Effectiveness of Digital Interventions for Improving Glycemic Control in Persons with Poorly Controlled Type 2 Diabetes: A Systematic Review,Meta-analysis,and Meta-regression Analysis

Abstract

Background:

Digital interventions may assist patients with type 2 diabetes in improving glycemic control. We aimed to synthesize effect sizes of digital interventions on glycated hemoglobin (HbA1c) levels and to identify effective features of digital interventions targeting patients with poorly controlled type 2 diabetes.

Materials and Methods:

MEDLINE, ISI Web of Science, and PsycINFO were searched for randomized controlled trials (RCTs) comparing the effects of digital interventions with usual care. Two reviewers independently assessed studies for eligibility and determined study quality, using the Cochrane Risk of Bias Assessment Tool. The Behavioral Change Technique Taxonomy V1 (BCTTv1) was used to identify BCTs used in interventions. Mean HbA1c differences were pooled using analysis of covariance to adjust for baseline differences and pre–post correlations. To examine effective intervention features and to evaluate differences in effect sizes across groups, meta-regression and subgroup analyses were performed.

Results:

Twenty-three arms of 21 RCTs were included in the meta-analysis (n = 3787 patients, 52.6% in intervention arms). The mean HbA1c baseline differences ranged from −0.2% to 0.64%. The pooled mean HbA1c change was statistically significant (−0.39 {95% CI: [−0.51 to −0.26]} with substantial heterogeneity [I ² statistic, 80.8%]) and a significant HbA1c reduction was noted for web-based interventions. A baseline HbA1c level above 7.5%, β = −0.44 (95% CI: [−0.81 to −0.06]), the BCTs “problem solving,” β = −1.30 (95% CI: [−2.05 to −0.54]), and “self-monitoring outcomes of behavior,” β = −1.21 (95% CI: [−1.95 to −0.46]) were significantly associated with reduced HbA1c levels.

Conclusions:

Digital interventions appear effective for reducing HbA1c levels in patients with poorly controlled type 2 diabetes.

Background

In 2017, more than 425 million adults were living with diabetes. This number is estimated to reach 629 million cases by 2045. More than 90% of this burden is due to type 2 diabetes.^1
–3 Type 2 diabetes is the second highest cause for obesity-related deaths, accounting for more than half a million deaths and 30.4 million disability-adjusted life years in 2015.⁴

Type 2 diabetes is a multifactorial metabolic disease linked with obesity, dietary behavior, and a sedentary lifestyle.^5,6 A recent trial conducted in the United Kingdom demonstrated remission to a nondiabetic state after changes in dietary behavior and significant weight loss in persons with type 2 diabetes.⁷ Hence, type 2 diabetes has recently been recognized as a potentially reversible metabolic state.⁸ However, the likelihood of a remission of the reversed state of the disease is still unclear. In addition, remission is less likely among persons with longer duration of type 2 diabetes.^7,9 Therefore, regular monitoring of blood glucose levels, as well as an optimal adherence to glucose-lowering medications, a healthy diet, and moderate-to-high-intensity physical activity (PA) remain important factors contributing to the prevention of macrovascular and microvascular complications of the disease.^10

–13

Failure to strictly adhere to medication, nutrition, and PA recommendations leads to hyper- and hypoglycemic levels^12,14,15 that worsen quality of life and increase the risk of mortality.^16,17 Ideally, tight glycemic control or maintaining glycated hemoglobin (HbA1c) levels between 5.7% and 6.5% is generally recommended to prevent complications and comorbidities. To help patients achieve tight glycemic control targets of HbA1c levels of 5.7%–6.5%,¹⁸ the American Association of Diabetic Educators (AADE) identified seven self-care behaviors (AADE7). Healthy eating, being physically active, monitoring, taking medication, problem solving, reducing risks, and healthy coping are the listed AADE7 self-care behaviors to guide diabetes education and care.¹⁹ The uptake of these self-care behaviors among patients can be strongly supported with digital interventions, such as text messaging and web-based and telemedicine interventions.^20

–23 By integrating digital technologies, e-health interventions help patients change their behavior toward regular monitoring of blood glucose levels, regular PA, a balanced diet, and other healthy lifestyle behaviors.^23

–29 Hence, diabetes-related behavioral and clinical outcomes can be improved through active engagement in e- and m-health interventions. In general, diabetes care is increasingly incorporating interactive digital e- and m-health interventions because the use of modern information and communication technologies comes with many advantages regarding the self-monitoring of the disease and self-regulation of lifestyle behaviors.^30

–34

HbA1c remains a surrogate marker of diabetes interventions after Stratton et al. demonstrated an independent log linear relationship between HbA1c- and diabetes-related complications.²⁶ Furthermore, recent reports suggest that interventions leading to a reduction in HbA1c of at least 0.3% among persons with type 2 diabetes are considered clinically significant.^24,29 Clinically significant reductions of HbA1c were achieved from various randomized controlled trials (RCTs) of digital interventions.^33

–37 Findings of several systematic reviews and meta-analyses on the effectiveness of digital interventions have also reported clinically significant HbA1c reductions, with varying level of effectiveness. For example, HbA1c reductions of −0.63%, −0.5%, −0.43% have been documented for videoconferencing,²⁵ mobile-based interventions,³¹ and interactive self-management interventions,²⁸ respectively.

Meta-analysis results also suggest that the changes in HbA1c levels were different across duration and mode of interventions.^21,32 A review on the effects of health information technology self-management interventions reported an aggregated HbA1c reduction of 0.36% at 6 and 0.27% at 12 months.²¹ In another review, all information technology-based interventions led to a reduction of 0.33%.³⁸ Participation in telemedicine, telecare, teleconsultation, and videoconferencing interventions was associated with HbA1c reductions of 0.31%,³⁹ 0.37%,⁴⁰ 0.54%,⁴¹ and 0.63%,²⁵ respectively. Furthermore, meta-analysis results suggest a reduction of HbA1c when participating in interactive self-management interventions by 0.43%,²⁸ whereas participation in computer-based interventions was only associated with a reduction of 0.2%³² and mobile-based interventions with a reduction of 0.5%.³¹ Two different reviews on the effects of mobile short message services reported an HbA1c level reduction of 0.22%³⁰ and 0.60%.⁴² It can be argued that these HbA1c changes are small³² but, in the long run, these small changes can help patients attain the target HbA1c level of less than 6.5% and thus prevent the risk of microvascular complications and diabetic-related deaths.^43,44

One limitation of the existing evidence of systematic reviews on the topic is the disregard of the influence of baseline HbA1c, the mean baseline HbA1c difference between control and intervention groups, and the pre–post correlation in the overall estimates of effect sizes for interventions. Results of subgroup and meta-regression analysis indicate that baseline HbA1c is associated with overall pooled effect sizes estimated using meta-analysis.^31,45 Available methodological literature on meta-analysis of a continuous outcome emphasizes the importance of accounting for baseline imbalance and pre–post correlations to determine precise and unbiased effect size estimates of a continuous outcome, such as HbA1c. However, the methodology to account for baseline imbalance and pre–post correlations is complex in the absence of individual participant data (IPD) and necessary summary data from published RCTs. Nevertheless, if relevant summary data are reported in RCTs, it is recommended to use analysis of covariance (ANCOVA) rather than change scores and final value effect size estimators.^46

–49

Importantly, e-health interventions targeting persons with type 2 diabetes are generally multicomponent behavioral interventions and complex in nature.⁵⁰ One way to simplify the complexity of reporting and analyzing the effect size of such interventions is by describing the active ingredients of the interventions by using the Behavioral Change Technique Taxonomy V1 (BCTTv1)^51,52 and/or the AADE7 self-care behaviors.¹⁹ The AADE7 self-care behaviors provide an evidence-based framework to identify contents of diabetes self-management interventions.¹⁹ The effectiveness of the active ingredients for reducing HbA1c levels in patients with poorly controlled type 2 diabetes has, to our knowledge, not yet been investigated. The results of our previous scoping review suggested the need for a detailed investigation of the individual and combined effects of BCTs on HbA1c and their role as mediators in Hba1c change.⁵³ Therefore, this systematic review and meta-analysis aimed to synthesize the effectiveness of digital interventions and identify BCTs associated with reductions of HbA1c levels.

