Evaluation of Clinical Metrics for Identifying Defective Physiologic Responses to Hypoglycemia in Long-Standing Type 1 Diabetes

Abstract

Repeated hypoglycemia exposure leads to impaired awareness of hypoglycemia (IAH) and the development of defective counterregulatory responses. To date, only pancreas or islet transplantation has demonstrated normalization of hypoglycemia awareness and the endogenous glucose production (EGP) response to defend against insulin-induced hypoglycemia in long-standing type 1 diabetes (T1D). This study aims to validate clinical metrics of IAH (Clarke score), hypoglycemia severity (HYPO score), glycemic lability (lability index), and continuous glucose monitoring (CGM) as predictors of absent autonomic symptom (AS) recognition and defective glucose counterregulation during insulin-induced hypoglycemia, thus enabling early identification of individuals with compromised physiologic defense against clinically significant hypoglycemia. Forty-three subjects with mean ± standard deviation age 43 ± 13 years and T1D duration 28 ± 13 years, including 32 with IAH and 11 with hypoglycemia awareness (Aware), and 12 nondiabetic control subjects, underwent single-blinded randomized-paired hyperinsulinemic–euglycemic and hypoglycemic clamp experiments. Receiver operating characteristic (ROC) curves and sensitivity analyses were performed to assess metric prediction of absent AS recognition and defective EGP responses to hypoglycemia. Clarke score and CGM measures of hypoglycemia exposure demonstrated good ability to predict absent AS recognition (area under the curve ≥0.80). A composite threshold of IAH–Clarke ≥4 with ROC curve-derived thresholds for CGM measures of hypoglycemia exposure showed high specificity and predictive value in identifying an absent AS response during the hypoglycemic clamp. Metrics demonstrated poor ability to predict defective glucose counterregulation by the EGP response, which was impaired even in the Aware group. Screening for IAH alongside assessment of CGM data can increase the specificity for identifying individuals with absent hypoglycemia symptom recognition who may benefit from further intervention.

Introduction

Novel exogenous insulin analogs and delivery systems aim to support an increasingly physiologic pattern of circulating insulin; however, risk of hypoglycemia persists and remains a significant barrier to the attainment of adequate glycemic control.¹ Severe hypoglycemia (SH), requiring help of a third party to recover,² may be experienced across the range of glycemic control assessed by HbA1c.^3,4 Recurrent exposure to hypoglycemia leads to a diminished magnitude and glucose concentration at which autonomic symptoms (AS) are generated in response to insulin-induced hypoglycemia, leading to the development of hypoglycemia unawareness.^5–7 Risk of impaired awareness of hypoglycemia (IAH) increases with diabetes duration and development of hypoglycemia unawareness is associated with up to 20-fold increased risk of experiencing SH events.⁸

In addition to IAH, defective counterregulatory responses develop with long-standing type 1 diabetes (T1D), including blunted sympathoadrenal epinephrine secretion and impaired endogenous (predominantly hepatic) glucose production (EGP).^5,7,9 Collectively described as hypoglycemia-associated autonomic failure (HAAF),¹ this syndrome is associated with increased risk for morbidity, mortality, and distress.^10–13

Medical management approaches to HAAF, including educational, psychobehavioral, and technology interventions focusing on strict avoidance of biochemical hypoglycemia have restored AS and hypoglycemia awareness but with no to modest improvement in the epinephrine response to insulin-induced hypoglycemia^14–19 and glucose counterregulation, assessed by EGP response, remaining defective.¹⁸ To date, only the complete and sustained avoidance of hypoglycemia attained by intrahepatic islet transplantation has demonstrated normalization of both AS recognition and the EGP response to defend against the development of hypoglycemia.^20,21 Thus, for many individuals with a history of IAH or recent severe hypoglycemic events, clinical measures that accurately predict the absence of AS or defective glucose counterregulation in response to hypoglycemia would be useful for identifying individuals with physiologic impairment that makes them dependent on technology for hypoglycemia avoidance and also potential candidates for considering β cell replacement.

This study aims to validate current clinical measures of hypoglycemia awareness, hypoglycemia severity, and glycemic lability alongside continuous glucose monitoring (CGM) metrics as predictors of absent AS recognition and defective glucose counterregulation to hyperinsulinemic–hypoglycemic clamp testing, and thus aid the identification of those who would benefit from consideration of focused hypoglycemia avoidance interventions.

Participants and Methods

Participants with C-peptide-negative T1D and disease onset <40 years, ages between 25 and 70 years, and actively involved in diabetes self-management were included in this study. Those with IAH had taken part in pancreatic islet transplantation (Cohort 1) or real-time CGM (RT-CGM) (Cohort 2) intervention studies at the University of Pennsylvania between 2008 and 2016.^18,20,21 The present analysis includes all participants who completed baseline glucose counterregulatory assessment and an additional group of individuals who were enrolled with intact awareness of hypoglycemia (Aware). Study protocols were approved by the Institutional Review Board of the University of Pennsylvania and all subjects provided written informed consent to participate.

Diabetes history, including assessment of hypoglycemia awareness (Clarke score²²) and number of SH events in the last year were obtained at screening. Eligible participants completed a 4-week hypoglycemia event diary for calculation of the hypoglycemia severity (HYPO) score and glycemic lability index (LI).^23,24

Individuals with at least a 10-year duration of diabetes, IAH (Clarke score ≥4) and at least one episode of SH in the past year associated with one of the following: severe problematic hypoglycemia [HYPO score ≥1047 (90th percentile of a group of individuals with T1D); marked glycemic lability (LI ≥433 mmol/L²/h.week⁻¹ (90th percentile)]; or a composite of HYPO score ≥423 (75th percentile) and LI ≥329 mmol/L²/h.week⁻¹ (75th percentile),²³ were recruited to the islet transplantation²⁰ and subsequent RT-CGM¹⁸ studies and represent the IAH groups (Cohort 1 and Cohort 2) in the present analysis. Individuals meeting all criteria except for experiencing an episode of SH in the past year could enroll in the RT-CGM study (Cohort 2) if spending at least 5% time <60 mg/dL by screening CGM.

