A meta-analysis of 9 RCTs found little benefit in self-monitoring of blood glucose levels on A1C outcomes in patients with diabetes mellitus not taking insulin.
: To perform a meta-analysis of randomized controlled trials (RCTs) and systematic reviews evaluating the efficacy of self-monitoring of blood glucose (SMBG) levels among patients with diabetes mellitus (DM).
: Meta-analysis of RCTs among patients with DM not taking insulin comparing patients with SMBG versus those without SMBG and reporting results as change in glycosylated hemoglobin (A1C) values.
: Prior systematic reviews and a PubMed search were used to identify studies. Data were extracted by trained physician reviewers working in duplicate. Trials were classified according to duration of the intervention, and random-effects meta-analysis was used to pool results.
: Three trials of SMBG of 3 monthsâ€™ duration were too heterogeneous to pool. Nine other trials were identified. Five trials of SMBG of 6 monthsâ€™ duration yielded a pooled effect estimate of a decrease in mean A1C values of -0.21% (95% confidence interval [CI], -0.38% to -0.04%). Four trials that reported outcomes of 1 year or longer yielded a pooled effect estimate of a decrease in mean A1C values of -0.16% (95% CI, -0.38% to 0.05%). Three trials reported hypogly- cemic outcomes, which were increased in the patients using SMBG, although this mostly involved asymptomatic or mild episodes.
: At most, SMBG produces a statistically significant but clinically modest effect in controlling blood glucose levels in patients with DM not taking insulin. It is of questionable value in helping meet target values of glucose control.
(Am J Manag Care. 2008;14(7):468-475)
Self-monitoring of blood glucose (SMBG) levels is proven effective at helping control glucose levels in patients with diabetes mellitus (DM) taking insulin. The usefulness of SMBG in patients with DM not taking insulin is unclear.
Our meta-analysis of 9 randomized controlled trials of SMBG found a statistically significant improvement in glycosylated hemoglobin outcomes at 6 months of -0.21%. Results at 3 months or 12 months were not significant.
At best, SMBG is an intervention of modest efficacy in patients with DM not taking insulin.
According to the World Health Organization,1 at least 180 million people worldwide have diabetes mellitus (DM). The National Diabetes Information Clearinghouse estimates that DM costs $132 billion in the United States every year.2 Given these estimates along with the projection that the worldwide incidence of DM will double in the next 20 years,1 research into better management of this chronic disease is important.
Glucose control along with blood pressure and lipid control is a main goal of DM management. In addition to pharmaceutical agents, an advocated method to achieve better glucose control is self-monitoring of blood glucose (SMBG) levels. For patients with DM taking insulin, SMBG has generally been accepted as necessary to help monitor the effect of insulin on daily blood glucose levels, with adjustments in dosage made in response to SMBG values. However, evidence supporting the use of SMBG for patients with DM not requiring insulin is not as clear. Self-monitoring of blood glucose levels is postulated to have a beneficial effect on glucose levels in patients not taking insulin by promoting dietary and lifestyle changes that a patient may make as a response to feedback provided by blood glucose results.3-6
Prior trials and systematic reviews have reached conflicting results, with 6 trials4,5,7-10 and 3 reviews11-13 reporting benefits and 3 trials3,6,14 and 3 reviews 15-17 reporting no benefits. With an estimated cost of $466 million for Medicare alone, establishing the efficacy or lack of efficacy of SMBG represents an important goal for providers and payers interested in optimizing management of this condition. Because new larger and better-quality trials of SMBG continue to appear, we undertook a metaanalysis to provide the best current estimate of the efficacy of SMBG in patients with type 2 DM not taking insulin.
We first identified prior systematic reviews and meta-analyses on this topic. We identified 6 reviews.11-13,15-17 We judged the search strategy and inclusion and exclusion criteria of the review by Welschen et al,12 which included studies published from 1996 to September 2004, to be comprehensive and acceptable as the basis to begin our review. We updated this review by searching PubMed from the end date of the prior search to July 2007. We searched PubMed for the following terms: randomized controlled trial AND diabetes mellitus, as well as type 2 AND blood glucose self-monitoring. In addition to our PubMed search and screening of references from prior reviews, we performed reference mining of retrieved articles and received articles from experts. During this process, another review13 became available, and we reference mined this review as well.
