Telemedicine Application in the Care of Diabetes Patients: Systematic Review and Meta-Analysis

Background The impact of telemedicine application on the management of diabetes patients is unclear, as the results are not consistent among different studies. The objective of this study is to conduct a systematic review and meta-analysis of randomized controlled trials (RCTs) assessing the impact of telemedicine interventions on change in hemoglobin A1c (HbA1c), blood pressure, LDL cholesterol (LDL-c) and body mass index (BMI) in diabetes patients. Methods Electronic databases MEDLINE, Cochrane Central Register of Controlled Trials and LILACS were searched to identify relevant studies published until April 2012, supplemented by references from the selected articles. Study search and selection were performed by independent reviewers. Of the 6.258 articles retrieved, 13 RCTs (4207 patients) were included. Random effects model was applied to estimate the pooled results. Results Telemedicine was associated with a statistically significant and clinically relevant absolute decline in HbA1c level compared to control (mean difference -0.44% [-4.8 mmol/mol] and 95% confidence interval [CI] -0.61 to -0.26% [-6.7 to -2.8 mmol/mol]; p<0.001). LDL-c was reduced in 6.6 mg/dL (95% CI -8.3 to -4.9; p<0.001), but the clinical relevance of this effect can be questioned. No effects of telemedicine strategies were seen on systolic (-1.6 mmHg and 95% CI -7.2 to 4.1) and diastolic blood pressure (-1.1 mmHg and 95% CI -3.0 to 0.8). The 2 studies that assessed the effect on BMI demonstrated a tendency of BMI reduction in favor of telemedicine. Conclusions Telemedicine strategies combined to the usual care were associated with improved glycemic control in diabetic patients. No clinical relevant impact was observed on LDL-c and blood pressure, and there was a tendency of BMI reduction in diabetes patients who used telemedicine, but these outcomes should be further explored in future trials.


Introduction
Diabetes mellitus (DM) has taken epidemic proportions throughout the world and is considered among the most challenging health problems in the 21st century [1]. The disease is associated with a number of health-related complications, high morbidity and mortality rates and thus imposes substantial social and economic burdens worldwide [2].
Challenges in the care of diabetic patients require innovative management strategies to improve glycemic and pressoric control, as these are frequently above the desired goals [3]. The optimal management requires an organized, systematic and coordinated approach [2]. Strong evidence demonstrates beneficial effects of patient monitoring and education, focused on a prominent role of the individual self-care with the support of healthcare professionals [4]. This brings into focus the great advances in telecommunication technology and information technology, which can be exploited to improve diabetes management [5,6]. Telemedicine can be a strategy for closer monitoring and intervention to achieve not only better metabolic control, but also to help in the global care of individuals with multiple chronic illnesses [3,7,8].
Over the last decade, several studies have addressed the feasibility and efficacy of telemedicine strategies on the management of diabetes patients [9]. Many studies have proved it to be feasible, but the real impact of this of intervention in general and specific clinical situations is still unknown and poorly documented, as the results are not consistent among different studies [6,10]. If proven beneficial, the intervention could be widely disseminated to clinical practice and might help to reduce the burden of the disease.
The objective of this study is to conduct a systematic review and meta-analysis of randomized controlled trials (RCTs) assessing the impact of telemedicine interventions combined with usual care compared to usual care alone on the management of adult patients with DM type 1 or 2, in terms of change in hemoglobin A1c, blood pressure, LDL cholesterol (LDL-c) levels or body mass index (BMI).

Methods
This systematic review and meta-analysis was conducted in accordance with the Preferred Reporting Items for Systematic Review and Meta-Analyses (PRISMA) statement [11]. The investigators wrote a protocol and registered it with the International Prospective Register of Systematic Reviews (identification number: CRD42012002779) in August 2012 [12].

Data sources and searches
A literature search with no language restriction was performed using MEDLINE (accessed by Pubmed), Cochrane Central Register of Controlled Trials and LILACS databases to identify relevant studies published until April 2012. On Pubmed, combinations of the MESH terms diabetes, medical informatics, clinical decision support system, computer assisted decision making, information systems and telemedicine were used. In other databases, different combinations of the same terms and the keywords mobile health, telehealth, telehealthcare, health information technology were used. The search was supplemented by the reference lists of the identified papers.