Individuals recruited to the Aware subgroup had at least a 5-year duration of diabetes, hypoglycemia awareness defined by Clarke score ≤2, absence of SH over the preceding 3 years, and no recent use of CGM. Individuals using CGM were excluded to avoid including participants in the Aware group who could have IAH masked by CGM alert and alarm features.¹⁸ Comparison was also made to a nondiabetic control group, matched to the T1D IAH participants for age, sex, race, and body mass index (BMI).^18,21

Additional study exclusion criteria included: use of other hypoglycemic agents within 4 weeks of enrolment; presence of obesity (BMI >30 kg/m²) or insulin resistance (>1.0 U/kg/day); or poor glycemic control (HbA_1c >10%). Further details are published elsewhere.^18,20

All T1D groups completed a period of blinded CGM (CGMS; Medtronic Minimed). IAH-Cohort 1 (islet study,^20,21) completed 3 days of monitoring with data uploaded and derived from Medtronic software. Cohort 2 (RT-CGM study¹⁸) and the Aware group completed a 7-day period of monitoring with a minimum 96 h, including at least 24 h of overnight glucose readings recorded. Individuals were advised to self-monitor blood glucose (OneTouch Ultra; LifeScan, Inc.) four times a day to maintain sensor calibration. Mean glucose, glucose standard deviation (SD), coefficient of variation, percentage of time in hypoglycemia (<54, <60, <70 mg/dL), hyperglycemia (>180, >250 mg/dL), time in range (70–180 mg/dL), and low and high blood glucose indices were calculated from CGM device downloads using HypoCount software (version 1.6 PRECISE Center, University of Pennsylvania, Philadelphia, PA).¹⁸

Assessment of glucose counterregulation

All participants underwent paired hyperinsulinemic stepped-hypoglycemic and euglycemic clamp studies within a 4-week period as previously described (Supplementary Fig. S1).^18,20,21 Subjects were blinded to glycemic condition with the order determined by block randomization. Participants underwent a 12-h period of fasting with near-normoglycemia maintained in participants with diabetes by an intravenous insulin infusion protocol. At 0700, t = −120 min, a primed (5 mg/kg fasting plasma glucose in mg/dL/90 for 5 min) continuous (0.05 mg/kg per min for 355 min) infusion of 6,6-²H₂-glucose (99% solution; Cambridge Isotopes Laboratories, Andover, MA) was initiated to assess EGP before and during the clamp experiments with baseline blood samples obtained at t = −20, −10, and −1 min. At t = 0 min a continuous infusion of insulin (1.0 mU/kg per min for 240 min) was administered to produce hyperinsulinemia.

Subsequently, a variable rate infusion of 20% dextrose (enriched with 2% 6,6-²H₂-glucose to reduce changes in plasma enrichment during the clamp) was administered to maintain euglycemia (plasma glucose targeted to 90 mg/dL) or to induce hypoglycemia with hourly glucose targets of 80, 65, 55, and 45 mg/dL. Blood samples were taken every 5 min, centrifuged, and measured to determine clamp glucose at the bedside using an automated glucose analyzer (YSI 2300; Yellow Springs Instruments, OH) to adjust the glucose infusion rate and achieve the desired plasma glucose concentration. An AS questionnaire was completed every 20 min with the sum of scores ranging between 0 (none) and 5 (severe) for six symptoms: anxiety; palpitation; sweating; tremors; hunger; and tingling used to quantitate the AS score.²⁵ Additional blood samples were taken at t = 20, 40, 60, 80, 100, 120, 140, 160, 180, 200, 220, and 240 min for biochemical analysis and verification of the plasma glucose concentration.

All samples were collected on ice into chilled tubes containing ethylenediaminetetraacetic acid (EDTA), with Protease Inhibitor Cocktail (Sigma-Aldrich, St Louis, MO) added to the tubes for peptide hormones, centrifuged at 4°C, separated, and frozen at −80°C for subsequent analysis. Plasma insulin, glucagon, and pancreatic polypeptide were measured in duplicate by double-antibody radioimmunoassays (for insulin and glucagon: Millipore, Billerica, MA; for pancreatic polypeptide: ALPCO Diagnostics, Salem, NH). Plasma epinephrine and norepinephrine were measured by high-performance liquid chromatography with electrochemical detection. Enrichment of 6,6-²H₂-glucose in plasma was measured by gas chromatography–mass spectrometry (GC-MS), where deproteinized sample glucose was converted to the glucose aldol penta-acetate derivative and analyzed by GC-MS using electron impact ionization²⁶ with results expressed as the glucose labeled (M + 2)/unlabeled (M + 0) peak areas and enrichments determined by interpolation from a standard curve consisting of standards of known enrichment.

Calculations and statistics

EGP was calculated as previously described.²¹ The rate of appearance of glucose during the clamps was calculated by the Steele nonsteady-state equation modified for the use of stable isotopes.²⁷ EGP was calculated from the difference between the rate of appearance of glucose in the plasma and the infusion rate of exogenous glucose.

Magnitude of counterregulatory responses during the hypoglycemic clamp were assessed by the mean of values over the last hour of study (Step 4, 45 mg/dL) with the nondiabetic control group included to demonstrate normal hormonal and glucose counterregulatory response to hypoglycemia. Responses during Step 4 of the hypoglycemic clamp were compared within each group to data from the euglycemic clamp by Wilcoxon matched-pair analysis and between-group differences were assessed by Kruskal–Wallis analysis of variance (ANOVA). Change in AS from baseline (delta) and EGP response for each individual was considered significant if the response surpassed the positive 95% confidence limit for the response of their group during the euglycemic clamp condition.