Data were independently abstracted by 2 general internists (AT, MR) trained in critical reading of the literature, with consensus resolution. The following data were abstracted from included trials: design; randomization and appropriateness; blinding and appropriateness; withdrawals and dropouts described; sample size enrolled and followed up; characteristics of the population, including percentage of women and race/ethnicity; mean, median, and range of ages; body mass index and duration of DM; reported comorbidities; sample size, intervention, and exposure data for each arm of the study (intervention and exposure data included components of the intervention, total number of visits, frequency of SMBG, number of days per week monitored, duration of treatment, and cotherapies); outcomes measured; intervals in which the outcomes were measured; and adverse events.
The mean (SD) A1C level was recorded by treatment arm for each reported follow-up point. For trials that reported a mean outcome but no standard deviation, we estimated the standard deviation by taking the weighted mean standard deviation across all other trials that reported standard deviations for the A1C level.18
We synthesized data among the articles that were determined to be clinically eligible. Duration of follow-up and frequency of SMBG were reviewed across studies to see if they were comparable.
Since the outcome of interest was the same across all trials, a mean difference was calculated for each time point that reported statistical data. The mean difference is the difference between the follow-up mean A1C level for the SMBG group and the follow-up mean A1C level for the control group. A negative mean difference indicates that the SMBG group has a lower mean A1C score than the control group. For our main analysis, we did not control for the baseline mean A1C level for each group (a difference of differences estimate) because there is evidence that this approach is susceptible to bias.20 We present results controlling for the baseline as a sensitivity analysis.
A pooled estimate was calculated by follow-up duration in the following categories: 3 to 6 months, 6 to 11 months, and 12 months or longer. The pooled estimate was calculated using the DerSimonian and Laird21 random-effects model. In addition, we calculated a pooled estimate stratified by highquality and low-quality trials.
Meta-regression analyses22 were performed to individually examine the effect of treatment frequency, quality score, and mean baseline A1C level on the mean difference. For trials with more than 1 follow-up duration, the long-term estimates were used.
Test of heterogeneity was performed using the I2 statistic.23 I2 values close to 100% represent very high degrees of heterogeneity. Publication bias was examined using Begg rank correlation24 and Egger regression asymmetry test.25 All analyses were conducted in STATA 9 (Stata Corp LP, College Station, Texas).26
In total, we examined 55 titles. Seventeen titles were identified from prior systematic reviews. The electronic update search identified 25 titles. An additional 12 titles were identified through reference mining. One was identified by a content expert (JEW). Of the titles identified through our electronic literature search, 14 were rejected as irrelevant to the project. This left 41 from all sources. Eleven articles were excluded at abstract review.
In total, we reviewed 30 articles. Initial screening of the articles resulted in 16 RCTs that measured the effect of SMBG with at least 3 months of follow-up. Seven were excluded, 1 because the trial presented duplicate data, 2 because they did not report A1C levels as an outcome, 1 because the trial included an unknown number of patients taking insulin, and the other 3 because the trials compared a control group of patients with SMBG versus an intervention group of patients with SMBG plus other components ().
Description of the Evidence
The 9 RCTs ranged in size from 29 to 988 subjects. All patients had type 2 DM, with mean durations of 3 to 13 years. All trials but 1 included only patients treated without insulin.
The mean ages of patients ranged from 50 to 66 years. Almost all trials included counseling and education with SMBG in the intervention group, but other components of the intervention varied (). All trials measured A1C level as an outcome; 5 trials assessed this at 6 months, and 4 trials assessed this at 1 year or later. Three other trials assessed this at 3 months but were too heterogeneous to pool. The quality of trials varied; most trials scored positively on less than half of the criteria on the Delphi list.19 Details of each trial are given in (quality criteria are available in an online table [eAppendix Table; available at www.ajmc.com]).
Improving Glycemic Control
We grouped trials based on the duration of the intervention. The individual and pooled results are shown in .
