ABSTRACT
Background
Severe hypoglycemia is a major acute complication of type 1 diabetes (T1D) and is associated with increased morbidity, mortality, and impaired quality of life. Identifying individuals at the highest risk remains essential for optimising preventive strategies in real‐world practice.
Methods
The SEHYPAN (SEvere HYpoglycemia in ANdalusia) study was a multicenter case–control analysis including adults with T1D treated with multiple daily insulin injections (MDI). Cases were individuals who required pre‐hospital emergency care for severe hypoglycemia between 2018 and 2022, each matched by sex, age, glucose‐monitoring method (SMBG/isCGM), reference health‐care area with controls who had not experienced severe events. Logistic regression models were used to identify independent predictors, and nomograms were generated for individualised risk estimation.
Results
A total of 1464 participants were analysed (799 cases and 665 matched controls). Cases had longer diabetes duration, more comorbidities, and higher rates of smoking and alcohol use (all p < 0.001). Two nomogram‐based models were developed: one for the overall cohort, including glucose monitoring modality, history of severe and nocturnal hypoglycemia, comorbid depression, alcohol use, and chronic conditions; and another specific to isCGM users, which also incorporated time in range and time below range. In the full cohort model, isCGM use was independently associated with lower odds of severe hypoglycemia. Both models showed good discrimination (AUC 0.75–0.83) and high sensitivity (≥ 0.75).
Conclusions
The SEHYPAN study identifies key predictors of severe hypoglycemia in adults with T1D on MDI therapy. The innovative nomogram‐based models provide personalised risk estimates that may enhance preventive care in everyday clinical practice.
Keywords: intermittently‐scanned continuous glucose monitoring, multiple daily insulin injections, nomogram, predictive model, severe hypoglycemia, type 1 diabetes mellitus
1. Introduction
Hypoglycemia is one of the most frequent acute complications of type 1 diabetes (T1D) and exerts a major influence on the overall quality of metabolic control [1]. It is commonly defined by a blood glucose threshold below 70 mg/dL, although it may also be classified according to symptom severity (mild, moderate, or severe). According to the American Diabetes Association (ADA), severe hypoglycemia corresponds to Level 3 events, characterised by altered mental and/or physical status requiring external assistance, irrespective of glucose level [2].
From an epidemiological perspective, the reported incidence of severe hypoglycemia varies according to the population and data sources considered [3, 4]. In a population‐based Spanish study focussing on events requiring emergency medical services, the incidence among individuals with diabetes was estimated at 810 episodes per 10,000 person‐years, underscoring the substantial burden of severe hypoglycemia in real‐world settings [5]. Annual prevalence rates of severe hypoglycemia as high as 35% have been reported in T1D populations [6], consistent with data from healthcare centres in our region [7]. Importantly, severe hypoglycemia is not limited to a minority of highly vulnerable patients, but remains a common problem across different age groups and treatment regimens. The burden of these events extends beyond their immediate clinical consequences, as they often lead to hospital admissions, increased use of emergency services, and long‐term impairment in patients' confidence in managing their diabetes [5]. Beyond the clinical burden, hypoglycemia is also associated with substantial healthcare costs, including both direct medical expenditures and indirect costs [8].
Despite its clinical relevance, robust evidence regarding the identification of risk factors for severe hypoglycemia, and the development of validated strategies to manage individuals at high risk, remains limited. This is partly due to the lack of a universally accepted definition of ‘severe hypoglycemia,’ and the frequent exclusion of high‐risk individuals from large T1D studies [9]. Moreover, while some risk factors have been proposed—such as longer disease duration, intensive insulin regimens, impaired hypoglycemia awareness, or comorbid conditions—most studies are constrained by methodological variability and limited external validity [10].
The SEHYPAN (SEvere HYpoglycemia in ANdalusia) study was designed to identify predictors of severe hypoglycemia among adults with T1D treated with MDI within the Andalusian Public Health System (APHS). The objective of our study was therefore to analyse and identify risk factors for severe hypoglycemia in adults with T1D treated with MDI, and to propose practical approaches for recognising patients at the highest risk. By focussing on a large real‐world cohort, we aimed not only to validate previously suggested determinants but also to refine patient stratification tools that could support the updating of preventive management strategies in clinical practice.
2. Methods
2.1. Study Design and Participants
We conducted an observational case–control study using data from the CES‐061 registry, which systematically records the activity of pre‐hospital emergency medical services (EMS) in Andalusia. CES‐061 is the official EMS provider within the APHS, responsible for urgent and emergency care in the out‐of‐hospital setting across the region (Figure 1).