Materials and Methods

The design, conduct, and reporting of this systematic review was guided by the Preferred Reporting Items for Systematic review and Meta-analysis (PRISMA) 2015 guideline.⁵⁴ The protocol of this systematic review was registered a priori (PROSPERO Registration No. 42016049940). The full description of the protocol for this systematic review can be accessed elsewhere.⁵⁵

Study inclusion criteria

Type of studies

Studies were included if (1) the design of the studies examining intervention effects was an RCT, including multiple arms RCT; (2) patients in the intervention have documented poorly controlled type 2 diabetes defined by an HbA1c level of >7.0%; (3) interventions were technology based, such as m-health (mobile health) interventions, web-based interventions, interventions delivered through the use of a personal digital assistant, a tablet, a computer, the Internet, telemedicine, videoconferencing, telehealth, or other forms of e-health; (4) HbA1c was reported as an outcome; (5) the control group received usual care, standard care, or existing care, and (6) if the study results were published in English. We used the American Diabetes Association (ADA) definition to define poorly controlled type 2 diabetes. Hence, having an HbA1c value of greater than 7.0% was considered poorly controlled type 2 diabetes.⁵⁶ Studies examining interventions that targeted either persons with type 1 diabetes or both type 1 and 2, and those including control groups receiving interventions other than usual care, were excluded from the review.

Search strategy for identification of studies

Studies published up to June 30, 2017, were searched in MEDLINE via PubMed, ISI Web of Science via Thomson Reuters, and PsycINFO via OvidSP using a comprehensive search strategy. The search terms suiting the different databases were created in collaboration with a research librarian. MeSH terms, keywords, and Boolean operators were used to develop the search strategy. The search was first completed in June 7, 2016, and updated on June 30, 2017.

Article screening

Two authors (M.M.K. and M.P.) screened titles and abstracts, as well as full-texts independently. If the two authors could not reach consensus, a third author (C.R.P.) was consulted to resolve disagreement. Covidence, a web-based screening tool, was used to document the screening process.⁵⁷ Information regarding the search and screening process is displayed in Figure 1.

FIG. 1.

Preferred reporting items for systematic reviews and meta-analyses flowchart for database search and study selection.

Data extraction process

Two authors extracted the following information: citation information (authors, titles, journals, year of publication), study location, study population (ethnicity, sex, age), study objectives, intervention type and delivery mode, AADE7 self-care behavior targeted, inclusion criteria, information on whether the intervention was guided by the use of behavioral science models or theories, individualization or tailoring of the interventions, and BCTs included in interventions. Moreover, sample size, intervention period, HbA1c values, and respective standard deviations (SDs), P-values, and 95% confidence intervals (CIs) were extracted for each study. The mean HbA1c change scores (SD), mean HbA1c difference (SD), type of statistical test (e.g., t-test, z-test), and data on intention-to-treat analysis (ITT) were collected for each study. If not reported in the articles, mean HbA1c change scores for both, control and intervention groups, were calculated for a particular time point (3, 4, 6, 8, 9, and 12 months). Based on the full description of the interventions in the articles reporting the study results or in study protocols, BCTs were identified and coded using the BCTTv1.⁵⁸

Two authors (M.M.K. and T.L.H.) read the description of the interventions to collect data about the seven AADE7 self-care behaviors addressed in each intervention. Two reviewers (M.M.K. and C.R.P.) experienced in using the BCTTv1⁵⁸ coded the description of intervention contents independently and meetings were held to reach consensus on which BCTs were coded for each individual intervention.

Quality assessment

The Cochrane Risk of Bias Assessment Tool for randomized control trials⁵⁹ was used to assess the quality of the included studies. Two authors (M.M.K. and M.P.) independently assessed the risk of bias, resolving differences with consensus. Covidence was used to semi-automate the process.⁵⁷ Using this tool, seven domains of risk of bias can be identified: allocation concealment, sequence generation, blinding of participants and personnel, blinding of outcome assessors, incomplete outcome data, selective outcome reporting, and other sources of bias. The terms low, high, or unclear risk of bias were used to label the quality of studies for each domain. The seventh domain, “other sources of bias,” was assessed following the recommendation by Fu et al. Hence, baseline balance of HbA1c levels between control and intervention groups, information on loss to follow-up, retention and attrition rates, and reported competing interests were considered.⁴⁶ Finally, the consensus quality ratings were exported to RevMan⁶⁰ to receive the final graphical representation of all risk of bias ratings.

Missing data

Missing data were obtained by contacting corresponding authors or computed based on the reported data. Using Excel functions, SD values, which were initially not reported for some of the studies, were calculated based on the reported 95% CIs, standard error (SE), or P-values.^46,61 Contacting corresponding author and computing missing SD values with reported data did not work for a study by Wakefield et al.⁶² Therefore, this missing SD value was imputed using arithmetic means by following an existing methodological guideline.⁴⁶ For one study,⁶³ the mean and SD values were calculated based on the reported median and range using Hozo's formula.⁶⁴ The pre–post correlation values, both for control and intervention groups, were calculated based on formulas described in previous methodological studies and formulas by Fu et al.⁴⁶ and Morris and DeShon.⁴⁸

Data syntheses and analysis

Simple analysis of final values, simple analysis of change scores, and ANCOVA effect size estimator are the main methods used to calculate effect sizes of a continuous outcome with a similar scale of measurement.^46

–49 Methodological guidelines show that adjusting for baseline imbalance and pre–post correlation is important in meta-analyzing continuous outcomes. The baseline HbA1c differences in the studies included in our review ranged from 0% to 0.64%, with only two RCTs having a mean HbA1 difference of 0%.^36,65 Assuming publication bias is negligible, meta-analysis of the baseline differences should be close to zero if the two treatment groups are balanced.⁴⁶ In our case, meta-analysis of baseline HbA1c mean difference was 0.14% (95% CI: −0.31 to 0.59). In addition, pre–post correlation ranged from 0.06³⁷ to 0.74.⁶⁶ For this reason, accounting for baseline imbalance and pre–post correlation was essential. In this review, the ANCOVA effect size estimator was preferred because it helps to adjust for baseline imbalance and pre–post correlations.^46

–49 Therefore, the effect size estimates were computed using the “black-belt” ANCOVA approach using the following equation: \documentclass{aastex}\usepackage{amsbsy}\usepackage{amsfonts}\usepackage{amssymb}\usepackage{bm}\usepackage{mathrsfs}\usepackage{pifont}\usepackage{stmaryrd}\usepackage{textcomp}\usepackage{portland, xspace}\usepackage{amsmath, amsxtra}\usepackage{upgreek}\pagestyle{empty}\DeclareMathSizes{10}{9}{7}{6}\begin{document}$${ \rm{ANCOVA}} = \left( {{ \rm{Yint}} - { \rm{Yctr}}} \right) - { \rm{ \beta }} \left( {{ \rm{Xint}} - { \rm{Xctrl}}} \right)$$ \end{document} , where β is a regression coefficient computed from pooled SD of the treatment \documentclass{aastex}\usepackage{amsbsy}\usepackage{amsfonts}\usepackage{amssymb}\usepackage{bm}\usepackage{mathrsfs}\usepackage{pifont}\usepackage{stmaryrd}\usepackage{textcomp}\usepackage{portland, xspace}\usepackage{amsmath, amsxtra}\usepackage{upgreek}\pagestyle{empty}\DeclareMathSizes{10}{9}{7}{6}\begin{document}$$( { \rm{SDy ) }}$$ \end{document} and control groups \documentclass{aastex}\usepackage{amsbsy}\usepackage{amsfonts}\usepackage{amssymb}\usepackage{bm}\usepackage{mathrsfs}\usepackage{pifont}\usepackage{stmaryrd}\usepackage{textcomp}\usepackage{portland, xspace}\usepackage{amsmath, amsxtra}\usepackage{upgreek}\pagestyle{empty}\DeclareMathSizes{10}{9}{7}{6}\begin{document}$${ \rm{ ( SDx}} )$$ \end{document} . Hence, \documentclass{aastex}\usepackage{amsbsy}\usepackage{amsfonts}\usepackage{amssymb}\usepackage{bm}\usepackage{mathrsfs}\usepackage{pifont}\usepackage{stmaryrd}\usepackage{textcomp}\usepackage{portland, xspace}\usepackage{amsmath, amsxtra}\usepackage{upgreek}\pagestyle{empty}\DeclareMathSizes{10}{9}{7}{6}\begin{document}$$ { { \beta = { \rm r } } } { \frac { { \rm { SDy } } } { { \rm { SDx } } } } $$ \end{document} .^46,47 Equations by McKenzie et al.⁴⁷ and Riley et al.⁴⁹ were used to calculate the variances of the final values, change scores, and ANCOVA effect size estimates. If there was no possibility to compute ANCOVA effect size estimates from the reported data, the estimate with smaller effect size obtained from change score and final value estimates was pooled with the ANCOVA effect size estimates following existing methodology.^46,47,49

Meta-analyses

Stata version 13 statistical software was used to perform the meta-analyses. The outcome data reported at study closure were used to perform the overall meta-analysis. HbA1c reductions of at least 0.3% were labeled as clinically significant.^24,29,67 For studies reporting the results of RCTs with three or more arms, relevant arms were considered in the pooled analysis if they were deemed combinable.

