Long-term Glycemic Variability and Risk of Adverse Outcomes: A Systematic Review and Meta-analysis
OBJECTIVE Glycemic variability is emerging as a measure of glycemic control, which may be a reliable predictor of complications. This systematic review and meta-analysis evaluates the association between HbA1c variability and micro- and macrovascular complications and mortality in type 1 and type 2 diabetes.
RESEARCH DESIGN AND METHODS Medline and Embase were searched (2004–2015) for studies describing associations between HbA1c variability and adverse outcomes in patients with type 1 and type 2 diabetes. Data extraction was performed independently by two reviewers. Random-effects meta-analysis was performed with stratification according to the measure of HbA1c variability, method of analysis, and diabetes type.
RESULTS Seven studies evaluated HbA1c variability among patients with type 1 diabetes and showed an association of HbA1c variability with renal disease (risk ratio 1.56 [95% CI 1.08–2.25], two studies), cardiovascular events (1.98 [1.39–2.82]), and retinopathy (2.11 [1.54–2.89]). Thirteen studies evaluated HbA1c variability among patients with type 2 diabetes. Higher HbA1c variability was associated with higher risk of renal disease (1.34 [1.15–1.57], two studies), macrovascular events (1.21 [1.06–1.38]), ulceration/gangrene (1.50 [1.06–2.12]), cardiovascular disease (1.27 [1.15–1.40]), and mortality (1.34 [1.18–1.53]). Most studies were retrospective with lack of adjustment for potential confounders, and inconsistency existed in the definition of HbA1c variability.
CONCLUSIONS HbA1c variability was positively associated with micro- and macrovascular complications and mortality independently of the HbA1c level and might play a future role in clinical risk assessment.
Current management of type 1 and type 2 diabetes uses the average glycemia measure HbA1c to monitor control. This rationale is based on trial and observational evidence that lowering HbA1c reduces the risk of the micro- and macrovascular complications of diabetes (1–4). Whether an average glycemic measure is most appropriate to assess the risk for complications is currently under debate. For example, one analysis of the Diabetes Control and Complications Trial indicated higher rates of retinopathy in the conventional treatment group than in the intensive treatment group over time in patients with similar average HbA1c values in the two groups (5), suggesting that additional factors other than mean HbA1c may be responsible for this increased retinopathy risk (5–7). Glycemic variability is now emerging as a possible additional measure of glycemic control, which may be a better predictor of complications than average glycemic measures.
Glycemic variability relates to fluctuations in glycemia. Short-term glycemic variability refers to within- or between-day fluctuations in an individual and includes multiple methods of assessment. Long-term glycemic variability refers to fluctuations over several weeks or months and is most commonly assessed by HbA1c variability. However, neither have a standardized method of measurement or definition (8). A recent meta-analysis concluded that HbA1c variability, assessed by the SD, is associated with renal disease in type 1 and type 2 diabetes (9). However, no systematic reviews or meta-analyses have evaluated the relationship between long-term glycemic variability and other complications in diabetes, despite contradictory literature providing evidence in support (6,10–15) and against (16–20) such a relationship.
Long-term glycemic variability is important for several reasons. First, unlike short-term glycemic variability, long-term glycemic variability may predict complications in both type 1 and type 2 diabetes (6,10–15,21–29). Second, HbA1c is routinely recorded in primary care for both types of diabetes, whereas measures of short-term variability are not (30,31). Finally, it could be a potentially modifiable risk factor. Through a systematic review and meta-analysis, we evaluated the evidence for the association of HbA1c variability with mortality and complications in type 1 and type 2 diabetes to gain insight into the clinical utility of this relationship to predict adverse outcomes.
Research Design and Methods
Data Sources and Searches
We searched Medline and Embase for articles published between 2004 and September 2014, using the search terms shown in the Supplementary Data. We updated the search in July 2015. C.S.K. conducted the initial search, and C.G. duplicated it. All resulting articles, including conference abstracts, were reviewed. Broad search criteria included diabetes terms, outcomes of interest terms, and exposure terms (HbA1c variability).
We included studies of patients with diabetes that evaluated HbA1c variability and adverse outcomes published within the past 10 years. No restrictions were placed on participant age or definition of HbA1c variability used. The main adverse outcomes of interest were renal disease (diabetic nephropathy, microalbuminuria, macroalbuminuria, renal failure, chronic kidney disease), diabetic retinopathy, diabetic neuropathy, cardiovascular macrovascular events (myocardial infarction, ischemic heart disease, heart failure, stroke, peripheral vascular disease), and death. We excluded reviews, editorials, and case reports and searched the bibliographies of included studies and relevant reviews for additional studies. Study titles and abstracts were initially screened independently by two reviewers (C.G. and S.A.), and full articles on potentially relevant studies were downloaded and reviewed for inclusion. Five reviewers discussed and decided on the final inclusion of studies for this review and meta-analysis (C.G., C.S.K., S.A., E.K., M.A.M.) (Fig. 1).
