Effectiveness of mobile applications in diabetic patients’ healthy lifestyles: review of systematic reviews.

Background: Diabetes mellitus (DM) is currently a major public health problem worldwide. It is traditionally approached by a clinical inpatient relationship between patients and healthcare professionals. However, the rise in of the use of new technologies, particularly mobile applications, is revolutionizing the traditional healthcare model with the introduction of telehealthcare. Objective: (1) Examine the mobile applications that address lifestyles to improve the metabolic control of adult patients with Diabetes Mellitus. (2) Describe the characteristics of the used mobile applications, identify the healthy lifestyles they target, and describe any of their adverse effects. Methods: Review of systematic reviews following Cochrane Collaboration and Joanna Briggs Institute guidelines. We included studies that used any mobile application to help patients improve DM self-management by focusing on healthy lifestyles. Studies needed to include a control group receiving regular care without using mobile devices. In May 2018, Medline, Embase, Cochrane, LILACS, PsychINFO, Cinahl and Science Direct were searched, updated in May 2019. The methodological quality of the studies was assessed by the Amstar-2 tool. Results: Seven systematic reviews of 798 articles were initially selected for the analysis. Interventions lasted 1-12 months. Twenty-three different mobile applications were identied. They were all related to eating and physical activity. Signicant changes were found in HbA1c values, body weight and BMI, but no clear improvement was observed in others like lipid prole, quality of life or blood pressure. No signicant adverse effects were identied. Conclusions: Clearly evidence appeared for using mobile applications to improve glycemic control in diabetic patients in the short term, but not for long-term benets. Thus carrying out further studies is necessary to learn about the long-term effectiveness of mobile applications to promote DM patients’ healthy lifestyles.

also attempts to answer the question: are mobile applications that deal with lifestyles to improve the metabolic control of adult patients with DM effective?
In line with this question, we consider that the main objective is to examine mobile applications that address lifestyles to improve the metabolic control of adult patients with DM. As secondary objectives, we aim to describe the characteristics of the used apps, identify the healthy lifestyle aspect they target, and describe any adverse effects of their uses.

Methods
Design A review of systematic reviews was performed. This design allowed us to compare and verify the ndings of relevant reviews in response to similar review questions, which facilitates a view and clear understanding of a broad theme area [31]. Its main purpose is to summarize evidence from many research sources. Compared to a systematic review limited to one treatment or one intervention, this type of review offers a broader view of many interventions. It is useful for health-related technology assessments whose objective is to inform about patterns for clinical practice where many handling options must be considered and evaluated.
To implement the review, we followed the Cochrane Collaboration [32] and Reviewers Manual of the Joanna Briggs Institute [33] guidelines. To prepare this report, we took into account the PRISMA proposal [34] recommendations.
The protocol of this systematic review was registered in PROSPERO with code CRD42019133685 [35].
Inclusion and exclusion criteria 1. Type of study: Systematic review and/or meta-analysis.
2. Population: Patients aged over 18 years diagnosed with DM regardless of the type of treatment followed. 3. Intervention: using any App to help patients improve DM self-management by a healthy lifestyles approach. The mobile application can be used exclusively or combined with other types of interventions. The study needed to describe the type of employed mobile application and the addressed lifestyle aspect. Studies in which text messages were sent via mobile phones, independently of the use of any mobile application being excluded. 4. Comparison: a control group receiving regular care without using mobile devices toward healthier lifestyles. The studies that did not report any primary or secondary outcomes were excluded. Studies had to include a comparative analysis of the outcomes measured at the baseline and those measured at the end of the intervention.

Sources of information and search strategy
To de ne the search strategy, we used a set of studies that was rst divided to facilitate the location and de nition of the main descriptors. The key terms and their synonyms were identi ed in the Medline (via PubMed), Cinahl and Google Scholar databases. The identi ed terms were combined to calibrate a search syntax in Medline by bearing the sensitivity and speci city criteria in mind.
Next a second search using all the identi ed keywords and index terms was performed in the following electronic databases: Medline, Embase, Cochrane, Lilacs, PsychINFO, Cinahl and Science Direct. Articles were selected in May 2018 and updated in May 2019. There were no limits in terms of language or year of publication. The syntax of the bibliographic search carried out in Medline is attached (Multimedia Appendix 1).
The search for unpublished studies was carried out in: Open Grey, ProQuest Dissertations & Theses Global, Theseo, Networked Digital Library of Theses and Dissertations (NDLTD).
Finally, the reference lists of all the identi ed reports and articles were searched for additional studies. The search was completed by hand searching and reverse searching in reference journals specializing in smoking, diabetes and e-health.
The Mendeley reference management software was used to sort the bibliographic citations obtained in the search to eliminate duplicate citations and to order all the studies to facilitate the analysis.

