1 INTRODUCTION

Hypoglycaemia is a considerable barrier to intensified blood glucose management in type 1 diabetes mellitus (T1DM). Recurrent low blood glucose can lead to neurocognitive and cardiovascular complications.1, 2 Continuous glucose monitoring (CGM) provides comprehensive information about glycaemic control, including: (a) time in range (TIR); (b) time above range (TAR); and (c) time below range (TBR). The targets for these metrics were developed by consensus.3

Previous observational studies have shown a positive linear association between coefficient of variation (CV; standard deviation [SD]/average glucose) and TBR. This same relationship has been mathematically modelled by Rodbard.4 His results demonstrate significant reduction in hypoglycaemia at %CV values of <34% to 36%. Newer data, however, suggest these findings may be less applicable at the extremes of glycaemic control.5, 6 This would explain the lack of universal consensus on target %CV. Recent guidelines from the International Consensus on Time-In-Range offer only a nonspecific recommendation to limit blood glucose variation.3 In older patients, the combination of %CV and glucose management indicator (GMI) better identifies increased risk of hypoglycaemia than either variable alone.6 This implies that %CV targets perhaps need to be individualized based on blood glucose control. Given these previous reports, the aim of this study was to examine the association between %CV and TBR in T1DM, and to assess the consistency of that association across the range of GMI values.

2 METHODS

We performed a retrospective chart review of 219 personal (“real-time”) CGM downloads from T1DM patients seen at the University of Washington Diabetes Institute between 2016 and 2019. We required at least 14 days of sensor wear (at ≥93% data sufficiency) for inclusion in the analyses. We included patients using the Dexcom G4-G6, the Medtronic Guardian 3 and the Medtronic Enlite devices. We omitted the Freestyle Libre CGM device because of its known tendency to overread hypoglycaemia at lower blood glucose levels.7 There were otherwise no exclusion criteria. The association between TBR (blood glucose < 3.9 mmol/L [70 mg/dL]) and %CV was examined using linear regression, with each modelled as a continuous variable. Nonlinear terms for %CV were considered by modelling %CV as a restricted cubic spline with knots at the 5th, 25th, 50th, 75th and 95th percentiles of the observed %CV values. GMI was considered as a potential effect modifier, that is, as a variable whose values lead to differing magnitudes of association between TBR and %CV. In addition, we assessed the association between HbA1c and TBR to see if %CV or HbA1c provided a better fit to the model for TBR.

3 RESULTS 3.1 Demographics

Our cohort comprised subjects aged 19 to 84 years (mean [SD] 48 [± 15] years). There was no gender predominance. The mean (range; SD) diabetes duration was 21 (4-64; 13) years. Fifty percent were insulin pump users. Forty percent had documented microvascular disease (defined as albuminuria, peripheral neuropathy and/or any retinopathy). Eight percent of patients had a reduced glomerular filtration rate (<60 mL/min/1.73 m2). Fifty-two percent had common diabetes comorbidities (including coronary artery disease, hyperlipidaemia and/or hypertension. The mean (range; SD) glycated haemoglobin (HbA1c) level was 55 (27-97; ± 5.5) mmol/L (7.2% [4.6%-11.0%; ± 0.9%]). The mean (range; SD) GMI was 7.2% (5.3%-10.4%; ± 0.7%). The mean (SD) duration of CGM wear for our analysis was 24.7 (± 5.5) days. Thirty-six percent of patients had TBR >4%. The majority of patients were prescribed a Dexcom brand sensor (n = 148 [68%]). Further demographic information is available in Table 1.

