Mapping the Evidence on the Effectiveness of Telemedicine Interventions in Diabetes, Dyslipidemia, and Hypertension: An Umbrella Review of Systematic Reviews and Meta-Analyses

1Diabetes is affecting 463 million people worldwide (aged between 20 and 79 years). Hypertension and hyperlipidemia (or hypercholesterolemia) are common comorbidities in patients with type 2 diabetes (T2D), and also show an increasing coprevalence. The risk of diabetes-associated complications can be reduced by continuous control of blood glucose, blood pressure (BP) lowering, and blood lipid profile. Current guidelines of the American Diabetes Association (ADA) recommend that most adults with diabetes achieve glycated hemoglobin (HbA1c) <7.0%, BP<140/90 mmHg (<130/90 for patients with increased cardiovascular [CV] risk), and low-density lipoprotein cholesterol (LDL-c) <100 mg/dL. Diabetes self-management education and support, defined as an interactive and continuous process intended to increase knowledge, skills, and abilities required for successful self-management of diabetes interventions, has proven to be effective. Similarly, hypertensive patients may benefit from the combination of self-monitoring with education or counseling in terms of increased medication adherence and improved BP control.

2The application of information and communication technologies (ICTs) in health care has been rapidly increasing worldwide. Telemedicine is defined by three characteristics: (1) using ICTs, (2) covering a geographical distance, and (3) involving professionals who deliver care directly to a patient or a group of patients. Owing to the need for individualized and continuous monitoring and self-management support for patients, chronic diseases are considered the ideal target conditions for the development and implementation of telemedicine approaches.

3However, detailed guidance is still lacking on how to choose and integrate tools for specific target groups in diabetes care. Earlier systematic reviews of high-quality review articles already uncovered key elements for technology-enabled self-management, such as (1) communication between a health care provider and patient, (2) patient-generated health data, (3) education, and (4) feedback, or they simply underlined the promising nature of telemedicine. However, the available overviews mainly focus on a specific target group, do not take into account the heterogeneity of telemedicine applications, or focus on a specific tool. Heterogeneous applications of the term telemedicine further limit the external validity of single studies. Owing to the differentiated phenotypes of applied telemedicine solutions, their components, and settings, as well as missing analyses of the quality of studies (and certainty of effects), evidence-based guidance on the best available digital intervention is challenging.

4Therefore, the primary objective of this umbrella review is to identify, synthesize, and critically appraise the evidence on the effectiveness of telemedicine solutions and their components on clinical outcomes—HbA1c, high-density lipoprotein (HDL), low-density lipoprotein (LDL), total cholesterol (TC), triglycerides (TGC), systolic BP (SBP), diastolic BP (DBP)—in patients with diabetes (type 1 diabetes [T1D] and T2D), hypertension, or dyslipidemia. Owing to the increasing number of available reviews and meta-analysis as well as the potential of addressing three prevalent chronic conditions with multiple digital interventions, the analysis was conducted as an umbrella review.

5The research question is based on the Population, Intervention, Control, Outcome, and Time (PICOT) criteria: In patients with diabetes, hypertension or dyslipidemia, what is the evidence for the effectiveness of telemedicine-supported chronic care on disease-specific clinical outcomes?

6We conducted an umbrella review using extensive automated and manual searches of the databases PubMed, EMBASE, and the Cochrane Library to identify relevant evidence on the effectiveness of telemedicine interventions on the three target diseases. Umbrella reviews summarize and contrast evidence from existing systematic reviews and meta-analyses by looking at specific outcomes across included records.

7The search was carried out in October 2018. PICOT-criteria (Table 1) for “population,” “intervention,” and “study design” were combined to develop the search strings (Multimedia Appendix 1). No time limitation was applied.

8Records that fulfilled the following eligibility criteria were included (Table 1): systematic reviews or meta-analyses of randomized controlled trials (RCTs; as this is regarded as highest level of evidence) evaluating the effectiveness of telemedicine in at least one of the target diseases (diabetes, hypertension, and/or dyslipidemia) in adults. No restrictions were made with respect to the kind of participating medical providers. We included all eligible articles in English language and with full text available.

