant). Subgroup analysis was used to observe the differences between intervention times.

Results
A total of 2460 articles were obtained through the search. After de-duplication, 1885 articles remained. After screening the titles and abstracts, a total of 147 studies were eligible. The PRISMA flow chart depicts the process of the literature screening ( Figure 1).

Characteristics of Included Studies
Among the 32 included studies , 14 were from the United States, 3 from China, 2 from Spain, 2 from Australia, 2 from India, and 1 each from Australia, Iran, the United Kingdom, Slovenia, Sub-Saharan Africa, Canada, Belgium, Norway, Poland, and Italy. The studies ranged from 3 to 24 months of intervention between 2009 and 2021. Telemedicine in these studies used a variety of platforms to communicate and deliver the interventions, including cell phones (17.5%), Internet (17.5%), text messaging (27.5%), apps (20%), glucosemonitoring devices (12.5%), and tablets (5%). Some studies used a health belief model (13.3%), a trans-theoretical model of behavior change (26.7%), motivational interviewing (26.7%), social cognitive theory (13.3%), or chronic disease health management (20%) as part of the intervention. Intervention providers included physicians (18.9%), nurses (32.4%), allied health professionals (health education providers 5.5%; diabetes educators 2.7%; registered dietitians 2.7%; community workers 2.7%; telephone counselors 2.7%), and school research teams (32.4%). In addition, the content of routine care provided varied across studies, but in the majority of these studies, routine care provided general diabetes management. The details are shown in Tables 1 and 2.

Intervention Group Control Group
Stephen

Literature Quality Evaluation and Risk Bias
Through the CASP inventory [20], the methodological quality of the included articles was assessed, and the results are presented in Table 3. Most of the studies were rated as moderate quality. Due to the specificity of the intervention, the vast majority of studies were not blinded to interventionists or the subjects. Table 3. Critical Appraisal Skills Program (CASP) checklist on the included studies. All studies had clear reporting of sequence generation and were therefore considered to be low risk. Because there was no apparent allocation concealment, five [29,37,44,51,52] studies were rated as high risk. Twenty [23][24][25][27][28][29]31,[33][34][35][36][37][44][45][46][47][50][51][52][53] studies were also rated as unclear bias because they were not blinded to interventionists or subjects. One [42] study was rated as high risk because it was not blinded to the outcome evaluator. Two [26,32] studies were rated as high bias risk because of incomplete outcome data. Figures 2 and 3 show the results of the risk bias.

Secondary Results Impact on Fasting Blood Glucose
Eight studies [36,37,[42][43][44]46,47,52] reported the results of fasting blood glucose (FBG). There were 538 cases in the intervention group and 545 cases in the control group, and the duration of intervention ranged from 3 to 24 months. Meta-analysis showed a statistically significant reduction in FBG levels in the intervention group compared to that in the control group (MD: −0.49, 95% CI [−0.86, −0.12], p = 0.01; I 2 = 36%, Figure 6).

Effects on Postprandial Blood Glucose
Four studies [37,41,44,52] reported the results of post-prandial blood glucose (PBG). There were 297 cases in the intervention group and 298 in the control group, and the duration of the intervention ranged from 3 to 6 months. Meta-analysis showed a statistically significant difference in reduced PBG levels in the intervention group compared to that in the control group (MD: −2.06, 95% CI [−2.80, −1.31], p < 0.00001; I 2 = 36%, Figure 7).

Effects on Postprandial Blood Glucose
Four studies [37,41,44,52] reported the results of post-prandial blood glucose (PBG). There were 297 cases in the intervention group and 298 in the control group, and the duration of the intervention ranged from 3 to 6 months. Meta-analysis showed a statistically significant difference in reduced PBG levels in the intervention group compared to that in the control group (MD: −2.06, 95% CI [−2.80, −1.31], p < 0.00001; I 2 = 36%, Figure 7).

Effect on Systolic Blood Pressure
Thirteen studies [25,28,29,34,43,44,[47][48][49][50]53] reported the results of systolic blood pressure. There were 1847 cases in the intervention group and 1853 cases in the control group, and the duration of the intervention ranged from 3 to 24 months. Meta-analysis showed that systolic blood pressure improved in the intervention group compared to that

Effects on Postprandial Blood Glucose
Four studies [37,41,44,52] reported the results of post-prandial blood glucose (PBG). There were 297 cases in the intervention group and 298 in the control group, and the duration of the intervention ranged from 3 to 6 months. Meta-analysis showed a statistically significant difference in reduced PBG levels in the intervention group compared to that in the control group (MD: −2.06, 95% CI [−2.80, −1.31], p < 0.00001; I 2 = 36%, Figure 7).