Following Cochrane recommendations, observed statistical heterogeneity was assessed with the Cochrane's χ ²-test (a P-value of less than 0.1 indicates statistically significant heterogeneity) and quantified by using I ². With I ² value of ≥50%, a random-effects model was used, else a fixed-effects model.⁶⁸

Sensitivity analyses were performed by excluding studies judged as having a “high risk” of bias for more than three dimensions of the risk of bias assessment tool. Several subgroup analyses were performed to estimate the effects of various intervention features (e.g., tailoring, mode of intervention, and BCTs included). The differences across subgroups were assessed using the random-effects model.

A series of univariate meta-regression analyses were performed by regressing intervention effect sizes across studies on intervention features (i.e., duration of intervention, mode of delivery, theory based, tailoring, baseline HbA1c inclusion criterion [HbA1c >7.0% vs. >7.5%], type of BCT [present or absent], and total number of BCTs included in the interventions). Then, multivariate meta-regression analyses were performed to identify effective BCTs and intervention features associated with HbA1c level. Following the recommendation by Borenstein et al.,⁶⁹ BCTs were added in the subgroup and meta-regression analyses if they were included in at least two studies.

Visual inspection of contour funnel plot was used to detect publication bias. In addition, Egger's test using a P-value of less than 0.1 was conducted to assess publication bias.⁷⁰ If publication bias was suspected, the “trim and fill” imputation method was used to estimate the number of missing studies in the funnel plot.⁷¹ Finally, the quality of evidence generated through meta-analysis was classified as high, moderate, and low using the GRADE approach.^72,73 The GRADEpro online tool was used to systematically evaluate the synthesized evidence.⁷⁴

Results

Study selection and characteristics

In the database search, 1669 titles and abstracts were retrieved, with only 22 studies fulfilling the inclusion criteria.^{35
–37,62,63,65,66,75

–89} Two studies^66,80 were three-arm RCTs, and 20 of them were two-arm RCTs. Because of this, 24 arms of 22 RCTs were considered in the meta-analysis. However, one RCT⁷⁶ had most of the required data missing, and hence excluded from the quantitative syntheses. Finally, 23 arms of the 21 RCTs were used in the meta-analyses (Fig. 1).

Studies were published between the years 2009 and 2017, with the majority conducted in the United States (n = 9). A total of 3787 subjects were included in the 23 arms of the 21 RCTs and followed for a mean duration of 7.29 months (SD = 3.05). One thousand nine hundred ninety-one (52.6%) participants were assigned to the intervention arms. The mean number of participants randomized into control and intervention groups was 82.6 (SD = 62.7) and 83.9 (SD = 62.12), respectively. On average, treatment retention rate at study closure was 89.4% (SD = 9.97, range = 25%, min. = 75% to max. = 100%).

Nearly two-thirds of interventions (n = 15; 65.22%) were web based delivered via smartphones, tablets, PDA, and computers. Five interventions implemented telehealth (21.74%) and three (13.04%) text messaging. Eleven of the 21 RCTs (52.4%) targeted patients with baseline HbA1c values greater than 7.5%, while the rest were targeted patients having an HbA1c level of greater than 7.0%.

Quality of studies

Less than half of the included studies described ITT with 12 studies not stating ITT procedures. Only one study was judged as having a low risk of bias on all of the risk of bias assessment dimensions.⁸⁹ Four studies (18.2%) were judged as having a high risk of bias on three of seven dimensions.^76,78,82,84 Eighteen (81.2%) studies adequately described the randomization procedure and were judged as having a low risk of bias with regard to this dimension.^{35
–37,62,63,65,66,75

–81,83,85,86,89} Seven studies (31.2%) did not adequately describe how the allocation was performed and were unclear for judgment (Fig. 2).^{63,65,77,82,86
–88}

FIG. 2.

Risk of bias assessment of individual studies and aggregated summary.

Regarding intervention-related adverse events, only eight studies^{36,37,62,66,77,80,85,89} reported that adverse events were assessed. All of these studies reported that there were no intervention-related adverse events. One study reported two deaths but not due to intervention participation⁶⁶ and one reported trouble among intervention participants with regard to using the digital devices or connecting with Bluetooth.⁸⁰

Only seven interventions were designed following behavioral health theories. The theories used were the “health belief model,”^75,83 the “trans-theoretical model of behavioral change,”⁷⁵ the “health action process approach,” the “theory of planned behavior,” and the “Bandura's theory of self-efficacy.”⁸³ In addition, “Green and Kreuter's PRECEDE-PROCEED model,”⁸¹ “cognitive behavioral therapy,” the “Reach Out” problem-solving model, and “motivational interviewing”^66,80 were used to guide intervention design. Seventeen interventions^{36,37,62,65,66,77

–83,85

–89} were tailored according to individual patient characteristics (Table 1).

Table 1.

Characteristics of Included Studies

Study	Location	Intervention	Intervention time endpoints, months	Tailoring	Theory	Study population	Baseline HbA1c inclusion criteria, %	AADE7 self-care behavior targeted	Adverse event
Mobile phone-delivered text message interventions
Agboola et al.⁷⁵	United States	Text to move (text message)	6	No	Yes, trans-theoretical model of behavior change	Spanish- or English-speaking low-income and ethnic minorities, type 2 diabetes patients	>7.0	Being active, healthy coping, taking medication, healthy coping	Not reported
Arora et al.⁶³	United States	Two daily text messages for 6 months. Education/motivation—1 text per day, medication reminders—3 per week, healthier living challenge—2 per week, trivia. Unidirectional text message	6	No	Yes, health belief model	English- or Spanish-speaking Latino and black type 2 diabetes patients	>7.5	Being active, healthy eating, monitoring, taking medication, reducing risks	Not reported
Capozza et al.⁷⁶	United States	Text message (Care4Life program) for education and motivation, medication adherence, glucose control, weight, and exercise	3 and 6	No, allowed patients to send text messages to providers	No	No specific population, adult patients with type 2 diabetes patients	>7.5	Being active, monitoring, taking medication, healthy coping, reducing risks, problem solving	Not reported
Fortmann et al.³⁵	Canada	Dulce Digital: An m-health SMS-based intervention	3 and 6	Culturally and ethnically tailored, but not tailored to individual characteristics	No	Underserved Hispanics with poor glycemic control, type 2 diabetes patients	≥7.5	Healthy eating, monitoring, taking medication, problem solving	Not reported
PDA-, tablet-, computer-, and/or smartphone-delivered web-based interventions
Cho et al.⁶⁵	South Korea	Internet diabetes management	3	Yes	No	No specific population, type 2 diabetes patients, Koreans	>7.0	Healthy eating, being active, monitoring, taking medication, problem solving, reducing risks	Not reported
Cho et al.⁷⁷	Korea	Healthcare provider mediated, remote coaching system via a PDA-type glucometer and the Internet	3 and 6	Tailored	No	Koreans, no specific population, patients with type 2 diabetes	7–10.0	Healthy eating, being active, monitoring, taking medication, problem solving, reducing risks	No adverse event detected
Egede et al.³⁶	United States	Telehealth and clinical decision support system	3 and 6	Tailored	No	18 Years or older, type 2 diabetes patients	≥8.0	Monitoring, taking medication, problem solving	No adverse event detected
Holmen et al.⁸⁰ (usual care vs. FTA-HC)	Norway	FTA (diabetes diary app) with HC	12	Tailored	Yes, cognitive behavioral therapy, the “Reach Out” problem-solving model, and MI	No specific population, adult patients with type 2 diabetes	>7.0	Healthy eating, being active, monitoring, problem solving, healthy coping	No adverse event, but only trouble with use of the digital devices
Holmen et al.⁸⁰ (usual care vs. FTA)	Norway	FTA (diabetes diary app) without HC	12	Tailored	Yes, cognitive behavioral therapy, the “Reach Out” problem-solving model, MI	No specific population, adult patients with type 2 diabetes	>7.0	Healthy eating, being active, monitoring	No adverse event, but only trouble with use of the digital devices
Kim et al.⁸²	China	Internet-based glucose monitoring system	3 and 6	Tailored	No	Male and female outpatients with type 2 diabetes patients	7.0–10.0	Monitoring	Not reported
Kleinman et al.⁸³	India	Smart phone app for patients and smart phone app and a web-based portal for providers	3	Tailored	Health belief model, health action process approach, theory of planned behavior, Bandura's theory of self-efficacy	No specific population, type 2 diabetes patients for >6 months	7.5–12.5	Monitoring, problem solving, reducing risks, taking, medication	Not reported
Ralston et al.³⁷	United States	Web-based care management	12	Tailored	Yes, Wagner's chronic care model	No specific population, adult patients with type 2 diabetes	>7.0	Healthy eating, monitoring, taking, medication, problem solving	No adverse event
Tang et al.⁸⁵	United States	Online disease management system	6 and 12	Tailored	Yes, Universal models of behavior change, MI, and chronic care model	No specific population, adult patients with type 2 diabetes	>7.5	Healthy eating, being active, monitoring, taking medication, reducing risks, problem solving	No adverse event
Tildesley et al.⁸⁶	Canada	IBGMS	3, 6 and 12	Tailored	No	No specific population, type 2 diabetes patients	>7.0	Monitoring, taking medication	Not reported
Torbjørnsen et al.⁶⁶ (usual care vs. FTA-HC)	Norway	FTA (diabetes diary app) with HC	4	Tailored	Yes, cognitive behavioral therapy, the “Reach Out” problem-solving model, MI	No specific population, adult patients with type 2 diabetes	>7.0	Healthy eating, being active, monitoring, problem solving, healthy coping	Two deaths, but unrelated with the interventions
Torbjørnsen et al.⁶⁶ (usual care vs. FTA)	Norway	FTA (diabetes diary app) without HC	4	Tailored	Yes, cognitive behavioral therapy, the “Reach Out” problem-solving model, MI	No specific population, adult patients with type 2 diabetes	>7.0	Healthy eating, being active, monitoring	Two deaths, but unrelated with the interventions
Wang et al.⁸⁷	China	Monitoring via computer/web/mobile phone connected to glucometer via cable	3 and 6	Tailored	No	No specific population, type 2 diabetes patients confirmed for over 1 year	7–10.0	Healthy eating, being active, monitoring	Not reported
Welch et al.⁸⁸	United States	Internet-based integrated diabetes management system	6	Tailored	No	Latino, type 2 diabetes patients	>7.5	Healthy eating, being active, monitoring, taking medication, problem solving, reducing risks, healthy coping	Not reported
Wild et al.⁸⁹	Scotland	Monitoring through computer/web based/mobile phone connected to glucometer via modem	9	Tailored	No	No specific population, type 2 diabetes patients aged more than 17 years	>7.5	Monitoring, taking medication, reducing risks	Adverse events were equally distributed between in intervention and control groups
Telehealth (communication with provider via telephone or video)
Dario et al.⁷⁸	Italy	Videoconferencing	12	Tailored	No	Italy, no specific population, type 2 diabetes patients	>7.0	Monitoring, reducing risks	Not reported
Hansen et al.⁷⁹	Denmark	Videoconferencing	8	Tailored	No	Danish-speaking type 2 diabetes patients	>7.5	Monitoring, problem solving, reducing risks	Not reported
Khanna et al.⁸¹	United States	Automated telephone support with dialogic telephone card	3	Tailored	Yes, Green and Kreuter's PRECEDE-PROCEED model	Spanish-speaking patients with type 2 diabetes	>7.5	Healthy eating	Not reported
Liou et al.⁸⁴	Taiwan	Web-based and videoconferencing	6	No	No	No specific population, adult type 2 diabetes patients	>7.0	Healthy eating, taking medication, reducing risks, healthy coping, problem solving	Not reported
Wakefield et al.⁶²	United States	Telemonitoring	3 and 6	Tailored	No	No specific population, subjects with established type 2 diabetes	>8.0	Monitoring, taking medication	No adverse events