9Relevant reviews or meta-analyses were excluded if their primary studies mainly assessed mortality, utilization of health services, the usability of the technology studied, or patients’ acceptance of or satisfaction with the telemedicine tools, or if no quantitative comparison based on clinical outcomes was reported. Studies evaluating interventions using automated feedback without involving a professional or those providing only monitoring of relevant parameters (without feedback) were excluded. In addition, studies evaluating telemedicine use of medical providers only or those in which the components of the intervention were not transparently described were excluded. Eligible records had to report a change in one of the specified clinical outcomes after a follow-up time of at least three months, as this period is in line with current treatment guidelines.

10Conference abstracts or protocols were excluded as well. Research was excluded if it focused on specific countries or regions or targeted specified populations (eg, minorities and pregnant women with diabetes). We excluded those studies for which updates of the evidence—indicated by the same group of authors and/or application of identical search string—were available.

11We further searched the reference lists of all relevant publications by hand, to identify any additional studies. After carrying out the title-abstract screening, we conducted a hand search in Google Scholar and the three most relevant journals in the field of digital health, as indicated by the highest number of potentially relevant publications (Multimedia Appendix 2).

12Two authors (PT and LH) independently screened the records, extracted data, and carried out the quality assessments. The quality assessment of records was done using the Oxford Quality Assessment Questionnaire (OQAQ) to eliminate records of low quality before data extraction. Any disagreement over the suitability of certain records was discussed among the raters and resolved by consensus.

13As the Grading of Recommendations, Assessment, Development, and Evaluation (GRADE) is the established tool for assessing the overall certainty of evidence by analyzing its risk of bias, imprecision, inconsistency, indirectness, and publication bias, it was used to assess the quality of included records. This assessment was performed by three independent researchers (PT, SO, and LH), using independent pairwise ratings. Disagreements were again resolved by discussion or, where not possible, by consulting the independent third coder.

14The results of the included records were extracted using a piloted, standardized data extraction form. According to the methodological considerations for conducting umbrella or meta-reviews, the results were reported descriptively and in tabular form.

15The presentation of data is descriptive; however, the results of meta-analyses and subgroup analyses were specifically analyzed to find effective components or modes of delivery (intensity and frequency) in subgroups or settings. In light of previous trials, a clinically relevant reduction of –0.5% in HbA1c is considered a suitable threshold (Table 2). The definition of clinically relevant reduction rates (direction of arrows) and the statistical significance (green) were used to compare interventions’ effectiveness (Tables 3-5).

16In terms of BP control, a −10 mmHg reduction in SBP or a −5 mmHg reduction in DBP is considered as clinically relevant. No exact clinical relevance margins for lipid profiles could be prespecified, as European guidelines recommend a risk-based approach with regard to the presence of CV risk or established CV disease.

17To compare overall treatment effects between baseline and follow-up, meta-analyses reporting treatment effects as mean differences (MD), standardized mean difference (SMD), Cohen d, and Hedge g were included. For heterogeneity testing, results of I2 statistics (indicating variation across studies that is not due to chance) were used. A value of <40% indicates a low, 30%-60% a moderate, and >75% a substantial-to-high level of heterogeneity.

18Overall, 3564 references were identified. After title-abstract screening, 119 records remained for further full-text analysis. Details of the extracted evidence are provided in the Multimedia Appendices 3-9. The most important reasons for exclusion were low quality (n=15) and applied interventions not matching the prespecified telemedicine definition (n=14; annex section V). Overall, 46 studies were included in this umbrella review (Figure 1). In Figure 1, the Preferred Reporting Items for Systematic Reviews and Meta-analyses flowchart shows the study selection process, covering the single steps of identification via a 2-step screening (title and abstract and full-text base) for eligibility and inclusion into the qualitative synthesis of this review.