We identified 3 trials6,9,27 that reported A1C outcomes at 3 months. The trials reported variable results. We did not pool the results of these 3 trials because their results were too heterogeneous, with an I2 statistic of 67%. We identified 5 trials3-5,14,27 that reported outcomes at 6 months. One trial4 reported a statistically significant improvement in A1C level, although a second trial 5 yielded a statistically significant result after adjusting for baseline differences. The random-effects pooled effect estimate of these 5 trials was a change in mean A1C values of -0.21% (95% confidence interval [CI], -0.38% to -0.04%). The I2 statistic for heterogeneity was 0%.
We identified 4 trials6-8,27 that reported outcomes at 1 year or longer. No study reported a statistically significant difference between groups in the mean A1C values, although one study 7 reported statistically significant benefits after adjusting for baseline differences in A1C values. The randomeffects pooled effect estimate of these 4 trials was a change in mean A1C values of -0.16% (95% CI, -0.38% to 0.05%). The I2 statistic for heterogeneity was 0%.
We performed several additional analyses. We compared studies scoring 5 or more Delphi items positively (which we called high quality) with those scoring fewer than 5 items positively (low quality). The pooled results showed no statistically significant differences between high-quality and low-quality studies.
We repeated our primary analysis using as the outcome the difference in A1C levels between groups adjusted for baseline A1C levels (whether to do such adjusting in the results of an RCT is controversial). When analyzed this way, there was much greater heterogeneity between studies, with I2 statistics of 49% and 73% for studies with 6-month and 12-month outcomes, respectively. However, despite this, our primary pooled results were remarkably similar, with a modest effect on A1C levels at 6 months of -0.23% (95% CI, -0.48% to 0.02%) (compared with a pooled result of 0.21% in the main analysis) and a nonsignificant effect at 12 months of -0.26% (95% CI, -1.00% to 0.48%). In the difference of differences analysis, high-quality studies reported lower estimates of effect than low-quality studies, an observation seen in other conditions.28
Meta-regression analysis on baseline A1C levels showed differential effectiveness (P = .06), with higher baseline A1C levels being associated with lesser efficacy of SMBG. Each 1% increase in A1C level was associated with a 0.19% decrease in efficacy of SMBG. Therefore, indirect evidence suggests that SMBG results in a smaller percentage of A1C level change for patients with higher baseline A1C values.
We attempted to identify other components of the intervention or characteristics of the patients associated with greater effectiveness. The trials did not have sufficient dissimilarity in intervention components to permit a meta-regression analysis (Table 1). Almost all studies included SMBG and counseling and education, rendering an assessment of the effect of one without the other impossible, and other intervention components were too sparsely distributed to support meta-regression analysis. Meta-regression analysis using the quality assessment (as a continuous variable or dichotomous variable at a threshold value of 5) also did not demonstrate differences between results.
The funnel plot did not support the existence of publication bias. Neither Begg correlation rank nor Egger asymmetry test yielded evidence of publication bias.
The principal finding of our meta-analysis was a modest but statistically significant improvement in A1C level at 6 months in patients with type 2 DM not requiring insulin when SMBG and education were added to management. The 3-month trials showed variability in results, leaving little to conclude. Similar though slightly smaller reductions in A1C level were found when pooling the 12-month trials. Although the pooled effect estimate at 12 months had a 95% CI that just crossed the null value, the effect estimate was not statistically different from the pooled effect estimate at 6 months.
The efficacy of SMBG has been studied previously. In one such review published by Welschen et al,12 6 RCTs were evaluated and were found to show an overall statistically significant effect of SMBG in decreasing A1C levels by 0.39% (95% CI, -0.56% to -0.21%). All but 1 of the trials in the review were also included in our meta-analysis (the study29 excluded from our analysis compared SMBG with urine testing of glucose levels). In a systematic review published for Agency for Healthcare Research and Quality, Balk et al13 included 5 RCTs and some nonrandomized cohort studies and found results similar to our meta-analysis, specifically a trend toward statistically significant but clinically modest reductions in A1C levels. In contrast, a recent cost-effectiveness analysis estimated that SMBG in a mixed population of patients with type 2 DM resulted in improved quality of life and a cost-effectiveness ratio of less than $10,000 per quality- adjusted life-year.30 However, this analysis was based on data from an observational cohort and did not include more recent RCT data reporting limited effectiveness of SMBG in primary care populations.