FIGURE 1.
Study population and inclusion criteria. Flow diagram showing the inclusion and matching criteria. Adults with type 1 diabetes mellitus (T1DM) on multiple daily insulin injections (MDI) who required pre‐hospital emergency medical services (EMS) for severe hypoglycemia between 2018 and 2022 were identified as cases (n = 799). Controls (n = 665) were matched by sex, age (± 5 years), glucose‐monitoring method (self‐monitoring of blood glucose [SMBG] or intermittently scanned continuous glucose monitoring [isCGM]), and reference health‐care area, with no history of severe hypoglycemia during the study period or the previous 5 years. EMS, Emergency medical services; isCGM, intermittently scanned continuous glucose monitoring; MDI, Multiple daily insulin injections; SMBG, Self‐monitoring of blood glucose; T1DM, Type 1 diabetes mellitus.
Cases were defined as adults (≥ 18 years) with T1D treated with MDI who experienced a severe hypoglycemic event between 1 January 2018, and 31 December 2022. Eligible events required EMS attendance by CES‐061, with confirmation of hypoglycemia in situ through capillary glucose measurement, and provision of on‐site medical care by an emergency team. Exclusion criteria for cases were: (1) pregnancy, (2) treatment with continuous subcutaneous insulin infusion (CSII), and (3) incomplete clinical data regarding pre‐hospital care.
Controls were identified from hospital‐based registries of patients with T1D at each participating centre within the APHS. For each case, one control was selected (1:1 ratio) according to predefined matching criteria: (1) same sex, (2) age within ± 5 years of the index case, (3) same glucose‐monitoring method (self‐monitoring of blood glucose [SMBG] or intermittently‐scanned Continuous Glucose Monitoring [isCGM]), and (4) same healthcare area of reference. Additional requirements for controls included: confirmed diagnosis of T1D, treatment with MDI, and no record of severe hypoglycemia either during the study period or in the preceding 5 years. Exclusion criteria for controls were incomplete clinical data in the variables required for analysis.
2.2. Variables
Data collection encompassed several domains. Sociodemographic and basic clinical characteristics were obtained from electronic medical records. Information on comorbidities and concomitant pharmacological treatments was collected from the User Database of the Andalusian Public Health System, which integrates longitudinal data on active prescriptions and chronic disease diagnoses. These variables were used to estimate the total number of chronic conditions and prescribed drugs per patient. Specific details regarding the index severe hypoglycemic episode were extracted from the CES‐061 registry, which documents all pre‐hospital emergency medical attendances in Andalusia.
Metrics of glycaemic control were retrieved from glucose‐monitoring systems, particularly isCGM, when available. These included a detailed ambulatory glucose profile (AGP) parameter, including time in range, time below and above range, glucose variability (CV), number and duration of hypoglycemic episodes, and sensor use.
2.3. Statistical Analysis
Descriptive statistics were used to summarise the baseline characteristics of cases and controls. Continuous variables were expressed as mean ± standard deviation (SD) and compared using Welch's t test. Categorical variables were expressed as number (percentage) and compared using chi‐square tests.
To develop predictive tools for severe hypoglycemia, we constructed two separate multivariable logistic regression models. The first model included the entire study population. The second model was restricted to participants using isCGM, allowing the incorporation of ambulatory glucose profile (AGP) parameters into the analysis. For the latter model, only AGP records with at least 70% sensor wear time were considered valid and included.
For each model, the dataset was randomly split using stratified partitioning into a training cohort (70%) and a validation cohort (30%) (Supporting Information S1). A binary logistic regression model was then developed in the training set. Variables included in the models were selected according to their statistical significance in the univariate analysis and their clinical relevance, with the objective of developing robust yet parsimonious predictive models. From these models, a nomogram was generated to provide a graphical tool for risk prediction. The nomogram scoring system was derived by proportionally scaling the regression coefficients (β) from the final multivariable logistic model. Each predictor was assigned a number of points according to its relative contribution to the outcome, and the total score was subsequently converted into an estimated probability of severe hypoglycemia using the logistic function. To facilitate clinical use, calibration curves were plotted to show the relationship between total score (x‐axis) and predicted probability (y‐axis).
