Beyond HbA1c: Comparing Glycemic Variability and Glycemic Indices in Predicting Hypoglycemia in Type 1 and Type 2 Diabetes

Transcription

1 DIABETES TECHNOLOGY & THERAPEUTICS Volume 20, Number 5, 2018 ª Mary Ann Liebert, Inc. DOI: /dia ORIGINAL ARTICLE Beyond HbA1c: Comparing Glycemic Variability and Glycemic Indices in Predicting Hypoglycemia in Type 1 and Type 2 Diabetes Suresh Rama Chandran, MD, MRCP, 1 Wei Lin Tay, MD, MRCP, 1 Weng Kit Lye, MSc, 2 Lee Ling Lim, MRCP, AM, 3 Jeyakantha Ratnasingam, MD, MMed, 3 Alexander Tong Boon Tan, MB ChB, FRCP, 3 and Daphne S.L. Gardner, BA, BMBCh (Oxon), FRCP 1 Abstract Background: Hypoglycemia is the major impediment to therapy intensification in diabetes. Although higher individualized HbA1c targets are perceived to reduce the risk of hypoglycemia in those at risk of hypoglycemia, HbA1c itself is a poor predictor of hypoglycemia. We assessed the use of glycemic variability (GV) and glycemic indices as independent predictors of hypoglycemia. Methods: A retrospective observational study of 60 type 1 and 100 type 2 diabetes subjects. All underwent professional continuous glucose monitoring (CGM) for 3 6 days and recorded self-monitored blood glucose (SMBG). Indices were calculated from both CGM and SMBG. Statistical analyses included regression and area under receiver operator curve (AUC) analyses. Results: Hypoglycemia frequency (53.3% vs. 24%, P < 0.05) and %CV (40.1% 10% vs. 29.4% 7.8%, P < 0.001) were significantly higher in type 1 diabetes compared with type 2 diabetes. HbA1c was, at best, a weak predictor of hypoglycemia. %CV CGM, Low Blood Glucose Index (LBGI) CGM, Glycemic Risk Assessment Diabetes Equation (GRADE)Hypoglycemia CGM, and Hypoglycemia Index CGM predicted hypoglycemia well. %CV CGM and %CV SMBG consistently remained a robust discriminator of hypoglycemia in type 1 diabetes (AUC 0.88). In type 2 diabetes, a combination of HbA1c and %CV SMBG or LBGI SMBG could help discriminate hypoglycemia. Conclusion: Assessment of glycemia should go beyond HbA1c and incorporate measures of GV and glycemic indices. %CV SMBG in type 1 diabetes and LBGI SMBG or a combination of HbA1c and %CV SMBG in type 2 diabetes discriminated hypoglycemia well. In defining hypoglycemia risk using GV and glycemic indices, diabetes subtypes and data source (CGM vs. SMBG) must be considered. Keywords: Hypoglycemia, Glycemic variability, Coefficient of variation, Continuous glucose monitoring, Self-monitored blood glucose. Introduction Hypoglycemia is the major impediment to the intensification of therapy in both type 1 and type 2 diabetes. Hypoglycemia is unpleasant, reduces productivity, and increases health expenditure. 1 More importantly, hypoglycemia is associated with significant morbidity 2 and increased mortality. 3 The self-reported prevalence of at least one episode of hypoglycemia over a 4-week period among people with type 1 and insulin-treated type 2 diabetes was 83% and 46.4%, respectively. 4 A hypoglycemic seizure or coma occurred in 11.8% of people with type 1 diabetes in the preceding year. 5 While treating hyperglycemia clearly reduces the risk of micro- and macrovascular complications in both type 1 and type 1 Department of Endocrinology, Singapore General Hospital, Singapore. 2 Centre for Quantitative Medicine, Office of Clinical Sciences, Duke-NUS Medical School, Singapore. 3 Division of Endocrinology, Department of Internal Medicine, University of Malaya, Kuala Lumpur, Malaysia. Conference Presentations: 1. Oral Presentation at EASD 2017, September 14, 2017, Lisbon, Portugal. 2. Poster Presentation at ADA 2017, June 10, 2017, San Diego, California, USA. 3. Oral Presentation at SGH 22nd Annual Scientific Meeting, April 07 08, 2017, Singapore. 1

2 2 RAMA CHANDRAN ET AL. 2 diabetes, 6 9 the risk of hypoglycemia increases as HbA1c approaches target levels. 6 First observed in the Diabetes Control and Complications Trial (DCCT) in those with type 1 diabetes, the finding is similar in those with type 2 diabetes, with rates of severe hypoglycemia doubling among intensively treated type 2 diabetes Guidelines have consequently proposed individualization of HbA1c targets, advocating higher HbA1c targets for those at higher risk of hypoglycemia. 13 While HbA1c remains widely used as a measure of mean glycemia, it may not be the best marker for predicting hypoglycemia. HbA1c alone explained a very small proportion of hypoglycemia risk among subjects in the DCCT. 14 Even after adjusting for HbA1c levels, intensively treated individuals still had a higher risk of hypoglycemia, implying that factors other than HbA1c played a major contributory role. 15 Among individuals with type 2 diabetes, observational studies demonstrate an inconsistent relationship between HbA1c and hypoglycemia rates, with some suggesting an inverse relationship 16 while others finding higher rates of hypoglycemia at higher HbA1c. 17 A post hoc analysis of the Action to Control Cardiovascular Risk in Diabetes (ACCORD) study showed that the risk of severe hypoglycemia was higher among those with poor glycemic control. 18 Importantly, those who experience severe hypoglycemia have significantly higher rates of death. 3 Therefore, relying on just HbA1c as a marker for glycemia fails to take into account hypoglycemia risk and its potential antecedent role in mortality. In contrast, the variability in glucose levels shows great promise as predictors of hypoglycemia Severe hypoglycemia episodes in type 1 diabetes subjects appear to be preceded by glucose variability. 22 DCCT data also demonstrated that standard deviation (SD) of capillary glucose as a measure of glycemic variability (GV) predicted hypoglycemia in addition to, and independent of, HbA1c. 14 Subsequently, SD was shown to be more predictive of hypoglycemia compared to mean glucose in both type 1 and type 2 diabetes. 23 To date, more than two dozen indices of glucose variability have been defined and validated, 24 each with its own advantages and disadvantages. More recently, Zinman et al. showed that dayto-day variability in fasting glycemia alone appeared to be associated with higher rates of hypoglycemia, severe hypoglycemia, and cardiovascular events. 