Association of impaired awareness of hypoglycemia and real-time continuous glucose monitoring with the occurrence and frequency of severe hypoglycemia in adults with type 1 diabetes mellitus: a real-world panel data analysis from the PR-IAH study

Abstract

Background

Patients with long-standing type 1 diabetes mellitus (T1DM) often experience severe hypoglycemia (SH); however, the protective and risk factors that influence its occurrence and frequency in the era of advanced diabetes technology remain unclear. This study aimed to investigate the association of impaired awareness of hypoglycemia (IAH) and real-time continuous glucose monitoring (rtCGM) with the incidence and frequency of SH in adults with T1DM.

Methods

This prospective, observational study recruited 311 adults with T1DM (mean age: 50.6 ± 14.7 years; male: 37.9%; diabetes duration: 17.9 ± 11.3 years; mean HbA1c: 7.7 ± 1.0%) from seven diabetes centers across Japan. All participants were aged ≥ 20 years and had been diagnosed with type 1 diabetes for at least 1 year. The primary outcomes were the incidence and frequency of SH, defined as an episode of hypoglycemia necessitating assistance from others. Logistic and Poisson fixed- or random-effects models were selected using the Hausman test and were applied to analyze the data. Data are presented as coefficients with 95% confidence intervals (CIs).

Results

The prevalence of SH was 5.4 (95% CI 3.6–7.7)/100 person-years during the 2-year follow-up period. The logit random-effects model for predicting the occurrence of SH revealed that diabetic peripheral neuropathy (DPN) tended to be associated with an increased risk (coefficient: 2.01, 95% CI − 0.02 to 4.04; P = 0.053), whereas IAH (coefficient: 1.08, 95% CI 0.49 to 1.66; P < 0.001) exhibited a significant association with an increased risk. rtCGM (coefficient: − 1.75, 95% CI − 2.49 to − 1.00; P < 0.001) was associated with a reduced risk. The Poisson random-effects model for predicting the frequency of SH revealed that DPN and the IAH score (coefficient: 0.21, 95% CI 0.06 to 0.35; P = 0.006) exhibited positive associations with an increased frequency of SH, whereas rtCGM (coefficient: − 1.60, 95% CI − 2.84 to − 0.37; P = 0.011) was associated with a reduced frequency of SH.

Conclusion

This panel data analysis demonstrated that IAH was associated with an increased incidence and frequency of SH, whereas rtCGM was associated with a decreased incidence and frequency of SH.

Trial registration

University Hospital Medical Information Network (UMIN) Center: UMIN000039475), approval date: February 13, 2020.

Background

Severe hypoglycemia (SH) is defined as an episode of hypoglycemia necessitating assistance from another individual for the administration of carbohydrates, glucagon, or other corrective measures [1, 2]. SH is associated with poor glycemic control and an increased risk of neuroglycopenic symptoms in patients with type 1 diabetes mellitus (T1DM), which can lead to cognitive impairment, confusion, coma, and death [3–8]. Impaired awareness of hypoglycemia (IAH), a phenomenon commonly observed among patients with T1DM, results in a significant increase in the risk of SH and a negative impact on the quality of life and glycemic control [9]. Notably, the physiological deficiencies in patients with T1DM impair the ability to counteract decreasing blood glucose levels, thereby predisposing them to hypoglycemia. Repeated episodes of hypoglycemia can lead to IAH, a condition characterized by a diminished ability to recognize hypoglycemia and take corrective action [10].

Advanced diabetes technologies such as continuous glucose monitoring (CGM) and sensor-augmented insulin pumps with low-glucose suspension systems have led to a decrease in the frequency and severity of hypoglycemia without compromising glycemic control [11]. Nevertheless, concerns regarding recurrent SH persist among patients with T1DM despite significant advancements in treatment options such as CGM and continuous subcutaneous insulin infusion (CSII) [12]. Furthermore, the risk and protective factors affecting the occurrence and frequency of SH remain underexplored in the era of advanced diabetes technology.