The glycemic threshold for each parameter was calculated for each subject as the plasma glucose concentration at which the parameter first exceeded the 95% confidence limit observed for that parameter at the corresponding time point in the euglycemic control experiments for that subject's group.^28–30 If the hypoglycemic clamp AS or EGP response never surpassed the positive 95% confidence limit of the euglycemic clamp, absent AS recognition or EGP response/glucose counterregulation were defined; for analysis of glycemic thresholds, the lowest observed glucose concentration was imputed when there was an absent response.¹⁶

The sensitivity, specificity, predictive value, and accuracy of established cutoffs of clinical metrics (Clarke, HYPO, and LI) to identify absent AS and EGP responses were assessed. Receiver operating characteristic (ROC) curves were generated using Prism software version 9.0.2 (GraphPad Software, San Diego, CA) and area under the curve (AUC) data interpreted to define metric ability to predict clamp response, with an AUC 0.80–0.89 defining a good test and <0.70 a nonuseful test.³¹ Youden indices were calculated using SAS software version 9.4 (SAS Institute, Inc., Cary, NC) to determine the optimal threshold of maximal sensitivity and specificity for CGM metrics to predict clamp response, and these thresholds were combined with clinical metrics to form composite measures for assessment of absent AS and EGP responses.

Statistical analysis was performed using Statistica software (StatSoft, Inc., Tulsa, OK). Comparison of demographic and CGM measures between groups was by Mann–Whitney U test, Kruskal–Wallis ANOVA or Chi-squared test as appropriate. Significance was considered at P ≤ 0.05 (two-tailed).

Results

Participant demographic, clinical, and CGM metrics

Forty-three individuals with T1D, including 32 participants with IAH (IAH-Cohort 1 and 2; Clarke ≥4) and 11 with awareness of hypoglycemia (Aware; Clarke ≤2), and 12 nondiabetic control participants were included in the study.

The demographics of the diabetes groups are shown in Table 1. The Aware group was significantly younger with a shorter duration of diabetes compared with the IAH cohorts. Weight, BMI, and insulin requirement were also greater, however, glycemic control, assessed by HbA1c, was comparable. By definition in terms of inclusion criterion, groups differed for Clarke score. Both IAH Cohorts demonstrated marked hypoglycemia severity and glycemic lability with median HYPO and LI scores greater than the 90th% of scores for a normative group of 100 individuals with T1D.²³ In contrast, the Aware group had a very low HYPO score indicating both reduced hypoglycemia frequency and severity (P ≤ 0.001).

Table 1.

Demographic, Glycemic, and Lability Indices

	IAH-Cohort 1 n = 19	IAH-Cohort 2 n = 13	Aware group n = 11
Female	10 (53%)	7 (54%)	4 (36%)
Age (years)	45 ± 9	46 ± 14	35 ± 13*
Diabetes duration (years)	30 ± 11	33 ± 14	19 ± 10*
Weight (kg)	70 ± 11	70 ± 11	83 ± 11*
BMI (kg/m2)	25 ± 3	25 ± 2	28 ± 2*
Insulin (unit/kg/day)	0.51 ± 0.14	0.50 ± 0.16	0.65 ± 0.09*
HbA1c (%) [mmol/mol]	6.7 ± 1.0 [50 ± 11]	7.1 ± 0.8 [54 ± 9]	7.0 ± 0.7 [53 ± 7]
Clarke score	6 (6–7)	6 (6–7)	1 (0–2)**
HYPO score	2366 (1058–4001)	1735 (1076–2605)	89** (36–104)
LI (mmol/l2/h.week−1)	682 (477–811)	540 (366–761)	438¥ (268–529)

Data presented as number (proportion), mean ± SD, and median (IQR).

Kruskal–Wallis ANOVA comparison of diabetes groups: ^*P < 0.05, ^**P ≤ 0.001, ^¥P = 0.05.

ANOVA, analysis of variance; BMI, body mass index; HYPO, hypoglycemia severity score; IAH, impaired awareness of hypoglycemia; IQR, interquartile range; LI, lability index; SD, standard deviation.

CGM data were obtained in all participants with diabetes. However, as CGM in Cohort 1 included only a maximum of 3 days of data from early CGM devices obtained at screening for islet transplantation (2008–2011),²⁰ their data, although not different from Cohort 2, demonstrated wider variability and has been excluded from the ROC curve analysis. Table 2 shows the blinded CGM data collected from the IAH-Cohort 2 and the Aware groups over up to a 7-day period of monitoring. Groups were comparable for average glucose, time in range, and glucose variability measures. IAH-Cohort 2 spent greater percentage of time in extremes of hypoglycemia <54 mg/dL (P = 0.04) and <60 mg/dL (P = 0.03), with a concomitantly greater low blood glucose index (LBGI) (P = 0.05) in comparison to the Aware group. Percentage of time spent in level 1 hypoglycemia (<70 mg/dL)³² trended toward significance between groups (P = 0.06). However, in contrast, no difference was seen in percentage of time spent in measures of hyperglycemia, including the high blood glucose index.

Table 2.