25 However, this variability measure was derived from just three fasting glucose values in 1 week each month, and the data were part of a post hoc analysis in a trial done for another purpose, potentially implicating the role of other confounders in the observed relationships. 26 Given that the blood glucose readings have a non-normal distribution and are skewed toward hyperglycemia, numerical transformations to the blood glucose scale have been performed to account for this in trying to define glucose variability. When glucose values are transformed into a nearly Gaussian distribution, the probability of glucose values below or above any arbitrary threshold can be calculated. 27 This has led to hypoglycemia-specific indices such as the Low Blood Glucose Index (LBGI), 28,29 GRADE% Hypoglycemia, 30 and Hypoglycemia Index, 31 all of which were designed to focus on hypoglycemia and not hyperglycemia. LBGI has consequently been shown to be predictive of severe hypoglycemic episodes. 28,32 Composite glycemic indices have also been proposed, including the Index of Glycemic Control (IGC), 24 Blood Glucose Risk Index (BGRI), 33 Glycemic Risk Assessment Diabetes Equation (GRADE) score, 30 Q-score, 34 Personal Glycemic Score (PGS), 35 Comprehensive Glucose Pentagon, 36 and the M-value, 37 all in attempts to defining glycemic status and excursions beyond HbA1c. While useful for tracking outcomes in research studies, complex indices of GV and glycemic scores remain difficult to retrieve and comprehend and this may limit widespread use in day-to-day clinical practice. Unlike most other indices of GV, coefficient of variation (CV) of glucose is an easy measure that can be used in the clinical setting to allow an overview of an individual s glycemia. CV is easily computed and derived by dividing the SD of glucose by the mean glucose and then expressed as a percentage (%CV). CV therefore incorporates both low and high blood glucose fluctuations to describe glucose fluctuations relative to, and independent of, mean glycemia. 38 This is in contrast to SD, which captures absolute GV, and is linearly correlated with mean glucose and HbA1c. 39 In addition, CV is independent of the length of exposure, 40 which is another important feature of GV. While acknowledging the qualities that promote the use of CV as an index of GV, the greater importance lies in how well it predicts hypoglycemia. Previous studies have found a substantial relationship between higher GV expressed as CV and increased hypoglycemia episodes, using continuous glucose monitoring (CGM) systems 27,41 and self-monitoring of capillary blood glucose, 42 in both type 1 and type 2 diabetes. 41,42 It has been suggested as a superior predictor of hypoglycemia to indices such as SD, mean amplitude of glycemic excursion (MAGE), and average daily risk range. 43 The aim of this study was to assess the use of GV (particularly %CV of glucose) and other hypoglycemia-specific glycemic indices (LBGI, GRADEHypoglycemia, and Hypoglycemia Index), and compare their relative performance in predicting hypoglycemia among subjects with both type 1 and type 2 diabetes. Furthermore, we aimed to evaluate the utility of GV and other glycemic indices derived from selfmonitoring of blood glucose (SMBG) as a marker for hypoglycemia. Delineating a threshold for these indices above which individuals at higher risk of hypoglycemia could be identified would then enable targeted intervention toward hypoglycemia-prevention strategies. Subjects, Materials, and Methods This cross-sectional study included a total of 160 subjects who underwent professional CGM (ipro 2 Ò ; Medtronic) for 3 6 days. Altogether, data from 60 type 1 diabetes subjects from Singapore General Hospital ( ), Singapore, and 100 type 2 diabetes subjects 44 from the University of Malaya, Malaysia, were included. All subjects were attending diabetes outpatient clinics in both settings. This study was approved by the respective Institutional Review Boards of both hospitals. Subjects were instructed to perform SMBG at least twice a day (12 h apart) to allow calibration of sensor glucose to capillary blood glucose levels. Demographic, anthropometric, and treatment details and HbA1c at the time of CGM study were retrieved. Raw data from CGM records were entered into the software Easy GV ( 45 to generate various indices. The first 12 h

3 GLYCEMIC VARIABILITY AND GLYCEMIC INDICES IN PREDICTING HYPOGLYCEMIA 3 of CGM data were excluded from analysis to allow time for a stable calibration factor to be generated and the tissue injury response to abate. SMBG data in the first 12 h were also excluded from analysis for consistency. Using EasyGV, GV and glycemic indices, including SD, CV, MAGE, continuous overall net glycemic action (CONGA1), as well as high blood glucose index (HBGI), LBGI, and mean glucose, were calculated. Hypoglycemia-specific glycemic indices Other indices of hypoglycemia, apart from LBGI, calculated included the GRADEHypoglycemia and the Hypoglycemia Index using formulae described by Hill et al. 30 and Rodbard. 31 GRADE is a method of quantifying glycemic profiles into a clinical risk score. 30 GRADEHypoglycemia CGM was calculated as the sum of all GRADE scores derived from glucose readings less than 70 mg/dl (3.9 mmol/l). GRADE- Hypoglycemia was chosen over GRADE%Hypoglycemia (GRADEHypoglycemia/Sum of all GRADE scores 100) as the former had equal or higher correlations with all measures of hypoglycemia in this study whether derived from CGM or SMBG (Supplementary Tables S8 and S9; Supplementary Data are available at /dia ). The higher correlation was likely due to the noninterference of GRADEHypoglycemia by GRA- DEEuglycemia and GRADEHyperglycemia. Hypoglycemia Index is the weighted average of hypoglycemia values, adjusted for mild, moderate, and severe degrees of hypoglycemia. 