A careful evaluation of SH is necessary owing to its rarity, even among patients with T1DM. Survival analyses such as Kaplan–Meier analysis and Cox regression have been conducted using only baseline data in cohort studies. Thus, capturing the temporal changes over time is difficult as the individual samples are not fixed. This limitation hinders the effective analysis of dynamic variations [13–15]. Panel data analysis offers several key benefits, such as enhancing causal analysis to provide more accurate causal inferences than cross-sectional data. Expansion of the dataset will facilitate precise estimation, thereby yielding more reliable statistical estimates. Notably, panel data analysis enables the study of long-term trends and dynamic relationships by facilitating the capture of temporal effects. Furthermore, it also mitigates multicollinearity, thereby reducing correlation issues between variables over time [16, 17].

In contrast to cohort studies, which can assess baseline effects, panel data analysis facilitates the analysis of the effects of the changes in variables over time on the outcomes. Thus, panel data analysis is suitable for evaluating the impact of changes in IAH and real-time continuous glucose monitoring (rtCGM). This study investigated the association of IAH and rtCGM with the incidence and frequency of SH in adult patients with T1DM.

Materials and methods

The National Hospital Organization (NHO) Central Research Ethics Committee (R2-0117002) approved the protocol of this observational study. Adult patients with T1DM were recruited from seven NHO collaboration centers across Japan between February 2020 and March 2022. Individuals aged ≥ 20 years who had been diagnosed with T1DM for > 1 year and were receiving care from one of the collaborating centers were eligible for inclusion. The recruitment process was continuous, and no fixed number of participants were recruited from each center. Certified diabetologists diagnosed diabetic retinopathy, diabetic nephropathy, and diabetic peripheral neuropathy (DPN) in accordance with the diabetes treatment guidelines. Patients who satisfied at least two of the following criteria were diagnosed with DPN: subjective symptoms in the lower limbs or feet, loss of the ankle jerk reflex, and decreased perception of vibration measured bilaterally at the medial malleoli using a C128 tuning fork [18]. The coefficient of variation of R-R intervals (CV-RR) was calculated, with a value < 2% considered indicative of diabetic cardiac autonomic neuropathy. The corrected QT (QTc) interval was computed using Bazett’s formula, and a QTc > 440 ms was classified as prolonged. SH was defined as an event necessitating assistance from another individual for the active administration of carbohydrates or glucagon or taking corrective measures [19]. The Gold, Clarke, and Pedersen–Bjergaard scores were used to assess IAH. The response to the question “Do you recognize the beginning of your hypoglycemic episodes?” was rated on a seven-point Likert scale (from 1 = ”always aware” to 7 = ”never aware”) in the Gold method. A score of ≥ 4 indicated IAH [20]. Data regarding diabetes treatments, including CSII and the use of sensor-augmented pumps (SAPs), were collected. In addition, data regarding glucose monitoring methods, such as intermittently scanned CGM (isCGM) and rtCGM, were also collected.

Fisher’s exact test was used to assess qualitative variables. Unpaired t test or the Mann–Whitney U test was used to assess quantitative variables, depending on the normality of data distribution. The Hausman test was used to differentiate between fixed- and random-effects. A logit fixed- or random-effects model was used to assess the association between explanatory variables and a binary outcome. A Poisson fixed or random-effects model was used to analyze the association between these variables and the frequency of rare events after adjusting for covariates. Model 1 was used to calculate the coefficient for CSII after adjusting for DPN and IAH. Model 2 was used to calculate the coefficient for SAP. Model 3 was used to calculate the coefficient for rtCGM. Participants with missing data were excluded from the analysis. Statistical significance was set at P < 0.05. Low P values (0.05 ≤ P < 0.10) indicated a “trend toward statistical significance.” All statistical analyses were conducted using R version 4.1.2 and Stata software version 16.