19Study designs included 16 systematic reviews, 7 meta-analyses, 19 records conducting both a systematic review and meta-analysis, three systematic reviews and meta-analyses with meta-regression, and one systematic review and network meta-analysis. The included meta-studies were published between 2009 and 2018, the majority was published after 2015 (Multimedia Appendix 5) and focused on diabetes. No high-quality reviews or meta-analyses were found targeting patients with dyslipidemia.

20An analysis of primary studies revealed significant overlaps among the 26 meta-analyses (Multimedia Appendix 5). The majority of primary studies were published after 2010 (Multimedia Appendix 5).

21On a scale of 0 to 18, the median OQAQ score of the 46 included studies was 16 (IQR 1), indicating that they were good quality systematic reviews and meta-analyses.

22Data from 16 systematic reviews were extracted (Multimedia Appendix 6). Diabetes was the chronic disease covered most often by the included reviews. A total of 5 systematic reviews dealt with T2D; however, only one systematic review dealt with T1D. Both types were studied together for a total of 4 times, four other systematic reviews did not specify which type of diabetes they focused on. Among the other diseases studied, hypertension was the most common. The results of the included systematic reviews were mixed, presenting a tendency for positive effects of telemedicine, in general, and digitally supported self-management using phones or apps on HbA1c and SBP/DBP. In contrast, the majority of studies evaluating telemonitoring and self-monitoring interventions found no significant improvements in HbA1c, fasting plasma glucose, or BP.

23Digital self-management in diabetes (T1D and T2D) was analyzed by 13 meta-analyses, of which 4 meta-analyses evaluated the effectiveness of mobile health (mHealth) and one meta-analysis evaluated the use of social network services (SNS). In digital self-management interventions, those including prescription (−0.75%, 95% CI −1.05 to −0.43; P=.013), teleconsultation (−0.62%; P<.001), and health information technologies on top of usual care (mostly based on tele-education; −0.57%, 95% CI −0.71 to −0.43; P<.005) showed significant clinically relevant mean reductions in HbA1c. Digital self-management interventions using mHealth showed significant clinically relevant reductions in HbA1c if they provided remote access to usual care (−0.55%, 95% CI −0.72 to −0.38; P<.001), used one or two features (−0.52%, 95% CI −0.76 to −0.28; P<.001), used SMS-based feedback (−0.64%, 95% CI −1.09 to −0.19; P=.005), included a potential risk intervention (−0.61%, 95% CI −0.95 to −0.27; P<.001), included a structured display (−0.69%, 95% CI −0.32 to −1.06; P=.008), provided medication management (−0.56%, 95% CI −0.99 to −0.13; P<.001), and provided lifestyle modification management (−0.52%, 95% CI −0.84 to −0.20; P<.001). SNS applied in diabetes self-management interventions proved to be effective if they were Web-based (−0.51%, 95% CI −0.68 to −0.34; P<.001) or combined Web-based SNS with mobile technologies (−0.54%, 95% CI −0.72 to −0.37; P<.001).

24Overall mean reductions in HbA1c of telemedicine interventions in patients with T1D ranged between −0.12% and −0.86%. Overall mean reductions were mostly not significant. Highest mean reductions were observed for the combination of tele-education with teleconsultation (−0.91%, 95% CI −1.21 to −0.61), although data on statistical significance were not provided. No significant clinically relevant reductions for population characteristics such as baseline HbA1c or age were identified in patients with T1D.

25Although varying in range (−0.01% to −1.13%), telemedicine significantly reduced HbA1c in patients with T2D.

26Table 3 summarizes the effectiveness of telemedicine in patients with diabetes by comparing intervention durations.

27Significant and clinically relevant reductions were found for short (≤3 months), middle (4-8 months), and long (>12 months) intervention durations. Digital health education, analyzed in the meta-analysis by Angeles et al, on average, reduced HbA1c above the predefined clinical relevance margin (HbA1c ≤−0.5%; Table 2). However, only the effects of interventions with a long-term study duration (12 months) were statistically significant (−0.55%, 95% CI −0.7 to −0.39; P<.001). Although three meta-analyses observed a tendency for higher reduction rates in shorter intervention durations, no general significant differences in reduction rates among intervention durations were found.

