Complications of Diabetes and Metrics of Glycemic Management Derived From Continuous Glucose Monitoring

1Diabetes is associated with microvascular and macrovascular complications, including nephropathy, retinopathy, neuropathy, and cardiovascular and cerebrovascular disease, all of which contribute to a burgeoning disease burden. The risk of cardiovascular disease mortality and incidence of stroke are 2 to 4 times higher in Americans with diabetes compared to those without diabetes. Diabetic nephropathy accounts for 38.6% of new cases of end-stage renal disease, and 11.7% of adults with diabetes reported vision disability due to diabetic retinopathy. Globally, up to 75% of all lower-extremity amputations are performed in individuals with diabetes. In 2019, diabetes-related costs were estimated to have totaled $760 billion globally and $161.4 billion in Europe, with these figures being primarily composed of preventable hospital-based admissions from diabetes complications. Beyond financial cost, diabetes has a major effect on quality of life because of the daily demands of disease self-management, and possible effect of living with diabetes-related complications.

2Since the release of findings from the landmark Diabetes Control and Complications Trial (DCCT) and United Kingdom Prospective Diabetes Studies (UKPDS), glycated hemoglobin A1c (HbA1c) has become the gold standard by which to assess the success of glycemic management, and is used to guide clinical decision-making in diabetes management. There is a substantial body of evidence linking increases in HbA1c with diabetes-related complications. Yet patients with identical HbA1c values can also have vastly different complications rates—for example, only 11% of the variation in retinopathy risk may be explained by total glycemic exposure (HbA1c and duration of diabetes) in the DCCT cohort. Another study found mean HbA1c to be only weakly correlated to the presence and severity of cardiovascular autonomic neuropathy (r = 0.22). In addition, HbA1c measurements may also be influenced by a number of common pathological and physiological factors unrelated to glucose levels such as age, race, iron-deficiency anemia, chronic renal failure, pregnancy, and medications. While HbA1c is an integral marker of glycemic exposure over the preceding 8 to 12 weeks, it cannot describe interday or intraday glucose fluctuations. Beck et al showed there can be a wide range of glucose profiles associated with any given HbA1c level. Limitations around HbA1c highlight the need for complementary methods to assess glucose levels in people living with diabetes (Fig. 1).

3Continuous glucose monitoring (CGM) devices test interstitial glucose levels every 5 minutes and can record and store these data. Since the initial introduction of CGM into clinical practice in 2000, the accuracy and sophistication of these devices has progressively increased. Over recent years, the use of CGM has become more widespread. In the United States their use in people with type 1 diabetes (T1D) has increased from approximately 7% in 2011 to 28% in 2017, according to the Type 1 Diabetes Exchange Clinic Network encompassing 81 diabetes centers. In Australia, CGM is used by 79% of people younger than 21 years. In 2017 the Advanced Technologies and Treatments for Diabetes (ATTD) Congress came to an international consensus that mean glucose, glucose management indicator (GMI), glycemic variability, TIR, time above range (TAR), and time below range (TBR) were the primary measurable outcomes of CGM; and that they should be measured for more than 14 days with at least 10 days of valid data. Glycemic variability itself can be measured by numerous formulas using CGM data, including SD, mean amplitude of glycemic excursions (MAGE), coefficient of variation for glucose (CV), high blood glucose index (HBGI), low blood glucose index (LBGI), area under the curve hypoglycemia, and continuous overall net glycemic action (CONGA). These metrics are summarized in Table 1. The congress recommends the use of CGM data to complement HbA1c monitoring in a wide range of people with diabetes.

4CGM-derived outcomes are strongly correlated to HbA1c and thus indirectly with diabetes-related complications by inference. Beck et al found an r = –0.73 correlation between HbA1c and TIR70-180. Hirsch et al and Vigersky et al correlated HbA1c with TIR at r = –0.75 and r = –0.84, respectively. However, given the recency of the advent of CGM as a clinical and research tool, there is less evidence directly linking CGM metrics to these complications in comparison with HbA1c. This literature review therefore aims to amalgamate, summarize, and assess the existing evidence directly linking CGM-derived metrics with diabetes-related complications.

