The association between hypoglycemia and glycemic variability in elderly patients with type 2 diabetes: cross-sectional study

Background This study aimed to explore predictive factors of time below target glucose range (TBR) ≥ 1% among patients’ characteristics and glycemic variability (GV) indices using continuous glucose monitoring data in elderly patients with type 2 diabetes. Methods We conducted a prospective observational study on 179 (71 female) Japanese outpatients with type 2 diabetes aged ≥ 65 years. The characteristics of the participants with TBR ≥ 1% were evaluated by multivariate logistic regression analysis. Receiver-operating characteristic (ROC) curve analyses of GV indices, comprising coecient of variation (CV), standard deviation, and mean amplitude of glycemic excursions, were performed to identify the optimal index for the identication of patients with TBR ≥ 1%.


Introduction
The number of elderly patients with diabetes increased to 111.2 million worldwide in 2019 [1], re ecting a substantial burden on the medical care system. The common features of these patients are the likelihood of having hypoglycemia associated with decreasing quality of life [2] and falls leading to bone fractures [3]. In addition, severe hypoglycemia was found to be associated with dementia [4], cardiovascular diseases and mortality risk [5] in these patients. Thus, elderly patients with diabetes require treatments that can prevent diabetic complications by stabilizing glycemic variability (GV) without hypoglycemia.
Consequently, it is important to identify hypoglycemia and clarify the characteristics of these patients with hypoglycemia using real-world data.
Recently, continuous glucose monitoring (CGM) has been used worldwide for daily clinical practice. CGM traces the dynamical levels of glucose in interstitial uid within subcutaneous fatty tissue over the whole day and night, making it possible to estimate GV and to detect hypoglycemia and hyperglycemia. An international consensus recommended the coe cient of variation (CV) as the main measure of GV, because it can predict hypoglycemia more accurately than other GV indices including the standard deviation (SD) [6]. Stable GV was de ned as CV < 36% and unstable GV de ned as CV ≥ 36%, re ecting the frequency of hypoglycemia in all age groups [7]. The international consensus also recommended percentage of time below target glucose range (TBR) < 1% as an appropriate CGM-related value for prevention of hypoglycemia in elderly patients with diabetes [8].
However, there are few clinical studies on the characteristics of elderly patients with diabetes and TBR ≥ 1%. Furthermore, the relationships between TBR ≥ 1% and GV indices in CGM for elderly patients with diabetes remain to be fully characterized.
This study aimed to explore predictive factors of TBR ≥ 1% among patients' characteristics and GV indices using continuous glucose monitoring data in elderly patients with type 2 diabetes.

Study design and participants
The present study was a secondary analysis of a previous study with a multicenter (four sites), observational, prospective, cross-sectional design that utilized CGM [9]. Brie y, patients aged ≥ 20 years were included in this study when they consented to undergo ambulatory CGM without regard to levels of HbA1c, duration of diabetes, sex, and diabetic complications. The exclusion criteria were as follows: patients with type 1 diabetes; patients who had been hospitalized within the past 3 months; patients with diabetic ketosis or in diabetic coma, with serious infection, within preoperative or postoperative periods, or with trauma during the past 6 months; and patients administrated steroid treatments, were lactating or pregnant, or had di culty consuming a normal diet. Patients provided clinical information (age, sex, anthropometric measurements, duration of diabetes in years, treatment regimen, and medical history), fasting blood samples, and CGM data. For the present study, we extracted data for patients aged ≥ 65 years among the enrolled patients.