28Short-term intervention durations (≤6 months) of digital self-management showed greater mean reductions (−0.56%; P<.001) compared with mid- and long-term durations (>6 months). Clinically relevant mean reductions in SNS were significant for short-term intervention durations (≤3 months) as well. Using Web-based SNS for digital self-management proved to be significantly effective in the three pooled follow-up measurements. Again, the greatest mean reductions were apparent during the short-term (≤3 months) follow-up (−0.54%, 95% CI −0.80 to −0.28; P<.001).

29Although telemedicine interventions using feedback functions significantly reduced HbA1c in several studies, the highest reduction rates were found when no personalized feedback was provided (−0.61%, 95% CI −1.40 to 0.19; P=.001). No difference in HbA1c change was found for the type of health care professionals providing the feedback (eg, nurses or physicians).

30In addition, feedback, provided either via human telephone calls (−1.13%, 95% CI −1.51 to −0.75; P<.05) or via the internet (−0.62%, 95% CI −0.82 to −0.42; P<.001), significantly reduced HbA1c to a clinically relevant extent (≤−0.5 change). Higher frequency of provider feedback also showed greater reductions in HbA1c (−1.12%, 95% CI −1.32 to −0-91; P<.001) when compared with mean reduction rates of interventions utilizing low frequency rates (−0.33%, 95% CI −0.59 to −0.07; P<.01) (Table 4).

31The meta-regression carried out by Huang et al also revealed that factors we previously disregarded, such as study location, sample size, and feedback methods, were associated significantly with changes in HbA1c. Their combination in multivariate meta-regression analyses explained almost 100% of the variance among studies.

32Subgroup analyses on the effectiveness of telemedicine in certain patient populations (Table 5) were carried out by 12 meta-analyses.

33Although differences were not always significant, those subgroups with higher baseline HbA1c (>7.5% or >8.0%) showed increased reductions rates. Only for interventions categorized as digital self-management using SMS, the reduction rates were higher (−0.71%, 95% CI −0.93 to −0.48; P<.001) in patients with lower baseline HbA1c (<8%) when compared with those with higher (≥8%) baseline HbA1c (−0.38%, 95% CI −0.53 to −0.24; P<.001). Using meta-regression methods, Kebede et al found significant reduction rates in HbA1c for baseline HbA1c >7.5% (beta=−.44, 95% CI −0.81 to −0.06; P=.031), self-monitoring of behavioral outcomes, such as diets and physical activity (beta=−1.21, 95% CI −1.95 to −0.46; P=.009), and for support in problem solving (beta=−1.30, 95% CI −2.05 to −0.54; P=.007).

34Significant differences for age groups were sparse, as only three meta-analysis found significant reduction rates in patients with T2D and both types combined. The meta-analysis by Saffari et al on SMS-based digital self-management found significantly greater mean reductions (P=.006) in HbA1c for patients younger than 55 years (−0.65%, 95% CI −0.88 to −0.41; P<.001) when compared with the older age group (−0.42%, 95% CI −0.56 to −0.27; P<.001). The greatest significant mean reductions were observed for patients with diabetes aged 41 to 50 years (−1.83%, 95% CI −3.17 to −0.48; P<.001) and those over 50 years (−1.05%, 95% CI −1.50 to −0.60; P<.001).

35For digital self-management, a shorter time since diagnosis (<8.5 years) was associated with significantly greater mean reduction in HbA1c (−0.83%, 95% CI −1.10 to −0.56; P=.007) when compared with patients being diagnosed more than 8.5 years ago (−0.22%, 95% CI −0.44 to 0.01; P=.007). Similarly, patients diagnosed less than 7 years ago showed higher mean reductions (−0.61%, 95% CI −0.79 to −0.42) compared with their counterparts (−0.37%, 95% CI −0.61 to −0.13; P=.03) after using SMS-based digital self-management.