Two additional factors suggest that SMBG may be of little importance as a glucose control intervention in patients with type 2 DM not receiving insulin. First, our meta-regression findings suggest that the efficacy of SMBG may be lower as the baseline A1C level is higher. This means that SMBG may be least effective for the patients who need it most. Second, in the study27 that most closely approximates the use of SMBG in a community-based primary care practice, there was no effect found.
To help interpret the significance of the effect of SMBG on A1C levels demonstrated in our review, it needs to be put in the context of other interventions to control blood glucose levels. In an exhaustive review,31 investigators concluded that many oral DM medications as monotherapy (thiazolidinediones, second-generation sulfonylureas, metformin, and repaglinide) had reductions in A1C levels of about 1.0%. This is about 5 times the effect produced by SMBG in our pooled estimate. Lower A1C reductions have been seen for other classes of hypoglycemic agents (0.5%-0.8% for alpha-glucosidase inhibitors, 0.8% for dipeptidyl peptidase IV inhibitors, and 0.3%-0.45% for amylin analogues), but these are still greater than our findings for SMBG. More comparable may be DM education. Norris et al32 performed a meta-analysis on the effect of DM self-management education on A1C level and found an overall 0.26% reduction in A1C levels a few months after the intervention. Some of the trials included in the meta-analysis involved SMBG along with education, which renders the individual effect of either intervention difficult to separate. Still, SMBG is closer to DM self-management in terms of the effect size than it is to most pharmaceuticals.
Our study has several limitations. An important limitation common to systematic reviews is the quality of the original studies. The sensitivity analysis of our main result did not yield any suggestion that the quality of the trials influenced our findings in a significant way. Another limitation is the heterogeneity of the studies. While there were some differences in the populations being assessed, the most important heterogeneity in this meta-analysis was the differing intervention components added to SMBG and the difference in the recommendations for frequency of SMBG, provider interaction or algorithm to adjust medications, and intensity of education. In addition, because education and counseling were invariably included with SMBG in the various intervention groups, the effect of SMBG alone was impossible to distinguish. Although our literature search procedures were extensive and included all articles identified in prior reviews plus additional articles, publication bias is a limitation. Our formal tests for publication bias did not indicate the presence of possible publication bias, but such tests do not exclude the possibility that such bias exists.
In conclusion, our analysis showed a statistically significant but clinically modest overall reduction in A1C levels when using SMBG in patients with type 2 DM not taking insulin. The results of our meta-regression analysis and the findings by Farmer and colleagues27 further limit the likelihood that SMBG is a particularly useful intervention. Patients, providers, and health plans will need to look elsewhere for interventions that have the kind of effects needed to bring A1C levels down to target values.
Author Affiliations: From the Division of Internal Medicine (AT, MR, BM, PGS), and the Division of Endocrinology, Diabetes and Metabolism (JEW), Veterans Affairs Greater Los Angeles Healthcare System, Los Angeles, and the RAND Corporation (MJS, AZ, PGS), Santa Monica, CA.
Funding Source: This work was funded in part by the Department of Veterans Affairs Health Services Research and Development Evidence Synthesis Activity Pilot Program Project ESP 05-226. The views expressed in this article are those of the authors and do not necessarily represent the views of the US Department of Veterans Affairs.
Author Disclosure: The authors (AT, MR, BM, MJS, AZ, PGS) report no relationship or financial interest with any entity that would pose a conflict of interest with the subject matter of this article. Dr Weinreb has received lecture fees from Sanofi-Aventis and has attended meetings and conferences on behalf of Novartis.