5A systematic literature search of PubMed and Embase was performed August 22, 2020, to identify studies demonstrating direct links between CGM-derived metrics of glycemic management, and diabetes-related complications. The search strategy is detailed in Table 2.

6Only studies from January 1, 2010 (inclusive) were considered because of the recent rapid development and usage of CGM, and the ATTD consensus on standardized CGM-derived metrics since 2017. Data applicable to T1D and type 2 diabetes (T2D) and Latent Autoimmune Diabetes in Adults (LADA) were included to comprehensively capture all microvascular and macrovascular diabetes-related complications. Furthermore, markers of diabetes complications (eg, carotid intima-media thickness [CIMT] as a proxy for the development of macrovascular disease), were also included for the same reason. Included studies were in humans of all ages and that were printed in English. Study types included were those presenting primary data.

7This review is designed to evaluate CGM, by which most devices measure glucose levels every 5 minutes. Studies using less frequent measurement regimens, such as the 7-point glucose profiles used in the DCCT, were excluded, as recent studies have suggested these glucose profiles are too infrequent to portray actual glycemic variability. Baghurst et al demonstrated that markers of acute glycemic variability, such as SD and MAGE, became unreliable if measured more than 2 to 4 hours and 1 hour apart, respectively.

8The metrics used to evaluate glycemic management were those proposed at the ATTD Congress: TIR, TAR, TBR, CV, and GMI. All markers of intraday glycemic variability were included, which is the domain that distinguishes CGM from other glucose-monitoring regimens. Consequently, metrics not reliant on CGM, such as visit-to-visit glucose variability or HbA1c variability, were excluded. Acutely unwell populations, for example postsurgery or poststroke patients, were excluded because of potentially altered glycemic management through mechanisms such as altered cortisol release. Studies in pregnant populations were also excluded. Publications involving insulin pump use were excluded because their effects on glycemic management may have altered the development of diabetes complications above the effect of CGM alone.

9A total of 1629 records were identified and exported to EndNote X9. Duplications were removed, and 1338 publications were screened by title and abstract (Fig. 2). The majority of papers excluded did not represent the outcome of interest. A total of 127 full-text papers were screened according to the defined criteria, and 27 met these criteria. A further 7 papers that met the criteria were identified: 6 via hand-searching reference lists, and 1 post hoc. The literature search identified conference abstracts that were subsequently hand-searched and included provided there was full-text availability. Data were then classified into T1D and T2D, and extracted into purpose-built tables.

10Of the 34 included papers, 30 used cross-sectional study designs and 4 were longitudinal cohort studies. Nine studies involved participants with T1D, 22 with T2D, and 3 had mixed populations. Of the 20 852 total participants, 663 had T1D, 19 909 had T2D, 192 had LADA, and 88 were control participants without diabetes. Thirteen papers were from China, 9 from Europe, 4 from South Korea, 3 from Japan and the United States, and 2 from Australia. In addition, one multinational study included data from the United States, Europe, and Cameroon.

11One study involving 32 participants addressed microvascular complications as a whole, showing associations with glycaemic variability (Table 3).

12Six studies addressed nephropathy, involving 1563 participants (Table 4). Four of the 5 studies investigating albuminuria demonstrated statistically significant associations with glycemic variability. The larger studies (n = 866 and n = 281) found these associations with TIR. One study associated reduced eGFR (estimated glomerular filtration rate) with high SD.

13Six studies addressed retinopathy, involving 6599 participants (Table 5). Four of the 5 studies investigating the presence of established retinopathy demonstrated statistically significant associations with glycemic variability. The larger studies (n = 3262 and n = 3119) found these associations for SD and reduced TIR. Both studies that addressed structural retinal changes in T1D found associations with glycemic variability, particularly LBGI.