Data Extraction and Quality Assessment
Data extraction was performed independently by two reviewers (C.G. and S.A.). Data collected were study design, participant characteristics, quality of study assessment, definitions of HbA1c variability, outcomes evaluated, and results. Discrepancies in extractions were discussed with two other reviewers (C.S.K. and Y.L.).
Data Synthesis and Analysis
We conducted a random-effects meta-analysis of the adjusted risk estimates (where available) with use of the inverse variance method in RevMan 5.3 (Nordic Cochrane Centre). Analysis was stratified according to the definition of HbA1c variability used, the method of analysis used, and the type of diabetes. In terms of HbA1c variability, studies were divided into those that reported a coefficient of variation (CV) and those that reported an SD. Within the two groups, the analysis was further divided according to whether the highest variability group was compared with the lowest variability group or whether variability was measured per incremental increase in CV or SD. Where possible, we chose to analyze results for the group with the greatest HbA1c variability against that of the one with lowest variability. If there were several groups with differing levels of variability, we conducted the meta-analysis based on the group with the greatest variability compared with the one with the least variability.
Both SD and CV are measures of variability. SD measures how much values differ from the group mean. CV is the ratio of SD to the mean, so it is a measure independent of the mean. The CV may be appropriate for parameters such as HbA1c where the variability is likely to increase as the mean increases. However, there is no standardized method of measuring HbA1c variability (8).
Where there were insufficient studies for pooling or significant heterogeneity that could not be explained, we performed narrative synthesis. We assumed similarity between risk ratios (RRs) and odds ratios (ORs) because adverse events are rare (32).
Statistical heterogeneity was assessed with the I2 statistic (33), where values of 30–60% represent a moderate level of heterogeneity. Six sensitivity analyses were performed. These included prospective studies; studies with a follow-up of >5 years; and studies that adjusted for duration of diabetes, number of HbA1c measurements, comorbidities, and baseline medications. Publication bias was assessed using funnel plots if there were >10 studies and no evidence of statistical heterogeneity in a particular meta-analysis (34).
Studies Included and Participant Characteristics
Figure 1 shows a flow diagram of the study selection process. Twenty studies in 87,641 participants met the inclusion criteria. Ten studies included participants from Europe (10,11,14,16–19,26,29,35,36), eight from Asia (12,13,15,18,20,24,27,28), four from North America (6,18,23,25), and one from Australasia (18). The number of participants in each study ranged from 234 to 35,891. Details of the study design and participants are shown in Table 1.
Type 1 Diabetes
Seven studies included 44,021 participants with type 1 diabetes (6,10,11,14,25,29,35). These comprised three retrospective cohort studies (11,14,25), two prospective cohort studies (10,29), one post hoc analysis of a randomized controlled trial (6), and one cross-sectional study (35). Most studies used data from secondary care apart from two (10,11) that used primary and secondary care data.
Type 2 Diabetes
Thirteen studies included 43,620 participants with type 2 diabetes (12,13,15–20,23,24,26–28,36). These comprised six retrospective cohort studies (13,15,20,23,28,36), five prospective cohort studies (12,16,17,19,26,27), and two post hoc analyses of randomized controlled trials (18,24). All studies used secondary care data apart from one of primary and secondary care data (19) and one of solely U.S. primary care data (23).
Quality Assessment of Included Studies
The quality assessment of included studies is shown in Table 2. For both type 1 and type 2 diabetes, the outcome assessment varied from blood and urine tests for diabetic nephropathy, to fundoscopy for retinopathy, to formal follow-up for cardiovascular events and death. The frequency of outcome evaluation differed depending on the study. All studies adjusted for mean HbA1c.
Type 1 Diabetes
The shortest follow-up was a mean of 5.2 years (11) and the longest, 23 years (14). The number of HbA1c measurements per patient ranged from a median of 4 (29) to 13 (10). Data from all studies were unclear about loss to follow-up. All studies used some form of adjustment for baseline covariates; however, five did not adjust for baseline diabetes medications (10,11,14,25,29), and none adjusted for baseline hypertension medication.