24 Hypoglycemia Index CGM was calculated using an LLTR (lower limit of target range) of 54 mg/dl (3 mmol/l) for consistency with the definition of hypoglycemia on CGM. 31 Using an LLTR of 80 mg/dl (4.4 mmol/l), as described in the original article, 31 resulted in lower correlations with measures of hypoglycemia (Supplementary Table S10). %CV SMBG was calculated by dividing SD of SMBG data by the mean of SMBG data and expressed as a percentage. LBGI SMBG was calculated using EasyGV. 45 GRADEHypoglycemia SMBG and Hypoglycemia Index SMBG were calculated using Excel spreadsheets with the same formulae for the corresponding CGM-derived indices. The LLTR 31 used for calculation of Hypoglycemia Index SMBG was 70 mg/dl (3.9 mmol/l). The use of 80 mg/dl (4.4 mmol/l) similarly resulted in lower correlations with measures of hypoglycemia (Supplementary Table S11). Definition of hypoglycemia and measures of hypoglycemia An episode of hypoglycemia on CGM was defined as a sensor glucose <54 mg/dl (3 mmol/l) lasting for at least 20 min (no maximum duration) and preceded by a sensor glucose 54 mg/dl (3 mmol/l) for at least 20 min. 46 The frequency of hypoglycemic episodes was standardized and reported per 24 h of CGM. The frequency of nocturnal hypoglycemic episodes was reported as episodes per 8 h ( h) of nighttime monitoring. Duration of hypoglycemic episodes was reported as minutes per episode. The duration of hypoglycemia per 24 h and nocturnal hypoglycemia per 8 h was expressed as percentages. Thus, the six measures of hypoglycemia derived from CGM were the episodes/24 h, nocturnal episodes/8 h nighttime, duration per episode (min), duration per nocturnal episode (min), overall hypoglycemia duration/24 h (%), and nocturnal hypoglycemia duration/8 h (%). Statistics Continuous variables are expressed as mean and SD. Categorical variables are expressed as count and percentage. Comparisons of means of measures of hypoglycemia were done using the two-sample Mann Whitney U test. Spearman s correlation was used to assess the correlation between the GV indices and measures of hypoglycemia. Multiple linear regression models were generated to predict each measure of hypoglycemia separately. HbA1c and %CV were included as independent variables in all the models due to their accepted relevance. Independent variables were assessed by univariate regression with each measure of hypoglycemia and only variables with a P-value 0.2 were included in the model. Other GV indices tested in separate multiple linear regression models, in place of %CV, included the LBGI, GRADEHypoglycemia, and Hypoglycemia Index. Logistic regression analyses were then performed to predict the dichotomous outcome variable hypoglycemia (absent/ present). The area under the receiver operating characteristic curve (AUC) was used to assess the diagnostic utility of %CV and other indices for hypoglycemia. A two-sided P-value <0.05 was considered statistically significant. Results The baseline characteristics of both groups are shown in Table 1. Type 1 diabetes subjects were younger (34 13 years vs years), leaner (body mass index [BMI] kg/m 2 vs kg/m 2 ), had a lower HbA1c (8.0% 1.6% vs. 8.6% 1.7%),andusedalowertotaldaily dose of insulin ( U/kg vs U/kg). Fifty-eight Table 1. Baseline Characteristics Characteristics Type 1 Type 2 P Age (years) <0.001 Sex, n (%) Male 31 (51.7) 54 (54) NS Female 29 (48.3) 46 (46) NS Race, n (%) Malays 5 (8.3) 34 (34) <0.05 Indian 15 (25) 34 (34) NS Chinese 37 (61.7) 28 (28) <0.05 Others 3 (5) 4 (4) NS BMI (kg/m 2 ) <0.001 Diabetes duration NS (years) HbA1c (%) [mmol/mol] TDD of insulin (IU/kg) <0.001 Sulfonylurea or 94 (94%) insulin use Basal bolus insulin 60 (100%) 37 (37%) Basal only 10 (10%) Premixed insulin 9 (9%) Prandial insulin only 1 (1%) BMI, body mass index; TDD, total daily dose.

4 4 RAMA CHANDRAN ET AL. percent of the subjects with type 2 diabetes were using insulin therapy and 94% were on either sulfonylureas or insulin. The differences in ethnicity distribution between the type 1 and type 2 groups were reflective of the different country of recruitment. Hypoglycemia as defined from CGM ( 54 mg/dl [ 3 mmol/l]) Twice as many subjects with type 1 diabetes had at least one hypoglycemic episode (53.3% vs. 24%) and at least one episode of nocturnal hypoglycemia (33% vs. 17%) during the study period compared with those with type 2 diabetes (Table 2). Each episode of hypoglycemia was of longer duration in type 1 diabetes than in type 2 diabetes subjects (mean 39.1 min vs min). Consequently, type 1 subjects spent a significantly larger percentage of time in hypoglycemia (2.75% vs. 0.55%). Approximately one in four type 2 diabetes subjects had a hypoglycemic episode within just a few days of monitoring, suggesting that the incidence of hypoglycemia is not trivial. Mean %CV CGM was significantly higher among type 1 diabetes subjects compared with those with type 2 diabetes (%CV CGM = 40.1% 10% vs. 29.4% 7.8%, respectively; P < 0.001). HbA1c and hypoglycemia In the group as a whole, HbA1c had no relationship with measures of nocturnal hypoglycemia and only a weak relationship with episodes/24 h, duration per episode, and overall duration/24 h (%) (r =-0.20, -0.17, -0.21, respectively, P < 0.05) (Supplementary Table S1). However, this was largely driven by the weak, mostly nonsignificant relationship between HbA1c and measures of hypoglycemia in type 2 diabetes subjects. When the groups were analyzed separately, there was no relationship between HbA1c and any measure of hypoglycemia in type 1 diabetes subjects and only a minimal relationship between HbA1c and overall hypoglycemia duration/24 h (%) in type 2 diabetes (r =-0.223, P = 0.03; Supplementary Table S1). Table 2. Hypoglycemia in Type 1 and Type 2 Diabetes Measures of hypoglycemia Type 1 Type 2 P Hypoglycemia 1 episode during 32 (53.3%) 24 (24.0%) <0.001 study Episodes/24 h <0.001 Duration/episode <0.001 (min) Duration/24 h (%) <0.001 Nocturnal (0000:0800) hypoglycemia 1 nocturnal episode during study 20 (33.3%) 17 (17%) Nocturnal episodes/8 h Duration/nocturnal episode (min) Duration/8 h (%) GV and glycemic indices derived from CGM and measures of hypoglycemia The relationship between glucose variability and glycemic indices, and