Model performance was evaluated in the validation cohort. Discrimination was assessed using receiver operating characteristic (ROC) curves and the area under the curve (AUC) with 95% confidence intervals. The optimal cutoff point for risk stratification was determined by applying the Youden index to the predicted probabilities generated by the model and identifying the threshold that maximises sensitivity and specificity. Calibration was examined by comparing observed versus predicted probabilities, complemented by statistical indices including the Brier score, calibration intercept, and slope, in addition to graphical inspection of calibration plots.
All statistical tests were two‐tailed, and a p value < 0.05 was considered statistically significant. The analyses were performed using R software (version 4.5.0; R Foundation for Statistical Computing, Vienna, Austria). Key packages included tidyverse for data management, tableone and rstatix for descriptive and univariate analyses, rms for logistic regression and nomogram construction, and pROC for ROC curve analysis.
2.4. Ethics
This study was conducted in accordance with the principles of the Declaration of Helsinki. The protocol was reviewed and approved by the Coordinating Committee on Biomedical Research Ethics of Andalusia (approval code: 0674‐N‐21).
3. Results
3.1. Univariate Analyses and Glycaemic Control
A total of 799 patients with severe hypoglycemia (cases) and 665 matched controls were included in the study. By design, the two groups were comparable in age (44.9 vs. 45.2 years, p = 0.716), sex distribution (58.8% vs. 57.7% male, p = 0.716), and use of isCGM (65.2% vs. 62.9%, p = 0.388), with no statistically significant differences in these matching variables (Table 1).
TABLE 1.
Baseline characteristics of cases and controls.
| Variable | Cases N = 799 | Controls N = 665 | Effect estimate (95% CI) a | p‐value | Test |
|---|---|---|---|---|---|
| Matching variables | |||||
| Age (years), mean (SD) | 44.90 (14.45) | 45.17 (14.38) | −0.28 [−1.77; 1.21] | 0.716 | Welch t‐test |
| Sex, n (%) | |||||
| Male | 470 (58.8%) | 384 (57.7%) | 0.96 [0.78; 1.18] | 0.716 | Chi‐square |
| Female | 329 (41.2%) | 281 (42.3%) | |||
| isCGM user, n (%) | 521 (65.2%) | 417 (62.9%) | 1.11 [0.89; 1.37] | 0.388 | Chi‐square |
| Basic clinical and sociodemographic variables | |||||
| Years since T1DM onset, mean (SD) | 37.85 (16.16) | 34.92 (16.82) | 2.93 [1.23; 4.63] | < 0.001 | Welch t‐test |
| Age at T1DM onset, mean (SD) | 19.46 (13.29) | 21.58 (13.07) | −2.12 [−3.49; −0.75] | 0.002 | Welch t‐test |
| Smoking status, n (%) | 244 (33.2%) | 135 (22.5%) | 1.72 [1.34; 2.19] | < 0.001 | Chi‐square |
| Harmful alcohol use, n (%) | 77 (15.7%) | 20 (5.2%) | 3.37 [2.02; 5.62] | < 0.001 | Chi‐square |
| Body mass index (BMI), mean (SD) | 26.23 (4.89) | 26.43 (4.39) | −0.19 [−0.70; 0.31] | 0.454 | Welch t‐test |
| Obesity, n (%) | 128 (18.2%) | 99 (16.6%) | 1.12 [0.84; 1.49] | 0.503 | Chi‐square |
| Dyslipidemia, n (%) | 351 (44.2%) | 300 (45.1%) | 0.96 [0.78; 1.19] | 0.769 | Chi‐square |
| Hypertension, n (%) | 239 (30.1%) | 200 (30.1%) | 1.00 [0.80; 1.25] | 1.000 | Chi‐square |
| Microvascular complications, n (%) | 282 (35.3%) | 201 (30.2%) | 1.26 [1.01; 1.57] | 0.044 | Chi‐square |
| Diabetic retinopathy, n (%) | 166 (20.9%) | 140 (21.1%) | 0.99 [0.77; 1.27] | 0.976 | Chi‐square |