Blinded Continuous Glucose Monitoring Data

	IAH-Cohort 2 n = 13	Aware group n = 11
Mean glucose (mg/dL) [mmol/L]	150 (130–159) [8.3 (7.2–8.8)]	159 (133–179) [8.8 (7.4–9.9)]
Glucose SD (mg/dL) [mmol/L]	64 (54–84) [3.6 (3.0–4.7)]	58 (46–78) [3.2 (2.6–4.3)]
Coefficient of variation (%)	45 (42–54)	42 (35–44)
%Time <54 mg/dL [3.0 mmol/L]	4 (2–9)	2 (0–2)*
%Time <60 mg/dL [3.3 mmol/L]	6 (5–12)	3 (1–4)*
%Time <70 mg/dL [3.9 mmol//L]	12 (6–20)	5 (2–7)
%Time >180 mg/dL [10.0 mmol/L]	30 (17–34)	34 (18–48)
%Time >250 mg/dL [13.9 mmol/L]	6 (0–12)‡	6 (3–18)
%Time in range 70–180 mg/dL [3.9–10.0 mmol/L]	56 (42–70)	64 (49–72)
LBGI	2.9 (1.7–4.1)	1.3 (0.7–2.0)*
HBGI	6.9 (3.6–9.4)	6.6 (3.5–11.2)

Data presented as median (IQR), ^‡n = 7.

IAH-Cohort 2 and Aware group data derived from HypoCount software following 7-day period of iPro/other monitoring, with comparison of data by Mann–Whitney U-test, ^*P < 0.05.

HBGI, high blood glucose index; LBGI, low blood glucose index.

Counterregulatory response and glucose thresholds to hypoglycemic clamp testing

Insulin administered during the hypoglycemic clamp resulted in a comparable level of hyperinsulinemia between groups (Fig. 1A). Plasma glucose was maintained to target 90 mg/dL during euglycemia and in the hypoglycemic condition a per protocol reduction in plasma glucose resulted in overlap of plasma glucose from Step 2 (65 mg/dL) onward (Supplementary Fig. S1 and Fig. 1B). There was no difference in plasma glucose between diabetes groups at the end of the hypoglycemic clamp (Step 4), [median (interquartile range [IQR]) IAH-Cohort 1: 44 (40–48); IAH-Cohort 2: 44 (42–46); Aware group: 43 (40–46) mg/dL]; however, the nondiabetic control group trended toward a higher glucose 47 (43–51) mg/dL (P = 0.08, Kruskal–Wallis ANOVA).

FIG. 1.

Plasma insulin (A) and glucose (B) during the hyperinsulinemic–hypoglycemic clamp experiments. Data presented as mean ± SEM of (▴) IAH-Cohort 1; () IAH-Cohort 2; () Aware group; and () Nondiabetic control group. The shaded area represents the 95% CI for data from the hyperinsulinemic–euglycemic clamp control experiments (n = 54). CI, confidence interval; IAH, impaired awareness of hypoglycemia; SEM, standard error of the mean.

The Aware group and nondiabetic control group demonstrated an overlapping marked increase in AS score from baseline (delta) during the hypoglycemic clamp (Fig. 2A), however, no significant increase in AS was seen in the IAH Cohorts whose Step 4 responses were comparable and not different from the euglycemic condition. The nondiabetic control group demonstrated a robust increase in EGP to hyperinsulinemic hypoglycemia in comparison to during hyperinsulinemic euglycemia (Fig. 2B). The Aware group also generated an EGP response to hypoglycemia greater than the IAH Cohorts and greater than euglycemic clamp experiments, however in contrast to that observed for AS, the magnitude of the response was less than the nondiabetic control group. Figure 2C and D shows the final lowering of glucose from 55 mg/dL into the final 45 mg/dL step of the hypoglycemic clamp is required to separate the AS and EGP responses between IAH and Aware groups.

FIG. 2.

Delta AS score and EGP over time (A, B) and by glucose (C, D) during the hypoglycemic clamp experiments. Data presented as mean ± SEM of (▴) IAH-Cohort 1; () IAH-Cohort 2; () Aware group; and () Nondiabetic control group responses. The shaded area represents the 95% CI for data from the hyperinsulinemic–euglycemic clamp control experiments (n = 54). Dashed line marks glucose of 54 mg/dL (3.0 mmol/L) the International Hypoglycemia Study Group consensus of clinically significant hypoglycemia.³² AS, autonomic symptoms; EGP, endogenous glucose production.

Indeed, comparison of the glucose thresholds for AS and EGP responses between groups showed that nondiabetic individuals generate both AS and EGP responses to hypoglycemia at a higher glucose concentration compared with individuals with T1D (P < 0.01 for the AS and P = 0.03 for the EGP response, respectively; Table 3). The Aware group also demonstrated a higher glucose concentration for AS recognition with no individuals failing to generate a symptom response during the clamp, in contrast to two-thirds of individuals in the IAH Cohorts (P < 0.001). However, despite generating robust AS responses to hypoglycemia, four individuals in the Aware group did not generate an EGP response during the hypoglycemic clamp and although, unlike the IAH Cohorts, the magnitude of EGP response was greater than in euglycemia, the glucose threshold at which an EGP response was generated was not different from that in the IAH Cohorts.

Table 3.

Glucose Thresholds for Autonomic Symptom and Endogenous Glucose Production (EGP) Responses and Number of Individuals Without Autonomic Symptom Recognition or EGP Response by the End of the Hypoglycemic Clamp Study

	IAH-Cohort 1 n = 19	IAH-Cohort 2 n = 13	Aware group n = 11	Non diabetic group n = 12	P a	P b
Glucose threshold for AS response (mg/dL) [mmol/L]	46 (42–52) [2.6 (2.3–2.9)]	46 (43–50) [2.6 (2.4–2.8)]	51 (49–56)* [2.8 (2.7–3.1)]	54 (50–58)* [3.0 (2.8–3.2)]	0.009	0.05
Absence of clamp AS recognition	12 (63%)	9 (69%)	0*	1 (8%)*	<0.001	<0.001
Glucose threshold for EGP response (mg/dL) [mmol/L]	46 (42–55) [2.6 (2.3–3.1)]	47 (42–54) [2.6 (2.3–3.0)]	46 (42–49) [2.6 (2.3–2.7)]	55 (51–62)† [3.1 (2.8–3.4)]	0.03	0.99
Absence of clamp EGP response	12 (63%)	8 (62%)	4 (36%)	0†	0.003	0.32

Data presented as median (IQR), number (proportion). Comparison of all groups (a) and diabetes groups alone (b) by Kruskal–Wallis ANOVA and Chi-squared test of proportions. Intergroup comparison by Mann–Whitney U test: ^*P < 0.05 in comparison to IAH-Cohort 1 and IAH-Cohort 2; and ^†P < 0.05 in comparison to IAH-Cohort 1, IAH-Cohort 2 and Aware group.