Study selection
This study included RCTs that compared any strategy of telemedicine application in adult care (i.e. patients ≥18 years) with DM type 1 or 2 in which there was a direct or indirect (i.e. via another healthcare provider) personalized feedback from a healthcare practitioner to the patient about the forwarded clinical data, with a control group not using telemedicine. We restricted ourselves to trials that studied outpatients, and that evaluated at least one of the following outcomes: HbA1c, blood pressure, BMI or LDL-c.
The strategies of telemedicine application included in this review were computerized systems for information exchange, video conferencing, and exchange of information via telephone or other mobile devices, short message service, or through the internet.
Exclusion criteria were as follows: (1) gestational diabetes; (2) studies in patients under 18 years; (3) RCT which did not include feedback from a healthcare practitioner to the patient or other healthcare provider about forwarded clinical data; (4) studies with less than 6 months of follow-up; (5) duplicate publications or substudies of included trials. In this case, the publication with longer follow-up was chosen.

Data extraction and quality assessment
Titles and abstracts of retrieved articles were independently evaluated by 2 investigators (J.X.M. and M.S.M.). Abstracts that did not provide enough information for analysis of the intervention or the methodology regarding the inclusion and exclusion criteria defined in this review were retrieved for fulltext evaluation.
Subsequently, the 2 investigators independently evaluated full-text articles and determined study eligibility. Disagreements were solved by consensus, and if disagreements persisted, by a third investigator (A.L.R.). The corresponding authors were contacted as needed to obtain data not included in the published report.
The two investigators independently conducted data extraction, and disagreements were solved by the third investigator. Study quality assessment included adequate sequence generation, allocation concealment, blinding of outcomes assessors, use of intention-to-treat analysis, and description of losses and exclusions [13]. Studies without clear descriptions of an adequate sequence generation, and studies with loss to follow-up greater than 20% if analyzed by intention to treat (ITT) and greater than 5% if not ITT were excluded.

Data synthesis and analyses
The primary outcome was absolute changes in HbA1c from baseline to follow-up. HbA1c is recognized as a valuable indicator of treatment effectiveness in patients with diabetes [2], because it reflects average glycemia over several months and is strongly correlated with diabetes complications [2,14,15].
Secondary outcomes included the absolute changes of LDLc, systolic blood pressure (SBP), diastolic blood pressure (DBP) and BMI from baseline to follow-up. They were obtained from original studies or calculated (differences in arithmetic means before and after the intervention) [16]. When deviation of the mean difference was not available, the authors were contacted. In case of no response or no availability of the requested information, variance was estimated by using the reported confidence intervals, reported p-values, or an imputation using the standard deviation at follow-up in both groups to calculate the standard error of difference. One study reported data on two intervention groups and one control group [3]. In meta-analysis, data of these intervention groups were combined and compared to the control group.
A random-effects model was used for pooling the included studies, as clinical heterogeneity was expected. Statistical heterogeneity of the treatment effect among studies was assessed using Cochran Q test, and the inconsistency I 2 test in which values greater than 50% were considered indicative of high heterogeneity. Between-study heterogeneity was performed using the chi-square statistic. A p-value <0.05 was considered statistically significant.
Sensitivity and subgroup analyses were pre-specified and performed only for the primary outcome. In sensitivity analyses, the effect size was examined by omitting studies individually and also excluding simultaneously studies with extreme results and those with the biggest losses. In this case, the effect sizes were compared by using z tests.
Subgroup analyses aimed to assess whether there was a difference in results of HbA1c in studies which included exclusively patients with DM type 1 vs. type 2, interventions which included changes in patient prescription through telemedicine vs. the ones which did not, intervention made by physicians vs. by nurses, interventions that lasted for 6 months vs. the ones that lasted at least one year. Although not prespecified in the protocol, a subgroup analysis regarding the average HbA1c on baseline was performed.
A funnel plot of the trial effect size by its standard error was constructed to assess the possibility of publication bias [17]. The symmetry of the plot was evaluated both visually and formally with Egger's test. The implications for our results were assessed by the fail-safe N and the trim-and-fill method [18].
Statistical analysis was performed using Comprehensive Meta-Analysis version 2.0.