| Diabetic kidney disease, n (%) | 133 (16.8%) | 76 (11.5%) | 1.56 [1.15; 2.11] | 0.005 | Chi‐square |
| Diabetic neuropathy, n (%) | 127 (16.0%) | 56 (8.4%) | 2.06 [1.48; 2.88] | < 0.001 | Chi‐square |
| Macrovascular complications, n (%) | 121 (15.2%) | 49 (7.4%) | 2.25 [1.58; 3.19] | < 0.001 | Chi‐square |
| Ischaemic heart disease, n (%) | 45 (5.7%) | 23 (3.5%) | 1.67 [1.00; 2.80] | 0.063 | Chi‐square |
| Cerebrovascular disease, n (%) | 57 (7.2%) | 11 (1.7%) | 4.59 [2.39; 8.83] | < 0.001 | Chi‐square |
| Peripheral artery disease, n (%) | 67 (8.4%) | 23 (3.5%) | 2.57 [1.58; 4.18] | < 0.001 | Chi‐square |
| History of diabetic ketoacidosis, n (%) | 29 (3.7%) | 8 (1.2%) | 3.12 [1.42; 6.87] | 0.005 | Chi‐square |
| History of prior severe hypoglycemia, n (%) | 425 (55.6%) | 136 (21.2%) | 4.65 [3.67; 5.89] | < 0.001 | Chi‐square |
| Hypothyroidism, n (%) | 165 (20.8%) | 116 (17.5%) | 1.24 [0.95; 1.61] | 0.133 | Chi‐square |
| Motor disability, n (%) | 48 (6.0%) | 18 (2.7%) | 2.31 [1.33; 4.01] | 0.003 | Chi‐square |
| Cognitive disability, n (%) | 18 (3.8%) | 4 (1.1%) | 3.48 [1.17; 10.36] | 0.031 | Chi‐square |
| Depression, n (%) | 126 (16.2%) | 45 (7.0%) | 2.58 [1.80; 3.69] | < 0.001 | Chi‐square |
| Number of chronic conditions in health registry, mean (SD) | 4.79 (3.10) | 3.80 (2.36) | 0.99 [0.70; 1.29] | < 0.001 | Welch t‐test |
| Number of prescribed active substances, mean (SD) | 5.73 (4.68) | 4.52 (3.55) | 1.20 [0.75; 1.65] | < 0.001 | Welch t‐test |
| Education level, n (%) | |||||
| Basic | 276 (57.9%) | 187 (42.1%) | < 0.001 | Chi‐square | |
| Medium | 108 (22.6%) | 122 (27.5%) | |||
| University | 93 (19.5%) | 135 (30.4%) | |||
| Basic diabetes education programme, n (%) | 750 (97.0%) | 629 (96.8%) | 1.09 [0.60; 1.99] | 0.902 | Chi‐square |
| Advanced diabetes education programme, n (%) | 552 (71.4%) | 513 (79.4%) | 0.65 [0.51; 0.83] | < 0.001 | Chi‐square |
| Engages in physical activity, n (%) | 211 (31.5%) | 211 (39.1%) | 0.71 [0.56; 0.91] | 0.007 | Chi‐square |
| Lives alone, n (%) | 75 (13.8%) | 51 (11.0%) | 1.29 [0.88; 1.88] | 0.223 | Chi‐square |
| Lives in nursing home, n (%) | 7 (0.9%) | 1 (0.2%) | 5.06 [0.62; 41.27] | 0.185 | Chi‐square |
| Diabetes related variables | |||||
| Nocturnal hypoglycemia, n (%) | 367 (51.8%) | 142 (23.6%) | 3.48 [2.74; 4.42] | < 0.001 | Chi‐square |
| Impaired hypoglycemia awareness, n (%) | 330 (47.6%) | 127 (20.2%) | 3.59 [2.81; 4.59] | < 0.001 | Chi‐square |
| Fear of hypoglycemia, n (%) | 243 (44.1%) | 118 (24.9%) | 2.37 [1.82; 3.10] | < 0.001 | Chi‐square |
| Total daily insulin dose, mean (SD) | 50.31 (23.31) | 52.48 (24.24) | −2.17 [−4.98; 0.65] | 0.131 | Welch t‐test |
| Insulin dose per kg, mean (SD) | 0.67 (0.26) | 0.70 (0.27) | −0.02 [−0.06; 0.01] | 0.136 | Welch t‐test |
| Basal insulin (%), mean (SD) | 53.43 (16.76) | 52.23 (15.97) | 1.20 [−0.82; 3.22] | 0.245 | Welch t‐test |
| Rapid insulin (%), mean (SD) | 42.53 (15.99) | 42.70 (15.46) | −0.17 [−2.13; 1.79] | 0.866 | Welch t‐test |
| HbA1c, mean (SD) | 7.68 (1.33) | 7.61 (1.17) | 0.06 [−0.07; 0.19] | 0.360 | Welch t‐test |
| Ambulatory glucose profile variables (AGP) | |||||
| Hypoglycemia alarm threshold, mean (SD) | 72.53 (8.68) | 72.92 (7.25) | −0.39 [−1.85; 1.07] | 0.601 | Welch t‐test |