TABLE 1. Patient demographics (N = 219) Demographic Value Patient age, years 48 (± 15; 19-84) Gender, n (%) Male 106 (48.5) Female 113 (51.5) T1DM duration, years 21 (± 13; 4-64) Insulin pump users, n (%) 109 (50) Insulin pump used, n (%) Medtronic 670G 22 (20) Medtronic 630G 50 (46) Medtronic 530G 3 (3) Medtronic paradigm 1 (1) Insulet Omnipod 26 (24) Tandem T:Slim X2 7 (6) Microvascular disease, n (%) 88 (40) Reduced GFR (<60 mL/min/1.73 m2), n (%) 18 (8) HbA1c, % 55 (+5.5; 27-97) mmol/L 7.2 (± 0.9; 4.6-11.0) GMI 7.2 (± 0.7; 5.3-10.4) Duration of CGM device wear, days 24.7 (± 5.5; 14-30) TIR, % 60 (± 17; 16-93) TBR <4%, % 36 CGM device, n (%) Dexcom G6 55 (25) Dexcom G5 78 (37) Dexcom G4 14 (6) Medtronic Guardian 3 69 (31) Medtronic Enlite 3 (1) Note: Data are mean (SD; range), unless otherwise indicated. Abbreviations: CGM, continuous glucose monitoring; GFR, glomerular filtration rate; GMI, glucose management indicator; HbA1c, glycated haemoglogin; T1DM, type 1 diabetes mellitus; TBR, time below range; TIR, time in range. 3.2 Association between %CV and TBR

There was a positive correlation between %CV and TBR with %CV modelled as a continuous linear variable (R2 = 0.24). However, the data suggested that the addition of nonlinear terms led to a better fit to the data. In particular, modelling %CV as a restricted cubic spline led to an R2 value of 0.35 (Figure 1A). Comparing the model with nonlinear terms to the model without nonlinear terms yielded a significantly improved fit (P < 0.0001). The model with %CV modelled as a continuous nonlinear function was further improved by including GMI as an effect modifier (R2 = 0.70), and a statistical test of interaction between the nonlinear function of %CV and GMI yielded a P value < 0.0001, that is, the nonlinear association between %CV and TBR varies with different values of GMI. Figure 1B shows the modelled association between %CV and TBR for select values of GMI; the values chosen for display were somewhat arbitrary, but span a GMI range that the vast majority of subjects achieve.

(A) Time below range (TBR) versus coefficient of variation percentage (%CV), with %CV modelled as a continuous linear function. (B) TBR vs %CV, with %CV modelled as a restricted cubic spline; the model regressing TBR on %CV also includes glucose management indicator (GMI) plus interaction terms between GMI and %CV modelled as a restricted cubic spline. As such, the association between TBR and %CV is different for each GMI value. The association is depicted in the figure for select values of GMI, these GMI values being somewhat arbitrary but spanning GMI values seen in the overwhelming majority of patients in this study. (C) TBR vs %CV, with %CV modelled as a restricted cubic spline; the model regressing TBR on %CV also includes glycated haemoglobin (HbA1c) plus interaction terms between HbA1c and %CV modelled as a restricted cubic spline. As such, the association between TBR and %CV is different for each HbA1c. The association is depicted in the figure for select values of HbA1c, these HbA1c values being somewhat arbitrary but spanning HbA1c values seen in the overwhelming majority of patients in this study

We looked separately at subjects who were using an insulin pump versus those who were administering multiple daily injections (MDI) and the results were similar. In particular, the R2 value was 0.71 when CV was modelled as a cubic spline using GMI and interaction terms amongst subjects who were on a pump versus an R2 value of 0.74 for subjects who were on MDI.

When GMI was replaced with HbA1c, results were similar, although the R2 value with HbA1c and interaction terms was lower (R2 = 0.54) than that using GMI instead. Figure 1C shows the modelled association between TBR and CV for select values of HbA1c.

3.3 Association between HbA1c and TBR

There was a negative correlation between HbA1c and TBR with HbA1c modelled as a continuous linear variable, but the magnitude of the association was lower than that for %CV (R2 = 0.15). The data suggested a slight improvement to the model with the addition of nonlinear terms (P = 0.07 when comparing models with nonlinear terms versus without nonlinear terms; R2 = 0.17). The model with HbA1c modelled as a continuous nonlinear function was further improved by including GMI as an effect modifier (R2 = 0.32), and a statistical test of interaction between the nonlinear function of HbA1c and GMI yielded a P value of 0.002, that is, the nonlinear association between HbA1c and TBR varied with different values of GMI.