Type 2 Diabetes
The shortest follow-up was 2 years (20) and the longest, a median of 15.9 years (13). The number of HbA1c measurements per patient ranged from 3 (19) to a median of 79 (13). In six studies, loss to follow-up was unclear; six studies had <10% of participants lost to follow-up, whereas one had lost 27.5% to follow-up (13). All the studies used some form of adjustment for baseline covariates; however, six did not adjust for baseline diabetes medication (15,19,20,23,24,27), and four did not adjust for baseline hypertension medication (13,16,17,20,27). Of the seven studies that did include hypertension medication (12,15,18,23,24,26,28), two adjusted for ACE inhibitor/angiotensin receptor blocker use (23,24). The definition of glycemic variability, outcome evaluated, and study follow-up and results are shown in Table 3.
Type 1 Diabetes
Three studies evaluated adverse outcomes by considering the impact of HbA1c CV (Supplementary Fig. 2) (11,14,35). There was no significant association between HbA1c CV and retinopathy (RR [95% CI] 1.34 [0.89–2.04], two studies) or microalbuminuria (1.04 [1.00–1.08], one study). The study by Hermann et al. (14), however, reported that HbA1c variability based on CV was associated with a 3.5% higher risk of diabetic retinopathy per 1-unit increase in HbA1c CV at 10 years disease duration.
Four studies evaluated adverse outcomes associated with HbA1c SD (Supplementary Fig. 2) (6,10,25,29). All showed a significant association of HbA1c SD and adverse outcomes. Highest to lowest variation SD group was associated with an increased risk of nephropathy (RR [95% CI] 1.92 [1.49–2.47]) and cardiovascular events (1.98 [1.39–2.82]). Incremental increases in SD were also associated with increased risk of nephropathy (1.86 [1.41–2.46]), microalbuminuria (1.56 [1.08–2.25], two studies), and retinopathy (2.11 [1.54–2.89]).
No studies evaluated HbA1c variability in type 1 diabetes and mortality. Sensitivity analyses for study type and studies that adjusted for duration of diabetes, number of HbA1c measurements, comorbidities, and baseline medications produced similar results to those recorded with inclusion of all studies (Supplementary Table 1).
Type 2 Diabetes
Studies reporting all-cause mortality as an outcome were not pooled due to high levels of heterogeneity, which was believed to be a result of differing follow-up durations and loss to follow-up. The outcome, therefore, was split according to short follow-up (<5 years) and long follow-up (≥5 years).
Six studies evaluated adverse outcomes by considering the impact of HbA1c CV (13,15,18,26,28,36), and nine studies considered HbA1c SD (12,13,15–18,20,24,26,27). Increase in HbA1c variability defined by high versus low CV groups was associated with increased risk of diabetic nephropathy (RR [95% CI] 1.58 [1.19–2.10]) and all-cause mortality in studies with ≥5 years of follow-up (2.89 [1.45–5.74]) and in those with <5 years follow-up (1.06 [1.01–1.11]) (Supplementary Fig. 3A). Incremental increases in CV were also associated with a significantly increased risk of nephropathy (1.03 [1.01–1.05]), macro/microvascular events (1.11 [1.02–1.21]), macrovascular events (1.18 [1.04–1.33]), and mortality with ≥5 years of follow-up (1.10 [1.03–1.16]) and <5 years of follow-up (1.31 [1.16–1.48]). No significant association was found between incremental increase in CV and microvascular events (1.07 [0.96–1.20]) (Supplementary Fig. 3B).
Considering HbA1c variability with SD, high versus low SD group was associated with increased risk of nephropathy (RR [95% CI] 1.24 [1.02–1.51]), all-cause mortality (2.34 [1.48–3.71], two studies), microalbuminuria (1.34 [1.15–1.57], two studies), macroalbuminuria (1.41 [1.03–1.93]), ulceration/gangrene (1.50 [1.06–2.12]), and mortality in studies with ≥5 years of follow-up (3.09 [1.45–6.58]) and in those with <5 years of follow-up (1.99 [1.11–3.55]) (Supplementary Fig. 4A). Incremental increase in SD was associated with an increased risk of nephropathy (1.22 [1.05–1.42], two studies), end-stage renal failure (1.53 [1.35–1.73]), microalbuminuria (1.20 [1.03–1.39]), macro/microvascular events (1.12 [1.02–1.22]), macrovascular events (1.21 [1.06–1.38]), cardiovascular disease (1.27 [1.15–1.40]), and mortality in studies with ≥5 years of follow-up (3.17 [1.43–7.03]) and in those with <5 years of follow-up (1.34 [1.18–1.53]). No significant association was found between incremental increase in SD and microvascular events (1.08 [0.96–1.21]) or retinopathy (1.03 [0.69–1.53], two studies) (Supplementary Fig. 4B).