Description of the studies
A flow diagram of search and selection is shown in Figure 1. The database search resulted in 6258 (6039 Pubmed, 117 Lilacs, 102 Cochrane Trials) articles and 29 other articles were identified from the references of included studies, totalizing 6287 articles. Of these, 6203 (5984 Pubmed, 117 Lilacs, 102 Cochrane Trials) were excluded after title and/or abstract analysis. Therefore, 84 full text articles were assessed for eligibility. Of these, 69 articles were excluded. Consequently, 15 articles with 13 studies were included in the systematic review.
The main characteristics of the included studies are summarized in Tables 1 and 2. All 13 studies selected for the review were published in English. Included studies had a total of 4207 patients, 56% were men, 2252 patients were randomized to telemedicine strategies and 1955 to usual care. Ten studies took place in United States and 3 in Europe. One study [19] received funding from industry alone, nine studies [3,10,[20][21][22][23][24][25] were funded by the government alone and three studies [4,26,27] received funding from both.
Interventions took place in different care settings: 8 studies took place in primary care, 4 studies secondary care and 1 included both primary and secondary care settings. The duration of the intervention was 6 months in 6 studies, 12 months in 6 and 18 months in one study.
The intervention was particularly targeted at monitoring clinical values, or monitoring combined with education. As per design of our meta-analysis, all studies included personalized feedback from a healthcare practitioner to the patient or other healthcare provider about the forwarded clinical data. The intervention had varying degrees of complexity. Telemedicine strategies included different combinations of transmission of monitoring data, videoconferencing, educational web site, educational reminders, email exchange, evidence-based message to the general physician, teleconsultation, telephone calls, short message service and forum discussion.
Eight trials reported the frequency of data transmission from patients to nurses (six trials) or physicians (two trials). Eight trials provided information about the frequency of feedback of information to patients. The majority of the trials succeeded in data transmission at a maximum frequency of once a week.
Five studies (2889 patients) assessed the effect of telemedicine strategies on LDL-c. There was a significant reduction of 6.6 mg/dL (95% CI -8.3 to -4.9 mg/dL; p<0.001). There was no evidence of a significant heterogeneity (I 2 24%; p value for heterogeneity 0.259) ( Figure 3A).
Only 2 studies (397 patients) assessed the effect on BMI, therefore it was not possible to perform a meta-analysis. Both studies demonstrated a non-significant reduction on BMI.

Secondary analyses
In sensitivity analyses, studies were individually omitted from the meta-analysis, and the effect size remained fairly the same (Table 4). Even when studies with extreme results and those with the biggest losses [10,22,26] were simultaneously omitted, there was no significant difference in HbA1c reduction (-0.44% vs. -0.56%, p=0.70).
It was impossible to stratify patients regarding baseline HbA1c because patient-level analyses were not performed. Studies with an average baseline HbA1c equal to or higher than 8.0% [64.0 mmol/mol] were compared to the ones with an average lower than 8.0% and demonstrated a higher impact of telemedicine in the first group: -0.64% (7.0mmol/mol) (95% CI Interventions which included changes in patient prescription through telemedicine (4 studies, 334 patients) were associated with a more pronounced HbA1c reduction than the ones which did not (7 studies, 3435 patients): -0.75% (-8.2 mmol/mol) (95% CI -1.05 to -0.43% [-11.5 to -4.7 mmo/mol]) vs. There was a tendency of a higher HbA1c reduction in interventions which lasted 6 months (6 studies, 741 patients) when compared to the ones that lasted at least one year (7 studies, 3466 patients; in only one study the follow-up was longer than one year): -0.57% (-6.2 mmol/mol) (95% CI -0.85 to -0.30% [-9.