| Percentage of sensor use, mean (SD) | 84.41 (22.43) | 89.65 (14.84) | −5.24 [−8.07; −2.41] | < 0.001 | Welch t‐test |
| Number of readings per day, mean (SD) | 12.05 (9.40) | 12.12 (8.18) | −0.07 [−1.43; 1.29] | 0.922 | Welch t‐test |
| GMI, mean (SD) | 7.21 (0.87) | 7.18 (0.76) | 0.03 [−0.09; 0.15] | 0.613 | Welch t‐test |
| Mean glucose, mean (SD) | 163.41 (38.34) | 162.13 (31.74) | 1.28 [−3.86; 6.42] | 0.625 | Welch t‐test |
| Glucose CV, mean (SD) | 42.09 (8.64) | 37.53 (6.62) | 4.56 [3.43; 5.69] | < 0.001 | Welch t‐test |
| Time in range 70–180 Mg/Dl, mean (SD) | 55.23 (16.82) | 60.67 (16.62) | −5.44 [−7.83; −3.05] | < 0.001 | Welch t‐test |
| Time above range 181–249 mg/dL, mean (SD) | 25.00 (13.03) | 26.41 (13.22) | −1.41 [−3.29; 0.47] | 0.142 | Welch t‐test |
| Time above range ≥ 250 mg/dL, mean (SD) | 13.48 (13.76) | 11.05 (11.55) | 2.43 [0.55; 4.31] | 0.011 | Welch t‐test |
| Time below range 55–69 mg/dL, mean (SD) | 6.91 (5.98) | 4.40 (4.13) | 2.50 [1.76; 3.25] | < 0.001 | Welch t‐test |
| Time below range ≤ 54 mg/dL, mean (SD) | 2.46 (3.81) | 0.85 (1.60) | 1.61 [1.17; 2.05] | < 0.001 | Welch t‐test |
| Hypoglycemia events, mean (SD) | 12.88 (12.07) | 10.13 (8.83) | 2.75 [1.10; 4.41] | 0.001 | Welch t‐test |
| Mean duration of hypoglycemia episodes (minutes), mean (SD) | 106.57 (54.09) | 83.27 (45.24) | 23.29 [15.32; 31.27] | < 0.001 | Welch t‐test |
Note: Baseline demographic, clinical, and diabetes‐related characteristics of adults with type 1 diabetes mellitus (T1DM) included as cases (n = 799) and matched controls (n = 665). Microvascular complications is a composite variable encompassing diabetic retinopathy, diabetic kidney disease, and diabetic neuropathy. Macrovascular complications is a composite variable encompassing ischaemic heart disease, cerebrovascular disease, and peripheral artery disease. Statistically significant p‐values, or those showing a trend toward significance (p < 0.1), have been marked in bold.
Abbreviations: AGP, ambulatory glucose profile; BMI, body mass index; CI, confidence interval; CV, coefficient of variation (of glucose); GMI, glucose management indicator; HbA1c, glycated haemoglobin A1c; isCGM, intermittently scanned continuous glucose monitoring; OR, odds ratio; SD, standard deviation; T1DM, type 1 diabetes mellitus.
Continuous variables are expressed as mean ± standard deviation (SD) and are compared using the mean difference with 95% confidence interval (CI) and the corresponding p‐value. Categorical variables are expressed as number (%) and are compared using the odds ratio (OR) with 95% CI and p‐value.
Regarding clinical and sociodemographic characteristics, cases had a longer duration of diabetes (37.9 vs. 34.9 years, p < 0.001) and an earlier age at onset (19.5 vs. 21.6 years, p = 0.002). Lifestyle differences were also evident: smoking (33.2% vs. 22.5%) and harmful alcohol use (15.7% vs. 5.2%) were both more frequent in cases (p < 0.001).
Cases showed a higher prevalence of chronic complications, especially diabetic kidney disease (16.8% vs. 11.5%, p = 0.005) and macrovascular disease (15.2% vs. 7.4%, p < 0.001), including cerebrovascular disease (7.2% vs. 1.7%). They also carried a greater comorbidity burden, with more motor disability (6.0% vs. 2.7%), depression (16.2% vs. 7.0%), and a higher number of prescribed drugs (5.7 vs. 4.5, p < 0.001).