14Peripheral neuropathy was investigated in 7 studies involving 2247 participants (Table 6). Four papers investigated the presence of peripheral neuropathy, while 3 studied markers of nerve conduction. Glycemic variability markers, particularly SD, MAGE, and reduced TIR, were associated both with the presence of peripheral neuropathy and abnormal nerve conduction across all papers. Cardiac autonomic neuropathy (CAN) was measured in 8 publications, involving 782 participants (Table 7). CAN was associated with glycemic variability in 7 of these publications, but some associations were directly contradicted in 4 publications. Reduced TIR was investigated in 2 studies and found to be associated with CAN in both.

15Thirteen studies addressed different aspects of macrovascular disease, which encompasses cardiovascular, cerebrovascular, and peripheral vascular disease, involving 10 206 participants: 412 with T1D and 9794 with T2D (Tables 8 and 9). Four papers investigated the presence of established macrovascular disease in which MAGE and reduced TIR were the predominant associations. The large (n = 6225) prospective cohort study identified TIR as an association for cardiovascular disease mortality and all-cause mortality. Glycemic variability was associated with abnormal echocardiography finding in two studies. Three studies used angiography to measure coronary artery disease in which glycemic variability, particularly MAGE, was an association. Cardiovascular disease risk factors were associated with glycemic variability in 1 out of 2 studies. Endothelial function was investigated in 1 paper and generally was not statistically significantly correlated with glycemic variability. CIMT was evaluated as a well-established proxy for cardiovascular and cerebrovascular disease in 4 publications. Associations with MAGE and SD were varied, but associations with TIR and TBR were unopposed.

16Overall, 22 out of 34 studies investigated the associations between CGM metrics and diabetes complications after adjusting for HbA1c (Table 10). Most (n = 19) showed various CGM metrics (namely MAGE, TIR, CV, and TBR) remained associated with diabetes complications after adjusting for HbA1c. Five studies found a glycemic variability metric to lose significance after adjustment for HbA1c.

17Complete findings from each individual study can be found in Tables S4.

18This review of 34 publications totaling 20 852 participants, which investigated the associations between 6 different domains of diabetes complications against more than 10 different markers of intraday glycemic variability, demonstrated associations with all included diabetes complications. Glycemic variability, particularly low TIR and high SD, was almost unanimously associated with nephropathy, retinopathy, and peripheral neuropathy. Associations with CAN were present but varied. Glycemic variability was also associated with the presence of cardiovascular disease, cardiovascular disease mortality, all-cause mortality, and CIMT, as well as echocardiography and angiography abnormalities.

19Of the CGM metrics recommended at the ATTD Congress, TIR was identified as the single most important CGM metric as it provides the most clinically practical information in 11 studies involving 14 319 participants. In nephropathy, TIR and glycemic variability metrics were strongly supported except for one cross-sectional study. The evidence for TIR and glycemic variability metrics in retinopathy was particularly strong given the study sizes (n = 6381). Glycemic variability was also associated with all severities of retinopathy, including preclinical neuroretinal abnormalities. Benbow et al found peripheral neuropathy to be the most common complication of diabetes, having a major effect on quality of life, and CAN is a particularly strong risk factor for mortality in T1D. Overall, the evidence suggests CAN is associated with decreased TIR, but not glycemic variability; but in peripheral neuropathy both glycemic variability—particularly MAGE—and lower TIR were supported as risk factors. These results highlight the importance of measuring these intraday parameters.

20Approximately three-quarters of individuals with diabetes die from a cardiovascular cause, and a diagnosis of diabetes may be as much a risk factor for poor cardiovascular disease outcomes as having coronary artery disease itself, even in T1D. Most evidence supported lower TIR as a risk factor for macrovascular disease. The largest study was of longitudinal study design and importantly provided strong evidence for TIR as a protective factor for cardiovascular disease mortality. The second largest was a cross-sectional study that also showed strong correlations between TIR and CIMT. Only one cross-sectional study that measured the presence of overt macrovascular disease found no statistically significant association with TIR.