Study Selection
The complete study selection process was carried out by peer review. In the event of disagreement between the two reviewers, a third party was invited to participate.
First, a selection of the studies was made based on their title and/or Abstract according to the prede ned inclusion criteria. Duplicates and those articles which, given their title and/or Abstract did not match our eld of interest, were eliminated. None of the authors of the review was blinded to the titles of journals or the authors or institutions of the study.
Second, a new selection was made by reading the full articles and verifying that they met the inclusion criteria.

Assessing methodological quality
After selection, studies were assessed as to their methodological quality with the Measurement Tool to Assess Systematic Reviews-2 (Amstar-2) [36].
Amstar-2 allows the assessment of systematic reviews, which include randomized controlled clinical trials (RCTs) and non randomized health interventions. This questionnaire consists of 16 items with several response categories: "yes", when the answer is positive; "no", when the item is not present or contains insu cient information to respond; "partial yes", when the item is partially present. Seven domains (items 2, 4, 7, 9, 11, 13 and 15) are considered critical because they can signi cantly affect a review's validity and conclusions. It classi es the quality of systematic reviews into four levels: high, moderate, low, very low. This methodological quality assessment was made in pairs. Although provided for in the protocol, it was not necessary to consult a third reviewer because an agreement between reviewers was reached. We included reviews of moderate or high methodological quality in our review.
Reviews with low or very low quality were excluded so as not to distort the evidence of the conclusions.

Data extraction and variables
The data from the studies were extracted, synthesized and recorded on an Excel sheet which included: Amstar-2 items, rst author, year, source, search strategy, number of included studies, total number of participants, study design, intervention time (months) and outcome measures. Data were extracted independently by two reviewers. The reliability and quality of the extracted data were ensured by cross-checking, rereading the complete studies and reviewing the collected data. were included in the review? 10 Did the review authors report on the sources of funding for the studies included in the review? 11 If meta-analysis was performed did the review authors use appropriate methods for statistical combination of results? 12 If meta-analysis was performed, did the review authors assess the potential impact of risk of bias in individual studies on the results of the meta-analysis or other evidence synthesis? 13 Did the review authors account for risk of bias in individual studies when interpreting/ discussing the results of the review? 14 Did the review authors provide a satisfactory explanation for, and discussion of, any heterogeneity observed in the results of the review? 15 If they performed quantitative synthesis did the review authors carry out an adequate investigation of publication bias (small study bias) and discuss its likely impact on the results of the review? 16 Did the review authors report any potential sources of conflict of interest, including any funding they received for conducting the review?
NA=Not applicable.

Characteristics and quality of the systematic reviews
The main characteristics of the seven systematic reviews included in this review are shown in Table 2. The complete analysis of their methodological quality is found in Table 1.
Although differences were observed in search methods, databases, inclusion and exclusion criteria, data extraction, quality assessments and statistical analyses, the seven systematic reviews generally provided an extensive description of the used methods, as well as the quality and general characteristics of the studies they included.
The studies included in the seven systematic reviews were conducted mainly in the United States, Asia and Europe. When we analyzed the original studies included in the reviews, we considered that 28 studies appeared in more than one systematic review.
The number of studies included in each analyzed review fell within the 8-19 range. In all cases, except for one study with a cross design featured in the review by Porter et al. [27], random clinical trials (RCTs) were included which assessed interventions based on using Apps to manage DM in adult patients. Interventions lasted between 1 and 12 months. The vast majority of the studies analyzed HbA1c as the primary outcome measure, which was also used to perform a meta-analysis in the reviews by Pal et al. [14], Bonoto et al. [41], Lunde et al. [48] and Wu et al. [49].