Discussion
This study confirms the effectiveness of various telemedicine interventions applied in primary health care. Meta-analysis showed that despite high heterogeneity, telemedicine interventions can improve patients' metabolic control more than conventional controls over 6 months, as is supported in other meta-analyses [54][55][56]. However, we did not

Discussion
This study confirms the effectiveness of various telemedicine interventions applied in primary health care. Meta-analysis showed that despite high heterogeneity, telemedicine interventions can improve patients' metabolic control more than conventional controls over 6 months, as is supported in other meta-analyses [54][55][56]. However, we did not find any significant effect on lipid metabolism such as body weight, HDL, TG, or TC, which is different from the results of several previous studies [55,57,58]. This may be because, on the one hand, most of the study interventions focused on the improvement of patients' blood glucose, and fewer studies provided a full range of interventions on patients' diet, exercise, and diabetes knowledge. One the other hand, lipid control in diabetic patients is mainly dependent on the use of drugs [59], so there is no difference between the two.

Blood Sugar Control
The results of the study showed that the telemedicine intervention resulted in a significant decrease in clinical glycated hemoglobin levels, with a mean decrease of 0.1% to 0.32%. HbA1c concentrations effectively reflected the patients' average blood glucose levels over the previous 8-12 weeks. Thus, the reduction in HbA1c concentrations reflects the positive impact of telemedicine on the long-term care of patients with T2DM. In the subgroup analysis regarding duration, the intervention group showed lower HbA1c than the conventional group only when the intervention duration was 6 months. In contrast, there was no significant improvement in glycosylated hemoglobin at short-term (3 months) or long-term (12 months) intervention times. This may be because T2DM is a slow-treating chronic disease and the HbA1c level reflects the average blood glucose over 2-3 months [60]. Therefore, there was no significant improvement in glycated hemoglobin for the short-term intervention, which is consistent with other studies [61,62]. In addition, the small number of studies of short-term interventions included in the articles of this study may be somewhat biased, and hence, more and higher quality studies may be needed for further evaluation. The decline in benefit at 12 months of intervention may be related to the patient's response to the intervention, which is consistent with the results of other studies [63,64]. Therefore, in studies of long-term interventions, patient engagement declines with the duration of the intervention, and continuous monitoring and individualized reminders are important to maintain the continuity of the intervention.

Self-Management
Good self-management can significantly reduce the occurrence and progression of T2DM complications [65]. In this study, we evaluated the patients' self-management abilities based on results of FBG and PBG, and the results showed that the blood glucose level was reduced in the intervention group compared to the control group, as is supported in other meta-analyses [66]. Moreover, significant improvements in self-efficacy occurred in the intervention group. Ten of these studies [22,29,30,33,35,[38][39][40]42,47], combined with theoretical models, improved the self-management ability of patients with T2DM by providing them with knowledge about blood glucose monitoring, diet, and exercise, and by providing them with timely personalized message reminders through the data obtained. However, since the number of each study is limited, it is not possible to determine whether there are interactions between different telemedicine technologies (e.g., text messaging, smartphones). Most research trials reported on glycosylated hemoglobin concentrations, and only a small number of study trials reported on other aspects regarding diabetes management (FBG, postprandial glucose, lipid metabolism, self-management ability, etc.). Therefore, great care should be taken in interpreting the results of other secondary outcomes. The meta-analysis also showed that telemedicine strategies were associated with other clinical benefits of glycemic control and that telemedicine measures contributed to improved self-management abilities, including systolic blood pressure and diabetes self-efficacy, in patients with T2DM.

Telemedicine
In the study by Natalie Robson et al. [67], phone-based telehealth interventions had the greatest impact on the self-management of people with type 2 diabetes. In the study of Puikwan A. Lee et al. [68], although the results of meta-analysis showed that telemedicine intervention was better than usual care, it was impossible to judge which intervention method would benefit patients the most. The results of this study showed that telemedicine intervention was superior to conventional care. In the subgroup analysis of different intervention methods, only remote monitoring and application software intervention significantly improved HbA1c in the intervention group, whereas there were no significant differences in mobile medicine, telephone communication, or SMS remote intervention. Interventions delivered through remote monitoring had the greatest impact on type 2 diabetes-related outcomes, followed by app interventions, M-Health, telephone counseling, and SMS interventions. At the same time, there was significant heterogeneity among individual studies, which may have been caused by the different methods and applications of telemedicine intervention. Therefore, it is difficult to determine which interventions are most effective in managing type 2 diabetes. A total of five [22,29,33,39,48] studies assessed patient engagement or perception of telemedicine in this review. Although these studies reported widespread patient acceptance of telemedicine interventions, only three studies formally tested subjects' satisfaction. Meta-analysis results showed that although the satisfaction of the intervention group was better than that of the control group, the difference was not statistically significant. At the same time, high-frequency and short-term interventions [25,41,47,53] can significantly improve the hemoglobin of patients. However, the effectiveness of the intervention decreases with the duration of the intervention, so the next step that should be considered is how to maintain a high level of patient compliance during prolonged intervention. Only one study [48] reported the costs associated with the intervention and no cost-benefit analysis was performed, making it impossible to draw firm conclusions about the cost-effectiveness of telemedicine interventions in type 2 diabetes.
In response to the ongoing COVID-19 pandemic, telemedicine has seen incremental growth [69], with one study [44] reporting improved patient glycemic control during the COVID-19 pandemic with telemedicine-based patient interventions using an app. We believe this number is too small to draw firm conclusions about telemedicine in diabetes care in the context of COVID-19. Although telemedicine can provide benefits to patients, many problems remain in its future development. For example, the professional and ethical challenges posed by telemedicine may affect both patients and physicians [70]. Patients' trust in telemedicine interventions may be reduced due to untimely phone calls or incorrect text messages. Moreover, patient compliance decreased with the prolongation of intervention time. In addition, in most countries the cost of telehealth interventions is borne by the patient, which can lead to low patient acceptance. Therefore, future research should explore a convenient and timely telemedicine system, improve the quality of intervention, and increase patient acceptance through the training of intervention teams, as well as maintain a certain frequency of intervention to increase patient compliance, ensure government funding to reduce telemedicine costs, and accelerate telemedicine adoption.

