Home Telemonitoring of Patients With Type 2 Diabetes: A Meta-Analysis and Systematic Review

We performed a systematic review and meta-analysis of randomized controlled trials (RCTs) published from 2015 to 2019 that assessed the impact of HTM (using digital tablet, smartphone, or a Web-based platform) compared with usual care (outpatient) on the management of adult patients with type 2 diabetes. Targeted outcomes included A1C, blood glucose, blood pressure, and BMI. To better compare different levels of technology and study characteristics such as study duration, we adopted a systematic comparative approach to examine the effectiveness of technologies using subgroup analyses. This review was performed according to the Cochrane Collaboration’s methodological guidelines and registered with the PROSPERO (the International Prospective Register of Systematic Reviews) (15). The review was reported according to accepted guidelines (Figure 1). All statistical analyses were performed using the dmetar, metafor, meta packages for R software, v. 3.6 (16–19).

To be included in this review, articles had to be written in English, published in a peer-reviewed journal, and targeting a population of adults (≥18 years of age) with a primary diagnosis of type 2 diabetes. Studies included RCTs with the aforementioned definition of HTM as an intervention that used comparators of standard outpatient care (usual care) or other telehealth interventions and reported A1C outcomes. Systematic reviews were excluded. In addition, if the same authors produced several publications of the same data source, the latest one was included and other versions were excluded.

To ensure that the results would be applicable to current practice, the search strategy was limited to RCTs published between 1 January 2015 and 31 January 2019. Electronic bibliographic databases included the Cochrane Library, PubMed, and EMBASE.

All data were extracted independently and in duplicate by two researchers using a standardized protocol. The abstracts of all articles identified were screened to determine whether they met the inclusion criteria. In the preliminary screening stage, all titles and abstracts were reviewed by two authors according to the inclusion criteria based on a priori selection criteria for eligibility. References that did not clearly meet all criteria were deferred for full text review. Any disagreements were resolved through discussion until consensus was reached.

One of five teams of reviewers, each consisting of two abstractors (one health services researcher and one clinician researcher), independently assessed full-text articles for study inclusion. Search terms included “diabetes mellitus,” “mobile applications,” “type 2 diabetes,” “telecommunications,” and “teleconference,” among others. A total of 2,835 studies in the databases were identified with our search terms, of which 319 were duplicates. Overall, 58 studies met the initial eligibility criteria. Studies such as observational studies, pilot studies, and protocol descriptions were not included. Figure 1 shows the systematic review study flowchart that demonstrates the inclusion and exclusion process.

In total, 20 RCTs (20–39) met the criteria of the systematic review of HTM interventions and study design. Discrepancies were discussed by the two reviewers until consensus was reached. Studies were reviewed again to confirm that they met inclusion criteria; those that did not were excluded. Eligible HTM interventions involved the transmission of self-monitored physiological vital measures (e.g., blood glucose and blood pressure) either via fully automatic transmission, fully manual uploading, or transmission with moderate patient interaction (such as patients pushing a button).

Relevant data from the included studies were extracted using the researchers’ designed forms. A standardized spreadsheet was used to extract all of the relevant data on study characteristics, intervention details, and outcomes. Specifically, the data included the first author, year of publication, country of origin, HTM intervention, study duration, control group information, outcomes, and results. We applied a Hartung, Knapp, Sidik and Jonkman random-effects meta-analysis, which is known to perform better than other forms of analysis when trials of similar sizes are combined (40,41). Effect size was represented by the Hedges’ g statistic (42), which directly corresponds to the standardized mean difference in A1C between the control and intervention arms of all included studies.

The primary outcome was change in A1C. To assess this measure with consistent units, a researcher translated all A1C values into percentages. Changes in blood pressure (systolic and diastolic) and BMI, which are also important clinical outcomes representing goals of treatment, were secondary outcomes. Blood pressure measurements were all converted to mmHg for a consistent statistical comparison. To further compare different patients’ baseline A1C (inclusion can have an impact on glycemia control), patients were considered to have controlled glycemia if they had an A1C <7%, whereas those having an A1C ≥7% were considered to have uncontrolled glycemia, in accordance with American Diabetes Association recommendations (43).

The Cochran Q statistic was used to estimate statistical heterogeneity among the included studies, and P <0.05 was considered statistically significant. In the presence of significant heterogeneity, the I² statistic was used to determine the level of heterogeneity based on the Higgins and Thompson’s standard (44).

Data from the selected studies were independently extracted by two reviewers using a data extraction form. A risk-of-bias assessment tool, summarized in the Cochrane Handbook for Systematic Reviews of Interventions, v. 6.0 (45), was applied to assess the quality of each study. The studies were evaluated separately based on seven domains: random sequence generation, allocation concealment, blinding of participants and personnel, blinding of outcome assessment, incomplete outcome data, selective reporting, and other bias. Two reviewers (X.Z. and R.P.) subjectively reviewed all selected studies and assigned a value of “low risk,” “unclear risk,” or “high risk” to these domains.