Sociodemographic differences were also observed: basic education was more common in cases (57.9% vs. 42.1%), whereas university education was less frequent (19.5% vs. 30.4%, p < 0.001). Participation in advanced diabetes education programs was lower among cases (71.4% vs. 79.4%). In terms of hypoglycemia‐related history, cases more often reported previous severe hypoglycemia (55.6% vs. 21.2%), nocturnal episodes (51.8% vs. 23.6%), impaired awareness (47.6% vs. 20.2%), and fear of hypoglycemia (44.1% vs. 24.9%) (all p < 0.001). No significant differences were observed between cases and controls regarding the type of basal insulin or rapid‐acting insulin analogue (Supporting Information S1).
Analysis of ambulatory glucose profile (AGP) parameters revealed that cases spent more time below range (≤ 54 mg/dL: 2.5% vs. 0.9%) and had both a higher number (12.9 vs. 10.1) and duration of hypoglycemic episodes (107 vs. 83 min). They also showed lower time in range (55.2% vs. 60.7%) and greater glycaemic variability (CV: 42.1% vs. 37.5%).
3.2. Multivariate Analysis and Predictive Models
Multivariable logistic regression analyses were performed to identify independent predictors of severe hypoglycemia. Two models were constructed: the first including all participants (n = 1464), and the second restricted to those using isCGM (n = 938) to integrate detailed glycaemic metrics.
In the full cohort, several variables remained independently associated with severe hypoglycemia after multivariable adjustment (Table 2). Nocturnal hypoglycemia (OR = 4.52; 95% CI 2.84–7.32; p < 0.001) and a history of previous severe hypoglycemia (OR = 4.12; 95% CI 2.62–6.58; p < 0.001) emerged as the strongest predictors. The number of chronic conditions also showed a significant effect (OR = 1.14; 95% CI 1.04–1.25; p = 0.005). Conversely, the use of isCGM was independently associated with a lower risk (OR = 0.29; 95% CI 0.10–0.73; p = 0.013), suggesting a protective influence compared with SMBG. Alcohol use disorder (p = 0.075) and depression (p = 0.071) showed trends towards significance.
TABLE 2.
Multivariable logistic regression models for the prediction of severe hypoglycemia.
| Predictor | OR | 95% CI | p‐value | |||
|---|---|---|---|---|---|---|
| A. Model including all participants (n = 1464) | ||||||
| Use of isCGM | 0.29 | [0.10; 0.73] | 0.013 | |||
| History of depression | 2.06 | [0.96; 4.67] | 0.071 | |||
| Nocturnal hypoglycemia episodes | 4.52 | [2.84; 7.32] | < 0.001 | |||
| History of severe hypoglycemia | 4.12 | [2.62; 6.58] | < 0.001 | |||
| Alcohol use disorder | 2.31 | [0.97; 6.25] | 0.075 | |||
| Number of chronic conditions (health registry) | 1.14 | [1.04; 1.25] | 0.005 | |||
| B. Model restricted to participants using flash glucose monitoring (n = 938) | ||||||
| Mean duration of hypoglycemia (isCGM) | 1.00 | [0.997; 1.010] | 0.317 | |||
| History of severe hypoglycemia | 3.50 | [2.05; 6.06] | < 0.001 | |||
| Nocturnal hypoglycemia episodes | 3.60 | [2.10; 6.29] | < 0.001 | |||
| Time below range < 54 mg/dL (%) | 1.15 | [0.980; 1.370] | 0.106 | |||
| Time in range 70–180 mg/dL (%) | 0.98 | [0.970; 0.999] | 0.036 | |||
| Number of chronic conditions per period | 1.17 | [1.06; 1.30] | 0.003 | |||
| Time below 70 mg/dL (%) | 1.02 | [0.953; 1.092] | 0.6 | |||
Note: Multivariable logistic regression analyses evaluating predictors of severe hypoglycemia in adults with type 1 diabetes. Odds ratios (OR) are presented with 95% confidence intervals (CI) and exact p‐values.
Abbreviations: CI, confidence interval; OR, odds ratio.