FTA-HC, Few Touch Application health counseling; HbA1c, glycated hemoglobin; IBGMS, Internet-based glucose monitoring system; m-health, mobile health; MI, motivational interviewing.

AADE7 self-care behaviors targeted in the interventions

Inter-rater agreement determined by using Cohen's kappa and prevalence adjusted bias kappa (PABAK) was 0.6 and 0.8, respectively, suggesting a high reliability. A mean of 4 (SD = 1.74, range 1–7) AADE7 self-care behaviors were addressed in interventions. Of the seven AADE7 self-care behaviors, “monitoring” was the most frequently included (in 21 of 24 intervention arms) followed by “healthy eating” and “taking medication,” which were addressed 16 and 15 times, respectively. However, “healthy coping” was included in only nine intervention arms.

Of the 22 interventions included in this review, only one intervention addressed seven of the AADE7 self-care behaviors.⁸⁸ Four interventions targeted six self-care behaviors each (Table 1).^65,76,77,85

BCTs used in e-health interventions targeting persons with poorly controlled type 2 diabetes

Inter-rater agreement determined by using percent agreement was 96.5% and PABAK was 0.93, suggesting a very good agreement. A total of 27 BCTs with a minimum of 4^36,84,89 and a maximum of 11 BCTs⁷⁵ were addressed in interventions. Of the 27 BCTs, “instruction how to perform behavior” was used most frequently, included in 21 intervention arms.^{35
–37,62,63,65,66,75
–77,79

–82,84
–86,88,89} Eight BCTs were included only once. None of the intervention arms had a similar combination of BCTs (Supplementary Table S1, available online at https://www.liebertpub.com/suppl/doi/10.1089/dia.2018.0216).

Impact of interventions in terms of reducing HbA1c

The pooled mean HbA1c difference suggests a statistically significant HbA1c reduction, −0.39 (95% CI: −0.51 to −0.26), favoring digital intervention groups. However, heterogeneity was high (I ² statistic: 80.8%).

Publication bias

Visual inspection of the contour funnel plot shows that the majority of the effect sizes of the interventions are in the significant region as well as in the upper left part of the plot, suggesting the predominance of published significant findings. Only one study falls on the nonsignificant region (P > 0.5) of the contour-enhanced funnel plot (Supplementary Fig. S1). This asymmetry of the plot might possibly be due to either the presence of publication bias or due to factors other than publication bias. To confirm this, Egger's test and then “trim and fill” test were performed. Egger's test indicated that there is not enough evidence of small-study effects (coefficient = −1.13, P = 0.182). Performing the “trim and fill” test also did not result in changes in the pooled effect size estimate. Therefore, both Egger's and the “trim and fill” tests indicated that the influence of publication bias is negligible.

Sensitivity analyses

Sensitivity analyses were conducted excluding four studies judged as having a “high risk” of bias in more than three dimensions of the risk of bias assessment. The pooled HbA1c MD and heterogeneity I ² statistics were not substantially changed, resulting in a pooled HbA1c MD of −0.38% (95% CI: [−0.51 to −0.24]) and I ² of 77.9%. However, removing eight studies having a “high” and an “unclear” risk of bias for allocation concealment resulted in a lower effect size estimate, −0.30% (95% CI: [−0.45 to −0.15]). I ² statistic was also lowered to 68.4%. Further sensitivity analysis conducted by removing the four studies with inadequate description of randomization yielded a lower effect size estimate, −0.29% (95% CI: −0.41 to −0.17).

Subgroup analyses by intervention features and BCTs

The pooled mean HbA1c difference was −0.52% (95% CI: [−1.04 to 0.00]), −0.41% (95% CI: −0.55 to −0.27), −0.21% (95% CI: [−0.65 to 0.22]) for text message-delivered, web-based, and telehealth interventions, respectively. Statistically significant pooled HbA1c reductions favoring the intervention group were only noted for web-based interventions. However, there was substantial statistical heterogeneity across the three intervention subgroups (Fig. 3).

FIG. 3.

Subgroup analysis of the effectiveness of digital interventions for reducing HbA1c-levels by mode of delivery. HbA1c, glycated hemoglobin.

A subgroup analysis on the duration of interventions yielded an ANCOVA-adjusted mean HbA1c difference of −0.30 (95% CI: −0.495 to −0.11), −0.59 (95% CI: −0.78 to −0.39), and −0.21 (95% CI: −0.35 to −0.075) for interventions having outcome endpoints after 3–4, 6–8, and 9–12 months, respectively. However, there was substantial heterogeneity in the 3–4 months (I ² = 89%) and 6–8 months (I ² = 85%) subgroups (Supplementary Fig. S2).

Additional subgroup analysis was performed to investigate the differences in mean HbA1c reduction for interventions that “included” versus “did not include” a specific BCT. Hence, we noted HbA1c mean differences favoring the intervention group for the presence of the following BCTs: “information about health consequences” (−0.77%), “instruction on how to perform behavior” (−0.35%), “self-monitoring of behavior” (−0.27%), “self-monitoring outcomes of behavior” (−0.15%), “adding objects to the environment” (−0.13%), and “feedback on outcomes of behavior” (−0.12%). However, as can be seen above, only two BCTs led to clinically significant HbA1c changes (delta > −0.3%) (Supplementary Table S2 available at https://www.liebertpub.com/suppl/doi/10.1089/dia.2018.0216).