The datasets generated and/or analyzed during this study are available from the corresponding author upon reasonable request.

Table 1 summarizes the 20 selected studies, including their country, duration, number of participants, baseline characteristics of participants, mean A1C change, and type of data transmission used. Seven were conducted in the United States, three in Europe, nine in Asia, and one in Australia. Of a total of 4,739 participants (median age 58.3 years), 2,358 were randomized to the telehealth group, and the remaining 2,381 received usual care. Most studies had been implemented using wireless technologies such as Bluetooth connectivity for monitoring. We focused on the methods of data transmission by separating the studies into three subgroups: 1) “transmission with moderate patient interaction (such as patients pushing a button” 2) “automatic transmission (patients did not actively upload data),” and 3) “patients manually transmit data.” Nine studies used automated transmission of patients’ vital signs (i.e., blood pressure), three used moderate patient interaction, and five required patients to manually upload to either a digital app or Web platform to transmit their vital signs. Three studies did not report their method of transmitting data. The median length of the intervention was 6 months, with one study having a 5-year follow-up period. All 20 studies provided participants with ongoing support after HTM data were collected within their follow-up period, either as consultation from a physician or as two-way communication via the app or platform. Feedback mechanisms varied based on the design of the technology used, and two of the relatively newer studies (those by Kim et al. [26] in 2019 and Lim et al. [27] in 2016) used algorithm-based reports and messages generating automatically for patients to view on their end once data were received.

All included studies used change in A1C as an outcome measure to evaluate the effectiveness of HTM for type 2 diabetes management. As previously mentioned, we chose the random-effects model because it pays more attention to small studies when pooling overall effects in a meta-analysis. With the exception of one large-sample study by Shea et al. (31), most of the selected studies had relatively small samples, with a median size of 56 participants in both groups.

The overall effect using the random effect model was significant (standardized mean difference Hedges’ g −0.42, 95% CI −0.59 to −0.26), with an overall decrease in A1C as shown in Figure 2. However, there was heterogeneity detected across studies (I² = 61.3%, P <0.01), indicating a moderate to high degree of heterogeneity.

A funnel plot with Egger’s test (46) was conducted and confirmed that there was no publication bias (bias = −0.73, P = 0.47). Because of the degree of heterogeneity, to further detect the true effect of HTM, we conducted a P_curve analysis (47) as an alternative way to assess publication bias and estimate the true effect reflected in our collected data. This approach rules out some cases in which researchers might have selectively removed outliers and chosen outcomes to reach a significant result, a practice sometimes referred to as “P hacking.” Our results indicated that there was a true effect in terms of the differences we found, with an overall observed power of 93% (95% CI 81–98%).

Blood pressure was reported in nine studies (Figures 3 and 4). We pooled all systolic and diastolic blood pressure results to look at the effectiveness of HTM intervention. One study (27) did not report systolic blood pressure. Both standardized mean differences appeared to indicate that the interventions reduced both systolic (Hedges’ g = −0.07, 95% CI −0.14 to −0.002], P = 0.044) and diastolic blood pressure (Hedges’ g = −0.10, 95% CI −0.16 to −0.03, P <0.005). Statistical heterogeneity was not detected for either diastolic (Q = 7.38, P = 0.49) or systolic (Q = 5.66, P = 0.77) blood pressure. Although statistically significant, clinically, blood pressure change within 1 mmHg is considered insignificant, especially within the context of HTM.

BMI was also examined using the random-effects model. Standardized mean differences in BMI between groups appeared to be insignificant (Hedges’ g = 0.09, P = 0.54).

Given the various telehealth technologies, as well as the different data transmission methods, used in the interventions, we performed subgroup meta-analyses to assess whether their impact on glycemic control differed.

Overall differences between transmission methods were found (Hedges’ g = −0.40, 95% CI −0.50 to −0.30), with moderate heterogeneity between studies (Q = 50.69, P = 0.0001, I² = 62%). Between-group mean differences showed that patients with uncomplicated interactions to transfer data (Hedges’ g = −0.65, 95% CI − 1.06 to −0.24) performed better in A1C than those with automatic data transmission (Hedges’ g = −0.44, 95% CI − 0.70 to −0.18) or manual transmission (g = −0.20, 95% CI − 0.43 to −0.03).

The median length of the selected studies was 6 months (range 12 weeks to 5 years). The test for subgroup differences using the random-effects model was not significant (Hedges’ g = −0.35, P = 0.76) for different study duration periods.

Of the 20 studies included in our analysis, 4 did not indicate an A1C inclusion criterion, 2 recruited patients with both controlled and uncontrolled glycemia, and 13 recruited only patients with uncontrolled glycemia. We further grouped the studies of uncontrolled glycemia into A1C levels at recruitment; three studies recruited patients with an A1C >7.5%, and two studies recruited those with an A1C >8% and >9%, respectively (Table 1).