A nomogram derived from this model (Figure 2A–B) provides an individualised risk estimation tool based on these predictors. The optimal cutoff defined by the Youden index was 71.5 points, corresponding to a predicted probability of 0.43 for severe hypoglycemia. At this threshold, the model achieved a sensitivity of 86.1%, specificity of 63.9%, positive predictive value (PPV) of 79.2%, and negative predictive value (NPV) of 74.3% in the training set, with similar performance in the validation cohort. The overall discriminative ability was good, with an AUC of 0.812 (95% CI 0.775–0.850; p < 0.0001) in training and 0.754 (95% CI 0.687–0.820; p < 0.0001) in validation. Calibration plots confirmed satisfactory agreement between observed and predicted probabilities (Brier = 0.137; slope = 0.746; intercept = 0.136) (Supporting Information S2).
FIGURE 2.
Nomograms for predicting the probability of severe hypoglycemia in adults with type 1 diabetes mellitus. Panels (A) and (B) display the nomogram and predicted probability curve for the total study population, while panels (C) and (D) show the corresponding models for users of intermittently scanned Continuous Glucose Monitoring. (A), (C). Each predictor variable contributes a specific number of points according to its position on the upper scale. The total score (sum of all points) corresponds to the estimated probability of severe hypoglycemia shown on the lower axis. The red marker indicates the optimal Youden cut‐off for risk discrimination. (B), (D). The lower panels show the relationship between total points and the predicted probability of severe hypoglycemia, allowing graphical estimation of individual risk. Dashed lines indicate the optimal threshold based on the Youden index. Additional details on model calibration and validation are provided in the Supporting Information S2. isCGM, intermittently scanned Continuous Glucose Monitoring; SMBG, Self‐monitoring of blood glucose; T1DM, Type 1 diabetes mellitus; TP, Total points; Youden, Youden index (optimal cut‐off point).
In the subgroup of participants using isCGM, a history of previous severe hypoglycemia (OR = 3.50; 95% CI 2.05–6.06; p < 0.001) and nocturnal hypoglycemia episodes (OR = 3.60; 95% CI 2.10–6.29; p < 0.001) remained the most powerful determinants of risk of severe hypoglycemia (Table 2). In addition, a higher number of chronic conditions was associated with increased odds of severe hypoglycemia (OR = 1.17; 95% CI 1.06–1.30; p = 0.003), while greater time in range (70–180 mg/dL) was independently protective (OR = 0.98; 95% CI 0.97–0.999; p = 0.036).
A nomogram derived from this model (Figure 2C–D) enables individualised estimation of risk among isCGM users. The optimal cutoff defined by the Youden index was 94.4 points, corresponding to a predicted probability of 0.49 for severe hypoglycemia. At this threshold, the model achieved a sensitivity of 84.2%, specificity of 67.3%, PPV of 77.6%, and NPV of 75.5% in the training set, with similar performance in the validation cohort. The discriminative ability of the model was excellent, with an AUC of 0.812 (95% CI 0.768–0.855; p < 0.0001) in training and 0.829 (95% CI 0.762–0.897; p < 0.0001) in validation. Calibration analysis demonstrated a close alignment between observed and predicted probabilities (Brier = 0.103; slope = 1.119; intercept = −0.119) (Supporting Information S3).
4. Discussion
This study provides a comprehensive analysis of risk factors for severe hypoglycemia in adults with T1D treated with MDI, using data from a large regional cohort that integrates emergency medical records and clinical information from the APHS. By combining population‐based data with advanced glucose‐monitoring metrics, our findings contribute to the identification of vulnerable patient profiles and propose practical predictive tools for clinical risk stratification. Severe hypoglycemia remains a major challenge in T1D, despite technological advances and structured education programs [11, 12, 13]. The unpredictable nature of these events, their association with increased morbidity, and their psychological burden justify the need for accurate, individualised risk assessment models [14, 15].
In our analysis, several clinical and behavioural variables were independently associated with severe hypoglycemia. A history of previous severe episodes and nocturnal hypoglycemia emerged as the strongest predictors, consistent with previous studies demonstrating that past events are the single most powerful determinant of future risk [16]. This finding reflects the persistence of individual susceptibility, likely driven by impaired counterregulatory responses, diminished hypoglycemia awareness, and behavioural patterns that are difficult to modify [17, 18]. The strong association with nocturnal events also highlights the role of unrecognised hypoglycemia during sleep, a period of blunted autonomic response and limited capacity for self‐correction [19]. These results reinforce the need for targeted interventions in patients with recurrent or nocturnal episodes, including technology‐assisted alarms, bedtime glucose optimisation, and individualised insulin adjustment.