Furthermore, subgroup analysis on the effect size differences shows that interventions implemented among patients with HbA1c levels of greater than 7.5% led to higher reductions (delta = −0.12%) of HbA1c levels relative to interventions among patients with HbA1c of greater than 7.0% (Supplementary Table S3).

Subgroup analysis on the baseline HbA1c inclusion criteria resulted in relatively bigger effect size for interventions targeting patients with HbA1c levels greater than 7.5% (i.e., −0.45% {95% CI: [−0.70 to −0.21]} vs. −0.33% {95% CI: [−0.478 to −0.18]}) (Supplementary Fig. S3 available at https://www.liebertpub.com/suppl/doi/10.1089?dia.2018.0216).

Identifying intervention features and BCTs associated with HbA1c reductions

The univariate meta-regression analysis, obtained by regressing the effect sizes of interventions on intervention features indicated that none of the features, except for duration of interventions, was significant. Interventions with 3- to 4-month duration (β = 0.42, P = 0.016, tau2 = 0.085, R ² = 21.7%) and a 6- to 8-month duration (β = −0.29, P = 0.03, tau2 = 0.089, R ² = 17.7%) displayed significant associations with the effect size indicating that 6–8 months of intervention duration resulted in a pronounced reduction of HbA1c levels. Only two BCTs, “feedback on behavior” (β = 0.29, P = 0.037, tau2 = 0.092, R ² = 15.4%) and “social support practical” (β = 0.42, P = 0.016, tau2 = 0.0085, R ² = 21.6%), were significantly associated with the effect size. Because the β coefficients were positive, the use of these BCTs did not demonstrate HbA1c reductions.

Multivariable meta-regression revealed that the presence/absence of nine BCTs and other intervention features in the model explained 79.8% of the variance in the effect size. Tailoring the interventions, β = 1.15 (95% CI: [0.14 to 2.17]), baseline HbA1c higher than 7.5, β = −0.44 (95% CI: [−0.81 to −0.06]), and the presence or absence of four BCTs were significantly associated with HbA1c levels. Hence, the BCTs “problem solving” (β = −1.30; 95% CI: [−2.05 to −0.54]), “feedback on outcomes of behavior” (β = 0.68; 95% CI: [0.08 to 1.28]), “self-monitoring outcomes of behavior” (β = −1.21; 95% CI: −1.95 to −0.46), and “prompts/cues” (β = 0.44; 95% CI: [0.03 to 0.85]) were significantly associated with the HbA1c levels. Of these, baseline HbA1c higher than 7.5%, “problem solving,” and “self-monitoring outcomes of behavior” were associated with reduced HbA1c-levels (Table 2).

Table 2.

Multivariable Meta-regression Model of Intervention Features and BCTs

Intervention features & BCTs	Coefficient	Std. Err.	t	P	[95% confidence interval]
Type of intervention
Web-based interventions	−0.39	0.18	−2.23	0.076	[−0.85 to 0.06]
Text message based	0.38	0.35	1.08	0.331	[−0.52 to 1.28]
Duration of intervention, months
3–4	−0.003	0.13	−0.02	0.987	[−0.35 to 0.34]
6–8	−0.17	0.16	−1.11	0.318	[−0.58 to 0.23]
Tailoring	1.15	0.39	2.91	0.033	[0.14 to 2.17]^a
Baseline HbA1c >7.5	−0.44	0.15	−2.97	0.031	[−0.81 to −0.06]^a
Theory use	0.05	0.15	0.33	0.752	[−0.33 to 0.42]
Problem solving	−1.30	0.29	−4.41	0.007	[−2.05 to −0.54]^a
Review outcome goals	−0.44	0.20	−2.22	0.077	[−0.96 to 0.07]
Self-monitoring of behavior	−0.22	0.24	−0.92	0.399	[−0.84 to 0.40]
Self-monitoring outcomes of behavior	−1.21	0.29	−4.15	0.009	[−1.95 to −0.46]^a
Feedback on outcomes of behavior	0.68	0.23	2.91	0.033	[0.08 to 1.28]^a
Instruction how to perform behavior	0.05	0.20	0.24	0.823	[−0.46 to 0.55]
Salience of consequences	−0.14	0.12	−1.12	0.312	[−0.45 to 0.18]
Prompts/cues	0.44	0.16	2.77	0.040	[0.03 to 0.85]^a
Adding objects to the environment	0.04	0.28	0.16	0.882	[−0.68 to 0.76]
Total number of BCTs	−0.02	0.05	−0.42	0.695	[−0.14 to 0.10]
Intercept	−0.40	0.36	−1.11	0.317	[−1.33 to 0.53]

Number of studies included in the model = 23, estimate of between-study variance tau2 = 0.022, adjusted R ² = 79.83, I ² = 97.76%, Joint test for all covariates F(17.5) = 5.34, P = 0.037.

Statistically significant at a P-value of 0.05.

ANCOVA, analysis of covariance; BCT, behavior change technique.

Grading the quality of evidence generated from the meta-analyses

Applying the GRADE principles,⁷⁴ the quality of evidence generated through this meta-analysis can be considered “moderate quality” (Supplementary Tables S4 and S5 available at https://www.liebertpub.com/suppl/doi/10.1089/dia.201.2018.0216).

Discussion

This systematic review is the first to demonstrate the effectiveness of digital interventions for reducing HbA1c levels in patients with poorly controlled type 2 diabetes. It is also the first review to account for baseline imbalance and pre–post correlations using an available robust statistical method, ANCOVA. The review also used a reliable taxonomy⁵² to identify effective BCTs used in digital interventions targeting persons with type 2 diabetes, as well as the well-established AADE7¹⁹ to unravel the effects of intervention components on HbA1c levels. These tools offer a great opportunity to handle heterogeneity across multicomponent and complex interventions.^51,53,90

In this review, we report clinically and statistically significant effects of PDA-, mobile phone-, or computer-delivered web-based interventions on HbA1c. A clinically significant HbA1c reduction is associated with lower rates of deaths, myocardial infractions, and reduced microvascular complications.²⁴

Similar to our results, clinically and statistically significant pooled HbA1c reductions were reported for Internet-based interactive self-management interventions²⁸ and mobile phone-based Internet interventions.³¹ Our findings thus support the previously reported evidence on beneficial effects of web-based interventions.³⁸ However, we could not show a statistical significant reduction of HbA1c levels after participation in text message and telehealth interventions. Contrary to our findings, previous results of meta-analyses reported significant HbA1c reductions after participation in telehealth^39,41 and text message interventions.⁴² This may possibly be due to the number of telehealth and text message interventions included in our meta-analyses, which was relatively low.

Sensitivity analyses performed by removing studies with a high risk of bias for more than three dimensions suggest that there was no change in the direction of the overall effect estimate. However, the two additional meta-analyses performed after dropping studies with an inadequate description of randomization and a high or unclear risk of bias regarding allocation concealment resulted in a lower effect size estimate. This supports the finding that studies with inadequate or unclear allocation concealment may report inflated treatment effect estimates.^91,92

Results of the subgroup analyses by duration of intervention suggest higher effect size estimates for longer intervention periods. This is likely due to the fact that it takes time for behavioral interventions to change patterns of thoughts and feelings toward behavior change and behavior itself, in turn leading to a change in HbA1c. Yet, the effect size estimate decreased 9–12 months into the intervention. This is in line with other recent meta-analysis results that reported similar effects of digital interventions on HbA1c reduction by duration.^21,28,67 However, a review by Cradock et al. reported a higher HbA1c reduction at month 3 compared with month 6 during the intervention period.⁶⁷ It should be noted though that only a small number of studies (four) were included and that baseline HbA1c levels were not taken into account in the subgroup analyses performed by Cradock et al.⁶⁷ Previous literature also suggests that more pronounced reductions of HbA1c occur among patients with higher HbA1c levels at the beginning of the intervention.^31,45,93 This was confirmed in both our subgroup and meta-regression analyses. Patients with an HbA1c level greater than 7.5% displayed higher effect estimates. Clinically, this supports the usefulness of digital interventions, particularly among patients with poor initial glycemic control.

Only interventions addressing the following two BCTs, “information about health consequences” and “instruction on how to perform behavior,” led to clinically significant HbA1c changes in patients with poorly controlled type 2 diabetes. Cradock et al. also reported a clinically significant effect of using the BCT “instruction on how to perform behavior,”⁶⁷ as did another meta-analysis reported by Avery et al.⁹⁴ Future meta-analyses, including more studies and larger study populations as well as concise intervention descriptions, are needed to validate these findings.

