Telecare for Patients With Type 1 Diabetes and Inadequate Glycemic Control: A randomized controlled trial and meta-analysis

The Mayo Foundation Institutional Review Board approved the study protocol.

Patients were eligible to participate in this study if they had 1) documented type 1 diabetes (with C-peptide ≤0.02 nmol/l) of >1 years’ duration and 2) inadequate glycemic control (HbA_1c ≥7.8%). Patients had completed a 3-day educational program to learn intensive insulin management (15), were using multiple daily insulin injections or insulin pumps, and were receiving usual diabetes care in a diabetes clinic. Patients were excluded if they were pregnant or planning pregnancy.

After confirming eligibility and obtaining written informed consent, the study coordinator obtained the patient allocation from the randomization center. A computer generated the allocation sequence using adaptive randomization (16) to minimize differences between the groups in baseline HbA_1c (≥9%), glucose goals (80–100 mg/dl [4.4–6.7 mmol/l] or 100–140 mg/dl [5.5–7.8 mmol/l]), and use of a portable insulin pump.

We trained all patients enrolled in this study to connect an Acculink modem to an Accu-Chek Complete glucometer (Roche Diagnostics, Indianapolis, IN) and the phone line and to transmit glucometer data to the research computer. Glucose analysis software on this computer assisted the study nurse with interpretation.

We asked all patients to monitor their blood glucose four times per day, 7 days per week, and to transmit the recorded glucometer data at least every 2 weeks. Patients allocated to the telecare arm received feedback within 24 h of transmission from a study nurse supervised by a clinical endocrinologist. Patients allocated to the control arm did not receive unsolicited feedback, but contacted the study nurse as frequently as necessary. All patients received face-to-face diabetes care at clinic visits every 3 months.

The primary outcome measure was HbA_1c (measured in the same reference laboratory using a high-performance liquid chromatography technique [BioRad, Hercules, CA] by personnel blinded to allocation) 6 months after randomization. Secondary outcome measures included proportion achieving HbA_1c ≤7% at 6 months, number and severity of hypoglycemic episodes (as defined in the DCCT [1]), the time the nurse spent in reviewing data and providing patients with feedback, and the time the physicians spent in supervising the study nurse. To assess the impact of the intervention on diabetes self-management, patients completed the valid and reliable Summary of Diabetes Self-Care Activities (SDSCA) questionnaire (17) at baseline and at 6 months after randomization.

We estimated the sample size of our study (15 participants per arm) using the distribution of HbA_1c at baseline (SD of 1.5%) in the DCCT (1), α of 5%, β of 20%, and an estimated difference in the HbA_1c change at the end of 6 months of 1.5% (1 SD). We analyzed all patients in the arms to which they were randomized. To take into account imbalances in HbA_1c at baseline, we assessed the effect of treatment (fixed factor) on 6-month HbA_1c (dependent variable) after accounting for baseline HbA_1c (as covariate) using a general linear model; we also tested whether there was interaction between baseline HbA_1c and treatment. We estimated the risk difference for count data (e.g., number of patients achieving HbA_1c ≤7%) and its 95% CI using the Wilson method (18). We estimated the median difference in the change in SDSCA scores from baseline to 6 months between the groups and its 95% CI using the bootstrap technique with 10,000 iterations (19) using Resampling Procedures version 1.3 (copyright 2002, D.C. Howell).

We included randomized controlled trials of telecare (transmission of glucometer data and feedback by health professional) in patients with type 1 diabetes compared to usual care, to other forms of data transmission, or to transmission without feedback. Studies that included other forms of telemedicine without glucometer transmission (i.e., video link and telephone consultation) or mixed patient populations (i.e., type 1 and type 2 diabetes) were not eligible. We conducted electronic searches in Medline, Embase, Cinahl, and HealthStar (from 1982 to June 2003) as well as on the Internet (Google). Working independently and in duplicate, two investigators reviewed all abstracts, selected studies for inclusion, and extracted data about study design and results (HbA_1c). Hypoglycemia episodes were incompletely reported, and we do not include those results in this report. We contacted two authors when data needed were missing; one responded with the requested information. Details of excluded studies and the review flowchart are available from the authors.