AS, autonomic symptoms; EGP, endogenous glucose production.

Supplementary Figure S2 presents additional counterregulatory hormone responses to clamp testing. The nondiabetic control group demonstrated robust glucagon, epinephrine, and pancreatic polypeptide responses to hypoglycemia. No diabetes group had a discernible glucagon response (Supplementary Fig. S2A). The epinephrine response to hypoglycemia was impaired in all diabetes groups relative to that seen in nondiabetic controls, although was greater in the Aware group than in the IAH Cohorts (Supplementary Fig. S2B). Indeed, IAH Cohorts demonstrated a significantly lower glucose concentration at which the epinephrine response was generated in comparison to the nondiabetic control group [median (IQR) glycemic threshold 52 (45–60) mg/dL in IAH Cohorts vs. 57 (55–72) mg/dL in nondiabetic controls, P < 0.05]. However, similar to the intermediate magnitude of epinephrine response generated in the Aware group, their epinephrine glycemic threshold (51 (48–61) mg/dL) was intermediate and not statistically different from either the nondiabetic control group or IAH Cohorts.

A pancreatic polypeptide response to hypoglycemia was observed in all groups, however, this response was marginal in IAH Cohorts, occurring at a median glycemic threshold of 47 mg/dL. Interestingly, the magnitude of Step 4 pancreatic polypeptide response in the Aware group was comparable to the nondiabetic control group response. However, the response curve appeared right shifted (Supplementary Fig. S2C) and assessment of the glycemic threshold at the point at which the euglycemic condition response was exceeded confirmed that the response occurred at a lower glucose concentration in comparison to the nondiabetic control group [median (IQR) glycemic threshold: 48 (40–52) mg/dL Aware Cohort vs. 60 (54–72) mg/dL nondiabetic controls, P < 0.001] (Supplementary Fig. S2C).

Metric prediction of clamp AS recognition and EGP response

Clarke score and CGM measures of hypoglycemia exposure, including percentage of time spent <54, <60, and <70 mg/dL and the LBGI demonstrated good ability to predict absence of AS recognition during the hypoglycemic clamp with area under the receiver operating curves of ≥0.80 (Fig. 3, Table 4). The Youden index was calculated for the CGM metric curves to determine the threshold value of maximal sensitivity and specificity to predict clamp symptom response. The HYPO score showed fair ability to predict symptoms with area under ROC curve of 0.79. Measures of glycemic lability, glucose variability, time in range, or hyperglycemia were poor predictors of symptom response. In addition, clinical and CGM metrics were unable to predict defective EGP response with all metric area under ROC curves ≤0.70 and although statistical significance was reached for LI prediction of EGP response, this was likely due to the wide spread of LI scores across groups, as the discriminatory index, assessed through AUC, remained modest (Fig. 3).

FIG. 3.

ROC curves demonstrating the ability of metrics to predict absent AS (A–F) or defective glucose counterregulation response (EGP) (G, H) to hyperinsulinemic–hypoglycemic clamp testing. (A) Clarke score; (B, G) HYPO severity score; (C) % Total time spent <70 mg/dL; (D) LBGI; (E) % Total time spent <54 mg/dL; (F) % Total time spent <60 mg/dL; (H) LI. HYPO, hypoglycemia severity score; LBGI, low blood glucose index; LI, lability index; ROC, receiver operating characteristic.

Table 4.

Receiver Operating Curve Data Demonstrating the Ability of Clinical and Continuous Glucose Monitoring Metrics to Predict Autonomic Symptom Recognition (Unaware/Aware) and Endogenous Glucose Production Response (Defective/Intact) to Hypoglycemic Clamp Testing

	Prediction of absent AS recognition		Prediction of defective EGP response
	AUC (95% CI)	P	AUC (95% CI)	P
Clarke score	0.81 (0.68–0.93)	0.005	0.51 (0.33–0.69)	0.46
HYPO score	0.79 (0.64–0.93)	0.03	0.67 (0.50–0.84)	0.22
LI	0.51 (0.33–0.69)	0.93	0.69 (0.52–0.85)	0.04
Mean glucose	0.73 (0.52–0.95)	0.09	0.65 (0.42–0.89)	0.29
Glucose SD	0.51 (0.24–0.78)	0.89	0.62 (0.37–0.87)	0.36
CV	0.61 (0.34–0.89)	0.17	0.57 (0.31–0.82)	0.85
%Time <54 mg/dL	0.80 (0.59–1.00)	0.05	0.58 (0.34–0.83)	0.25
%Time <60 mg/dL	0.82 (0.63–1.00)	0.04	0.54 (0.29–0.79)	0.34
%Time <70 mg/dL	0.80 (0.59–1.00)	0.05	0.56 (0.32–0.81)	0.32
%Time in range 70–180 mg/dL	0.53 (0.25–0.81)	0.82	0.52 (0.27–0.77)	0.84
%Time >180 mg/dL	0.70 (0.47–0.93)	0.19	0.62 (0.38–0.86)	0.37
%Time >250 mg/dL	0.72 (0.43–1.00)	0.25	0.70 (0.42–0.98)	0.64
LBGI	0.81 (0.61–1.00)	0.05	0.58 (0.33–0.83)	0.26
HBGI	0.65 (0.41–0.88)	0.20	0.63 (0.38–0.88)	0.48

n = 43 for clinical metrics and n = 23 for CGM metrics.