Results

A total of 311 adult patients with T1DM (mean age: 50.6 ± 14.7 years; males: 37.9%; duration of diabetes: 17.9 ± 11.3 years; mean glycated hemoglobin [HbA1c] levels: 7.7 ± 1.0%) were enrolled in the present study (Table 1). The prevalence of IAH at baseline was 12.5%. The usage rates of CSII, SAP, isCGM, and rtCGM at baseline were 35.0, 20.3, 31.2, and 22.2%, respectively. No significant differences were observed between the groups at baseline in terms of age, male ratio, duration of diabetes, HbA1c levels, and SAP or CGM usage (isCGM and rtCGM). However, the ratio of participants with DPN who had IAH at baseline was higher (31.2 vs. 15.7%, P = 0.045). In contrast, the CSII usage was significantly lower (20.5 vs. 37.1%, P = 0.048).

Table 1.

Baseline characteristics of the participants (n = 311)

Variables	Mean ± SD or %
Age, years	50.6 ± 14.7
Male, %	37.9
Diabetes duration, years	17.9 ± 11.3
HbA1c, %	7.7 ± 1.0
Diabetic complications
Diabetic peripheral neuropathy, %	17.5
Diabetic retinopathy, %	24.7
Diabetic nephropathy,	21.4
Impaired awareness of hypoglycemia, %	19.1
Glucose monitoring
isCGM, %	31.2
rtCGM, %	22.2
Treatment
CSII, %	35.0
SAP, %	20.3
ECG
QTc (Bazett’s formula), ms	417 ± 29
> 440 ms, %	17.1
CV-RR, %	3.3 ± 1.8
< 2%, %	23.0

Mean ± standard deviation SD) or number (percent)

CSII continuous subcutaneous insulin infusion, CV-RR coefficient of variance of the heart rate variation, ECG electrocardiogram, IAH impaired awareness of hypoglycemia, isCGM intermittently scanned continuous glucose monitoring, rtCGM real-time continuous glucose monitoring, SAP sensor-augmented pump, QTc corrected QT interval

The prevalence of SH was 5.4 (95% CI 3.6–7.7)/100 person-years during the 2-year follow-up period. The univariable logistic random-effects model revealed that DPN and IAH were associated with an increased risk of developing SH. In contrast, SAP usage and rtCGM were associated with a decreased risk of developing SH. Notably, HbA1c, CSII, and isCGM exhibited no association with the incidence of SH.

No significant association was found between the presence of retinopathy or nephropathy and the development of SH. At baseline, there were no significant differences in the proportion of individuals with CV-RR < 2% or QTc > 440 ms between the groups with and without SH.

Model 1, which was used to predict the incidence of SH using a logit random-effects model, revealed that DPN and IAH exhibited significant associations with an increased risk. In contrast, CSII exhibited no such association. Model 2 revealed that DPN and IAH exhibited significant associations with an increased risk, whereas SAP usage was associated with a reduced risk. Model 3 revealed that DPN tended to exhibit an association with an increased risk (coefficient: 2.01, 95% CI − 0.02 to 4.04; P = 0.053), whereas IAH (coefficient: 1.08, 95% CI 0.49 to 1.66; P < 0.001) exhibited a significant association with an increased risk. Notably, rtCGM use (coefficient: − 1.75, 95% CI − 2.49 to − 1.00; P < 0.001) was associated with a reduced risk (Fig. 1 and Table 1).

Fig. 1.

Logit random-effects model for predicting the incidence of severe hypoglycemia. a Model 1; b Model 2; c Model 3. Coefficients are presented with 95% confidence intervals (Coef [95% CI]). *P < 0.05. ^†0.05 ≤ P < 0.10. CI confidence interval, csii continuous subcutaneous insulin infusion, dpn diabetes peripheral neuropathy, iah impaired awareness of hypoglycemia, rtcgm real-time continuous glucose monitoring, SAP sensor-augmented pump. Model results (logit random-effects model)—Model 1: dpn 2.05 (0.17, 3.92)*; iah 1.02 (0.59, 1.44)*; csii − 1.47 (− 3.26, 0.32), Model 2: dpn 2.06 (0.02, 4.11)*; iah 1.06 (0.44, 1.69)*; sap − 2.20 (− 2.95, − 1.46)*, Model 3: dpn 2.01 (− 0.02, 4.04)†; iah 1.08 (0.49, 1.66)*; rtcgm − 1.75 (− 2.49, − 1.00)*

Table 2.