21Glycemic variability showed mixed results for macrovascular disease and tended to relate more to microvascular complications. The ATTD Congress decided on CV as the consensus marker of glycemic variability, as SD is flawed in that it is significantly influenced by mean glucose. Yet CV seemed to have the least statistically significant associations with complications of diabetes, although it was measured only in 12 studies. While TIR and CV are emerging as popular CGM metrics, the relationship between the two should also be elucidated as it is possible to have a highly variable CGM trace with a high TIR, and conversely a minimally variable trace with low TIR (Fig. 3). Three studies investigated this in T2D, finding TIR remaining associated with albuminuria, retinopathy, and CAN after adjustments for glycemic variability metrics such as SD, MAGE, and CV. No studies examined the association of glycemic variability metrics after adjustment for TIR. Further research is required to clarify the nature of the relationship between TIR and CV with overt diabetes-related complications. Ideally these would involve large-scale longitudinal studies or randomized controlled trials (RCTs) with consistent CGM data that are long enough to capture adequate numbers of outcome events. However, such studies would be expensive and difficult to standardize for confounders such as exercise, physical activity, and treatment. Rather than individually randomized trials, evidence from CGM data using artificial intelligence may assist in interpreting observational data from individuals using CGM in collaboration with producers of CGM.

22Another reason for the lack of definitive outcomes is this review’s strict inclusion criteria of CGM use, as opposed to other less frequent blood glucose level (BGL) measurement regimens. While the accuracy of these infrequent measurement regimens is questionable compared to CGM, there is a wealth of data that must be considered. The DCCT was a large multicenter RCT spanning 10 years and involving 1440 participants. Reanalysis from Beck et al found that every 10% increase in TIR reduced the risk of retinopathy and microalbuminuria by 64% and 40%, respectively. However, the DCCT commenced in 1982 when therapeutic regimens and technologies were very different compared with today. It did not use CGM, and BGLs were measured only via 7 finger-prick samples collected in 1 day, every 3 months.

23Since diabetes complication risk is only partly explained by HbA1c, intraday glycemic variability is important to measure in conjunction with average glycemia because of its independent pathogeneses summarized by Nusca et al. While chronic sustained hyperglycemia can cause excessive protein glycation, acute hyperglycemic fluctuations may cause increased oxidative stress, inflammation, endothelial dysfunction, and altered gene expression. One case-control study involving 27 participants with T2D investigated the effect of glycemic variability on endothelial function (measured by flow-mediated dilatation) and oxidative stress (measured by plasma 3-nitrotyrosine and 8-iso-PGF2α urinary excretion rates). They found that oscillations in BGLs resulted in statistically significantly more endothelial dysfunction and oxidative stress than constant glucose levels, even if the average glucose was higher in the stable group. These effects lasted even beyond the return to euglycemia; thus, highlighting the importance of glycemic variability. A case-control study confirmed the strong association (r = 0.86) between MAGE and 8-iso-PGF2α urinary excretion rates, while another showed that transient hyperglycemia can induce lasting epigenetic changes in the promotor region of NFκB (a proinflammatory gene) in vitro and in mice. It is also established that high glycemia variability is associated with more frequent episodes of hypoglycemia, which contributes to a range of adverse effects including cardiovascular morbidity and mortality.

24A narrative review by Livingstone et al explored the glucose variability hypothesis—the notion that glucose variability contributes additional risk of diabetes complications after adjusting for HbA1c. While they agreed mechanisms for glycemic variability causing complications exist, they concluded that there were insufficient data at the time to substantiate this hypothesis. However, this was based primarily on the DCCT/EDIC (Epidemiology of Diabetes Interventions and Complications), which was greatly limited in that it used 7-point glucose profiles to estimate intraday glycemic variability, rather than using 5-minute data points from CGM. Our data (Table 10) showed that CGM-derived metrics of glycemic variability tended to remain associated with diabetes complications after adjusting for HbA1c (and indeed many other potential confounders). However, only one longitudinal study investigated this association, finding CV was associated with baroreflex sensitivity (CAN) after adjusting for HbA1c. Therefore, further longitudinal data are required to investigate the glucose variability hypothesis.