AUC, area under curve; CI, confidence interval.

A sensitivity analysis was performed to assess the ability of established clinical metric cutoffs to predict clamp response (Table 5). ROC curve derived Youden index thresholds for CGM metrics demonstrated improved specificity, negative predictive value, and accuracy to predict absent AS recognition during the hypoglycemic clamp than clinical metrics, however combining the Clarke score threshold of IAH (</≥4) with CGM metrics further improved the specificity and positive predictive value of the tests. Indeed, composite scores of Clarke ≥4 with %Time <54 mg/dL ≥2.21 and Clarke score ≥4 with %Time <60 mg/dL ≥4.54 demonstrated 89% sensitivity, 87% specificity, 80% positive predictive value, 93% negative predictive value, and 88% accuracy to identify individuals with absent AS recognition to hypoglycemic clamp testing. In contrast, established cutoffs for clinical metrics were poor predictors of absent EGP response and due to the weak ROC curve data further assessment of CGM or composite metric scores to predict an absent EGP response were not pursued.

Table 5.

Sensitivity Analysis Assessing the Ability of Clinical, Continuous Glucose Monitoring, and Composite Metrics to Predict Autonomic Symptom Recognition (Unaware/Aware) and Endogenous Glucose Production Response (Defective/Intact) to Hyperinsulinemic–Hypoglycemic Clamp Testing

	Sensitivity	Specificity	PPV	NPV	Accuracy	P
Prediction of clamp absent AS recognition
Clinical metrics
Clarke score (</≥4)	100	50	66	100	74	<0.001
HYPO score (</≥90th%)	86	64	69	82	74	<0.01
LI (</≥90th%)	71	32	50	54	51	0.82
Composite HYPO/LI (</≥75th%)	86	59	67	81	72	<0.01
CGM metrics
%Time <54 mg/dL ≥2.21	89	73	67	92	79	<0.01
%Time <60 mg/dL ≥4.54	89	73	67	92	79	<0.01
%Time <70 mg/dL ≥9.42	78	87	78	87	83	<0.01
LBGI ≥2.19	78	87	78	87	83	<0.01
Composite metrics
Clarke ≥4 and HYPO >90th%	86	64	69	82	74	<0.001
Clarke ≥4 and %Time <54 mg/dL ≥2.21	89	87	80	93	88	<0.001
Clarke ≥4 and %Time <60 mg/dL ≥4.54	89	87	80	93	88	<0.001
Clarke ≥4 and %Time <70 mg/dL ≥9.42	78	93	88	88	88	<0.001
Clarke ≥4 and LBGI ≥2.19	78	93	88	88	88	<0.001
Prediction of clamp defective EGP response
Clinical metrics
Clarke score (</≥4)	83	37	63	64	63	0.13
HYPO score (</≥90th%)	75	58	69	65	67	0.03
LI (</≥90th%)	79	42	63	62	63	0.14
Composite HYPO/LI (</≥75th%)	75	53	67	63	65	0.07

n = 42 for clinical metrics and n = 23 for CGM metrics. HYPO score (HYPO) 90th%: ≥1047 and 75th%: ≥ 423. LI 90th%: ≥433 and 75th%: ≥ 329.

CGM, continuous glucose monitoring; LI, lability index; NPV, negative predictive value; PPV, positive predictive value.

Discussion

This cross-sectional study utilizes the gold standard paired hyperinsulinemic–hypoglycemic and euglycemic clamp assessment of the counterregulatory response to insulin-induced hypoglycemia to determine practical values of clinical and CGM metrics, which best predict absent AS (hypoglycemia unawareness) and EGP (defective glucose counterregulation) responses in T1D. This evaluation of clinical and CGM metrics is important for interpreting the physiologic significance of these readily available measures as hypoglycemic clamp testing is intensive and not readily performed in many centers even within the research setting. This analysis demonstrates that absent AS recognition during hyperinsulinemic hypoglycemia (terminal glucose of ∼45 mg/dL), may be predicted by IAH (Clarke score) and CGM measured hypoglycemic exposure and, thus, can be readily implemented as an assessment tool in the clinic that may enhance the specificity for identifying those who are likely to have “true hypoglycemia unawareness” and may, therefore, require more urgent attention toward hypoglycemia avoidance interventions, including consideration of β cell replacement.

In our study, the older age and longer duration of diabetes of the IAH Cohorts align closely to that which has been previously reported.³³ The utility of awareness scores in identifying individuals at risk of SH is well established^22,34 and the Clarke score has been previously shown to correlate well with hyperinsulinemic–hypoglycemic clamp symptom response. However, individuals with intermediate symptom and epinephrine response are often misclassified resulting in difficulties with classifying IAH based on one score alone or comparing outcomes between centers.³⁵ The present analysis confirms that a Clarke score ≥4 is an excellent screening tool with high sensitivity for predicting absent AS recognition, supporting current practices to targeted hypoglycemia avoidance. However, the Clarke score also captures a number of individuals with preserved although impaired AS recognition during the hypoglycemic clamp and the specificity to predict absent symptoms, thus those in most urgent need of intervention, is improved by coassessing with hypoglycemia exposure derived from a brief period of CGM.

