This is an open-access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work, first published in the Journal of Medical Internet Research, is properly cited. The complete bibliographic information, a link to the original publication on http://www.jmir.org/, as well as this copyright and license information must be included.
Brief automated messages have the potential to support self-management in people with type 2 diabetes, but their effect compared with usual care is unclear.
To examine the effectiveness of interventions to change lifestyle behavior delivered via automated brief messaging in patients with type 2 diabetes.
A systematic literature review of controlled trials examined the impact of interventions, delivered by brief messaging, and intended to promote lifestyle change in people with type 2 diabetes, on behavioral and clinical outcomes. Bibliographic databases searched included Medline, Embase, CINAHL, PsycINFO, and ISI WoK. Two reviewers independently screened citations. We extracted information on study risk of bias, setting (high versus low- and middle-income countries) and intervention characteristics (including use of theory and behavior-change techniques). Outcome measures included acceptability of the interventions and their impact on 1) determinants of lifestyle behavior (knowledge about diabetes, self-efficacy, attitudes towards self-management), 2) lifestyle behavior (diet, physical activity), and 3) clinical and patient-reported outcomes. Where possible, we pooled data using random-effects meta-analyses to obtain estimates of effect size of intervention compared to usual care.
We identified 15 trials (15 interventions) meeting our inclusion criteria. Most interventions were delivered via short message service text messaging (n=12) and simultaneously targeted diet and physical activity (n=11). Nine interventions consisted of unidirectional messages, whereas six consisted of bidirectional messages, with patients receiving automated tailored feedback based on self-reported data. The acceptability of the interventions, and their impact on lifestyle behavior and its determinants, were examined in a low proportion of trials, with heterogeneous results being observed. In 13 trials (1155 patients) where data were available, there was a difference in glycated hemoglobin of -0.53% (95% CI -0.59% to -0.47%) between intervention groups compared to usual care. In five trials (406 patients) there was a non-significant difference in body mass index of -0.25 kg/m2 (95% CI -1.02 to 0.52). Interventions based on unidirectional messages produced similar effects in the outcomes examined, compared to those based on bidirectional messages. Interventions conducted in low- and middle-income countries showed a greater impact than those conducted in high-income countries. In general, trials were not free of bias and did not use explicit theory.
Automated brief messages strategies can improve health outcomes in people with type 2 diabetes. Larger, methodologically robust trials are needed to confirm these positive results.
The number of people with type 2 diabetes worldwide is currently estimated to be 387 million, and is expected to increase to 592 million by 2035 [
Available evidence suggests that better control of blood glucose, blood pressure, and cholesterol levels would delay the onset of complications, and thereby prevent premature deaths among those already diagnosed with diabetes [
Recommendations for supporting diabetes self-management are now widely incorporated in clinical practice guidelines [
A number of systematic reviews have provided evidence for the effectiveness of mobile-phone based interventions on self-management of long term conditions [
The primary objective of this systematic review was to examine the effectiveness of brief messages in improving glycemic control through promoting healthy eating and increasing physical activity, compared to usual care. We focused specifically on interventions delivered via mobile devices to people with type 2 diabetes. Secondary objectives included 1) examining the extent to which interventions have used explicit theory, 2) examining the behavior-change techniques used, 3) examining the acceptability of the interventions, 4) examining their impact on lifestyle-change behavior and its determinants, 5) examining their impact on other clinical (blood pressure, lipids, and weight) and patient-reported outcomes, 6) comparing the impact of unidirectional vs bidirectional messages, and 7) exploring the specific impact of the interventions in HIC and LMIC.
The study was planned, conducted, and reported according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines [
Specific search strategies were designed for the following databases (
The search strategy combined Medical Subject Headings terms and free-text keywords (
We included controlled trials examining the impact of interventions intended to promote lifestyle changes on diet and physical activity among people with type 2 diabetes. All interventions were delivered by brief messaging using mobile devices, and were compared on behavioral and clinical outcomes, versus usual care.