We pooled the standardized mean difference in change of HbA_1c from baseline from randomized trials included in the systematic review using random-effects meta-analysis (DerSimonian and Laird method) on Revman 4.2 (Cochrane Collaboration, 2003). We pooled the SD of the change (baseline − end of study) in HbA_1c in each group; if these data were unavailable, we pooled the SD of HbA_1c at baseline in each group (20). We quantified heterogeneity, the proportion of between-study variability due to study differences (i.e., not due to random error), using the I² method (21). A priori, we decided to examine differences in patients studied (enrollment of patients with well-controlled versus poorly controlled diabetes), interventions (frequency of transmission and feedback), length of follow-up (<6 months and ≥6 months), and methods (allocation concealment and loss to follow-up) to try to explain heterogeneity of results.

Figure 1 describes the flow of the 31 enrolled patients through the study. HbA_1c at 6 months was not available for three patients, two in the telecare intervention arm and one in the control arm. Table 1 describes patient characteristics at baseline; there was an imbalance in HbA_1c. There was a significant between-group difference in the 6-month HbA_1c (telecare 7.8 ± 1.3% vs. control 8.2 ± 1.2%) after taking into account the difference in HbA_1c at baseline (P = 0.03). There was no significant interaction between baseline HbA_1c and treatment (P = 0.1).

Four of 14 (29%) patients in the telecare intervention arm and 1 of 15 (7%) in the control arm had 6-month HbA_1c ≤7% (risk difference 21.9%, 95% CI −4.7 to 50.5). There were no episodes of ketoacidosis and three episodes of severe hypoglycemia in each group. After taking into account the difference in insulin doses at baseline, there were no differences in insulin doses (total, basal, and bolus doses) between the groups at 6 months (P = 0.8, 0.4, and 0.8, respectively). However, telecare patients had more documented dose changes than the control group during the duration of the study (14 vs. 6, P = 0.04); most of these changes occurred during the first 3 months of the study (9 vs. 3, P = 0.007).

There was no difference between telecare and control groups in the mean proportion of transmitted values above goal per patient (64%, 95% CI 57–71 vs. 66%, 59–73) and in the mean proportion of transmitted values <60 mg/dl (3.3 mmol/l) per patient (9%, 5–13 vs. 9%, 6–11).

During the first 2 months of the trial, patients in the telecare group transmitted a median of six times (interquartile range [IQR] 5–9), whereas the control group transmitted five times (IQR 4–9). During the last 2 months of the trial, telecare patients transmitted five times (IQR 4–8) and control patients 4 times (IQR 3–7). During the first month, patients in both groups monitored 4.1 times per day (95% CI 3.7–4.5). During the last month of the trial, telecare patients self-monitored a mean of 3.6 times per day (95% CI 2.8–4.4) and control patients 3.5 times per day (95% CI 2.9–4).

At 6 months, there was a median 10.7% increase from baseline in the blood glucose testing subscale of the SDSCA in the telecare intervention arm and a median 0% change in the control group (median difference in percentage change between telecare and control arms: 10.7%, 95% CI 0–32.1). Patients in both groups reported self-monitoring their blood glucose a median of 6 days per week (IQR 3–7) at baseline and 7 days per week at 6 months (IQR 5–7). Patients in both groups reported self-monitoring their blood glucose four times per day a median of 4 days per week (IQR 3–7) at baseline and 5 days per week at 6 months (IQR 3–7). There were no differences in the change in scores (6 month − baseline) of the SDSCA between the two groups for the general diet, specific diet, exercise, foot care, smoking status subscales, and adherence to insulin doses.

The clinical endocrinologists spent a median of 9 min per patient (IQR 3–17) reviewing data transmission with the study nurse and 0 min (IQR 0–6) reviewing data from control patients. The nurse spent as a median of 76 min reviewing data from each telecare patient during the 6 months of the study (IQR 43–107) and 12 min (IQR 9–22) reviewing data from control patients. Finally, the nurse spent a median of 68 min (IQR 36–119) providing feedback to telecare patients and 18 min (IQR 13–31) providing feedback to control patients. In other words, during the 6 months of the trial, the average telecare patient spent a total of 1 h on the phone discussing diabetes care with a nurse who had reviewed data for 2.4 h (10 min of which with a clinical endocrinologist); the average usual care patient spent only 17 min on the phone with a nurse who had reviewed data for 30 min. We found a significant correlation (Pearson 0.56, P < 0.001) between nurse feedback time and the number of documented insulin dose changes.