4 DISCUSSION

Glucose variation remains an unclear risk for diabetes complications and, although CGM shows that %CV and hypoglycaemia are associated, there is no clear consensus on individual %CV targets.8 The 2019 International Consensus published a %CV target of <36% but noted that “some studies suggest lower %CV targets (<33%) provide additional protection against hypoglycaemia for those receiving insulin or sulfonylureas”.3

Our results show that %CV and GMI provide important information regarding the risk of hypoglycaemia in T1DM, but the impact of one on the risk of hypoglycaemia is dependent on the value of the other. The same %CV can predict a different TBR at various levels of glycaemic control. As average blood glucose decreases, it takes less variation to increase TBR. This also means that some patients who have more hyperglycaemia can tolerate higher %CV without an increased risk in hypoglycemia. Based on these data, we recommend an individualized approach to selecting a target %CV for CGM patients. Further, our results indicate that, while HbA1c is negatively correlated with TBR (higher HbA1c = lesser TBR), using %CV generates a better model, thus explaining more of the variability in TBR than HbA1c. Since %CV is a function of mean glucose and standard deviation of glucose, this improved fit implies that knowledge of the standard deviation adds more to explaining TBR than HbA1c alone (a derivation of mean glucose).

The strengths of this study include the use of “real-time” CGM in a large cohort of T1DM patients who had extended disease duration and common diabetes complications/comorbidities. The greatest limitation of our study is the homogenous White population from the Pacific Northwest in the United States.

In conclusion, using a %CV target alone to reduce hypoglycaemia is too simplistic as this metric depends on glycaemic control. Both clinicians and researchers need to appreciate that %CV is impacted by all of the different therapeutic and behavioural issues that affect diabetes patients. We intend to integrate GMI into our practice to better determine appropriate %CV targets for individual patients.

ACKNOWLEDGMENTS

The work of J.E.P. was supported by a University of Washington Housestaff Association resident research grant.

The data that support the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions.

REFERENCES

1Amiel SA. The consequences of hypoglycaemia. Diabetologia. 2021; 64(5): 963- 970. doi:10.1007/s00125-020-05366-3 2Andersen A, Jørgensen PG, Knop FK, Vilsbøll T. Hypoglycaemia and cardiac arrhythmias in diabetes. Ther Adv Endocrinol Metab. 2020; 11:2042018820911803. 3Battelino T, Danne T, Bergenstal RM, et al. Clinical targets for continuous glucose monitoring data interpretation: recommendations from the international consensus on time in range. Diabetes Care. 2019; 42(8): 1593- 1603. 4Rodbard 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(10): 868- 876. 5Helleputte S, De Backer T, Calders P, Pauwels B, Shadid S, Lapauw B. The added and interpretative value of CGM-derived parameters in type 1 diabetes depends on the level of glycemic control. Endocr Pract. 2021; 27(1): 44- 50. 6Toschi E, Slyne C, Sifre K, et al. The relationship between CGM-Derived metrics, A1C, and risk of hypoglycemia in older adults with type 1 diabetes. Diabetes Care. 2020; 43(10): 2349- 2354. 7Galindo RJ, Migdal AL, Davis GM, et al. Comparison of the FreeStyle libre pro flash continuous glucose monitoring (CGM) system and point-of-care capillary glucose testing in hospitalized patients with type 2 diabetes treated with basal-bolus insulin regimen. Diabetes Care. 2020; 43(11): 2730- 2735. 8Rodbard D. Glucose variability: a review of clinical applications and research developments. Diabetes Technol Ther. 2018; 20: S25- s215.