An exploratory factor analysis by Sepulveda et al., demonstrated that the Clarke score reflects antecedent SH experience in addition to awareness status.³⁶ Indeed, the absence of self-reported SH in the past year has been shown to be highly predictive of hypoglycemic clamp symptom recognition ≥54 mg/dL.³⁷ Consequently, in this analysis, it is important to note that the Clarke score may be a good predictor of absent clamp AS response both through identification of an individual with impaired awareness but also an individual who may have experienced recent SH. The ability of alternate awareness scores, such as Gold score, to predict clamp response was not assessed in this study, however, these scores are known to be highly correlated³³ with the Clarke score demonstrating superior sensitivity in defining IAH through agreement with reduced AS reported on the Edinburgh Hypoglycemia Scale.^35,38 However, as per study inclusion of individuals with IAH by Clarke score and recent SH, our aim was to identify a population with defective counterregulatory responses through HAAF thus, utilizing the Clarke score, which considers an SH component may be of greater value.

In this study we assessed outcomes at the end of the hypoglycemic clamp (target glucose 45 mg/dL), which although consistent with early methodologies,^5,6 is a lower level than used in other reports.³⁷ Yet this approach appears to have contributed toward a higher sensitivity of our metrics to detect true abrogation of AS recognition.³⁷ Indeed, Figure 2C and D demonstrates that counterregulatory AS and EGP responses only begin to separate out at around 54 mg/dL (Fig. 2C, D), the International Hypoglycemia Study Group consensus target glucose to avoid and report based on cardiovascular and mortality data.³² Thus, our data support the clamp methodology of lowering glucose to a final step below 50 mg/dL to determine true absence of AS recognition and EGP responses.

Interestingly, however, despite this methodology, the majority of individuals with IAH did not reach an AS or EGP glycemic threshold by the end of the clamp and although we demonstrated that the magnitude of Step 4 counterregulatory responses were abnormal, imputing the glycemic threshold as the lowest clamp glucose concentration, even if no response was achieved, likely led to an underreporting of differences in glucose thresholds between the IAH and Aware groups. This was accounted for in the study methodology with assessment of metric prediction of the presence/absence of response by the end of the clamp, rather than response at a certain glucose concentration. Any further lowering of glucose below 45 mg/dL was not considered safe.

It is well recognized that antecedent hypoglycemia leads to impaired counterregulatory response to future hypoglycemia and the development of HAAF.^1,6 Our finding of greater CGM hypoglycemia exposure in individuals with IAH defined by Clarke score is consistent with other reports.^39,40 A secondary analysis of CGM outcomes in the HypoDE study demonstrated that measures of CGM hypoglycemia exposure, including: LBGI; percentage and number of glucose readings <54 and <70 mg/dL, can predict the persistence of IAH (Clarke score) and the future recurrence of SH over a 26-week follow-up period.^41,42 Indeed, these metrics were stronger predictors of persisting IAH with areas under the ROC curves of ≥0.78.⁴² Although our analysis presents data of a much smaller cross-sectional study, by including a cohort of individuals with hypoglycemia awareness, we have observed similar findings to Hermanns et al., with metrics of hypoglycemia, including the clinically derived Clarke and HYPO scores representing strong predictors of absent AS recognition to clamp testing (area under ROC curves ≥0.79).

Indeed, similar ROC curve derived cutoffs for percentage of time <54 mg/dL and the LBGI predict the persistence of IAH in the HypoDE study and absent AS response in our study (<54 mg/dL: ≥2.4% vs. ≥2.21%; LBGI: ≥2.0 vs. ≥2.19, HypoDE compared with the present study, respectively).⁴²

The LBGI was devised to assess hypoglycemia risk and stratify future risk over a subsequent 6-month period.⁴³ Published thresholds predicting low, moderate, and high risks were based on glucose self-monitoring-derived LBGI scores and although still of value, CGM-derived scores underestimate risk.⁴⁴ Nevertheless, Hermanns et al., identified CGM-derived LBGI thresholds of >1.6 and >2.0 to predict risk of SH and persistence of unawareness, respectively.⁴² This is consistent with our Aware cohort LBGI median (IQR) of 1.3 (0.7–2.0). In addition, the LBGI in our IAH Cohort is over twofold greater than the Aware group, and ROC curve analysis demonstrates that it is a strong predictor of absent AS recognition with a CGM-derived cutoff ≥2.19 defining 78% sensitivity, 87% specificity, and 83% accuracy to identify an absent symptom response.

Although other groups have reported both higher glucose coefficients of variation in groups of individuals with IAH defined by Clarke and/or Gold score,^39,40 and positive associations between SD and hypoglycemia events,^45,46 variability measures did not differ in this analysis, were poor predictors of counterregulatory response, and are perhaps less predictive in a population with longer duration and high variability at baseline.

Clinical and CGM metrics were unable to predict defective EGP response to hypoglycemic clamp testing. This is perhaps unsurprising as despite normal AS recognition in the Aware group, the EGP response to hypoglycemia was impaired. During hyperinsulinemic–hypoglycemic clamp testing counterregulatory glucagon and epinephrine responses must overcome the inhibitory effect of the hyperinsulinemia to increase glycogenolysis and gluconeogenesis toward subsequent EGP. Consistent with the loss of glucagon response to hypoglycemia observed soon after onset of T1D,¹ cohorts with established T1D do not generate any discernible glucagon response.¹⁵ However, all groups generated an epinephrine response, which was greater in the Aware than IAH Cohorts but less than in nondiabetic controls. Previous groups have reported impairment in epinephrine response in hypoglycemia-aware individuals^15,47 and the epinephrine response appears to be inversely associated with the degree of hypoglycemia exposure (Supplementary Fig. S3),⁵⁶ but weakly correlated with AS that predominate from sympathetic neural origin⁴⁸ (Supplementary Fig. S4).

Epinephrine and glucose counterregulatory EGP response is highly correlated and so the partial epinephrine response in the Aware group reported in this study did not result in an EGP response comparable to the nondiabetic control group. It will be interesting to see whether normalization of EGP response can be attained by improvement of epinephrine secretion during medical intervention studies or will require a restored glucagon response as demonstrated in islet transplant recipients.²⁰ Nevertheless, these data suggest that the EGP response to hypoglycemia is absent in most individuals with IAH and if present, response is marginal and generated at a much lower glucose concentration.