Mobile applications and the lifestyles they address
The identi ed mobile Apps and lifestyles are provided in detail in Table 3.
Twenty-three different apps were identi ed. The only App that was included in all seven systematic reviews was "BlueStar Diabetes", which offers free access for Apple and Android, and addresses food. We were unable to identify the commercial name of one App.
Regarding the lifestyles covered by Apps, eight included different questions about DM self-control in relation to food, ve to physical activity, and 10 combined both these components.
Most research into the effectiveness of Apps was conducted with individuals with DM2. Although their effectiveness for DM1 was also studied, no systematic reviews included studies about Gestational Diabetes (GD).  [14] performed a meta-analysis of the 11 studies, which provided su cient data on HbA1c levels. They found statistically signi cant differences between the results of the intervention groups and the control groups, with lower HbA1c values in the intervention subjects. The impact of the intervention was signi cantly stronger (P <.001) in the three studies that used mobile phones, with an effect on HbA1c of -5,5 mmol/mol or -0.5% (95%CI: -0.74 to -0.26; P=.59; I²=0%). However, the effects of interventions seemed to disappear with time as the analysis of the results of the studies with a duration equal to or longer than 6 months was not statistically signi cant.
In four of the nine studies they analyzed, Porter et al. [27] found a statistically signi cant improvement in HbA1c of diabetics in the intervention group compared to the control group, while the other studies observed no signi cant differences between the two groups.
In the review by Bonoto et al. [41], in six of the 12 studies that included HbA1c values, a statistically signi cant difference was found in the reduction of this parameter in favor of intervention groups. Likewise, the meta-analysis carried out in this review showed the effectiveness of using mobile applications to control diabetes, with an average difference of -0.44% (95%CI: 0.59 to -0.29; P<.10; I²=32%), which was statistically signi cant (P<.001).
Veazie et al. [45] identi ed two Apps to control DM1 (Glucose Buddy and Diabeo Telesage) and three others for DM2 (Gather Health, BlueStar and WellTang), whose use demonstrated better statistics and clinical signi cance for HbA1c than the control groups (CG).
Lunde et al. [48] included three studies to evaluate the effectiveness of short-term apps, and four long-term studies. The result was a signi cant decrease in HbA1c in both cases, although the quality of the evidence was poor for the short-term and moderate for the long-term Apps.
To assess the effect of interventions based on using Apps on diabetic patients' healthy lifestyles, we also analyzed changes in body weight or BMI as a result of the intervention. Lunde et al. [48], included three studies in which the IG reduced weight signi cantly [50][51][52]. In the review by Pal et al. [14], only two of the ve studies using mobile devices in their interventions reported changes in body weight and BMI [51,54]. However, these authors found no signi cant differences between the intervention groups and control groups. Bonoto et al. [41] did not observe any signi cant differences in their combined analysis of the results of four studies with data on changes in participants' body weight [mean difference: -0.39 (95%CI: -1.43 to 0.66; P=.47; I²=0%)]. Similarly in their review, Porter et al. [27] described that none of the six studies with body weight or BMI data found signi cant changes linked with the intervention. Only one study indicated a slight reduction in the BMI of all the groups, but did not provide any data [53]. Finally, none of the participants in the studies analyzed by Veazie et al. [45] reported improvements in body weight or BMI.