The main component of eligible interventions was the provision of information via brief messaging systems, characterized by automated messages (including computer-generated messages following an algorithm), which were tailored/custom-made personalized messages, or bulk messages. Messages had to be received via devices using mobile technology, such as mobile phones, smart-phones or hand-held computers. Messaging systems were those using the following technologies: SMS, automated email, or software apps. These systems could be used to provide information through one-way (unidirectional) messages or to facilitate two-way communication (interactive or bidirectional). Unidirectional messages were conceptualized as messages sent from the providers or researchers to the participants. Bidirectional messages were conceptualized as those involving a two-way communication regarding self-monitoring data that was sent by the participants, who in return received real-time automated brief messages providing tailored feedback. Eligible interventions targeted healthy eating, physical activity, or both.
Inclusion criteria for study design specifically focused on controlled clinical trials, including randomized controlled trials (RCTs), cluster RCTs, non-randomized controlled trials, and crossover studies. Only studies with control or comparator arms that consisted of patients receiving usual (standard) clinical care, or a minimal intervention (ie an intervention that is unlikely to produce any effect, such as sending non-health related messages, but allows blinding participants to condition allocation) were eligible for inclusion. Eligible studies included adult (at least 18 years old) patients with type 2 diabetes mellitus (with or without comorbid conditions) and had to be set in the community or in any primary, secondary, or tertiary care setting.
Studies reporting at least one of the following outcome measures were included: acceptability by recipients, determinants of change in lifestyle (namely knowledge, attitudes and self-efficacy on lifestyle modifications), lifestyle behaviors that impact on diabetic control (diet and physical activity), and clinical outcomes (glycated hemoglobin [HbA1c], body mass index [BMI] or body weight, lipids, blood pressure, and waist circumference).
Trials were excluded if they examined the use of messages created by a clinician based on a clinical judgement of a patient’s disease status (ie, not automated), had a proportion of patients with type 2 diabetes lower than 90%, evaluated a multifaceted intervention in which brief messages were not the main component of the intervention, or were published only in the form of conference abstracts. No language restrictions were applied.
A preliminary screen for eligibility was followed by retrieval and assessment of full texts of the selected articles. Studies that met the inclusion criteria were included for data extraction. All citations were independently screened by two reviewers. Any disagreements were solved by consensus with a third reviewer.
Structured forms were used to extract data about the trial design, trial setting (HIC vs LMIC according to the World Bank classification [
The extent to which the trials used theory explicitly in the development and evaluation of the interventions was assessed using an established coding-scheme which contains 19 items. This scheme assessed whether a theory was mentioned, how theories were used in intervention design and in the selection of intervention techniques, how intervention evaluations tested theory, and the implications of the results for future theory development [
We used the Cochrane Collaboration’s tool for risk of bias assessment [
We examined and synthesized the acceptability of the interventions, and their impact on 1) determinants of change in lifestyle, 2) behavior (diet and/or physical activity), and 3) clinical outcomes. Outcomes in all studies were examined and classified as measuring one of these three domains. Variables that measured other domains were not included in the analysis. For all pooled outcomes we used subgroup analyses to examine potential differences between types of messages (unidirectional vs bidirectional) and settings (HIC vs LMIC).
From each study we extracted the mean and standard deviation (SD) of HbA1c levels and BMI, contacting study authors when the information was not available. We transformed this information into weighted mean difference (95% CI), and pooled the data using random-effects models. Where SD of the change between intervention and control group for an outcome was not provided, we derived them from baseline and final SDs, assuming a correlation of 0.5 [
Search results are summarized in the PRISMA flow diagram (
Flowchart of articles included at each stage of the screening process.