Publication Bias Assessment
The subject impression of the funnel plot test suggested an asymmetry ( Figure 2). Toward the bottom of the graph there are more studies on the left (indication a higher impact), which is consistent with the possibility that some studies are missing   ). However, the failsafe N for our pooled analysis is 273000, which is reassuring since it is very unlikely that there are over 1000 unpublished or undiscovered studies for every 1 study we found. Additionally, the trim-and-fill method revealed that publication bias did not interfere with the interpretation of results The method suggested that one study is missing, and the imputed effect size was -0.41% (95% CI -0.58 to -0.23%), which is similar to the original effect, suggesting that the apparent publication bias is insufficient to affect our results or interpretations in a meaningful way (Figure 4).

Discussion
This systematic review (13 studies, 4207 patients) indicates that in diabetes patients, telemedicine strategies concomitant to the usual care are associated with a mean HbA1c decline of −0.44% (-4.8 mmol/mol) when compared to the usual care alone. HbA1c is recognized as a valuable indicator of treatment effectiveness in patients with diabetes [2], because it reflects average glycemia over several months and is strongly correlated with diabetes complications [2,14,15]. Additionally, our analyses demonstrate that there was no clinical relevant impact on LDL-c (an effect size of only 6.6 mg/dL) and no impact on blood pressure. It was impossible to perform a metaanalysis to assess the impact on BMI, but there is a tendency of its reduction with the use of telemedicine.
The lack of a clinical significant impact on LDL-c and the absence of impact on blood pressure should not be interpreted as a definite finding. Regarding hypertension, a recent RCT which included a large sample of hypertensive patients (7.3% of them had diabetes) who were willing to monitor their blood pressure and self-titrate medication has shown that selfmanagement of hypertension in combination with telemonitoring of blood pressure measurements resulted in significant reduction in blood pressure in 6 and 12-month follow-up when compared to usual care [27]. No study in the present meta-analysis included an intervention similar to that one. Therefore, the telemedicine strategy design might have not been appropriate to generate a significant impact on hypertension and LDL-c, and the intervention might be more effective in a motivated sample of patients, as was the case. This should be in mind when designing new studies to address these outcomes.
Our findings also demonstrated that the impact of telemedicine strategies in reducing HbA1c was more pronounced in diabetes type 1. Type 1 diabetes patients are unique in that they have absolute insulin deficiency and consequently the have more pronounced glycemic variability than type 2 diabetes patients [2,28]. Their management always involves insulin administration, whereas type 2 diabetes patients may be managed using lifestyle changes and oral hypoglycemic agents. These differences could influence the impact of individual's self-management strategies and affect outcome measures [29]. Additionally, type 2 diabetes is more common in older adults. Compared with younger adults, older adults report greater anxiety about using computers, lower use of technology and less confidence in their technological abilities [6] Therefore, some patients may have found difficulties in dealing with the telemedicine applications, for example, to upload blood glucose results to the website or to access online educational materials.
Interventions which included changes in patient prescription through telemedicine were associated with a greater benefit on HbA1c reduction than the ones which did not, and there was no   difference regarding intervention made by physicians or nurses. These findings are important to address recommendations on telemedicine design strategies in clinical practice. The lack of difference between physicians and nurses impacts significantly in decreasing the costs of the intervention if applied to the healthcare system. Another important finding was the tendency of a higher HbA1c reduction in interventions which lasted 6 months when compared to the ones that lasted at least one year. Although it did not reach statistical significance, it suggests a trend of decreasing intervention impact over time. This finding suggests that in clinical practice contact through telemedicine and positive motivation should be intensified over time, in order not to decrease the impact on glycemic control. Additionally, a more participatory design approach, involving target users in   developing personalized interventions, might enhance persistence in usage and adherence to treatment [30].