Secondary Outcome Measures
Health-related quality of life (HRQL) was analyzed as a secondary outcome measure of interventions based on the use of Apps. Of the 16 studies included in the review by Pal et al. [14], ve provided results on HRQL. However, these authors did not observe any signi cant improvement in the HRQL of the intervention subjects compared to the control subjects. Bonoto et al. [41] found that three studies yielded positive and statistically signi cant changes in both the quality of life of and satisfaction with the treatment of the patients in the intervention group. The improvements reported by the participants using Apps included perceiving hyperglycemia episodes, social relationships, feeling less fear of hypoglycemia and that Apps helped them to control their treatment and to maintain healthier dietary habits [54][55][56]. Conversely, Veazie et al. [45] found no evidence for improvement in the quality of life of the diabetic patients who used mobile Apps.
Some of the studies included in the seven analyzed reviews provided results on the lipid pro le of diabetic subjects, which may be related to changes in their dietary habits and lifestyles. In connection with this, Pal et al. [14] carried out a meta-analysis with data from seven studies and found that the effect of the intervention on the participants' lipid pro le was not signi cant (P = .17), with a mean difference to the control group of -0.11 (95%CI:-0.28 to 0.05; P=.03; I²=57%). A signi cant improvement in the lipid pro le was observed in only one intervention study based on using mobile devices, speci cally a reduction in TC, LDL-C and TG levels. Bonoto et al. [41] and Akbari et al. [47] also found no signi cant differences in their meta-analysis of TC, HDL-C, LDL-C and TG levels. Porter et al. [27] and Veazie et al. [45] identi ed one study which observed a signi cant lowering (P = .04) of triglyceride levels, but not the remaining lipids, in the participants who used a mobile app to control DM1 (Diabetes Interactive Diary) [55].
Regarding changes in diabetic patients' blood pressure, one study included in the review by Pal et al. [14] found a statistically signi cant decrease in SBP (127±14 mmHg to 120±19 mmHg; P=.001) and DBP (78±10 mmHg to 74±8 mm Hg, P<.001) in the intervention group [57]. Bonoto et al. [41] did not observe any signi cant differences for the intervention and control groups of four studies in participants' SBP and DBP. Similarly, no intervention analyzed in the review by Veazie et al. [45] revealed participants' improved blood pressure.
Finally in a combined analysis of fasting blood glucose results from four studies, Bonoto et al. [41] reported no signi cant differences between the intervention and control groups. In the review by Porter et al. [27], two studies described a signi cantly more marked reduction (P <.01) in fasting blood glucose levels for the intervention group than for the control group [56,59], while two other studies did not observe any signi cant differences between these two groups [56,58]. Veazie et al. [45] identi ed one app (WellTang), whose use for DM2 demonstrated better fasting blood glucose values than for the control subjects.

Adverse effects of interventions
One important aspect to bear in mind when assessing the e ciency of Apps designed to improve DM self-control is that adverse effects may appear.
Of all the studies included in the seven reviews analyzed herein, a few describe adverse reactions reported by the participants of the intervention groups. Pal et al. [14] found a non signi cant increase in the frequency of mild hypoglycemia episodes in the intervention group, with no differences in severe or nocturnal hypoglycemia episodes [54]. Five studies included in the review by Bonoto et al. [41] reported hypoglycemia episodes. One reports averages of 30 and 33 mild episodes in the the intervention and the control group, respectively, and a severe episode in the control group [56]. In three other studies, no signi cant differences were observed between groups [55,60,61]. In a fth study, the relative risk of severe hypoglycemia episodes was lower in IG (0:14; 95%CI: 0.07-0.029) [54]. One of the studies in the review by Wu et al. [49] found no statistically signi cant differences between the hypoglycemic events of IG and CG [56]. Two mobile apps (Diabetes Diary and Diabetes Interactive Diary) reviewed by Veazie et al. [45] showed improvements in hypoglycemic episodes of DM1.