The results of the multivariable meta-regression analysis indicate that nine BCTs, as well as additional intervention features in the model, explained more than three-fourths of the variance in the effect size. Baseline HbA1c above 7.5% and the presence of the two BCTs “problem solving” and “self-monitoring outcomes of behavior” were associated with significant reductions in HbA1c levels. Contrary to the results of a meta-analysis by Kassavou and Sutton, our results suggest that interventions using the BCT “problem solving” had a higher beneficial pooled effect.⁹⁵ It is known that meta-regression models provide robust results when a greater number of studies and fewer covariates are taken into account.⁹⁶ Future meta-regression analyses therefore ought to pool a larger number of trials to develop relatively stable and precise meta-regression results.

Although tailoring the interventions and “feedback on outcomes of behavior” were significantly associated, these associations were inverse, which indicates that the presence of tailoring and this BCT do not lead to reductions in HbA1c levels. A systematic review of reviews by Greenwood et al. reported interventions with two-way communication, patient-generated data tracking and analysis, tailored education and individualized feedbacks were most effective.⁹⁰ There was also no evidence of an association between HbA1c levels and the use of theories for designing interventions. Although the use of theories ideally offers scientific explanations of the process of change and is helpful for linking observed changes in outcomes with active intervention ingredients, the existing evidence regarding the benefits of theory use for intervention development is mixed.⁹⁷

Our results did not suggest an association between the number of BCTs addressed in interventions and changes in HbA1c. In contrast, two previous systematic reviews demonstrated an association between a greater number of BCTs included in behavioral and self-management interventions and reductions in HbA1c levels.^67,94 The substantial variation in the breadth and depth of BCT descriptions included in the articles could partially explain this finding. Greater quality of intervention descriptions enhances reliability and validity of characterization of the multicomponent interventions and improves reliability and the power of results.^53,98

Limitations

The study has several limitations. Our search was limited to three main databases. However, during the preliminary search, we did not observe major differences in search results when using EMBASE, CINAHL and Cochrane library. Therefore, we concluded that our search in PubMed, ISI Web of Science, and PsycINFO covered the relevant articles. We also did not search in unindexed databases and gray literature.

Previous research suggests that each additional intervention increases the unspecific “attention factor” for patients. Testing a digital intervention against a standard care might therefore overestimate the specific effect of digital interventions, since the “attention factor” is not well controlled. In addition, higher dropout rates were reported from previous digital interventions. This might be due to self-selection indicating that people who like digital media stay in the interventions and people who have difficulties with digital interventions may drop out from the study, leading to overestimation of the results. However, in this review, a high retention rate was observed from individual studies.

Mapping and differentiating the intervention content to determine which BCTs were addressed in interventions rely on the quality and depth of descriptions available for various interventions.⁹⁹ In addition, it is based on a subjective judgment. We tried to minimize this limitation by taking online training in using the BCTTv1, applying consensus ratings, and using a third experienced reviewer to resolve any disagreement.

Two studies^33,34 that passed our inclusion criteria were not included in the meta-analysis. This is mainly because of two limitations related to methodology and missing values. Because the design of these studies was cluster RCT, the effect sizes and SEs were needed to be adjusted for design effect by considering intracluster correlation (ICC). This analysis is more complicated when cluster RCTs have high baseline imbalance and did not report ICC. Hence, application of ANCOVA was not feasible for these two studies. We considered change scores or final values as alternatives. However, SD for change scores in each intervention endpoint (3, 6, 9, and 12 months) for both control and intervention groups were not available from the two trials. For this reason, we did not include these studies in our meta-analysis. We suggest future interventions to report details of summary data. Future research is needed on methodological guideline for handling continuous outcomes from cluster RCTs with a high baseline imbalance.

Conclusions

In conclusion, the results of this systematic review and meta-analysis indicate that participation in digital interventions, particularly web-based interventions, favorably influences HbA1c levels among patients with poorly controlled type 2 diabetes. Intervention effects were more pronounced among patients with higher baseline HbA1c levels and greater effects were observed 6–8 months into an intervention. Moreover, the results of the meta-regression analyses suggest that baseline HbA1c >7.5% and the two BCTs “problem solving” and “self-monitoring outcomes of behavior” were associated with a reduction of HbA1c levels. Hence, considering these two BCTs in future interventions may lead to clinically meaningful reductions in HbA1c.

The effort to adjust for baseline imbalance and pre–post correlation relies on the level of detail of reporting available for individual studies. We suggest to authors of future intervention studies, particularly with baseline imbalance, to report detailed information that allows authors of systematic reviews to calculate ANCOVA effect size estimates or, ideally, to provide access to IPD.

Footnotes

Acknowledgments

We are grateful to Professor Richard D. Riley, Dr. Jochen Wilhelm, Dr. James E. Pustejovsky, Professor Vanessa Didelez, and Dr. Fleur Fritz for the methodological support provided particularly to the first author. We also acknowledge our research librarian, Lara Christianson, for her support in developing the search strategy. In addition, we are very grateful to all authors of the studies included in this meta-analysis particularly to Caroline R. Hansen, Sarah Wild, Richard Parker, Nora Kleinman, Professor Leonard Egede, Dr. Addie Fortmann, Heidi Holmen, Dr. James Ralston, Dr. Raman Khanna, and Elisa Visentin for providing us with the data we requested and simplifying our effort in handling the missing data.

Availability of Data

The data collected for this study are available from the corresponding author and can be received on request.

Authors' Contributions

M.M.K. conceptualized and designed the study, conducted the systematic literature search with the help of research librarian, performed the title and abstract screening, quality assessment, data extraction, data analysis and interpretation of the data, and wrote the article.

H.Z. participated in the conception and the design of the study and the development of the methodology, and critically revised the article for important intellectual content.

M.P. conducted the title and abstract screening, quality assessment, and participated in the draft of the article.

T.L.H. contributed to the extraction of the data and critically revised the article.

C.R.P. participated in the conception, data extraction, and critically revised the article for important intellectual content.

All authors read and approved the final version of the article.

Author Disclosure Statement

No competing financial interests exist.

References

IDF: International Diabetes Federation Diabetes Atlas, 7 ^th ed. IDF, Brussels, Belgium, 2015.

Darnton-Hill

, Nishida

, James

: A life course approach to diet, nutrition and the prevention of chronic diseases. Public Health Nutr, 2004; 7:101–121.

IDF: IDF Diabetes Atlas, 8 ^th ed. International Diabetes Federation, IDF, Brussels, Belgium, 2017.

Bovet

, Chiolero

, Gedeon

: Health effects of overweight and obesity in 195 countries. N Engl J Med, 2017; 377:1495–1496.

Despres

, Poirier

: Diabetes: looking back at look AHEAD—giving lifestyle a chance. Nat Rev Cardiol, 2013; 10:184–186.

Taylor

: Pathogenesis of type 2 diabetes: tracing the reverse route from cure to cause. Diabetologia, 2008; 51:1781–1789.

Lean

MEJ

, Leslie

, Barnes

, et al.: Primary care-led weight management for remission of type 2 diabetes (DiRECT): an open-label, cluster-randomised trial. Lancet, 2018; 391:541–551.

Taylor

: Type 2 diabetes: etiology and reversibility. Diabetes Care, 2013; 36:1047–1055.

Steven

, Hollingsworth

, Small

, et al.: Weight loss decreases excess pancreatic triacylglycerol specifically in type 2 diabetes. Diabetes Care, 2016; 39:158–165.

10.

American Diabetes Association: Foundations of care and comprehensive medical evaluation. Diabetes Care, 2016; 39(Suppl 1):S23–S35.

11.

IDF: Diabetes Atlas, 6 ^th ed. International Diabetes Federation, IDF, Brussels, Belgium, 2013.

12.

WHO: Global Report on Diabetes. Geneva, Switzerland: World Health Organization, 2016.

13.

Diabetes Prevention Program Research Group: The Diabetes Prevention Program (DPP): description of lifestyle intervention. Diabetes Care, 2002; 25:2165–2171.

14.

Brown

, Bussell

: Medication adherence: WHO cares?. Mayo Clin Proc, 2011; 86:304–314.

15.

Tuomi

, Santoro

, Caprio

, et al.: The many faces of diabetes: a disease with increasing heterogeneity. Lancet, 2014; 383:1084–1094.

16.

Girach

, Manner

, Porta

: Diabetic microvascular complications: can patients at risk be identified? A review. Int J Clin Pract, 2006; 60:1471–1483.

17.