Predictors of severe hypoglycemia occurrence and frequency in adults with type 1 diabetes

Variable	SH occurrence		SH frequency
	Coef. (95% CI)	P value	Coef. (95% CI)	P value
HbA1c	0.04 (– 0.41, 0.0.48)	0.875	0.25 (– 0.14, 0.64)	0.215
DPN	2.27 (0.37, 4.16)	0.019*	0.86 (0.02, 1.69)	0.044*
IAH/IAH score	4.07 (2.14, 6.00)	< 0.001	0.22 (0.08, 0.36)	0.002*
CSII	– 0.79 (– 1.89, 0.31)	0.158	– 0.87 (– 1.82, 0.08)	0.072†
SAP	– 1.22 (– 2.00, – 0.44)	0.002*	– 1.09 (– 2.32, 0.13)	0.080†
isCGM	– 0.35 (– 1.25, 0.55)	0.440	– 0.03 (– 0.72, 0.66)	0.931
rtCGM	– 1.1 (– 1.28, – 0.91)	< 0.001*	– 1.07 (– 2.23, 0.09)	0.070†

Univariable logit and Poisson random-effects models. Coefficient (95% CI)

*P < 0.05. ^†0.05 ≤ P < 0.10.

CI confidence interval, Coef coefficient, CSII continuous subcutaneous insulin infusion, IAH impaired awareness of hypoglycemia, isCGM intermittently scanned continuous glucose monitoring, rtCGM real-time continuous glucose monitoring, SAP sensor-augmented pump

Model 1, which used a Poisson random-effects model to predict the frequency of SH, revealed that DPN and the IAH score exhibited positive associations with an increased frequency of SH, whereas CSII exhibited a negative association with the frequency of SH. Model 2 revealed that DPN and the IAH score exhibited positive associations with an increased frequency of SH, whereas SAP usage was associated with a reduced frequency of SH. Model 3 revealed that DPN and the IAH score (coefficient: 0.21, 95% CI 0.06 to 0.35; P = 0.006) exhibited positive associations with an increased frequency of SH, whereas rtCGM (coefficient: − 1.60, 95% CI − 2.84 to − 0.37; P = 0.011) was associated with a reduced frequency of SH (Fig. 2 and Table 2).

Fig. 2.

Poisson random-effects model for predicting the frequency of severe hypoglycemia. a Model 1; b Model 2; c Model 3. Coefficients are presented with 95% confidence intervals (Coef [95% CI]). *P < 0.05. ^†0.05 ≤ P < 0.10. CI confidence interval; csii continuous subcutaneous insulin infusion, dpn diabetes peripheral neuropathy, iah impaired awareness of hypoglycemia-gold score, rtcgm real-time continuous glucose monitoring; SAP, sensor-augmented pump. Model results (Poisson random-effects model): Model 1: dpn 0.95 (0.12, 1.74)*; iahgs 0.19 (0.05, 0.33)*; csii − 1.54 (− 2.56 0.53)*, Model 2: dpn 0.88 (0.07, 1.68)*; iahgs 0.21 (0.07, 0.36)*; sap − 1.82 (− 3.15, − 0.49)*, Model 3: dpn 0.83 (0.03, 1.62)*; iahgs 0.21 (0.06, 0.35)*; rtcgm − 1.60 (− 2.84, − 0.37)*

Discussion

The present study is the first to demonstrate that IAH increases the incidence and frequency of SH in adult patients with T1DM, whereas rtCGM decreases the incidence and frequency of SH in adult patients with T1DM through panel data analysis.

DPN and IAH

The present study revealed that DPN is associated with an increased risk of SH. Previous cross-sectional and observational studies have indicated an association between DPN and SH [21–23]. Flatt et al. reported that peripheral neuropathy was more prevalent among individuals with SH than those without SH during the 24-month follow-up of the HypoCOMPASS study [24]. Furthermore, DPN has been associated with cognitive impairment among adult patients with T1DM [25]. Adult patients with T1DM and IAH often exhibit autonomic symptoms [26], which may contribute to diminished awareness of hypoglycemia. Animal studies have reported that repeated episodes of hypoglycemia can induce DPN [27]. IAH increased the incidence and frequency of SH in the present study. A retrospective analysis of adult patients with T1DM from the T1DM Exchange Registry/online community revealed that SH and IAH continued to occur, irrespective of the usage of CGM or automated insulin delivery (AID) [28]. Further large-scale, long-term studies must be conducted in the future to explore these associations and their clinical implications.