A study by Penno et al. (17) reported additional nonsignificant associations with any lower-limb vascular event, any cerebrovascular event, any coronary event, acute myocardial infarction, any cardiovascular disease, and stroke. This study could not be included in the meta-analysis because raw data were not provided. Data on the significant association of HbA1c CV and all-cause mortality reported by Lang et al. (36) (RR [95% CI] 1.02 [1.01–1.03]) were not included in the meta-analysis because all participants had incident chronic heart failure, increasing heterogeneity with other studies and affecting external validity.
Cummings et al. (23) reported a significant worsening of one more chronic kidney disease stages with an average excess HbA1c >7% (53 mmol/mol) (OR 1.173 [95% CI 1.031–1.335]). Hirakawa et al. (18) also used other variability measures, including HbA1c variation independent of the mean, HbA1c residual SD, and HbA1c average real variability. All were significantly associated with macrovascular complications, macro/microvascular complications, and mortality based on data from ADVANCE (Action in Diabetes and Vascular Disease: Preterax and Diamicron MR Controlled Evaluation) (hazard ratio [HR] [95% CI]: variation independent of the mean HbA1c 1.17 [1.04–1.32], 1.11 [1.02–1.2], 1.30 [1.15–1.46]; residual SD HbA1c 1.20 [1.07–1.35], 1.10 [1.01–1.19], 1.33 [1.19–1.49]; average real variability HbA1c 1.21 [1.07–1.37], 1.11 [1.02–1.21], 1.38 [1.22–1.55]). Skriver et al. (19) defined HbA1c variability as the mean absolute residual around the line connecting index value and closing value. They reported that for index HbA1c ≤8% (64 mmol/mol), variability >0.5 was associated with increased all-cause mortality (HR 1.3 [95% CI 1.1–1.5]) per HbA1c percentage point variability. However, for individuals with index HbA1c >8% (64 mmol/mol), no association between HbA1c variability and mortality could be identified.
Sensitivity analyses for study type and studies that adjusted for duration of diabetes, number of HbA1c measurements, baseline medications, and comorbidities produced similar results to those that included all studies (Supplementary Table 2). There were too few studies in the meta-analysis to assess publication bias.
Glycemic variability is emerging as a measure of glycemic control that may be an important predictor of complications in patients with diabetes. Our analysis suggests that greater HbA1c variability, irrespective of the definition used, is associated with adverse outcomes in several micro- and macrovascular end points and mortality. We report that HbA1c variability in type 1 and type 2 diabetes is associated with renal and cardiovascular disease. The former is supported by 10 studies using both CV and SD as a measure of HbA1c variability (6,10,12,16,24–29). Only one small cross-sectional study in a pediatric cohort using CV did not report this significant association (35). The latter is supported by two studies using SD (10,12). Retinopathy appears to be associated with HbA1c variability in type 1 diabetes (6) but not in type 2 diabetes (16,20). However, this was shown with SD as the measure of variability (6) and not with CV (11,14). Four studies addressed the relationship with mortality in type 2 diabetes (13,15,18,19), with significant associations reported for SD and CV (13,15,18). Post hoc analysis of the ADVANCE data set showed HbA1c variability defined by CV and SD to be associated with macrovascular events and combined macro/microvascular events but not with microvascular events in type 2 diabetes (18). These findings were independent of mean HbA1c, suggesting that HbA1c variability may be a useful additional risk stratification tool in both type 1 and type 2 diabetes.
The present results add to the findings of a significant association between HbA1c SD and renal disease reported in the 2014 systematic review and meta-analysis by Cheng et al. (9). This meta-analysis of eight articles assessing the relationship between HbA1c variability and renal disease in type 1 and type 2 diabetes has several limitations. Studies were excluded that did not report HR [including the study by Penno et al. (16)], measures of variability other than SD or CV were not considered, and different renal outcomes/end points were pooled.
The present results also differ from the previous systematic reviews of short-term glycemic variability and the risk of complications in diabetes (21,22). In previous studies, short-term glycemic variability was assessed by a variety of methods, such as SD, CV, and mean amplitude of glycemic excursions of daily glucose readings, including self-monitoring of blood glucose, continuous blood glucose monitoring, fasting plasma glucose, and postprandial glucose (21). These studies found no consistent evidence of a relationship between short-term glycemic variability and the risk of any complications in type 1 diabetes. However, in six studies involving patients with type 2 diabetes, both previous reviews found a positive association between glucose variability and retinopathy. In general agreement with these two reviews, we found a positive relationship between glycemic variability and cardiovascular disease in type 2 diabetes. The present finding of a significant association between HbA1c variability and all-cause mortality in type 2 diabetes is consistent with the findings of Nalysnyk et al. (22) but not those of Smith-Palmer et al. (21).