Avramopoulos

, Moulis

, Nikas

: Glycaemic control, treatment satisfaction and quality of life in type 2 diabetes patients in Greece: The PANORAMA study Greek results. World J Diabetes, 2015; 6:208–216.

18.

The National Collaborating Centre for Chronic Conditions: Type 2 diabetes: the management of type 2 diabetes. NICE clinical guideline. In: NICE Guideline [NG28]. London: Royal College of Physicians, 2015.

19.

Tomky

, Cypress

, Dang

, et al.: AADE7 self-care behaviors. Diabetes Educ, 2008; 34:445–449.

20.

Chen

, Chuang

, Chang

, et al.: Evaluating self-management behaviors of diabetic patients in a telehealthcare program: longitudinal study over 18 months. J Med Internet Res, 2013; 15:e266.

21.

Heitkemper

, Mamykina

, Travers

, Smaldone

: Do health information technology self-management interventions improve glycemic control in medically underserved adults with diabetes? A systematic review and meta-analysis. J Am Med Inform Assoc, 2017; 24:1024–1035.

22.

Farmer

, Gibson

, Tarassenko

, Neil

: A systematic review of telemedicine interventions to support blood glucose self-monitoring in diabetes. Diabet Med, 2005; 22:1372–1378.

23.

Baron

, McBain

, Newman

: The impact of mobile monitoring technologies on glycosylated hemoglobin in diabetes: a systematic review. J Diabetes Sci Technol, 2012; 6:1185–1196.

24.

Tricco

, Ivers

, Grimshaw

, et al.: Effectiveness of quality improvement strategies on the management of diabetes: a systematic review and meta-analysis. Lancet, 2012; 379:2252–2261.

25.

, Zhou

, Kelley

, et al.: Does telemedicine improve treatment outcomes for diabetes? A meta-analysis of results from 55 randomized controlled trials. Diabetes Res Clin Pract, 2016; 116:136–148.

26.

Stratton

, Adler

, Neil

, et al.: Association of glycaemia with macrovascular and microvascular complications of type 2 diabetes (UKPDS 35): prospective observational study. BMJ, 2000; 321:405–412.

27.

Farmer

, McSharry

, Rowbotham

, et al.: Effects of interventions promoting monitoring of medication use and brief messaging on medication adherence for people with type 2 diabetes: a systematic review of randomized trials. Diabet Med, 2016; 33:565–579.

28.

Cheng

, Sit

, Choi

, et al.: Effectiveness of interactive self-management interventions in individuals with poorly controlled type 2 diabetes: a meta-analysis of randomized controlled trials. Worldviews Evid Based Nurs, 2017; 14:65–73.

29.

Avery

, Flynn

, van Wersch

, et al.: Changing physical activity behavior in type 2 diabetes: a systematic review and meta-analysis of behavioral interventions. Diabetes Care, 2012; 35:2681–2689.

30.

Farmer

, McSharry

, Rowbotham

31.

Liang

, Wang

, Yang

, et al.: Effect of mobile phone intervention for diabetes on glycaemic control: a meta-analysis. Diabet Med, 2011; 28:455–463.

32.

Pal

, Eastwood

, Michie

, et al.: Computer-based interventions to improve self-management in adults with type 2 diabetes: a systematic review and meta-analysis. Diabetes Care, 2014; 37:1759–1766.

33.

Quinn

, Shardell

, Terrin

, et al.: Cluster-randomized trial of a mobile phone personalized behavioral intervention for blood glucose control. Diabetes Care, 2011; 34:1934–1942.

34.

Quinn

, Shardell

, Terrin

, et al.: Mobile diabetes intervention for glycemic control in 45-to 64-year-old persons with type 2 diabetes. J Appl Gerontol, 2016; 35:227–243.

35.

Fortmann

, Gallo

, Garcia

, et al.: Dulce digital: an mHealth SMS-based intervention improves glycemic control in Hispanics with type 2 diabetes. Diabetes Care, 2017; 40:1349–1355.

36.

Egede

, Williams

, Voronca

, et al.: Randomized controlled trial of technology-assisted case management in low income adults with type 2 diabetes. Diabetes Technol Ther, 2017; 19:476–482.

37.

Ralston

, Hirsch

, Hoath

, et al.: Web-based collaborative care for type 2 diabetes a pilot randomized trial. Diabetes Care, 2009; 32:234–239.

38.

Alharbi

, Alsubki

, Jones

, et al.: Impact of information technology-based interventions for type 2 diabetes mellitus on glycemic control: a systematic review and meta-analysis. J Med Internet Res, 2016; 18:e310.

39.

Verhoeven

, van Gemert-Pijnen

, Dijkstra

, et al.: The contribution of teleconsultation and videoconferencing to diabetes care: a systematic literature review. J Med Internet Res, 2007; 9:e37.

40.

Zhai

, Zhu

, Cai

, et al.: Clinical- and cost-effectiveness of telemedicine in type 2 diabetes mellitus: a systematic review and meta-analysis. Medicine (Baltimore), 2014; 93:e312.

41.

Huang

, Tao

, Meng

, Jing

: Management of endocrine disease effects of telecare intervention on glycemic control in type 2 diabetes: a systematic review and meta-analysis of randomized controlled trials. Eur J Endocrinol, 2015; 172:R93–R101.

42.

Saffari

, Ghanizadeh

, Koenig

: Health education via mobile text messaging for glycemic control in adults with type 2 diabetes: a systematic review and meta-analysis. Prim Care Diabetes, 2014; 8:275–285.

43.

UK Prospective Diabetes Study (UKPDS) Group: Intensive blood-glucose control with sulphonylureas or insulin compared with conventional treatment and risk of complications in patients with type 2 diabetes (UKPDS 33). Lancet, 1998; 352:837–853.

44.

Svensson

, Baggesen

, Johnsen

, et al.: Early glycemic control and magnitude of HbA1c reduction predict cardiovascular events and mortality: population-based cohort study of 24,752 metformin initiators. Diabetes Care, 2017; 40:800–807.

45.

Welch

, Garb

, Zagarins

, et al.: Nurse diabetes case management interventions and blood glucose control: results of a meta-analysis. Diabetes Res Clin Pract, 2010; 88:1–6.

46.

, Vandermeer

, Shamliyan

, et al.: Handling continuous outcomes in quantitative synthesis. In: Methods Guide for Effectiveness and Comparative Effectiveness Reviews. Rockville, MD, 2008.

47.

McKenzie

, Herbison

, Deeks

: Impact of analysing continuous outcomes using final values, change scores and analysis of covariance on the performance of meta-analytic methods: a simulation study. Res Synth Methods, 2016; 7:371–386.

48.

Morris

, DeShon

: Combining effect size estimates in meta-analysis with repeated measures and independent-groups designs. Psychol Methods, 2002; 7:105–125.

49.

Riley

, Kauser

, Bland

, et al.: Meta-analysis of randomised trials with a continuous outcome according to baseline imbalance and availability of individual participant data. Stat Med, 2013; 32:2747–2766.

50.

van Vugt

, de Wit

, Cleijne

, Snoek

: Use of behavioral change techniques in web-based self-management programs for type 2 diabetes patients: systematic review. J Med Internet Res, 2013; 15:e279.

51.

Michie

, Fixsen

, Grimshaw

, Eccles

: Specifying and reporting complex behaviour change interventions: the need for a scientific method. Implement Sci, 2009; 4:40.

52.

Michie

, Richardson

, Johnston

, et al.: The behavior change technique taxonomy (v1) of 93 hierarchically clustered techniques: building an international consensus for the reporting of behavior change interventions. Ann Behav Med, 2013; 46:81–95.

53.

Kebede

, Liedtke

, Möllers

, Pischke

: Characterizing active ingredients of eHealth interventions targeting persons with poorly controlled type 2 diabetes mellitus using the behavioral change technique taxonomy. J Med Internet Res, 2017; 19:e348.

54.

Moher

, Liberati

, Tetzlaff

, et al.: Preferred reporting items for systematic reviews and meta-analyses: the PRISMA statement. PLoS Med, 2009; 6:e1000097.

55.

Kebede

, Christianson

, Khan

, et al.: Effectiveness of behavioral change techniques employed in eHealth interventions designed to improve glycemic control in persons with poorly controlled type 2 diabetes: a systematic review and meta-analysis protocol. Syst Rev, 2017; 6:211.

56.

American Diabetes Association: Standards of medical care in diabetes—2014. Diabetes Care, 2014; 37(Suppl 1):S14–S80.

57.

Covidence: Systematic Review Software. Melbourne, Australia: Veritas Health Innovation.

58.

Michie

, van Stralen

, West

: The behaviour change wheel: a new method for characterising and designing behaviour change interventions. Implement Sci, 2011; 6:42.

59.