measures of hypoglycemia was examined using Spearman s correlation (Supplementary Tables S2 S4). %CV CGM, LBGI CGM, GRADEHypoglycemia CGM, and Hypoglycemia Index CGM showed a significant and consistent correlation with all the measures of hypoglycemia in both type 1 and type 2 diabetes. Hypoglycemia Index CGM had the highest correlations followed by GRADEHypoglycemia CGM, LBGI CGM, and %CV CGM (Supplementary Tables S2 S4). When the group was divided into tertiles of HbA1c, the relationships between these GV indices and measures of hypoglycemia were retained across each HbA1c tertile, in both type 1 and type 2 diabetes (data not shown). Among type 2 subjects, these relationships were retained whether the subjects were on insulin or oral drugs. Only six type 2 diabetes subjects were on non-sulfonylurea drugs; further analysis in this group was not feasible. Given that the GRADEHypoglycemia, Hypoglycemia Index, and LBGI are intrinsically derived from low blood glucose readings, it was unsurprising that these correlated well with measures of hypoglycemia. In type 1 diabetes, multiple linear regression analyses demonstrated that %CV CGM, LBGI CGM, GRADEHypoglycemia CGM, and Hypoglycemia Index CGM were each significant predictors of all measures of hypoglycemia (only models for %CV CGM are shown in Table 3; model r 2 for %CV CGM : , LBGI CGM : , GRADEHypoglycemia CGM : , Hypoglycemia Index CGM : , all P < 0.01). In contrast, HbA1c was not a significant predictor. In type 2 diabetes, HbA1c contributed only weakly to the prediction of the frequency of hypoglycemia, while %CV CGM and the three glycemic indices were individually the strongest predictors of all indices of hypoglycemia independent of age, gender, race, BMI, diabetes duration, sulfonylurea/insulin use, and total daily dose of insulin (only model for %CV CGM is shown in Table 3; model r 2 for %CV: , LBGI CGM : , GRADEHypoglycemia CGM : , Hypoglycemia Index CGM : , all P < 0.01). A similar picture was seen in logistic regression analysis. For example, in type 1 diabetes, only %CV CGM was a significant predictor of hypoglycemia (odds ratio [OR] of 1.38 [confidence interval [95% CI]: , P = 0.002]), independent of the above covariates, while HbA1c was not. Among subjects with type 2 diabetes, both %CV CGM and HbA1c were significant predictors for an episode of hypoglycemia (%CV CGM OR = 1.27 [95% CI: , P < 0.001] and HbA1c OR = 0.43 [95% CI: , P = 0.003], respectively). GV and glycemic indices derived from SMBG in predicting hypoglycemia During the study period, type 1 diabetes subjects recorded an average of readings/24 h, while those with type 2 diabetes recorded an average of readings/24 h. At this frequency of blood glucose monitoring, %CV derived from SMBG (%CV SMBG ) correlated strongly with %CV derived from CGM (%CV CGM ) (0.81, P < 0.001). The correlation (r) of LBGI CGM, GRADEHypoglycemia CGM, and Hypoglycemia Index CGM with their corresponding SMBG-derived

5 GLYCEMIC VARIABILITY AND GLYCEMIC INDICES IN PREDICTING HYPOGLYCEMIA 5 Table 3. Multivariate Linear Regression Analysis to Predict Hypoglycemia in Type 1 and Type 2 Diabetes Episodes/24 h Duration/episode Nocturnal episodes/8 h Duration/nocturnal episode B (95% CI) b P B (95% CI) b P B (95% CI) b P B (95% CI) b P Independent variables Type 1 diabetes CV% 0.04 ( ) 0.73 < ( ) 0.58 < ( ) 0.60 < ( ) 0.49 <0.001 HbA1c (-0.10 to 0.03) ( to 3.11) (-0.05 to 0.04) (-8.62 to 10.60) Age (-1.78 to 0.38) Diabetes duration 0.01 ( ) Gender ( to 5.81) ( to 23.60) TDD ( to 87.21) ( to 96.8) BMI (-8.30 to 1.57) Type 2 diabetes CV% 0.01 ( ) 0.46 < ( ) 0.41 < ( ) 0.43 < ( ) HbA1c (-0.04 to ) (-8.62 to 4.65) (-0.02 to 0.0) (-8.80 to 0.05) Age 0.34 (-0.91 to 1.59) Diabetes duration 0.14 (-1.37 to 1.66) SU use (-0.05 to 0.04) Insulin use 0.03 (-0.02 to 0.09) The models were adjusted for the following independent variables age, gender, race, BMI, diabetes duration, sulfonylurea and insulin use (type 2), and TDD (type 1). Variables with a P-value 0.2 in univariate analysis were entered in the model. CV% and HbA1c were included as independent variables in all the models. B, unstandardized coefficient; b, standardized coefficient; BMI, body mass index; CI, confidence interval; CV, coefficient of variation; SU, sulfonylurea; TDD, total daily dose (U/kg). indices (both type 1 and type 2 diabetes subjects combined) was lower at 0.66, 0.52, and 0.48, respectively (all P < 0.001). In type 1 diabetes subjects, the four CGM-derived indices (%CV CGM, LBGI CGM, GRADEHypoglycemia CGM, and Hypoglycemia Index CGM ) correlated well (r = ) with all measures of hypoglycemia (Supplementary Table S3). However, the use of SMBG-derived indices resulted in a substantial decrease in the correlation of these indices with hypoglycemia (Supplementary Table S6). Only %CV SMBG appeared to remain most closely correlated with all measures of hypoglycemia. Similarly, in type 2 diabetes, the CGM-derived indices of LBGI CGM, GRADEHypoglycemia CGM, and Hypoglycemia Index CGM correlated well with all measures of hypoglycemia (r = , P < 0.001), with %CV CGM having the lowest correlation (r = , P < 0.001) (Supplementary Table S4). However, using the SMBG-derived indices resulted in a large decrease in their correlation with hypoglycemia (r = ) (Supplementary Table S7). Receiver operator curve analysis Using CGM-derived indices, Hypoglycemia Index CGM was the best discriminator of an episode of hypoglycemia detected on CGM in both type 1 and type 2 diabetes (AUC 0.99 for both, Figs. 1 and 2), followed by LBGI CGM and GRADEHypoglycemia CGM, then %CV CGM (AUC 0.88 and 0.80, respectively for type 1 and type 2 diabetes). Using SMBG-derived indices, %CV SMBG had the highest AUC (0.88) to discriminate an episode of hypoglycemia from CGM among type 1 diabetes subjects. It is noteworthy that there was no decrease in AUC for %CV whether derived from CGM or SMBG in type 1 diabetes. The overall AUCs of SMBG-derived indices in discriminating an episode of hypoglycemia on CGM in type 2 diabetes were much lower with LBGI SMBG having the highest AUC (0.75), followed by %CV SMBG (AUC 0.66). Discussion Our findings demonstrate that GV and glycemic indices are better predictors of hypoglycemia than HbA1c. The performance of these indices varies depending on the subtype of diabetes and the data source (CGM vs. SMBG). Hypoglycemia Index CGM was the best predictor of hypoglycemia in both type 1 and type 2 diabetes when derived from CGM. However, when the indices were derived from SMBG, %CV SMBG stood out as the best predictor of hypoglycemia in type 1 diabetes and LBGI SMBG in type 2 diabetes. In contrast, HbA1c had no relationship to hypoglycemia risk in type 1 diabetes, and only a weak relationship in type 2 diabetes. In defining thresholds for GV and glycemic indices to predict hypoglycemia, diabetes subtype and the data source (CGM vs. SMBG) need to be considered. Traditional glycemic targets are HbA1c-centric and are often used to describe the degree of hyperglycemia. The mantra of targeting higher HbA1c targets for those at greater risk of hypoglycemia 13,47 is contradicted by the observation that HbA1c in itself explains only a small portion of the hypoglycemia risk. 15 This is once again demonstrated in this study; no consistent or significant relationship between HbAa1c and hypoglycemia could be demonstrated in type 1 diabetes and only a weak correlation to 24-h hypoglycemia

6 6 RAMA CHANDRAN ET AL. FIG. 1. Comparison of AUC of glycemic variability indices derived from CGM and SMBG in predicting hypoglycemia in CGM among type 1 diabetes subjects. AUC, area under receiver operator curve; CGM, continuous glucose monitoring; SMBG, self-monitored blood glucose. was found in type 2 diabetes. These data would be consistent with previous findings among those with type 1 diabetes 48 and similarly among those with type 2 diabetes, in whom selfreported severe hypoglycemia episodes were found to be common across all HbA1c categories. 49 We caution against taking a simplistic approach of attributing higher hypoglycemia risk to those with lower HbA1c and assuming protection from hypoglycemia by maintaining a higher HbA1c target. In contrast, as GV increases, hypoglycemia risk increases steadily and consistently, similar to findings from a study by Jin et al. 50 We looked at the association between several FIG. 2. Comparison of AUC of glycemic variability indices derived from CGM and SMBG in predicting hypoglycemia in CGM among type 2 diabetes subjects. AUC, area under receiver operator curve; CGM, continuous glucose monitoring; SMBG, self-monitored blood glucose.

7 GLYCEMIC VARIABILITY AND GLYCEMIC INDICES IN PREDICTING HYPOGLYCEMIA 7 indices of glycemia (mean glucose, SD, MAGE, CONGA1, %CV, LBGI GRADEHypoglycemia, and Hypoglycemia Index and HBGI) and measures of hypoglycemia. The performance of %CV and other glycemic indices in predicting hypoglycemia varied according to diabetes subtype and the data source (CGM vs. SMBG). When derived from CGM, Hypoglycemia Index CGM had the highest correlation ( ) with all six measures of hypoglycemia in type 2 diabetes and with half of measures of hypoglycemia in type 1 diabetes. The rest of the measures (of nocturnal hypoglycemia) correlated best with LBGI CGM and GRADEHypoglycemia CGM ( ) in type 1 diabetes (Supplementary Tables S3 and S4). While Hypoglycemia Index CGM appeared to be the best correlated with most indices of hypoglycemia, it is important to note the lower LLTR used (54 mg/dl, <3 mmol/l); a higher LLTR would not have performed as well (Supplementary Table S10). Among SMBG-derived indices, %CV SMBG had the highest correlation ( ) with all six measures of hypoglycemia in type 1 diabetes and LBGI SMBG had the highest correlation ( ) with 4/6 of the measures of hypoglycemia in type 2 diabetes (Supplementary Tables S6 and S7). CGM may not be accessible to everyone, and we therefore questioned if it was possible to screen for those at higher risk of hypoglycemia (diagnosed on CGM) using GV and glycemic indices obtained from SMBG alone. Sensor readings are retrieved every 5 min (288 sensor readings/24 h), while calibration capillary blood readings were performed just times/24 h in our study. The expectation might have been a poor correlation between CGM- and SMBG-derived indices since calibration capillary blood readings typically do not incorporate postprandial and nocturnal readings. Surprisingly, %CV SMBG correlated well with %CV CGM (r = 0.81, type 1 and type 2 combined) and could possibly have been higher had there been postprandial capillary glucose readings. In contrast, LBGI SMBG, GRADEHypoglycemia SMBG, and Hypoglycemia Index SMBG were not well correlated with the corresponding CGM-derived indices (max r = 0.66). These three indices are inherently focused on the hypoglycemia range, unlike %CV. When these indices are derived from SMBG, underrepresenting asymptomatic and nocturnal hypoglycemia, a large decrease in correlation with their corresponding CGM-derived indicesisinevitable. Type 1 diabetes: SMBG-derived indices Given the strong correlation between %CV SMBG and %CV CGM, the ability of %CV to discriminate those with hypoglycemia would be expected to be similar using either of these indices. Indeed, in type 1 diabetes, %CV SBMG and %CV CGM generated equivalent areas under the curve from receiver operator curve (ROC) analysis (AUC = 0.88 and 0.88, respectively), implying similar discriminatory capabilities in predicting hypoglycemia. As expected from the poorer correlation between the SMBG-derived and CGMderived indices of LBGI, GRADEHypoglycemia, and Hypoglycemia Index, the corresponding SMBG-derived indices had lower AUCs ( ) compared with CGM-derived indices ( ) (Fig. 1). In fact, %CV SMBG became the best discriminator of hypoglycemia among these four indices (AUC 0.88). Type 2 diabetes: SMBG-derived indices In type 2 diabetes, the SMBG-derived indices were generally less well correlated with their CGM counterparts, poorly correlated with all measures of hypoglycemia (Supplementary Table S7), and consequently did not perform as well in discriminating those with hypoglycemia (AUC ) (Fig. 2). As discussed earlier, SMBG usually fails to