Biochemical analyses and data collection
Patients' weight and height were measured using a calibrated scale. Body mass index (BMI) was calculated as weight in kilograms divided by height in meters squared. Further information including age, sex, treatment regimen, and medical history was taken with a questionnaire administered by the attending physicians. Blood samples were collected for this study after an overnight fast to measure the levels of HbA1c, fasting plasma glucose (FPG), C-peptide (CPR), and estimated glomerular ltration rate. The C-peptide index (CPI) was obtained by the following formula: 100 × fasting CPR (ng/mL) / plasma glucose (mg/dL), and applied to the re ection of endogenous insulin secretion [10].
Between 2018 and 2019, all patients went through ambulatory CGM for 14 consecutive days by means of the same technology (FreeStyle Libre Pro Sensor; Abbott Diabetes Care, Alameda, CA, USA). We analyzed the CGM data for patients with at least 4 days of recorded data available. Data for the rst and last days of wearing the device were excluded from the analysis because of concerns regarding the accuracy of the CGM system during attachment and detachment [11]. Several indices for GV were calculated using GlyCulator2 software [12], as follows: CV (100 × [SD of glucose] / [mean glucose]), SD, and mean amplitude of glycemic excursions (MAGE) [13]. Moreover, we calculated three key measurements of CGM, comprising percentage of time within target glucose range (TIR; 70-180 mg/dL), percentage of time above target glucose range (TAR; >180 mg/dL), and percentage of TBR (< 70 mg/dL), in accordance with the international consensus recommendation [8].

Statistical analysis
As describes above, international consensus recommended TBR < 1% as an appropriate CGM-related value for prevention of hypoglycemia in elderly patients with diabetes [8]. Therefore, to determine the characteristics of elderly patients with diabetes who developed hypoglycemia, we assigned the patients to two subgroups: TBR ≥ 1% and TBR < 1%. For clinical factors related to TBR, comparisons between the two groups were performed by the Mann-Whitney U-test for means of continuous variables and Fisher's exact test for proportions of categorical variables. Data are shown as mean ± standard deviation.
To analyze the characteristics of elderly patients with diabetes and TBR ≥ 1%, signi cant variables at value of P < 0.05 in univariate analysis were examined in multivariate logistic regression analysis for each GV index. To further determine the best GV index for identifying TBR ≥ 1% in elderly patients with diabetes as well as the optimal cut-off point for each GV index corresponding to TBR ≥ 1%, receiveroperating characteristic (ROC) curve analyses were carried out for the GV indices. All tests were two-sided, and P < 0.05 was taken to show statistical signi cance. All statistical analyses were carried out using JMP 14 software (SAS Inc., Cary, NC, USA).

Results
Among these 311 patients, 27 were excluded since they met at least one exclusion criteria described previously 9 . Furthermore, 105 patients aged < 65 years were excluded. The remaining 179 patients (71 females) were considered eligible and included in the analyses (Fig. 1).
The 179 elderly patients with type 2 diabetes were divided into two subgroups: TBR ≥ 1% (n = 57) and TBR < 1% (n = 122). Table 1 shows the anthropometric and biochemical characteristics of the full analysis group and the two subgroups. Compared with the TBR < 1% group, the TBR ≥ 1% group had signi cantly higher rate of insulin treatment and history of stroke or transient ischemic attack (TIA), and lower FPG, HbA1c, and CPI levels. For the three GV indices, CV, SD and MAGE were signi cantly higher in the TBR ≥ 1% group compared with the TBR < 1% group (Table 2). TAR in the TBR ≥ 1% group was signi cantly lower than that in the TBR < 1% group although there was no signi cant difference in TIR between the two groups ( Table 2). There were no indications of multicollinearity between these variables (data not shown).