36Mean reductions of both SBP and DBP were also found in T2D patients. Toma et al found highly significant mean reductions in patients with both T1D and T2D for SBP (−3.47 mmHg, 95% CI −5.01 to −1.94; P<.001) and DBP (−1.84 mmHg, 95% CI −2.98 to −0.70; P=.002) because of Web- and mobile-based SNS interventions. Evaluating the effect of digitally supported dietary interventions in patients with chronic diseases, Kelly et al also reported significant mean reductions in SBP (−5.91 mmHg, 95% CI −11.14 to −0.68; P=.003) in the diabetes subgroup (although showing high heterogeneity between the two studies; I²=69%). Although no information on statistical significance was provided, Lee et al showed greatest mean reductions in SBP for the telemedicine subgroups focusing on tele-education (−4.05 mmHg, 95% CI −5.64 to −1.10), as well as those combining tele-education and telemonitoring (−3.91 mmHg, 95% CI −10.07 to 2.25). Analyzing the data of four studies, Cui et al found nonsignificant reductions for both DBP (−1.76 mmHg, 95% CI −3.6 to 0.07; P=.06) and SBP (−2.62 mmHg, 95% CI −5.6 to 0.36; P=.08). Digitally supported dietary interventions in patients with diabetes resulted in significant mean reductions in SBP (−5.91 mmHg, 95% CI −11.14 to −0.68; P=.003). However, none of the presented reduction rates reached the predefined clinical relevance margin.

37Only 8 of the included studies reported on lipid profiles; 4 in T1D/T2D patients, 2 in T2D patients, 1 in T1D patients, and 1 in several chronic diseases. On the basis of these studies, evidence on the effectiveness of telemedicine in lowering LDL-c or TGC or increasing high-density lipoprotein cholesterol (HDL-c) in patients with diabetes is scarce and heterogeneous. Marcolino et al found evidence that digital self-management applications for both diabetes types can reduce LDL-c levels; however, although significant, the effect was small (−6.6 mg/dL, 95% CI −8.3 to −4.9; P<.001; I²=24%). In addition, for both types of diabetes, Toma et al found evidence for a significant improvement in TC (−5.74 mg/dL, 95% CI −9.71 to −1.78; P<.005; I²=53%), HDL (1.90 mg/dL, 95% CI 0.24 to 3.57; P=.02; I²=19%), and TGC (−11.05 mg/dL, 95% CI −20.92 to −1.18; P<.03; I²=0%). Reductions in LDL (−1.15 mg/dL, 95% CI −5.19 to 2.88; P=.58; I2=47%) were not significant. Again, for patients with T2D, the pooled analysis of Lee et al found little and rather inconsistent effects, be it for LDL-c, HDL-c, TC, and TGC.

38A total of 3 of the included meta-analyses focused on patients with hypertension. Although the two analyses of Omboni et al focused on home BP monitoring, Liu et al examined the effect of internet-based interventions. Liu et al reported a significant overall mean reduction in SBP (−3.8 mmHg, 95% CI −5.63 to −2.06; P=.001) and DBP (−2.1 mmHg, 95% CI −3.51 to −0.65; P<.05). Owing to the identified heterogeneity for SBP (I²=61%), the authors carried out a subgroup analysis, revealing that mean change in SBP was greater in long-term interventions (6-12 months; −5.8 mmHg, 95% CI −4.3 to −4.1) when compared with short-term interventions (<6 months; −3.47 mmHg, 95% CI −5.2 to −1.7). However, data on statistical significance were not provided. The results of Omboni et al show significant mean reductions in SBP when using ambulatory measurement (−2.28 mmHg, 95% CI −4.32 to −0.24; P<.05). In their more recent analysis, they included studies evaluating additional features such as combined data transmission to physician, feedback, advice, and medication regulation. This time, they observed significant mean reductions (−3.48 mmHg, 95% CI −5.31 to −1.64; P<.001).

39The quality assessment of outcomes using the GRADE framework revealed the following levels of certainty (Multimedia Appendix 8). Of the 219 HbA1c outcomes assessed overall, 170 (77.63%) outcomes were rated as very low evidence and 42 (19.18%) outcomes were rated as low evidence. All of the 42 outcomes measuring SBP or DBP resulted in very low ratings of overall certainty (Table 6).