detect asymptomatic and nocturnal hypoglycemia, which CGM captures, accounting for the poorer correlation for indices that focus on hypoglycemia. Despite %CV SMBG being more closely correlated with %CV CGM (r = 0.76), it did not perform as well as %CV CGM in discriminating hypoglycemia (AUC = 0.66 vs. 0.80). This could be the result of fewer SMBG readings among type 2 subjects, resulting in sparse coverage of the full 24 h and greater random variability in the estimate of SD and hence %CV. However, it is more likely related to the overall lower %CV in type 2 diabetes than in type 1 diabetes (%CV = 29.4% 7.7% vs. 40.1% 10%, respectively, P < 0.001). Using an HbA1c level of <10.1% and a %CV SMBG of >33% provides a sensitivity of 61% and specificity of 79% for detecting hypoglycemia. LBGI SMBG had the highest AUC in discriminating hypoglycemia on CGM in type 2 diabetes. Using an LBGI SMBG of 1.0 predicted hypoglycemia on CGM with 83% sensitivity and 62% specificity. Applying a combination of HbA1c <10.1% and an LBGI SMBG 1.0 did not improve the performance of LBGI SMBG any further (sensitivity 87%, specificity 59%). In type 2 diabetes, a combination of the two predictor variables (%CV and HbA1c) or LBGI SMBG alone might be helpful. These suggested cutoffs in both diabetes subtypes will need to be validated in prospective studies. Clinical utility of glycemic indices derived from SMBG Although most glycemic indices may be calculated and incorporated into CGM and SMBG software for easy reporting, the clinical applicability of indices such as LBGI SMBG, GRADEHypoglycemia SMBG, and Hypoglycemia Index SMBG needs to be questioned since these are derived from glucose readings in the hypoglycemic range and the SMBG readings used in the calculation of these indices would in themselves have already demonstrated hypoglycemia to the clinician. It would be helpful to the clinician for an SMBG-derived index to predict asymptomatic or nocturnal hypoglycemia in the absence of obvious hypoglycemia on SMBG. For this reason, although LBGI SMBG appeared to perform better than %CV SMBG in discriminating type 2 diabetes subjects with hypoglycemia, we chose to focus on %CV since it encompasses both high and low glucose levels. Even in excluding those with SMBG <3.5 mmol/l (obvious hypoglycemia on SMBG), only %CV SMBG remained significantly higher among those who had asymptomatic/nocturnal hypoglycemia detected on CGM in type 1 diabetes, and tended to significance in type 2 diabetes (%CV SMBG hypoglycemia vs. no hypoglycemia on CGM: type 1 diabetes: 45.9% 7.1% vs. 34.4% 6.2%, P = 0.001, type 2 diabetes: 32.7% 10% vs. 28.7% 8.1%, P = 0.08). This implies that even in those without low glucose readings (<3.5 mmol/l) on SMBG, a higher %CV SMBG indicated a higher propensity to hypoglycemia. Indeed, just a 1% increase in %CV increased the risk of hypoglycemia in type 1 diabetes by 38% (OR 1.38,

8 8 RAMA CHANDRAN ET AL. 95% CI: ) and in type 2 diabetes by 26.9% (OR: 1.27, 95% CI: ). The need for different thresholds for diabetes subtypes The different %CV thresholds for each diabetes subtype proposed herein differ from a previous study by Monnier et al. 41 in which a single %CV CGM threshold of 36% was proposed to predict hypoglycemia in both diabetes subtypes. Within our type 1 diabetes group, a %CV CGM threshold of >36% would have given 81% sensitivity and 72% specificity to detect hypoglycemia. Using the Youden s J statistic, a %CV CGM of more than 41% (sensitivity 72% and specificity 96%) and %CV SMBG of more than 44% (sensitivity 81.3% and specificity 89%) provided the best discrimination of subjects with hypoglycemia among type 1 diabetes in our cohort. However, using the same threshold as suggested by Monnier et al. for our type 2 diabetes group, just 12/24 (50%) type 2 individuals with hypoglycemia would have been detected. Despite a higher proportion of sulfonylurea/insulin users in our study (94% vs. 57% [groups 2b and 3 in Monnier et al. 41 ]) and higher median %CV in the type 2 diabetes subjects in our study compared with Monnier et al., a lower %CV threshold was still found to be optimal for discriminating hypoglycemia among type 2 diabetes subjects. Even in using the equivalent %CV CGM (rather than %CV SMBG ) threshold, the optimal %CV CGM threshold was still lower in the type 2 diabetes group at 27%. Jin et al. similarly suggested using a different (lower) cutoff of %CV for detection of hypoglycemia among type 2 diabetes subjects compared with type 1 subjects, 50 related to the intrinsic differences in mean %CV between the diabetes subtypes. We reinforce the need for different %CV cutoffs for the two diabetes subtypes. Interventions to reduce GV It would appear that measures of GV or glycemic indices derived from SMBG alone could be used to identify those with higher glucose variability and at greater risk of hypoglycemia. %CV SMBG in type 1 subjects, and a combination of HbA1c and %CV SMBG or LBGI SMBG in type 2 subjects could be used to select those at higher risk of hypoglycemia to undergo CGM to detect asymptomatic hypoglycemia and intervene to reduce GV and hypoglycemia. Interventions to reduce GV should include education on structured blood glucose testing, individualizing blood glucose targets, empowering self-management through education of carbohydrate counting with insulin dose adjustments, and greater precision in insulin dose delivery with the use of pump therapy and/or sensor-augmented pump therapy. 