These differing risk prediction results for short- and long-term glycemic variability may indicate differing pathological mechanisms. Short-term glycemic variability has been postulated to induce oxidative stress, inflammatory cytokines, and endothelial damage (37–41), mechanisms linked to diabetes complications (42,43). Additional mechanisms that may explain the association of HbA1c variability and adverse events include cellular metabolic memory (44–47), insulin resistance (10,48), sensitivity of HbA1c for detecting glycemic variability (44), and the exponential relationship between HbA1c and risk of microvascular complications (16,44).
Confounding factors rather than a causal relationship may explain the association of HbA1c variability with complications. These include poor medication compliance and self-management (10,12,28); multimorbidity (28); certain medications, such as steroids and antipsychotics (49); poor quality of life and lack of support (50,51); and infections (10).
Eight studies indicated that HbA1c variability was superior at predicting diabetes-related complications than mean HbA1c (6,10,12,13,15,17,24,25). Only one study found a significant association of mean HbA1c with diabetes-related complications but not with HbA1c variability (16,17). Further research is required to assess whether HbA1c variability might be clinically useful for risk stratification and whether it might be a valuable therapeutic target.
To our knowledge, this systematic review and meta-analysis is the first of HbA1c variability in diabetes and risk of mortality and complications other than renal disease. Limitations of the analysis are exclusion of non-English-language articles and studies before 2004. However, inclusion of studies earlier than the past 10 years may not be generalizable to current practices because current therapies (long-acting insulins, GLP-1 agonists, and dipeptidyl peptidase-4 inhibitors) were not available before 2004. Because of the small number of available studies, we were unable to use meta-regression to assess study characteristics as moderators. The heterogeneity estimates vary from very high to zero, and arguably, highly heterogeneous studies should not be meta-analyzed in the first place. However, homogeneity has been shown to be rare and often falsely assumed, especially for small meta-analyses, sometimes leading to false conclusions (52). From a statistical point of view, it is better to identify heterogeneity (which is likely present anyway), which can then be successfully accounted for in a random-effects meta-analysis model (53). Some limitations are inherent to the available literature, including the observational nature of studies, retrospective design of some, unclear or short follow-up periods, exclusion of patients deemed as having too few HbA1c measurements (13,14,16,17,23,28), and the nonadjustment for different numbers of HbA1c measurements, duration of diabetes, comorbidities, and baseline medications. In addition, there is no accepted method of assessing HbA1c variability, and a single definition of outcomes was not used.
The present findings support the need for further studies investigating the relationship between HbA1c variability and diabetes complications. More-sophisticated measures of HbA1c variability are needed as well as consensus about how such variability should be defined, including adjustment for differing intervals between HbA1c measurements and addressing the temporality of the variance problem (54). The present findings suggest that HbA1c variability may be a useful risk stratification tool in both type 1 and type 2 diabetes.
In conclusion, this meta-analysis shows significant associations between HbA1c variability and all-cause mortality, renal disease, and cardiovascular disease in type 2 diabetes and retinopathy, renal disease, and cardiovascular disease in type 1 diabetes. These relationships are independent of mean HbA1c, and in the majority of studies, variability was more predictive of adverse outcomes than mean HbA1c.
Funding. C.G. is funded by a National Institute for Health Research Academic Clinical Fellowship.
Duality of Interest. No potential conflicts of interest relevant to this article were reported.
Author Contributions. C.G. searched databases, selected studies, extracted data, and wrote the manuscript. C.S.K. searched databases, selected studies, extracted and analyzed data, and contributed to writing the manuscript. S.A. selected studies and extracted data. I.B. contributed to the discussion. E.K. and M.A.M. selected studies, contributed to the discussion, and reviewed and edited the manuscript. P.K.M. and G.H. reviewed and edited the manuscript. Y.L. extracted and analyzed data, contributed to the discussion, and reviewed and edited the manuscript. M.K.R. contributed to the design and reviewed and edited the manuscript.
This article contains Supplementary Data online at http://care.diabetesjournals.org/lookup/suppl/doi:10.2337/dc15-1188/-/DC1.
- Received June 13, 2015.
- Accepted September 22, 2015.
- © 2015 by the American Diabetes Association. Readers may use this article as long as the work is properly cited, the use is educational and not for profit, and the work is not altered.