Each trial assessed only one intervention. Thirteen trials used an RCT design, whereas the remaining two used a cluster-RCT [
On average, interventions lasted 7 months (SD=4). The majority of trials (n=12) used SMS technology to deliver the messages, whereas the rest were based on graphical information presented to the patients [
Twelve behavior-change techniques were used in total. Most interventions used multiple techniques. Most frequently used techniques consisted of 1) providing information about the consequences of inaction, 2) providing instructions about how to perform a behavior, and 3) providing feedback on performance – each of which were used in eight interventions. Whereas all studies examined the impact of the interventions on clinical outcomes, behavior-change was only measured in six of them, with a wide range of instruments being used.
Only a minority of the trials presented low risk of bias, and none was completely free of bias (
Five trials examined whether unidirectional messaging interventions were acceptable to participants. Three [
Four trials examined participants’ acceptability of bidirectional messaging interventions, consistently observing high acceptability in terms of easiness to use systems, usefulness, and general satisfaction [
The impact of unidirectional messages on behavior-change determinants was only examined by two trials. One study observed significant improvement in diabetes knowledge and self-efficacy, but not in self-management attitudes [
Four trials examined the impact of unidirectional messages on diet and physical activity. Two studies reported no effects [
Data from thirteen trials reporting the impact of the interventions on HbA1c [
A second meta-analysis examined the impact of the interventions on BMI. Five trials [
Other clinical outcomes were too heterogeneous to pool. Unidirectional messaging interventions led to significant reductions of blood pressure in one of the two trials examining this outcome [
Weighted mean difference in size of effect of intervention compared with “no treatment” for glycated hemoglobin. HbA1c, glycated hemoglobin; CI, confidence interval; N, number of participants; SD, standard deviation
The proportion of studies reporting positive effects was consistently higher for trials carried out in LMIC compared to HIC in all domains examined, including acceptability of the interventions (100% in LMIC vs 57% in HIC), impact on determinants of behavior-change (100% vs 0%), on physical activity and diet (67% vs 0%), and on clinical outcomes (100% vs 55%, see
The extent to which the trials explicitly used theory in relation to a number of criteria is reported in
Where theory was explicitly mentioned, two trials [
This systematic review identified 15 controlled trials examining the effectiveness of interventions to promote healthy eating and physical activity in people with type 2 diabetes, delivered via automated brief messaging sent to mobile devices. The interventions predominantly used SMS technology, addressed both diet and physical activity, and were not based on theoretical models of behavior-change. Our meta-analyses showed that automated brief messaging produced a clinically important and statistically significant effect on glycemic control (pooled effect on HbA1C= -0.53%,
This is the first systematic review specifically examining the impact of automated brief messages on self-management behavior in people with type 2 diabetes. Additional novel aspects of this review include an assessment of the extent to which interventions were based on theory, an assessment of the behavioral change techniques used, and an examination of the relative impact of this type of intervention in countries with different levels of economic development. Relevant trials were identified using a comprehensive search strategy and a large number of bibliographic sources.
In terms of limitations, our meta-analyses were restricted to glycemic control and BMI. Although we intended to conduct meta-analyses on other relevant outcomes (namely diabetes knowledge, attitudes toward diabetes self-management, and change in diet and physical activity-related behavior), these parameters were seldom measured, which represents a gap in evaluations to date. Finally, although formal tests on publication bias seemed to exclude its presence, we cannot completely rule out its existence.