In this study, we assessed the presence of DPN; however, international diagnostic criteria classify DPN into categories such as possible, probable, and confirmed, and also emphasize the evaluation of severity and symptom burden using numeric rating scales [29–31]. Future studies incorporating standardized severity assessments and symptom rating scales are needed to validate and expand upon these findings.

rtCGM and SAP

rtCGM was associated with a reduced incidence and frequency of SH in the present study, whereas isCGM was not. Mustonen et al. conducted a retrospective study and reported that the isCGM reduced the incidence of SH requiring emergency medical services among 642 patients with T1DM in Finland [32]. Similarly, the WISDM randomized controlled trial (RCT) demonstrated that compared with standard self-monitoring of blood glucose monitoring (SMBG), rtCGM resulted in a modest but statistically significant reduction in overall hypoglycemia and a decrease in the incidence of SH episodes (10 in the CGM group vs. 1 in the SMBG group) over a 6-month period among 203 older adults with T1DM in the United States [33, 34]. In contrast, an RCT conducted by Davis et al. revealed that compared with standard SMBG, isCGM was not effective in reducing the risk of recurrent SH among 59 individuals with T1DM or insulin-treated type 2 diabetes mellitus in Australia [35]. The ALERTT1 study conducted by Visser et al. demonstrated that switching from isCGM to rtCGM reduces the risk of incidence of SH [36]. Hásková et al. reported that rtCGM was superior to isCGM in terms of reducing the risk of hypoglycemia and that rtCGM was associated with improvements in the duration of diabetes and awareness of hypoglycemia among patients with T1DM [37]. The HypoDE study conducted by Heinemann et al. demonstrated that rtCGM reduced the number of hypoglycemic events among patients with T1DM and IAH or a history of SH receiving multiple daily injections [38]. The variations across the findings of these studies may be attributed to the differences in study design and population characteristics.

CSII exhibited no association with a reduced incidence of SH in the present study; however, it exhibited a negative correlation with the frequency of SH. Wohland et al. suggested that CSII may lower the risk of SH [39]. Sensor-augmented insulin pump therapy with predictive low-glucose suspension can potentially reduce the risk of recurrent SH [40–43]. Further research on the use of hypoglycemic alerts and trend arrows [40] must be conducted to confirm these findings.

The JDRF-CGM study reported a reduction in SH incidence from 27.7 to 15.0 events per 100 person-years during the 6-month follow-up period; however, this difference did not reach statistical significance (P = 0.08) [44]. Similarly, Lin et al. [45] conducted a cross-sectional study involving 135 patients with type 1 diabetes using rtCGM and found a high prevalence of IAH, which was associated with increased risk of SH. Although their findings suggest a potential benefit of rtCGM in reducing SH, causality cannot be inferred due to the cross-sectional nature of the study. Importantly, Lin et al.’s analysis did not adjust for other key confounders, such as DPN, which may also contribute to SH risk. In contrast, our study employed a longitudinal cohort design with repeated measures, allowing us to capture changes in complications and treatments over time. Using panel analysis (random-effects logit and Poisson models), we were able to account for within-subject variability and assess independent risk factors more robustly. We believe this methodological strength enhances the validity of our findings compared to prior cross-sectional studies. Strengths of this study include its longitudinal cohort design and the application of panel data analysis using random-effects models, which enabled the evaluation of within-individual changes in treatment modalities and diabetes-related complications over time. This methodological approach offers more robust evidence for identifying independent risk factors for SH compared to cross-sectional studies. Furthermore, our analysis adjusted for key clinical variables, including DPN and IAH, enhancing the validity of the findings.