51 In type 2 diabetes, in addition to education on the use of sulfonylureas, insulin, and meal timings, greater consideration could be given toward the use of therapies associated with reduced GV and hypoglycemia risk such as GLP-1 analogs, DPP-IV inhibitors, SGLT2 inhibitors, and basal insulin Study strengths The present analyses have several strengths, including the comparison of four indices derived from both CGM and SMBG against six measures of hypoglycemia in both type 1 and type 2 diabetes populations. The type 2 diabetes group consisted of not just those on intensive insulin regimens but also on oral glucose-lowering agents, allowing for greater applicability. Unlike in earlier studies 25,55 in which few SMBG readings per month were used to calculate GV, in this study, there was an average of 4 6 SMBG readings/day over the CGM period, enabling a 24-h representation of GV, which was expressed as %CV. Other strengths include the comparison of GRADEHypoglycemia and GRADE%Hypoglycemia, as well as use of multiple parameters for Hypoglycemia Index. Study limitations This study is limited in its retrospective design and nonrandom selection of study subjects. Type 1 diabetes subjects underwent CGM for clinical indications, of which asymptomatic hypoglycemia was one, and hence more likely to be at higher risk of hypoglycemia. Type 2 diabetes subjects underwent nonrandom consecutive recruitment for the purposes of another study. 44 The findings in this study, including the thresholds identified, need to be borne out in a prospective, randomly selected population. The use of sulfonylurea is generally high among Asian populations with type 2 diabetes. 56 The applicability of these findings in other type 2 diabetes populations needs validation, particularly considering the increasing use of newer oral glucose-lowering agents with reduced hypoglycemia risk. A third limitation would be the reliability of the CGM readings, particularly in the hypoglycemic range. However, we sought to minimize these inaccuracies by excluding the first 12 h of CGM data and defining episodes of hypoglycemia using stringent criteria. Conclusions GV and glycemic indices are better predictors of hypoglycemia but their performance varies according to diabetes subtype and data source. HbA1c contributes minimally to hypoglycemia risk in type 2 diabetes and has no relation to hypoglycemia in type 1 diabetes. The assessment of glycemic status should go beyond HbA1c to incorporate GV and glycemic indices to define stable and good glycemic control. Measures of GV such as %CV or LBGI, even from just SMBG readings, could help screen individuals at higher risk of hypoglycemia, allowing stratification toward targeted interventions to reduce hypoglycemia risk. Author Disclosure Statement No competing financial interests exist. References 1. Goh S-Y, Abusnana S, Emral R, et al.: Health economic impact of hypoglycemia among 7,289 insulin-treated patients with diabetes: results from an International survey in 9 countries. Diabetes Res Clin Pract 2016;120:S Frier BM: Hypoglycaemia in diabetes mellitus: epidemiology and clinical implications. Nat Rev Endocrinol 2014; 10: Bonds DE, Miller ME, Bergenstal RM, et al.: The association between symptomatic, severe hypoglycaemia and mortality in type 2 diabetes: retrospective epidemiological analysis of the ACCORD study. BMJ 2010;340:b Khunti K, Alsifri S, Aronson R, et al.: Rates and predictors of hypoglycaemia in people from 24 countries with

9 GLYCEMIC VARIABILITY AND GLYCEMIC INDICES IN PREDICTING HYPOGLYCEMIA 9 insulin-treated type 1 and type 2 diabetes: the global HAT study. Diabetes Obes Metab 2016;18: Weinstock RS, Xing D, Maahs DM, et al.: Severe hypoglycemia and diabetic ketoacidosis in adults with type 1 diabetes: results from the T1D Exchange clinic registry. J Clin Endocrinol Metab 2013;98: Diabetes Control and Complications Trial Research Group, Nathan DM, Genuth S, et al.: The effect of intensive treatment of diabetes on the development and progression of long-term complications in insulin-dependent diabetes mellitus. N Engl J Med 1993;329: UK Prospective Diabetes Study (UKPDS) Group: Intensive blood-glucose control with sulphonylureas or insulin compared with conventional treatment and risk of complications in patients with type 2 diabetes (UKPDS 33). Lancet Lond Engl 1998;352: Nathan DM, Cleary PA, Backlund J-YC, et al.: Intensive diabetes treatment and cardiovascular disease in patients with type 1 diabetes. N Engl J Med 2005;353: Holman RR, Paul SK, Bethel MA, et al.: 10-year follow-up of intensive glucose control in type 2 diabetes. N Engl J Med 2008;359: Action to Control Cardiovascular Risk in Diabetes Study Group, Gerstein HC, Miller ME, et al.: Effects of intensive glucose lowering in type 2 diabetes. N Engl J Med 2008; 358: Duckworth W, Abraira C, Moritz T, et al.: Glucose control and vascular complications in veterans with type 2 diabetes. N Engl J Med 2009;360: ADVANCE Collaborative Group, Patel A, MacMahon S, et al.: Intensive blood glucose control and vascular outcomes in patients with type 2 diabetes. N Engl J Med 2008;358: American Diabetes Association: 6. Glycemic Targets. Diabetes Care 2017;40 (Supplement 1):S48 S Kilpatrick ES, Rigby AS, Goode K, et al. Relating mean blood glucose and glucose variability to the risk of multiple episodes of hypoglycaemia in type 1 diabetes. Diabetologia 2007;50: The Diabetes Control and Complications Trial Research Group: Hypoglycemia in the diabetes control and complications trial. Diabetes 1997;46: Miller CD, Phillips LS, Ziemer DC, et al.: Hypoglycemia in patients with type 2 diabetes mellitus. Arch Intern Med 2001;161: McCoy RG, Van Houten HK, Ziegenfuss JY, et al.: Increased mortality of patients with diabetes reporting severe hypoglycemia. Diabetes Care 2012;35: Miller ME, Bonds DE, Gerstein HC, et al.: The effects of baseline characteristics, glycaemia treatment approach, and glycated haemoglobin concentration on the risk of severe hypoglycaemia: post hoc epidemiological analysis of the ACCORD study. BMJ 2010;340:b Perea V, Amor A, Giménez M, et al.: Glycemic variability measures in a group of subjects with type 1 diabetes and repeated severe and non-severe hypoglycemia. J Diabetes Sci Technol 2013;7: Cox DJ, Kovatchev BP, Julian DM, et al.: Frequency of severe hypoglycemia in insulin-dependent diabetes mellitus can be predicted from self-monitoring blood glucose data. J Clin Endocrinol Metab 1994;79: Siegelaar SE, Holleman F, Hoekstra JBL, et al.: Glucose variability; does it matter? Endocr Rev 2010;31: Kovatchev BP, Cox DJ, Farhy LS, et al.: Episodes of severe hypoglycemia in type 1 diabetes are preceded and followed within 48 hours by measurable disturbances in