Evidence using the random-effects model demonstrated that differences with regard to baseline A1C were significant (Hedges’ g = −0.33, 95% CI − 0.45 to −0.21, P <0.0001). The baseline group with A1C >9% performed best among the groups (Hedge’s g = −0.62, 95% CI −1.23 to −0.013), followed by the group with baseline A1C >8% (Hedges’ g = −0.53, 95% CI −1.04 to −0.01).

This systematic review of 20 studies demonstrated some promising results regarding HTM in terms of glycemic and blood pressure control for patients with type 2 diabetes.

In light of the proliferation of telehealth, the relevance of these results is especially timely. Before the COVID-19 global health pandemic, widespread implementation of HTM was traditionally hampered by regulations, reimbursement limitations, and other policy issues (2). Today, in the COVID-19 era, HTM has become a mainstream method of chronic care management. It is therefore of utmost importance that we fully understand the impact of HTM on clinical outcomes.

Our findings are in accordance with several recent systematic reviews, such as an analysis by Wu et al. (14) of 19 RCTs. In addition to looking at A1C and blood pressure, our study specifically explored data transmission methods (i.e., whether data were transmitted fully automatically, fully manually, or with moderate interaction by patients). Previous studies (e.g., Lee et al. [10]), specifically focused on technology type (e.g., telephone vs. smartphone vs. Web platform). Our novel approach explored the human side of home monitoring, emphasizing the patient’s role. Our results suggest that patients with a definite but uncomplicated role in data uploading procedures had greater A1C reductions over time than those whose data were transmitted completely manually or fully automatically. Further, patients seeing and attending to (uploading) their vital sign data without having to manually input the data achieve better outcomes. They also do better than those whose vital signs are automatically transmitted, requiring no patient action.

This finding makes sense from a behavioral framework perspective. According to the Health Belief Model (48), an individual course of action depends on subject’s perception of benefits of and barriers to this action; the decision-making process often involves the subject’s use of a cue or trigger to action. Such prompts can be the result of either external factors such as health care providers or internal factors (such as physiological cues). Our findings also align with Situated Learning Theory, which proposes the concept of patient learning through active participation rather than through regular educational lectures/notes (49). Asking patients to actively engage while not requiring them to do too much manual work appears to build self-efficacy in managing type 2 diabetes.

We also explored the length of HTM interventions. There were no significant differences between interventions with a duration <6 months and those of longer duration. These findings support previous results from Shea et al. (31), who found the HTM treatment effect to increase the most from baseline to 12 months and then remain steady in the second, third, and fourth years. Perhaps for patients requiring chronic care management assistance, what is learned after 1 year is maintained thereafter. This finding suggests that interventions within the first 6 months to 1 year are consequential, as health learning peaks at this point. This strategy should be explored further in future studies of type 2 diabetes.

There are several limitations to our review. First, we report moderate to high heterogeneity across studies. Although our P_curve analysis was added as additional evidence, it is still possible that the heterogeneity of studies may be an issue, as results were not fully conclusive with regard to the existence of substantial heterogeneity. Thus, although our findings are strong, we cannot definitively conclude that HTM is effective in general or that certain methods of data transmission are better than others. Second, although our findings are similar to those of Lee et al. (10) that HTM performed better among patients with a mean baseline A1C >8% regardless of insulin, we found that those with higher baseline A1C levels (>8%) experienced further reductions using HTM. Although this finding is somewhat expected (i.e., regression toward the mean), it was based on only two studies, and future research should include RCTs recruiting patients with type 2 diabetes who have a wide range of baseline A1C levels (i.e., both controlled and uncontrolled glycemia) and with larger sample sizes.

Compared with usual care, the addition of HTM appears to significantly improve A1C in patients with type 2 diabetes and especially in those with severely uncontrolled glycemia at baseline. Although there was substantial heterogeneity, the pooled analyses showed that HTM lowered A1C by 0.42% over 6 months and by 0.28% beyond 6 months. Meanwhile, pairwise subgroup comparison showed that longer intervention duration does not lead to significantly greater glycemic control.

No potential conflicts of interest relevant to this article were reported.

X.Z. and R.P. wrote the manuscript. M.W., K.F., V.P., L.S., G.W.-K., A. Marziliano, C.N., A. Makaryus, R.Z., L.T., and A. Myers read and coded all of the included articles. M.W., K.F., and V.P. helped to collect the data. Data were cleaned and analyzed by X.Z. L.S., G.W.-K., A. Marziliano, C.N., A. Makaryus, R.Z., and A. Myers reviewed and edited the manuscript. T.S. searched key terms in peer-reviewed journals in the library. R.P. oversaw this research and provided insights.