40The main reasons for low-quality assessment results in both outcome categories were as follows:

41Unclear or high-risk of bias: Missing allocation concealment, missing blinding of patients, study personnel and outcome assessors, high risk of selection bias and reporting bias (intention-to-treat analysis), and high or unclear losses to follow-up.

42Inconsistency: High heterogeneity in subgroup analysis, inconsistent confidence intervals crossing the mark for no effect.

43Indirectness: Differences in populations (type of diabetes, baseline HbA1c, age, duration of diabetes, and gender), differences in interventions (devices used, components combined, feedback intensity and frequency, and professional or professionals involved), and differences in settings (community, hospital, and primary care) in the pooled subgroups.

44Imprecision: Large confidence intervals and small effect sizes mostly because of small sample sizes.

45Publication bias: Visual and statistical or missing publication bias assessment; the reasons for the increased risk of publication bias mostly referred to the overrepresentation of smaller studies with higher effect sizes (favoring telemedicine). Furthermore, one reason is the paucity of data on mid- and long-term effects (6-12 months).

46Underreporting of relevant information: Reporting of study duration, dropouts/missing data, and follow-up time. Guidance on this matter was further complicated as some authors did not make a clear distinction between study duration and follow-up.

47Only for two outcomes (0.92%) measuring HbA1c, overall certainty was judged as moderate (Tables 5 and 6). In 6 (5 in HbA1c and 1 in DBP) cases, the outcomes of subgroup analyses were not assessed using GRADE, as results of only one trial were used by the authors of meta-analyses to pool data.

48As the initial search did not identify records primarily targeting patients with dyslipidemia and subgroup analyses on HDL, LDL, TC, and TGC were sparse, no grading of lipid outcomes was performed.

49High-level evidence from the 46 included meta-analyses and systematic reviews suggests that telemedicine interventions can be effective in improving clinical outcomes in patients with diabetes. Observed reduction rates are comparable with those of nonpharmacological eg, nutrition intervention or increased physical activity) and some pharmacological interventions (−0.5% to −1.25%). The observed reduction rates are encouraging, bearing in mind that the United Kingdom Prospective Diabetes Study (UKPDS) revealed that a 0.9% decrease in HbA1c was associated with a 25% reduction in microvascular complications, a 10% decrease in diabetes-related mortality, and a 6% reduction in all-cause mortality.

50In patients with diabetes, significant differences between telemedicine interventions and for certain population characteristics were identified. Telemedicine interventions embedded in frequent and intense patient-provider interactions and interventions with short durations (≤6 months) showed greater benefits. In addition, higher reduction rates were found for recently diagnosed patients and those with higher baseline HbA1c. However, quality assessment using GRADE revealed that overall and subgroup-specific certainty of evidence is low to very low. Therefore, the identified reduction rates have to be dealt with caution when translating them into evidence-based recommendations for treatment guidelines.

51Telemedicine was not found to have a significant and clinically meaningful impact on BP. Assessing the certainty of SBP and DBP outcomes, GRADE only revealed very low ratings. No records primarily targeting patients with dyslipidemia were found.

52According to the recent consensus report of the ADA and European Association for the Study of Diabetes, the application of telemedicine in diabetes is associated with a modest improvement in glycemic control. The European Society of Cardiology/European Society of Hypertension (ESC/ESH) guidelines for the management of arterial hypertension also report that telemonitoring and mobile phone apps may lead to improved outcomes for patients with hypertension. Our umbrella review updates this assessment of the effectiveness of telemedicine with special regard to intervention components, population characteristics, and it provides an in-depth assessment of the certainty of evidence. A brief summary of the study results can be found in Textbox 1.

53Telemedicine has the potential to improve clinical outcomes in patients with diabetes. Mixed results were found for patients with hypertension, none for those with dyslipidemia.

54Specific characteristics of the intervention (eg, high frequency and intensity of feedback/interaction and short treatment duration) and the patient (age <55 years, high baseline HbA1c, and recent diagnosis) seem to be associated with increased benefits in patients with diabetes.