Discussion
In this study, we rst demonstrated that the clinical characteristics of elderly patients with diabetes were not associated with TBR ≥ 1%. However, high GV indices contributed to TBR ≥ 1%. Among the three GV indices, CV was the most appropriate index for predicting TBR ≥ 1% in Japanese elderly outpatients with type 2 diabetes using CGM.
The strength of our study is that all patients in the analysis were aged ≥ 65 years, with the mean age being 73.9 years (Table 1). A previous cross-sectional study showed that low level of fasting CPR was associated with high risk of hypoglycemia in 98 outpatients with type 2 diabetes using CGM [14]. While that study focused on patients of all ages who were all treated with insulin, our study only enrolled elderly patients regardless of insulin therapy. For this reason, our results were different from ones in that study. Taken together, the ndings suggest that the predictive markers for hypoglycemia in patients with diabetes differ between elderly patients and adult patients of all ages. Recently, an observation study involving CGM for 281 adult type 2 diabetes outpatients aged 40-75 years, including 181 patients aged 65-70 years, revealed that low HbA1c level and use of hypoglycemic agents related to hypoglycemia including sulfonylureas, glinides and insulin were associated with high TBR [15]. Although the subjects in that study were aged < 75 years, our study enrolled older patients (> 75 years) to provide more re ective data for real-world clinical practice. In addition, our study has the novelty of focusing on whether the clinical characteristics had associations with TBR ≥ 1% or < 1%, as the measure recommended by the international consensus to prevent hypoglycemia in elderly patients with diabetes. The result of our study showed that none of the clinical characteristics of these patients had correlations with TBR ≥ 1%. Elderly patients are at high risk of hypoglycemia related to drugs because of liver or renal dysfunction [16].
Furthermore, these patients are at high risk of severe hypoglycemia because of their characteristic of being likely to experience hypoglycemia unawareness [17]. Thus, it is desirable to identify the existence of hypoglycemia by making effective use of CGM regardless of any characteristics in elderly patients with diabetes.
Our study showed signi cant associations between TBR ≥ 1% and GV indices. Unstable GV was found to be associated with hypoglycemia in a previous observational study that performed self-monitoring of blood glucose in 335 adult patients (254 with type 1 diabetes and 81 with type 2 diabetes in all age groups) and computed the Average Daily Risk Range, one of the indices for GV [18]. Another observational study that performed CGM in 88 patients (20 with type 1 diabetes and 68 with type 2 diabetes in all age groups) showed that age had positive correlations with SD and MAGE in patients with type 2 diabetes [19]. These ndings indicate that elderly patients with diabetes are at high risk of unstable GV. Although our study only focused on elderly patients with diabetes, a signi cant association between TBR ≥ 1% and GV was found. Taken together, unstable GV has an association with hypoglycemia in elderly patients with diabetes as well as in patients in all age groups.
Although CV was an indirect index that can predict hypoglycemia without being in uenced by hyperglycemia, both SD and MAGE had a bias toward hyperglycemia due to the fact that these markers are absolute measures [20,21]. In addition, CV was more strongly associated with a key CGM measure of hypoglycemia than other indices such as SD and MAGE in our previous cross-sectional study in 284 Japanese patients with type 2 diabetes in all age groups [9]. Similarly, the present study showed that CV was a more accurate predictor of TBR ≥ 1% than SD and MAGE, even in elderly patients with type 2 diabetes.
Our study showed that low TAR had a signi cant association with TBR ≥ 1%. While few studies have investigated the relationship between TAR and TBR in patients with type 2 diabetes, one study conducted CGM in 85 children and adolescents with type 1 diabetes and showed a weak inverse correlation between TAR and TBR [22]. This nding is compatible with our results.
There are some limitations in the present study. First, factors that can minimize GV remain to be determined because this study had a relatively small sample size. To identify such factors, a larger prospective longitudinal study is needed. Second, the validity of the CGM measurement quality could be a limitation. In this regard, these inaccuracies were minimized by excluding data from the rst and last days of wearing the device [23].

Conclusions
In conclusion, there were no associations between the clinical characteristics of Japanese elderly patients with type 2 diabetes and TBR ≥ 1% in a real-world clinical practice. Similar to previous studies on patients in all age groups, CV had the most signi cant association with hypoglycemia among the GV indices examined. In elderly patients with diabetes, it is preferable to perform evaluations using CGM to con rm good blood glucose pro les without hypoglycemia.

Not applicable
Availability of data and material The datasets used and/or analyzed during the current study are available from the corresponding author on reasonable request.

Funding
This study did not supported by anywhere.

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download. STROBEchecklist.docx