Supporting previous data,^15,48 we demonstrate that normal AS response to hypoglycemia is not dependent on a normal adrenomedullary epinephrine response and AS occur at a higher glucose concentration in Aware than in those with IAH.^5,14 Although our aim was to recruit an Aware group matched to the IAH Cohorts for demographic variables, including diabetes duration, this was a challenge with the resulting Aware group younger with a shorter diabetes duration (duration ≥10 years in all but one individual), known to be protective of IAH.³³ Nonetheless, despite a shorter duration of diabetes in the Aware group, counterregulatory responses were not normal with diminished magnitude of epinephrine, impaired EGP, and a lowered glucose concentration for pancreatic polypeptide response confirming that both sympathetic and parasympathetic neural counterregulatory impairment can occur independent of neuropathy and may progress with diabetes duration.⁴⁹

Indeed, in vivo studies by Whim and others, have shown that repeated hypoglycemia exposure leads to suppression of catecholamine release from adrenal chromaffin cells through depletion of tyrosine hydroxylase activity⁵⁰ that may account for the impaired catecholamine responses observed in this study and implicated in the induction of HAAF.

The limitations of this study include its small size, especially with respect to the CGM metrics. However, we believe the exclusion of the highly variable CGM data derived from early generation CGM has reduced the occurrence of type 2 statistical error. Nonetheless, CGM data were still captured over only a 7-day period of recording with consensus reports recommending at least 14 days of measurement⁵¹ to correlate well with 3-month data.⁵² In addition, it has been shown that capturing reliable time below range may require an even longer duration of monitoring⁵³ although perhaps with only a marginal improvement in estimation.⁵² Utilizing a mathematical approach to REPLACE-BG study data,⁵⁴ Camerlingo, et al., demonstrated a 2.1% SD of time in hypoglycemia estimation error for a 7-day CGM period.⁵⁵ However, this population had low risk of SH with two-thirds reporting a Clarke score of 0 at baseline^54,55 and it is likely that estimation error may be reduced in a population with higher exposure.⁵³ The present study, demonstrates proof of concept to support the screening and identification of individuals with defective symptom response.

Indeed, the alignment of the CGM-derived thresholds with the masked 28-day CGM data obtained in the HypoDE analysis of individuals with IAH and/or recent SH is supportive.⁴² Nonetheless, further validation of CGM duration with hypoglycemia thresholds derived from a longer period of observation (e.g., 14 days) alongside prospective collection of SH events would be of value in a larger sample.

Focused on clinical predictors of HAAF in long-standing T1D, we did not assess individuals with short diabetes duration. Supplementary Figure S5 demonstrates a negative association between counterregulatory response and diabetes duration, and it is likely that counterregulatory responses in Aware individuals with shorter diabetes duration are preserved, however, the extent to which these metrics predict defective symptom and EGP response in those with IAH and short diabetes duration warrant further study. As a cross-sectional study, we are unable to comment on pathophysiological processes underlying the development of HAAF, associations with SH, or whether the avoidance/occurrence of a certain degree of hypoglycemia exposure would abolish/contribute to defective counterregulatory symptom or EGP responses.

Neuroglycopenic symptom measures were not collected in this study so any confounding influence of preserved/absence of neuroglycopenic symptoms on participant responses to the AS questionnaire cannot be excluded; however, the nearly identical AS responses observed in the Aware cohort and nondiabetic controls make this unlikely. Nevertheless, this study has benefitted from robust collection of clamp data under equivalent methodology, including the use of euglycemic clamp control experiments across study cohorts. In addition, study numbers are similar to those presented in earlier clamp studies.¹⁵

This study demonstrates that absent AS recognition can be predicted by IAH (Clarke score) and recent hypoglycemia exposure on CGM. Defective EGP response could not be predicted, however, is abnormal in the majority of individuals with long-standing T1D regardless of awareness status. In the clinical setting, screening for IAH by Clarke score and evaluating hypoglycemia exposure on CGM can identify individuals with defective counterregulatory responses to hypoglycemia who would benefit from further technology approaches or consideration of β cell replacement toward the reversal of HAAF.

Authors' Contributions

A.J.F. was responsible for data analysis and preparing the first draft of the article. E.C. assembled the dataset and performed initial exploratory analyses. A.J.P. and C.D.-B. performed study visits and along with E.C. and H.L.N. were responsible for data collection. H.L.N., H.W.C., J.S.M., and R.J.G. supported data analysis. M.R.R was responsible for study design and is the guarantor of this study, has full access to the data, and takes responsibility for the integrity and accuracy of the data analysis. All authors reviewed and edited the article.

Author Disclosure Statement

No competing financial interests exist.

Funding Information

This work was funded by Public Health Services research grants R01 DK091331 (to M.R.R.), U01 DK070430 (University of Pennsylvania Center for Clinical Islet Transplantation), UL1 TR001878 (University of Pennsylvania Center for Human Phenomic Science), P30 DK19525 (University of Pennsylvania Diabetes Research Center Radioimmunoassay and Biomarkers Core), and T32 DK00734 (University of Pennsylvania Training Grant in Diabetes, Endocrinology, and Metabolic Diseases); the Charles B. Humpton, Jr. Endowed Fellowship in Diabetes Research (to A.J.F.); and the Human Metabolism Resource and Metabolic Tracer Resource of the University of Pennsylvania Institute for Diabetes, Obesity, and Metabolism.

Supplementary Material

Supplementary Figure S1

Supplementary Figure S2

Supplementary Figure S3

Supplementary Figure S4

Supplementary Figure S5