Discussion
This work aimed to provide a broad view of the research conducted about using mobile Apps that address lifestyles to control and manage DM.
As far as we know, this is the rst review of systematic reviews (umbrella review) that considers that the effect of Apps on improving the metabolic control of DM patients.
Although many systematic reviews were found by the literature search, very few met su cient quality criteria to be included in the present work.
Given the variability of Apps and outcome measures, a decision was made to not include the data of the original studies in a meta-analysis.
The seven systematic reviews that acted as the basis for our study showed a clear bene t of mobile Apps that deal with lifestyles to improve DM patients' short-term glycemic control. These data need to be interpreted cautiously because other reasons could have intervened in HbA1c lowering, such as the persuasion of Apps to modify lifestyles [49] or health professionals' access via Apps that include remote communication tools [41]. Indeed the results of a recent pragmatic multicenter clinical random controlled trial did not indicate any differences between the IG (intervention group) and the CG for the primary clinical outcome of glycemic control measured by HbA1c, which also occurred in secondary outcomes like quality of life and behaviors stemming from medical healthcare uses [65].
Finding a convincing explanation for this phenomenon might be a complex matter. Our ndings coincide with those of other authors who stated that most studies did not take into account the basis of behavioral health theories when developing Apps [66].
The scienti c literature about behavioral health models in which Apps participate reveal very little discussion about behavioral health theories or models that provide a basis on which to support intervention [67]. One possible convincing explanation could lie in the theory of controlling interventions based on other theories. This theory postulates that a synergic association exists between receiving information about someone's behavior (via "self-control" or feedback") and obtaining a strategy with which to act on this information ("planning action" or "information as to how and where to perform behavior"). The former provides a sign and/or motivation for the latter.
The review by Pal et al. [14] does not provide su cient evidence with which interventions improve cardiovascular risk factors (blood pressure, lipid pro le and body weight) or cognitive, behavioral or emotional outcomes.
Our ndings about long-term effects coincide with those published by other authors by identifying only one positive short-term effect on HbA1c. Indeed very few studies have found a positive long-term effect of Apps on HbA1c [68].
In the subgroups analysis, modi cation of lifestyles would have a stronger effect on DM2 patients than DM1 patients. This could be explained by the DM1 control depending largely on questions about administering insulin and not about amending lifestyles, which are the main cause for DM2 to appear [6]. Wu et al. consider that to ensure Apps being very e cient, a speci c design is necessary for all DM subtypes [49]. This must also be done extensively for GD.
Using Apps to control DM seems to reinforce the self-control perception by providing DM patients with better health information and education.
It can also increase patient security as to how to deal with their disease by mainly reducing their fear of not knowing how to treat possible hypoglycemia episodes [49,58], and to improve their quality of life [41]. Nonetheless, the impact of these Apps on long-term outcomes, such as quality of life, high blood pressure or frequent diabetes complications (neuropathy and retinopathy), remain unclear. More rigorous longer terms studies are still required to carefully consider the potential of the interaction between patients and health professionals/study personnel (Veazie et al. [45] and Lunde et al. [48]), along with their effects on DM-related mortality [69].
All the studied Apps focused on two lifestyles, which are most important to control DM well: food and physical exercise. This makes sense as both questions are dealt with simultaneously in clinical practice because they are closely related [6]. However, no Apps that addressed other lifestyles with a high prevalence, like smoking habit, were included [70].
Apart from the characteristics that are typical of Apps to deal with food and physical exercise, others were described because they appear to contribute to improve DM patients' glycemic control to a great extent, such as being able to store and feedback data about blood glucose, support to control doses and therapy with medicines being met and, nally, access to communicate with health professionals.
Among all the studied works, very few described the adverse effects of using Apps. It is necessary to bear in mind that despite potential bene ts for patients, the Apps used to calculate insulin doses imply the risk of incorrect dose recommendations that range from those that lead to suboptimum disease control to potentially lethal consequences [71].
Indeed until quite recently, the growth and early adoption of technology both tend to lie mainly in healthcare suppliers' hands. Nowadays, the digital era has extended patients' access to technology. Most of the population has access to portable devices, mobile Apps, with better access to electronic medical records and health data over the Internet. The main leading role of technology in health management, and greater user accessibility and autonomy, are changing the patient's position from a passive person receiving health care to someone who also participates in managing his/her health.

Limitations
Some limitations appear to interpret and extrapolate the ndings of this review. The few data that cover a period longer than 12 months is a major limitation if we consider the chronic condition of DM, and many of its complications are mani ested after the disease has been longstanding. Further evidence is necessary to contemplate studies that include longer follow-up peridos.
Evaluating the quality of the studied basic reviews was done using AMSTAR for its clear validity and reliability, which helps to identify the best quality evidence for each outcome without prejudicing the quality of the original studies that acted as the basis of the studied reviews. Given this scale's subjectivity component, the evaluation was done both step-wise and independently to minimize this risk.
Another important limitation arose when comparing studies as a result of the low clinical homogeneity of the seven included systematic reviews. Differences were observed in the literature search methods, the inclusion and exclusion criteria, the evaluation of the quality of individual studies, the extraction of primary and secondary outcome measures, and the analysis of the results. The meta-analyses included in ve reviews analyzed only a few outcome measures.

Conclusions
The outcomes of this review will support the use of Apps to improve short-term glycemic control in DM patients. All the examined Apps centered on dealing with food and physical exercise. No signicifant adverse effects were identi ed for users of Apps.
Apps' bene cial long-term effect on health for diabetic patients was much weaker. Therefore, more in-depth research is necessary as far as the design, features and effectiveness of studied mobile Apps are concerned to encourage healthy lifestlyes for DM patients. All the authors were involved in drafting the manuscript. All the authors contributed to develop the selection criteria, the bias risk assessment strategy and the data extraction criteria. AAMQ developed the search strategy. ACF provided her experience in technology assessment. FJRC provided his experience in research into DM. All the authors read, provided comments on and approved the nal manuscript.

Con icts of interest
None declared.