Limitations of the study

The strength of this study lies in the use of a validated self-administered questionnaire and a Poisson random-effects model, which is well-suited for analyzing rare events. Nevertheless, the present study has some limitations. IAH was assessed using the Gold method, despite the availability of other validated methods, such as the Clarke, Pedersen, and HypoA-Q methods [46, 47]. Further evaluation using the Clarke, Pedersen, and HypoA-Q methods is needed to clarify these issues. Furthermore, the C-peptide levels were not measured in the present study, despite evidence suggesting that residual C-peptide is associated with new-onset and persistent IAH among individuals with T1DM [48]. Further studies incorporating C-peptide measurements must be conducted in the future to investigate these associations and confirm the findings of the present study.

A meta-analysis by Teo et al. (2022), which included 21 studies with a total of 2,149 individuals with T1DM, demonstrated that CGM significantly reduced HbA1c levels compared to SMBG, with a mean difference of – 0.23% [49]. However, CGM did not significantly affect the incidence of SH (P = 0.13). With respect to severe hypoglycemia, meta-analysis by Pala et al. (2019) showed that CSII did not produce a significant reduction of risk in comparison with traditional insulin injections. Among 926 adults with type 1 diabetes from the T1D Exchange Registry, AID users spent more time in range; however, 27.9% did not meet TIR targets, 15.5% experienced severe hypoglycemic events, and 16.0% had CGM-detected level 2 hypoglycemia [50]. These findings suggest that despite the use of advanced diabetes technologies, a substantial proportion of individuals still struggle to achieve optimal glycemic control and remain at risk for SH. Future studies that include sensor adherence are needed to further address this issue.

In this study, SH was defined based on self-reported episodes in accordance with established guidelines; however, detailed information regarding episode severity—such as hospitalization, confirmed glucose levels, or treatment modalities—was not clearly documented. Therefore, further evaluation incorporating objective indicators such as ambulance use, hospitalization records, blood glucose measurements, and medical interventions including intravenous glucose or nasal glucagon administration is warranted to validate and characterize the severity of SH more precisely.

Conclusion

This panel data analysis revealed that IAH was associated with an increased incidence and frequency of SH, whereas the rtCGM was associated with a decrease in the incidence and frequency of SH. The findings of the present study suggest that rtCGM may help reduce the incidence and frequency of SH. DPN and IAH were associated with an increased incidence of SH. Switching from SAP to AID systems is feasible and safe in patients with T1DM at high risk of developing hypoglycemia [51, 52]. Diabetes education programs, such as blood glucose awareness training and the use of trend arrows, in combination with rtCGM and AID have shown potential benefits in the prevention of SH [53–55]. Further studies must be conducted in the future to gather insights into IAH and evaluate targeted interventions aimed at preventing and reducing the frequency of SH in high-risk individuals.

Author contributions

All authors contributed to the study conception and design. Material preparation, data collection and analysis were performed by Naoki Sakane, Ken Kato, Sonyun Hata, Erika Nishimura, Rika Araki, Kunichi kouyama, Masako Hatao, Yuka Matoba, Yuichi Matsushita. Naoki Sakane and Akiko Suganuma wrote the original draft. Takashi Murata, Seiko Sakane, Masayuki Domichi reviewed and wrote the article. All authors read and approved the final manuscript.

Funding

The PR-IAH study received funding from the National Hospital Organization Clinical research (NHO) (Grant number: H31-NHO (Endocrinology and Nephrology)-01).

Data availability

The datasets generated and/or analyzed during the present study are not publicly available; however, they may be obtained from the corresponding author upon reasonable request.

Declarations

Conflict of interest

The authors declare no conflicts of interest associated with this manuscript.

Ethical approval

This study conformed to the standards of the Declaration of Helsinki. Approval of the research protocol: This study was approved by the National Hospital Organization Central Review Board (NHOCRB/ R2-0117002, Feb. 7, 2020). Informed consent or substitute for it was obtained from all patients for being included in the study. Approval date of Registry and the Registration No. of the study/trial: Trial registration number: University hospital Medical Information Network (UMIN) Center: UMIN000039475).

Informed consent

Informed consent or substitute for it was obtained from all patients for being included in the study.

Animal studies

Not applicable.