55An assessment of the overall certainty using GRADE resulted in low and very low ratings, indicating that effects have to be dealt with caution.

56Looking at the characteristics of the telemedicine applications analyzed by the included meta-analyses, those encompassing frequent and intense patient-provider communication interactions showed greater benefit in HbA1c reduction. This was especially true for the combination of tele-case management with either teleconsultation (−1.20%, 95% CI −2.30 to −0.10; P<.001) or telemonitoring (−0.54%, 95% CI −2.44 to −0.06) in patients with T2D. Similarly, analogue disease self-management education interventions are known to be more effective in terms of HbA1c reduction when they offer additional support (eg, structured dietary or exercise interventions). On the basis of the analysis by Kastner et al, the combination of case management and self-management in addition to education provides potential for reduced HbA1c levels when compared with education and plain care coordination. Therefore, continuous and frequent communication, either via intensive feedback or psychological support, seems most promising.

57With a longer duration of follow-ups, the quality of evidence steadily declines because of considerable risk of bias and heterogeneity of study populations and interventions included. As for digital self-management, the evidence base is larger yet more diverse, as SMS (1 meta-analysis), social networks (1 meta-analysis), and mHealth apps (4 meta-analysis) can be used. However, the quality of evidence for digital self-management is low to very low, irrespective of the basal technology or the type of diabetes.

58In our analysis, some application types were found to reduce BP, for example, in SBP after tele-education (−4.05 mmHg, 95% CI −5.64 to −1.10), as well as strategies combining tele-education and telemonitoring (−3.91 mmHg). In patients with diabetes, Web- and mobile-based SNS interventions significantly reduced DBP (−3.47 mmHg), and digitally supported dietary interventions led to significant mean reductions in SBP (−5.91 mmHg). Although these reduction rates did not reach clinical relevance of ≥10 mmHg in SBP or ≥5 mmHg in DBP, they are similar to the expected reduction rates of nonpharmacologic interventions in patients with hypertension. Our results support the identified potential of telemonitoring and mobile phone apps in home BP self-monitoring, articulated in the current ESC/ESH guideline because of the additional advantages in memorizing, reviewing, and transmitting BP measurements.

59On the basis of the identified potential of telemedicine to provide individual self-management support, it is likely that embedded or additional components may have an additive and/or sustained impact on clinical outcomes. As such, recent evidence identified social media, gamification, and machine learning models as successful strategies to improve clinical outcomes and prevent disease-related complications.

60According to the included meta-analyses, telemedicine interventions are more effective for patients with T2D, higher baseline HbA1c, and a more recent diagnosis of diabetes. The increased potential for newly diagnosed patients was also identified by systematic reviews and landmark trials such as the UKPDS. As for hypertension, the results did not allow for population-specific analyses, which might be because of the rather passive interventions studies, such as telemonitoring.

61With the exception of a baseline BMI <30 kg/m2 (considered in one meta-analysis), all population-specific subgroup analyses were of low or very low evidence, the latter being more prevalent. This is also true for differences among age groups, for which no significant evidence was found. However, there was a tendency for higher reduction rates of HbA1c in younger patient cohorts with diabetes. Owing to the increased risk of elevated BP levels (> 130/80) and long-term risk of CV events, the current ESC/ESH guideline suggests treatment in younger adults (<50 years). In terms of age-specific BP control, ADA suggests focusing on DBP in patients under 50 years.

62Overall, as the results concerning population characteristics are diverse and of low to very low quality, our analysis did not find enough high-level evidence to recommend telemedicine for the treatment of patients with both hypertension and diabetes.

63Only reviews or meta-analyses reporting lipid outcomes in patients with diabetes were found. The extracted results on lipid outcomes are sparse and too heterogeneous to draw a conclusion on the effectiveness of telemedicine on these outcomes. With special regard to the effects of statins, as the first-line agents used to decrease cholesterol in the management of dyslipidemia and hypertension, the extracted effects of telemedicine on lipid profiles can be considered as minor. However, recent evidence underlines the promising potential of mobile phone-based self-monitoring interventions in patients with lipid metabolism disorders, because of the combination of therapy and lifestyle behavior changes.