Two reviewers working independently considered six trials published in seven manuscripts (22–28) eligible for review, with perfect interobserver agreement (Table 2). One additional eligible study was presented as a poster at the 63rd Scientific Sessions of the American Diabetes Association in June 2003 (29). Meta-analyses of these eight studies (including the trial reported in this work) showed that telecare was not significantly different from usual care (pooled HbA_1c change from baseline: 0.2%, 95% CI −0.2 to 0.6%) (Fig. 2). We could not explain heterogeneity with any of our hypotheses stated a priori. One-third of the variability in the pooled estimate came from between-study differences (I² = 34%) that were completely accounted for by excluding the only randomized controlled trial in children with type 1 diabetes (I² after exclusion of this study = 0%). The result from this latter study was significantly different from the pooled HbA_1c change from baseline from the studies of adults with type 1 diabetes (P = 0.02), which was 0.4%, 95% CI 0–0.8. The result was similar after excluding the randomized controlled trial of telecare in pregnant patients with type 1 diabetes (0.5%, 95% CI 0.05–0.9).

In this randomized trial of telecare versus modem transmission in patients with type 1 diabetes on an intensive insulin regimen but failing to achieve glycemic control (as judged by their HbA_1c levels), we found telecare to have a small impact on glycemic control. We saw a trend in enhanced adherence to self-monitoring (greater in the telecare group) and better glycemic control in both groups. The main difference in care delivery between the two groups was in the health professional’s time. Greater nurse feedback time was associated with more insulin dose changes, and significantly more insulin dose changes occurred in the telecare group.

Our study has several strengths. This is the first study to offer modem transmission to the control group to isolate the effect of clinician feedback in the telecare context. It is also a randomized trial conducted with adequate allocation concealment and in accordance with the intention-to-treat principal. We focused our attention on the difficult group of patients with type 1 diabetes on intensive insulin therapy who were failing to achieve recommended glycemic goals despite intense education and usual specialist diabetes care. Limitations include a follow-up period limited to 6 months, loss of 3 of 31 patients to follow-up, the imbalance in baseline HbA_1c, and the borderline statistical significance of our results.

Our study is unique in that we asked patients in the control group to regularly transmit their glucose levels, thus controlling for modem transmission per se. Our study shares the limitations of previous studies, which include enrolling a small sample, following them for a brief period (maximum follow-up was 12 months in two studies [26,29], 6 months in four others, and 3 months in another one [24]), and losing about 10% of patients to follow-up.

The strongest inference about the true magnitude of telecare effect on glycemic control comes from our meta-analysis. Taken together, the literature on telecare among adult patients with type 1 diabetes effectively rules out larger effects of ≥1% in HbA_1c. The results are consistent with a small significant effect on diabetes control, but this inference is dependent on a post hoc subgroup analysis that one could question.

Clinicians and policymakers have limited information (summarized here) about the efficacy of telecare in patients with type 1 diabetes. Given the wide CIs that include negligible treatment effects in spite of pooling all of the available evidence, stronger inferences about effectiveness would require additional studies.

While it is possible for telecare to replace usual care for some patients (25), particularly those with limited access to health professionals, the available evidence does not inform us about the patient subgroups most likely to benefit. A recently published randomized trial of intensive insulin therapy versus usual care among young patients with type 1 diabetes documented greater improvement in HbA_1c from intensive treatment (including intensive face-to-face advice and support) in the subgroup of patients least competent in self-management (30). It is plausible that delegation of routine diabetes support to nonphysician clinicians (31) and to telecare to replace routine face-to-face visits among patients with adequate self-management skills who are doing well may free up resources (e.g., clinical encounter appointment slots with a diabetologist) to deliver intensive advice and support to patients with limited self-management skills and who are not doing well.

In summary, telecare (the transmission of glucometer data followed by nurse-mediated feedback and support) augmenting usual care among patients with type 1 diabetes and inadequate glycemic control had a small effect on glycemic control compared with the transmission of glucometer data without feedback in the context of usual care. A meta-analysis of the available evidence on telecare in patients with type 1 diabetes was consistent with the findings of our trial and effectively ruled out larger treatment effects (change in HbA_1c ≥1%). Taken together, the evidence is insufficient to recommend telecare or to identify a subgroup of patients most likely to benefit from this technology.

The Mayo Foundation funded this study with a research award to Y.C.K. Roche Diagnostics donated modems and glucometer equipment for both study groups. Other than these specific contributions, these funding sources had no other role in the planning, conduct, analysis, or reporting of this study.

We are especially grateful to Stephen D. Weigand, MS, for his helpful suggestions; to Ann Oberg, PhD, for her assistance with randomization; to Pat Erwin from the Mayo Library for conducting expert searches for the systematic review; to Deborah Toobert, PhD, and Russell Glasgow, PhD, for their permission to use the SDSCA questionnaire; and to Robert A. Rizza, MD, for his critical review of the study protocol.