This study was a rigorous systematic literature review that focused exclusively on RCTs in which telemedicine strategies included a personalized feedback from a healthcare practitioner to the patient or to other healthcare provider. Although only 5 studies had blinded assessment of outcomes, the outcomes were objective measures, with a low risk of bias in their assessment. Previous systematic reviews with less methodological rigor have not performed meta-analysis or failed to detect a significant impact of different telemedicine strategies on glycemic control [6,26,[30][31][32]. Methodological limitations may have biased the results, including inaccurate or incomplete indexing in the database, or a weak search strategy, leading to incomplete accounting of existing studies; limited or no assessment of study quality; and inclusion of weak studies. One systematic review demonstrated a significant decrease in HbA1c, but with a lower impact (effect size -0.22% [-2.4 mmol/ mol]; 95% CI -0.35 to -0.08% [-3.8 to -0.9 mmol/mol]) [33]. The inclusion of RCTs independent of study quality and the possibility of selection bias on observational studies may have influenced the results.
The evidence presented here could support the implementation of telemedicine in the care of diabetes patients. It has already been shown that telemedicine help improving patient care and access to specialized care, in particular in rural areas with limited accesses to health care facilities [34], and it is expected to facilitate a productive interaction between the patient and the health care provider in order to achieve improved treatment results and lower treatment costs [35]. This review did not address the cost-effectiveness of the intervention. Another systematic review, specific on teleconsultation, reported controversy about that, but costeffectiveness was not measured uniformly among the various studies, and some of them were observational studies without a control group [30]. This question is of utmost importance for decision makers and therefore should be addressed in further clinical trials.
In order to make these results applicable to clinical practice, it is also important to take into account that telemedicine may not be appropriate for all patients. There are some factors that are described as important elements for the success, such as awareness of the significance of the disease and its consequences, willing to participate actively in the treatment and in preventing complications and positive motivation [4]. Patient computer skills and adherence to the technology are also important depending on the complexity of the telemedicine strategy [36].
The main limitation of this meta-analysis is the heterogeneity. To address this, we have performed analyses to identify clinical and methodological differences between studies. Some factors could not be addressed, as the type of telemedicine intervention, because it was too diverse among studies. As the analyses were not based on individual data, it was not possible to stratify according to the basal HbA1c, and it is expected a higher impact in individuals with poor metabolic control. The inclusion of patients with good metabolic control may have underestimated the impact of the intervention, so it could be expected an even better result if we could stratify by the baseline HbA1c levels. The subgroup analysis comparing studies with an average HA1c lower or higher than 8.0% (64.0 mmol/mol) is limited, because it is not possible to effectively separate individuals with better and worst metabolic control. However, as it demonstrated a significantly higher impact on HbA1c in the subgroup of studies with baseline HbA1c equal to or higher than 8.0%, it provides an estimative of a higher impact of the intervention in patients with worst metabolic control.
Although there was evidence of publication bias, trim-and-fill method revealed that publication bias did not interfere with the interpretation of results, as the imputed effect size was similar to the original effect, suggesting that the apparent publication bias is insufficient to affect our results or interpretations in a meaningful way.
It is important to note that the endpoint HbA1c has some limitations. As a mean, it may show normal results in individuals who alternate hypo and hyperglycemias [37]. Besides this, it is necessary that laboratories use HbA1C assay methods certified by the National Glycohemoglobin Standardization Program (NGSP) as traceable to the DCCT (Diabetes Control and Complications Trial) reference [38,39]. Finally, some clinical situations can lead to a HbA1C falsely high or falsely low, such as pregnancy, hemoglobinopathies, disorders of erythrocyte turnover, renal failure [37,40].

Conclusions
In conclusion, this systematic review and meta-analysis has shown that telemedicine strategies combined with the usual care are associated with improved glycemic control in diabetic patients. In this review, there was no clinical relevant impact on LDL-c and blood pressure, and there was a tendency of BMI reduction in diabetes patients who used telemedicine, but the impact on these outcomes should be further explored in future trials.
A logical next step for future trials would thus be to assess whether the improved glycemic control observed with telemedicine interventions will translate into improved clinical outcomes. The optimal design of telemedicine systems is uncertain, so future research should also focus on what type of telemedicine intervention has the largest impact on diabetes management, and how to adapt systems to the individual patient needs and resources of specific health care systems. The evaluation of the cost-effectiveness of the intervention should also be addressed, which is relevant to decision makers and impacts on the applicability of the evidence in clinical practice.