Higgins

JPT

, Altman

, Gøtzsche

, et al.: The Cochrane Collaboration's tool for assessing risk of bias in randomised trials. BMJ, 2011; 343:d5928.

60.

Review Manager (RevMan) [Computer Program]. Version 5.3. Copenhagen: The Nordic Cochrane Centre, The Cochrane Collaboration, 2014.

61.

Cochrane Handbook [article online]. 2011. http://handbook-5-1.cochrane.org (accessed May 20, 2018 )

62.

Wakefield

, Koopman

, Keplinger

, et al.: Effect of home telemonitoring on glycemic and blood pressure control in primary care clinic patients with diabetes. Telemed J E Health, 2014; 20:199–205.

63.

Arora

, Peters

, Burner

, et al.: Trial to examine text message-based mHealth in emergency department patients with diabetes (TExT-MED): a randomized controlled trial. Ann Emerg Med, 2014; 63:745–754.e6.

64.

Hozo

, Djulbegovic

, Hozo

: Estimating the mean and variance from the median, range, and the size of a sample. BMC Med Res Methodol, 2005; 5:13.

65.

Cho

J-H

, Kwon

H-S

, Kim

H-S

, et al.: Effects on diabetes management of a health-care provider mediated, remote coaching system via a PDA-type glucometer and the Internet. J Telemed Telecare, 2011; 17:365–370.

66.

Torbjørnsen

, Jenum

, Småstuen

, et al.: A low-intensity mobile health intervention with and without health counseling for persons with type 2 diabetes, part 1: baseline and short-term results from a randomized controlled trial in the Norwegian part of RENEWING HEALTH. JMIR Mhealth Uhealth, 2014; 2:e52.

67.

Cradock

, ÓLaighin

, Finucane

, et al.: Behaviour change techniques targeting both diet and physical activity in type 2 diabetes: a systematic review and meta-analysis. Int J Behav Nutr Phys Act, 2017; 14:18.

68.

Higgins

, Thompson

, Deeks

, Altman

: Measuring inconsistency in meta-analyses. BMJ, 2003; 327:557–560.

69.

Borenstein

, Hedges

, Higgins

JPT

, Rothstein

: Notes on subgroup analyses and meta-regression. In: Introduction to Meta-Analysis, Hoboken, New Jersey, USA. John Wiley & Sons, Ltd., 2009, pp. 205–212.

70.

Egger

, Davey Smith

, Schneider

, Minder

: Bias in meta-analysis detected by a simple, graphical test. BMJ, 1997; 315:629–634.

71.

Duval

, Tweedie

: A nonparametric “Trim and Fill” method of accounting for publication bias in meta-analysis. J Am Stat Assoc, 2000; 95:89–98.

72.

Guyatt

, Oxman

, Vist

, et al.: GRADE: an emerging consensus on rating quality of evidence and strength of recommendations. BMJ, 2008; 336:924–926.

73.

GRADE Handbook for Grading Quality of Evidence and Strength of Recommendations [article online]. Updated October. 2013. http://gdt.guidelinedevelopment.org/app/handbook/handbook.html (accessed October 20, 2017 )

74.

GRADEpro Guideline Development Tool [Software]. McMaster University, 2015 (developed by Evidence Prime, Inc.). 2015. gradepro.org (accessed July 9, 2018 )

75.

Agboola

, Jethwani

, Lopez

, et al.: Text to move: a randomized controlled trial of a text-messaging program to improve physical activity behaviors in patients with type 2 diabetes mellitus. J Med Internet Res, 2016; 18:e307.

76.

Capozza

, Woolsey

, Georgsson

, et al.: Going mobile with diabetes support: a randomized study of a text message-based personalized behavioral intervention for type 2 diabetes self-care. Diabetes Spectr, 2015; 28:83–91.

77.

Cho

, Kim

, Yoo

, et al.: An Internet-based health gateway device for interactive communication and automatic data uploading: clinical efficacy for type 2 diabetes in a multi-centre trial. J Telemed Telecare, 2017; 23:595–604.

78.

Dario

, Toffanin

, Calcaterra

, et al.: Telemonitoring of type 2 diabetes mellitus in Italy. Telemed J E Health, 2017; 23:143–152.

79.

Hansen

, Perrild

, Koefoed

, Zander

: Video consultations as add-on to standard care among patients with type 2 diabetes not responding to standard regimens: a randomized controlled trial. Eur J Endocrinol, 2017; 176:727–736.

80.

Holmen

, Torbjornsen

, Wahl

, et al.: A mobile health intervention for self-management and lifestyle change for persons with type 2 diabetes, part 2: one-year results from the Norwegian randomized controlled trial RENEWING HEALTH. JMIR Mhealth Uhealth, 2014; 2:e57.

81.

Khanna

, Stoddard

, Gonzales

, et al.: An automated telephone nutrition support system for Spanish-speaking patients with diabetes. J Diabetes Sci Technol, 2014; 8:1115–1120.

82.

Kim

, Sun

, Yang

, et al.: Randomized, open-label, parallel group study to evaluate the effect of internet-based glucose management system on subjects with diabetes in China. Telemed J E Health, 2016; 22:666–674.

83.

Kleinman

, Shah

, et al.: Improved medication adherence and frequency of blood glucose self-testing using an m-Health platform versus usual care in a multisite randomized clinical trial among people with type 2 diabetes in India. Telemed J E Health, 2017; 23:733–740.

84.

Liou

, Soon

, Chen

, et al.: Shared care combined with telecare improves glycemic control of diabetic patients in a rural underserved community. Telemed J E Health, 2014; 20:175–178.

85.

Tang

, Overhage

, Chan

, et al.: Online disease management of diabetes: engaging and motivating patients online with enhanced resources-diabetes (EMPOWER-D), a randomized controlled trial. J Am Med Inform Assoc, 2013; 20:526–534.

86.

Tildesley

, Mazanderani

, Chan

JHM

, Ross

: Efficacy of A1C reduction using Internet intervention in patients with type 2 diabetes treated with insulin. Can J Diabetes, 2011; 35:250–253.

87.

Wang

, Zhang

, Feng

, et al.: Telemedicine in the management of type 2 diabetes mellitus. Am J Med Sci, 2017; 353:1–5.

88.

Welch

, Zagarins

, Santiago-Kelly

, et al.: An internet-based diabetes management platform improves team care and outcomes in an urban Latino population. Diabetes Care, 2015; 38:561–567.

89.

Wild

, Hanley

, Lewis

, et al.: Supported telemonitoring and glycemic control in people with type 2 diabetes: the Telescot diabetes pragmatic multicenter randomized controlled trial. PLoS Med, 2016; 13:1–16.

90.

Greenwood

, Gee

, Fatkin

, Peeples

: A systematic review of reviews evaluating technology-enabled diabetes self-management education and support. J Diabetes Sci Technol, 2017; 11:1015–1027.

91.

Schulz

, Chalmers

, Hayes

, Altman

: Empirical evidence of bias. Dimensions of methodological quality associated with estimates of treatment effects in controlled trials. JAMA, 1995; 273:408–412.

92.

Wood

, Egger

, Gluud

, et al.: Empirical evidence of bias in treatment effect estimates in controlled trials with different interventions and outcomes: meta-epidemiological study. BMJ, 2008; 336:601–605.

93.

Health Quality Ontario: Behavioural interventions for type 2 diabetes: an evidence-based analysis. Ont Health Technol Assess Ser, 2009; 9:1–45.

94.

Avery

, Flynn

, Dombrowski

, et al.: Successful behavioural strategies to increase physical activity and improve glucose control in adults with type 2 diabetes. Diabet Med, 2015; 32:1058–1062.

95.

Kassavou

, Sutton

: Automated telecommunication interventions to promote adherence to cardio-metabolic medications: meta-analysis of effectiveness and meta-regression of behaviour change techniques. Health Psychol Rev, 2018; 12:25–42.

96.

Thompson

, Higgins

: How should meta-regression analyses be undertaken and interpreted?. Stat Med, 2002; 21:1559–1573.

97.

Michie

, Wood

, Johnston

, et al.: Behaviour change techniques: the development and evaluation of a taxonomic method for reporting and describing behaviour change interventions (a suite of five studies involving consensus methods, randomised controlled trials and analysis of qualitative data). Health Technol Assess, 2015; 19:1–188.

98.

Borrelli

: The assessment, monitoring, and enhancement of treatment fidelity in public health clinical trials. J Public Health Dent, 2011; 71:S52–S63.

99.

Johnston

, Johnston

, Wood

, et al.: Communication of behaviour change interventions: can they be recognised from written descriptions?. Psychol Health, 2018; 33:713–723.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.28 MB