Limitations and Strengths
First, the study found high heterogeneity in some outcomes, and the high heterogeneity may have been due to differences in telemedicine interventions, subjects, the timing of interventions, interventionists, and intervention frequency. Second, because we found a small number of studies reporting diabetes self-management ability, the findings are inconclusive regarding the effectiveness of self-management and require large sample testing and further investigation. Third, there are fewer studies with long-term follow-up, and the findings may contain only short-term effects of improving T2DM and lack evaluation of long-term effects (development of complications, mortality, etc.). In addition, the cost-effectiveness of the intervention becomes uncertain as the duration of the intervention increases. Fourth, the nature of the intervention led to the fact that most of the studies were not blinded to the interventionists or subjects. Fifth, the inclusion of studies from different countries resulted in significant differences in baseline primary health care systems. For example, government-funded telemedicine services are common in primary health care in the United States and Europe [71], but in some developing countries, especially South Africa, even basic services are not available due to low government budgets [72]. This may affect the compliance of patients. Sixth, because the term "telemedicine" is very nonspecific, conclusions may not be generalizable. Finally, we identified slight missing corners in some of the funnel plots, which may have been caused by the presence of small sample sizes and low-quality studies in our included studies. However, the Egger and Begg tests revealed that there was no significant publication bias in the included studies, and sensitivity analyses did not show any significant change in the effect values of the study results. Therefore, this does not affect our conclusions.
This meta-analysis has multiple strengths. First, we assessed the quality of evidence for all included articles using the CASP [20] method. Second, we did not restrict the search language, so articles in all countries and languages were eligible for this study. Third, we strictly followed the recommendations of the meta-analysis evaluation method [73][74][75]. Fourth, we had no time limit on previous published articles, so no earlier published articles were missed.
The meta-analysis of this study supports that telemedicine interventions can be effective in improving glycemic control in patients with T2DM in primary health care. This information is critical for decision-makers to develop interventions based on desired outcomes and to maximize the use of resources in the primary health care setting.

Conclusions
The results of this study suggest that compared with non-telemedicine care, telemedicine intervention in primary health care may improve glycemic control and self-management in patients with type 2 diabetes. In particular, the six-month intervention period was most effective, and interventions with remote monitoring or apps were most effective. Therefore, telemedicine interventions are an effective supplement to traditional face-to-face counseling in primary health care settings, and telemedicine can provide convenience for homebound patients with T2DM who do not have timely access to medical resources. The next step should focus on assessing the acceptability and feasibility of telemedicine implementation on a large scale and whether it is effective in reducing healthcare costs.
Supplementary Materials: The following supporting information can be downloaded at: https: //www.mdpi.com/article/10.3390/ijerph19074173/s1, Figure S1: Funnel plot evaluating publication bias with regard to HbA1c change for all studies; Figure S2: Funnel plot evaluating publication bias with regard to wieght change for all studies; Figure S3. Funnel plot evaluating publication bias with regard to SBP change for all studies; Figure S4. Funnel plot evaluating publication bias with regard to HDL change for all studies; Figure S5. Funnel plot evaluating publication bias with regard to LDL change for all studies Author Contributions: A.Z. and J.W. contributed to study concept and design; A.Z. and X.W. contributed to the data extraction; A.Z., S.Z. and Z.Z. screened the articles; A.Z. and J.W. discussed the themes of the review and contributed to the analysis and interpretation of the data; A.Z. contributed to the drafting of the manuscript; J.W., Z.G. and X.W. contributed to the critical revision of the manuscript for important intellectual content; C.W. contributed to the language editing. All authors have read and agreed to the published version of the manuscript.