Authorship Information: Concept and design (PGS); acquisition of data (AT, MR, BM, MJS, AZ); analysis and interpretation of data (AT, MR, JEW, MJS, AZ, PGS); drafting of the manuscript (AT, MR, BM, PGS); critical revision of the manuscript for important intellectual content (AT, JEW, PGS); statistical analysis (AT, MJS, AZ); obtaining funding (PGS); administrative, technical, or logistic support (BM); and supervision (PGS).
Address correspondence to: Ali Towfigh, MD, Division of Internal Medicine, Veterans Affairs Greater Los Angeles Healthcare System, 11301 Wilshire Blvd, Mailcode PACC, Los Angeles, CA 90073. E-mail: email@example.com.
1. World Health Organization. Media centre: diabetes. September 2006. http://www.who.int/mediacentre/factsheets/fs312/en/index.html. Accessed May 2, 2001.
3. Fontbonne A, Billault B, Acosta M, et al. Is glucose self-monitoring beneficial in nonâ€“insulin-treated diabetic patients? Results of a randomized comparative trial. Diabetes Metab. 1989;15(5):255-260.
5. Schwedes U, Siebolds M, Mertes G. Meal-related structured self-monitoring of blood glucose: effect on diabetes control in nonâ€“insulin-treated type 2 diabetic patients. Diabetes Care. 2002;25(11):1928-1932.
7. Rutten G, van Eijk J, de Nobel E, Beek M, van der Velden H. Feasibility and effects of a diabetes type II protocol with blood glucose selfmonitoring in general practice. Fam Pract. 1990;7(4):273-278.
9. Jaber LA, Halapy H, Fernet M, Tummalapalli S, Diwakaran H. Evaluation of a pharmaceutical care model on diabetes management. Ann Pharmacother. 1996;30(3):238-243.
11. Jansen JP. Self-monitoring of glucose in type 2 diabetes mellitus: a Bayesian meta-analysis of direct and indirect comparisons. Curr Med Res Opin. 2006;22(4):671-681.
13. Balk E, Teplinsky E, Trikalinos T, et al. Applicability of the evidence regarding intensive glycemic control and self-monitored blood glucose to Medicare patients with type 2 diabetes. Technology Assessment Report. September 7, 2007. http://www.cms.hhs.gov/determinationprocess/downloads/id40TA.pdf. Accessed March 12, 2008.
15. Davidson MB. Counterpoint: self-monitoring of blood glucose in type 2 diabetic patients not receiving insulin: a waste of money. Diabetes Care. 2005;28(6):1531-1533.
17. Faas A, Schellevis FG, Van Eijk JT. The efficacy of self-monitoring of blood glucose in NIDDM subjects: a criteria-based literature review. Diabetes Care. 1997;20(9):1482-1486.
19. Verhagen AP, de Vet HC, de Bie RA, et al. The Delphi list: a criteria list for quality assessment of randomized clinical trials for conducting systematic reviews developed by Delphi consensus. J Clin Epidemiol. 1998;51(12):1235-1241.
21. DerSimonian R, Laird N. Meta-analysis in clinical trials. Control Clin Trials. 1986;7(3):177-188.
23. Higgins JP, Thompson SG, Deeks JJ, Altman DG. Measuring inconsistency in meta-analyses. BMJ. 2003;327(7414):557-560.
25. Egger M, Davey Smith G, Schneider M, Minder C. Bias in meta-analysis detected by a simple, graphical test. BMJ. 1997;315(7109):629-634.
27. Farmer A, Wade A, Goyder E, et al. Impact of self monitoring of blood glucose in the management of patients with nonâ€“insulin treated diabetes: open parallel group randomised trial. BMJ. 2007;335(7611):132-140.
29. Allen BT, DeLong ER, Feussner JR. Impact of glucose self-monitoring on nonâ€“insulin-treated patients with type II diabetes mellitus. Randomized controlled trial comparing blood and urine testing. Diabetes Care. 1990;13(10):1044-1050.
31. Bolen S, Feldman L, Vassy J, et al. Systematic review: comparative effectiveness and safety of oral medications for type 2 diabetes mellitus [published correction appears in Ann Intern Med. 2007;147(12):887]. Ann Intern Med. 2007;147(6):386-399.