blood glucose. J Clin Endocrinol Metab 2000;85: Saisho Y, Tanaka C, Tanaka K, et al.: Relationships among different glycemic variability indices obtained by continuous glucose monitoring. Prim Care Diabetes 2015;9: Rodbard D: Interpretation of continuous glucose monitoring data: glycemic variability and quality of glycemic control. Diabetes Technol Ther 2009;11 Suppl 1:S55 S Zinman B, Marso SP, Poulter NR, et al.: Day-to-day fasting glycaemic variability in DEVOTE: associations with severe hypoglycaemia and cardiovascular outcomes (DEVOTE 2). Diabetologia 2018;61: Rutter MK: Devoting attention to glucose variability and hypoglycaemia in type 2 diabetes. Diabetologia 2018;61: Rodbard D: Hypo- and hyperglycemia in relation to the mean, standard deviation, coefficient of variation, and nature of the glucose distribution. Diabetes Technol Ther 2012;14: Kovatchev BP, Cox DJ, Gonder-Frederick LA, et al.: Assessment of risk for severe hypoglycemia among adults with IDDM: validation of the low blood glucose index. Diabetes Care 1998;21: Kovatchev BP, Straume M, Cox DJ, et al.: Risk analysis of blood glucose data: a quantitative approach to optimizing the control of insulin dependent diabetes. J Theor Med 2000;3: Hill NR, Hindmarsh PC, Stevens RJ, et al.: A method for assessing quality of control from glucose profiles. Diabet Med J Br Diabet Assoc 2007;24: Rodbard D: A semilogarithmic scale for glucose provides a balanced view of hyperglycemia and hypoglycemia. J Diabetes Sci Technol 2009;3: Cox DJ, Gonder-Frederick L, Ritterband L, et al.: Prediction of severe hypoglycemia. Diabetes Care 2007;30: Clarke W, Kovatchev B: Statistical tools to analyze continuous glucose monitor data. Diabetes Technol Ther 2009; 11 Suppl 1:S45 S Augstein P, Heinke P, Vogt L, et al.: Q-Score: development of a new metric for continuous glucose monitoring that enables stratification of antihyperglycaemic therapies. BMC Endocr Disord 2015;15: Hirsch IB, Balo AK, Sayer K, et al.: A simple composite metric for the assessment of glycemic status from continuous glucose monitoring data: implications for clinical practice and the artificial pancreas. Diabetes Technol Ther 2017;19(S3):S38 S Vigersky RA, Shin J, Jiang B, et al.: The comprehensive glucose pentagon: a glucose-centric composite metric for assessing glycemic control in persons with diabetes. J Diabetes Sci Technol 2018;12: Schlichtkrull J, Munck O, Jersild M: The M-valve, an index of blood-sugar control in diabetics. Acta Med Scand 1965; 177: Le Floch J-P, Kessler L: Glucose variability: comparison of different indices during continuous glucose monitoring in diabetic patients. J Diabetes Sci Technol 2016;10: Rodbard D: Clinical interpretation of indices of quality of glycemic control and glycemic variability. Postgrad Med 2011;123: Service FJ: Glucose variability. Diabetes 2013;62:

10 10 RAMA CHANDRAN ET AL. 41. Monnier L, Colette C, Wojtusciszyn A, et al.: Toward defining the threshold between low and high glucose variability in diabetes. Diabetes Care 2017;40: Qu Y, Jacober SJ, Zhang Q, et al.: Rate of hypoglycemia in insulin-treated patients with type 2 diabetes can be predicted from glycemic variability data. Diabetes Technol Ther 2012;14: DeVries JH: Glucose variability: where it is important and how to measure it. Diabetes 2013;62: Lim LL, Brnabic AJ, Chan SP, et al.: Relationship of glycated hemoglobin, and fasting and postprandial hyperglycemia in type 2 diabetes mellitus patients in Malaysia. J Diabetes Investig 2017;8: EasyGV University of Oxford, Nuffield Department of Primary Care Health Sciences. [cited 2017 May 22]. Available from: (accessed January 18, 2018). 46. International Hypoglycaemia Study Group: Glucose concentrations of less than 3.0 mmol/l (54 mg/dl) should be reported in clinical trials: a joint position statement of the American Diabetes Association and the European Association for the Study of Diabetes. Diabetes Care 2017;40: Chiang JL, Kirkman MS, Laffel LMB, et al.: Type 1 diabetes through the life span: a position statement of the American Diabetes Association. Diabetes Care 2014;37: ter Braak EW, Appelman AM, van de Laak M, et al.: Clinical characteristics of type 1 diabetic patients with and without severe hypoglycemia. Diabetes Care 2000;23: Lipska KJ, Warton EM, Huang ES, et al.: HbA1c and risk of severe hypoglycemia in type 2 diabetes: the Diabetes and Aging Study. Diabetes Care 2013;36: Jin S-M, Kim T-H, Bae JC, et al.: Clinical factors associated with absolute and relative measures of glycemic variability determined by continuous glucose monitoring: an analysis of 480 subjects. Diabetes Res Clin Pract 2014;104: Tan HK, Little SA, Leelarathna L, et al.: Low blood glucose avoidance training improves glycemic variability in adults with type 1 diabetes complicated by impaired awareness of hypoglycemia: HypoCOMPaSS Trial. Diabetes Care 2016;39:e56 e Bajaj HS, Venn K, Ye C, et al.: Lowest glucose variability and hypoglycemia are observed with the combination of a GLP-1 receptor agonist and basal insulin (VARIATION Study). Diabetes Care 2017;40: Park SE, Lee BW, Kim JH, et al.: Effect of gemigliptin on glycaemic variability in patients with type 2 diabetes (STABLE study). Diabetes Obes Metab 2017;19: Rodbard HW, Peters AL, Slee A, et al.: The effect of canagliflozin, a sodium glucose cotransporter 2 inhibitor, on glycemic end points assessed by continuous glucose monitoring and patient-reported outcomes among people with type 1 diabetes. Diabetes Care 2017;40: Bode BW, Hiese T, Pieber T, et al.: Higher rate of confirmed hypoglycemia are associated with greater withinsubject variation in fasting plasma glucose in patients with type 1 or type 2 diabetes: a metanalysis. AACE Abstr 2013; Goh S-Y, Hussein Z, Rudijanto A: Review of insulinassociated hypoglycemia and its impact on the management of diabetes in Southeast Asian countries. J Diabetes Investig 2017;8: Address correspondence to: Suresh Rama Chandran, MD, MRCP Department of Endocrinology Singapore General Hospital 20 College Road, Level 3, The Academia Singapore Singapore suresh.rama.chandran@singhealth.com.sg