25Large-scale longitudinal studies have shown that CGM-users have lower HbA1c levels, less hypoglycemia, and more TIR compared to non-CGM cohorts. Thus, it would be expected for CGM use to translate longitudinally to reduced complications risk. CGM offers the unique ability to measure glucose levels in situations such as during sleep or exercise. Some systems have no requirement for daily capillary finger-prick tests, which patients appreciate. A survey of 3461 people with T1D or T2D identified TIR as the second most important of all factors that had a “big impact” on daily life with diabetes, while food choices were number one. CGM also alleviates the fear of hypoglycemia, which is a major barrier to exercise, dieting, and intensified treatment regimens, allowing for early glycemic management. This early management may be crucial in preventing future complications due to the “legacy effect.” A more than 66 000 person-year follow-up of the UKPDS RCT studied the effects of intensive glycemic management on the development of diabetes complications. It demonstrated that early intensive glycemic management resulted in risk reductions that persisted far beyond the transient HbA1c level differences between groups. Importantly, the integration of CGM also allows for more sophisticated use of insulin pump therapy, the gold standard for glycemia management.

26Health care systems must also adapt to allow for widespread CGM use. This includes education and training around consensus reporting and interpretation of results, their implications on treatment adjustments, as well as follow-up guidelines. The ATTD Congress recommends the use of ambulatory glucose profile reports to aid this interpretation. Guidelines for when and how often to use CGM also requires further investigation, as Vigersky et al found that periodic 14-day courses of CGM every 3 months would still adequately inform changes and responses to treatment as well as lasting behavioral changes. This may be aided by further cost-effectiveness analysis, which as it stands is extremely variable with different devices, patterns of use, and populations.

27This review had several limitations. First, there was considerable heterogeneity between studies, in the selection of study participants, treatment adjustments, and reporting of data. Thus, a meta-analysis was not possible, making the data more difficult to interpret. Second, 30 out of 34 papers were cross-sectional study designs, meaning that causal relationships between CGM-derived measurements and outcomes cannot be proven. Third, the ATTD Congress suggests longer than 14-day periods of CGM. Most studies used only 48 to 72 hours, which may not be representative of long-term control and therefore undermine any associations with long-term complications. This may not be the case in T2D populations, which have less glycemic variability than in T1D, and thus, patterns of glycemia are more reproducible on a day-to-day basis. Shorter monitoring periods in this cohort are therefore more likely to be representative of longer-term patterns of glycemia, which would better correlate with complication development. Fourth, some studies included individuals with short diabetes durations, which reduces the likelihood of diabetes-related complications, and therefore underestimates the effect of each marker of glycemic variability on the development of diabetes complications long term. Fifth, there was a distinct lack of data relating to T1D, which is important given the large use of CGM particularly in youth with T1D. While 11 studies involved populations with T1D, this totaled only 663 participants. Sixth, many studies inferred the presence of disease via measuring risk factors or markers of disease rather than definitive outcomes. Finally, while specific populations such as pregnancy were excluded to increase the generalizability and applicability of this review’s findings, this may have led to the exclusion of relevant studies. In particular, publications involving insulin pumps were excluded because their effects on glycemic management may have altered the development of diabetes complications above that of the effect of CGM alone. While this allowed for better isolation of the effects of CGM on diabetes complications, it unrealistically excludes the effect that therapy may have on markers of glycemic variability. Databases such as Web of Science and gray literature were not searched, which may similarly exclude studies. The chief strength of this review includes the number of studies and broad range of complication outcomes included (Figs. 4 and 5).

28Recent technological advancements in CGM present an exciting prospect for the future of cost-effective and equitable diabetes management. This literature review aimed to summarize the existing evidence for the direct link between CGM-derived metrics of glycemic management and complications of diabetes. As per the recommendations from the ATTD Congress, TIR was consolidated as an important metric; however, evidence was weaker for CV. While higher glycemic variability and lower TIR tended to be associated with diabetic complications, future research, particularly in the form of longitudinal studies, meta-analyses, and RCTs, are required to better evaluate relationships between these CGM-derived metrics and all diabetes complications, especially in T1D. Future studies should also consider the effect of closed-loop pump therapy on the development of diabetes complications.