64Robust systematic reviewing methods were used to generate an overview of high-quality evidence on the effects of telemedicine in three prevalent chronic conditions. The protocol of this umbrella review was presented to the research community. However, this study has several limitations, starting with the initial search and inclusion process. Although a comprehensive and piloted search strategy has been used, it is possible that some relevant studies were missed, if the exact search terms were used neither by the authors nor by the databases examined (Multimedia Appendix 10). The search within three different databases, complemented by a comprehensive hand search within the most important journals in the field of telemedicine, the use of MeSH terms, and a broad set of keywords, may have limited this risk of selection bias.

65In addition, some full-text articles were excluded because of their definition and application of the term “telemedicine,” which did not comply with standardized definitions, such as the one provided by Sood et al. Although the technology applied to deliver telemedicine has made tremendous advances during the past 10 years, our thorough application of the telemedicine definition and subgroup analyses using the GRADE assessment ensures comparability of intervention types. Intensive full-text assessment was applied to limit the bias of falsely including/excluding systematic reviews and meta-analyses because of mislabeling and inadequate delimitations of efficacy and effectiveness, as studies focusing on efficacy were excluded. As telemedicine is mostly embedded in low-risk interventions, mortality as an outcome was not considered. Although internationally recommended to be reported in addition to changes in HbA1c, parameters such as the time below, in, or above range, the number of hypoglycemic episodes, and quality of life were only reported by a few study authors and therefore did not allow for evidence-based guidance on this matter. A reason may be the publication date (median=2011) of the primary studies (Multimedia Appendix 6), which is before these recommendations were made.

66We also included different types of statistical analyses, including meta-analysis, network-meta-analysis, and meta-regression. Although the majority reported MD, there was a considerable methodological heterogeneity. This was because of the application of fixed- and random-effects models, as well as the reporting of SMD, Hedge g, or Cohen d instead of MD. Comparing the aggregated results of those statistical values without considering their weight (on the basis of the number of studies or number of patients per subgroup analysis) may have biased our analysis. However, this process was impeded by inconsistent reporting of baseline data such as the number of trials and participants in subgroup analysis. In addition, it is likely that reporting bias within the included systematic reviews and meta-analysis also affected our analysis. When studying the funnel plots, we also observed a tendency toward overrepresented smaller studies with higher effect sizes (favoring telemedicine), thereby increasing the risk of publication bias within some of the included analyses.

67Owing to the multimodal and individualized nature of digital interventions, the low GRADE results, especially the increase I2, are not surprising. In addition, we found significant overlaps among the primary studies of the included records (Multimedia Appendix 6). The results of the subgroup analysis therefore need to be considered with exceptional care before recommending certain intervention components for certain populations. However, as GRADE is the established procedure to evaluate the certainty of evidence when developing or updating guidelines, new quality assessment tools appropriate for the tailored and hybrid design of digital interventions should be developed. Along with the need for rather adaptive study designs, there is growing criticism on the suitability of RCTs for evaluating the effectiveness of digital interventions. In light of current efforts to support the clinical effectiveness, quality, and economic value of new technologies by using new assessment frameworks, our analysis underlines the challenges in this endeavor. In addition, future assessments on the clinical effectiveness should also include consolidated core outcome sets and patient-reported outcomes. However, as stated by the included records, longer study durations and more rigorously designed studies are needed for these future research needs.

68The results of this umbrella review indicate that telemedicine has the potential to improve clinical outcomes in patients with diabetes. Evidence extracted from systematic reviews and meta-analyses of RCTs showed subgroup-specific effectiveness rates favoring certain intervention and population characteristics. However, as indicated by the low GRADE ratings, evidence on the effectiveness of telemedicine in the three chronic conditions can be considered as limited.

69Future updates of clinical care and practice guidelines should carefully assess the methodological quality of studies and assess the overall certainty of subgroup-specific outcomes before recommending telemedicine interventions for certain patient populations.