The positive findings observed in our diabetes-specific review are consistent with findings from reviews examining a wide range of conditions. For example, a recent meta-analysis observed that SMS messages produced a small, positive, significant effect (g=0.29) on a broad range of healthy behaviors in patients with different types of long-term conditions [
The estimated 0.53% reduction in HbA1c observed in our meta-analysis is clinically important, as evidence suggests that every percentage point decrease in HbA1c over 10 years is associated with a risk reduction of 21% for deaths related to diabetes, 14% for myocardial infarctions, and 37% for microvascular complications [
Interventions based on the use of one-way messages produced a very similar effect to those based on two-way messages, which can be more tailored and usually require more complex technology, and are more resource intensive. This similarity is also consistent with findings from a recent trial evaluating text messages to improve treatment adherence in people with hypertension. This result is also consistent with findings from our systematic review on messages to improve adherence to diabetes medication, which observed that interventions exclusively based on brief messages produced a similar effect compared to more complex interventions combining messaging with monitoring strategies [
More than half of the trials did not include measures of behavior-change, and those that did reported mixed results. Previous systematic reviews also reported mixed results. For example, Cassimatis et al [
Although we did not observe a statistically significant reduction in BMI, we cannot exclude a small reduction. Since the text messages specifically targeted diet and physical activity behavior, we expected a greater effect. However, there were only a low number of trials reporting BMI as an outcome. Evidence from previous systematic reviews examining the impact of text messages on weight reduction is mixed, with some suggesting lack of consistent effects [
The interventions seem to have been acceptable to the recipients. There are many features related to mobile technology that may engage patients with the intervention. Some of these features include ease of use, convenience (eg messages need not be retrieved immediately), mobility (eg read at home or away), and frequent reinforcement (eg can read more than once). However, measures of acceptance and usability of the interventions assessed in the review were not obtained using a validated tool.
Our results suggested that the interventions were more effective in LMIC than in HIC. As far as we know, this is the first study comparing the impact of automated brief messages on long-term condition self-management between countries with different levels of economic development. Two recent reviews evaluated the impact of mHealth interventions in patients with long-term conditions living in LMICs, concluding that they are cost-effective and can produce a positive impact on clinical outcomes, health-related quality of life [
The studies in our review consistently supported the use of brief messages to promote healthier lifestyle behavior in patients with type 2 diabetes. However, available evidence is limited by several factors. First, most of the trials presented moderate or high risk of bias, mainly due to small sample sizes and inadequate blinding. To confirm the positive findings observed in our review, methodologically robust trials of greater size are very much needed. Second, although all the interventions specifically aimed to improve lifestyle behavior, behavior-change was measured in less than half of the studies. Where measured, a wide range of instruments were used, most having been designed
Additional research needs include an estimation of the cost-effectiveness of the interventions, an examination of their long-term impact, an understanding of what circumstances are effective (which features of the underlying health system and target population are helpful, and which features mitigate against them working), assessment of intervention safety, and an examination of their potential contribution to more comprehensive, multifaceted interventions [
Interventions based on the use of automated brief messages sent to mobile devices to promote lifestyle behavior can improve glycemic control in patients with type 2 diabetes, both in developed and developing countries. Larger and methodologically robust trials are needed to confirm these positive findings.
Bibliographic searches - registry of searches.
Bibliographic searches - search strategy (Medline).
Characteristics of the identified trials and interventions.
Cochrane summary risk of bias for the included trials (n=15).
Cochrane individual risk of bias for the included trials (n=15).
Main results of the studies identified.
Weighted mean difference in size of effect of intervention compared with “no treatment” for body mass index.
Differences between low- and middle- income countries and high-income countries in the impact of the interventions.
Degree of use of theory in the development of the interventions.
body mass index
glycated hemoglobin
high-income country
low- and middle-income countries
mobile health
Preferred Reporting Items for Systematic Reviews and Meta-Analyses
randomized controlled trial
standard deviation
short message service
AJF, CA, IR, DPF and PM designed the study. NR undertook the literature search. IR, PM, and CA were involved in the short-listing of identified studies, with input from AJF. Data extraction was undertaken by IR and PM, with input from AJF. Statistical analyses were undertaken by IR. All authors contributed to the final manuscript.
This work was funded through an NIHR Senior Investigator Award to AJF, who also received funding from the NIHR Oxford Biomedical Research Centre.
This review was carried out in partial fulfilment of CA’s fellowship at the University of Oxford under the Commonwealth Academic Fellowship scheme.
The funders had no role in study design, data collection or analysis, decision to publish, or preparation of the manuscript. The views expressed are those of the authors and not necessarily those of the funders.
None declared.