Comparative effectiveness of telemedicine strategies on type 2 diabetes management: A systematic review and network meta-analysis

The effects of telemedicine strategies on the management of diabetes is not clear. This study aimed to investigate the impact of different telemedicine strategies on glycaemic control management of type 2 diabetes patients. A search was performed in 6 databases from inception until September 2016 for randomized controlled studies that examined the use of telemedicine in adults with type 2 diabetes. Studies were independently extracted and classified according to the following telemedicine strategies: teleeducation, telemonitoring, telecase-management, telementoring and teleconsultation. Traditional and network meta-analysis were performed to estimate the relative treatment effects. A total of 107 studies involving 20,501 participants were included. Over a median of 6 months follow-up, telemedicine reduced haemoglobin A1c (HbA1c) by a mean of 0.43% (95% CI: −0.64% to −0.21%). Network meta-analysis showed that all telemedicine strategies were effective in reducing HbA1c significantly compared to usual care except for telecase-management and telementoring, with mean difference ranging from 0.37% and 0.71%. Ranking indicated that teleconsultation was the most effective telemedicine strategy, followed by telecase-management plus telemonitoring, and finally teleeducation plus telecase-management. The review indicates that most telemedicine strategies can be useful, either as an adjunct or to replace usual care, leading to clinically meaningful reduction in HbA1c.


Supplementary Online Content
Literature search strategy Table S2 Description of interventions and outcomes Table S3 Selected baseline demographics and details of intervention strategies employed Table S4 Key intervention component of studies included in the current review Table S5 Details of educational content delivered to participants in studies examining teleeducation Table S6 Details of self-monitoring frequency and recommended schedule Table S7 Details of the action(s) taken based upon results from glucose monitoring Table S8 Results of subgroup and sensitivity analysis for primary outcome Table S9 Results of the network meta-analyses Table S10 Inconsistency estimates Table S11 Summary of effect sizes comparing meta-analysis from other studies and current study Figure S1 Flow chart depicts the process of study selection Figure S2 Risk of bias assessment appraisal of randomized controlled studies on telemedicine using the Cochrane risk of bias tool Figure S3 Evidence network for outcomes assessed Figure S4 Surface under curve ranking Figure S5 Comparison adjusted funnel plot eReferences List of studies included in the current review

Holmen, 2014
Education only Y Group medical visits to physician and health educators facilitation

Table S9 -Results of the network meta-analyses
All estimates are presented as weighted mean difference (WMD) with the corresponding 95% confidence intervals. Comparisons between treatments should be read from column to row and the estimate is in the cell in common between the column-defining treatment and the row-defining treatment. Weighted mean difference that are negative favor the column-defining treatment. To obtain odds ratios for comparisons in the opposite direction, reciprocals should be taken. All significant results are presented in bold and underlined.

Table S9a: Pairwise meta-analyses and network meta-analyses of the various telemedicine strategies on glycosylated haemoglobin. Results of effectiveness of intervention are presented as mean difference ( 95% confidence interval) for each pairwise comparison of intervention, based on direct evidence alone (lower-left triangle) or direct and indirect evidence (upper right triangle)
Usual care              In all network meta-analysis, transitivity is the crucial assumption since it permits the use of indirect evidences, i.e. permits the comparison of treatment that never been directly contrasted. To estimate the inconsistency or transitivity, we used the difference between direct and indirect estimates (called inconsistency factor IF) and the corresponding 95% CI for each IF in each closed triangular loop. An inconsistent loop is those that present inconsistency factors with 95% confidence intervals incompatible with zero. For example, for the triangular loop of evidence including usual care-teleeducation-telecase-management for total cholesterol, the weighted mean difference for estimated effect differs between direct and indirect evidence by 0.03, however the 95% confidence interval includes zero indicating the possibility of no difference.
To further supplement our analyses on transitivity, we used the design-by-treatment interaction model which provides a single inference, using the χ 2 test, about the plausibility of assuming consistency throughout the entire network

Figure S2: Risk of bias assessment appraisal of randomized controlled studies on telemedicine using the Cochrane risk of bias tool
To assess the risk of bias for studies included in the current study, we used the Cochrane Risk of Bias tool. This tool assessed the specific bias of seven domains including methods for generating the random sequence, allocation concealment, blinding of participants and investigators, blinding of outcome assessment, incompleteness of outcome data and selective outcome reporting. The adjudication of the risk of bias is achieved by answering prespecified questions about the methods reported by each study in relation to the risk domain, and results are represented in a risk of bias table.

Figure S3 Evidence network for outcomes assessed
The width of lines for each connection in the evidence network is proportional to the number of randomized controlled studies that compared each pair of treatment. The sizes of nodes are proportional to the number of patients. The numbers represent the number of randomized controlled studies which contribute to the direct comparison.

Figure S3a: Evidence network of treatment comparison for primary outcome
The tau value for network meta-analysis for is 0.52 (moderate heterogeneity)