The number of chronic conditions per patient also showed a significant and independent association with severe hypoglycemia. This finding aligns with the concept that multimorbidity and polypharmacy contribute to metabolic instability and impaired self‐management [20, 21]. Comorbid conditions—such as cardiovascular disease, nephropathy, and neurologic impairment—may increase vulnerability through both physiological and cognitive mechanisms. In particular, cerebrovascular disease and motor disability could limit the recognition of hypoglycemia symptoms or the ability to respond promptly [22, 23]. The observed trend towards higher risk among individuals with depression further supports the interaction between mental health and metabolic outcomes [24, 25]. Depression is known to reduce self‐care capacity, increase treatment inertia, and alter appetite or sleep patterns, all of which may precipitate glycaemic fluctuations [26, 27]. Similarly, the association between alcohol use disorder and hypoglycemia risk, although of borderline significance, is clinically meaningful, given the inhibitory effect of alcohol on hepatic gluconeogenesis and the tendency to underestimate carbohydrate needs during drinking episodes [28].
Although education level was not included in the multivariable model due to collinearity with other factors, its univariate association with severe hypoglycemia underscores the importance of diabetes literacy [20]. Limited understanding of carbohydrate counting, insulin titration, and hypoglycemia recognition may compromise safety in routine self‐management. Educational interventions should therefore be adapted to the individual's level of health literacy, emphasising practical aspects such as symptom recognition and glucose monitoring, among others [29].
In our study, the use of isCGM was an independent protective factor against severe hypoglycemia, consistent with findings from our population‐based analysis using the same CES‐061 registry definition, where isCGM implementation reduced severe hypoglycemic events by 27% (rate ratio 0.72 [0.66–0.80]). These results reinforce its protective role at both individual and population levels [30].
In this sense, the model restricted to isCGM users adds valuable insights into the contribution of continuous glucose metrics to hypoglycemia prediction. Among these parameters, both TBR—encompassing Level 1 (54–69 mg/dL) and Level 2 (< 54 mg/dL) hypoglycemia—and the mean duration of hypoglycemic episodes were associated with an increased risk of severe events, while TIR emerged as an independent protective factor. These findings are consistent with growing evidence that TIR and TBR, rather than HbA1c alone, more accurately reflect the quality and stability of glycaemic control [31, 32]. In particular, a higher proportion of TIR, together with a lower proportion of TBR, has been linked to a reduced frequency of severe hypoglycemia across different populations and monitoring technologies, reinforcing their clinical relevance as therapeutic targets [33]. The integration of these continuous glucose metrics into predictive modelling represents a step forward in the characterisation of individual risk profiles and the refinement of personalised management strategies [34].
Beyond identifying risk factors, our study provides a practical and clinically oriented tool for individualised risk prediction of severe hypoglycemia in adults with T1D. The nomograms derived from both logistic regression models showed good discrimination (AUC 0.75–0.83) and satisfactory calibration, with balanced sensitivity (84%–86%) and specificity (64%–67%), ensuring accurate identification of high‐risk individuals in preventive settings. The inclusion of variables easily obtainable in daily practice—such as prior hypoglycemia, nocturnal events, comorbidity burden, and TIR—facilitates their integration into routine clinical workflows [35].
Methodologically, our approach shares conceptual similarities with the nomogram developed by Han et al. for T2D, although the populations and analytical frameworks differ substantially [36]. While their model used Cox proportional hazards regression in a heterogeneous T2D cohort, our study focused on adults with T1D treated with multiple daily injections and employed logistic regression optimised for event prediction. This design enabled the incorporation of continuous glucose metrics from isCGM, including TIR and TBR, providing a dynamic characterisation of glycaemic stability beyond traditional HbA1c‐based indicators. Together, these elements position our model as an innovative, data‐driven tool aligned with precision diabetology, capable of supporting early intervention and personalised follow‐up in real‐world care.
Several recent studies have sought to predict the risk of severe hypoglycemia in individuals with T1D using statistical or artificial intelligence—based approaches. Freeman et al. applied a random forest algorithm in older adults with T1D, identifying impaired awareness of hypoglycemia, fear of hypoglycemia, CGM‐derived coefficient of variation, percentage of time below 70 mg/dL (TBR), and cognitive performance as key predictors [37]. Similarly, Lara‐Abelenda et al. developed an explainable multimodal fusion model integrating clinical, psychological, and CGM time‐series data, highlighting the contribution of impaired awareness, fear, depression, and glucose variability metrics [38].
Our findings are consistent with these observations, as variables reflecting both clinical vulnerability and glycaemic instability were among the strongest determinants of risk in our models. In particular, the association of nocturnal hypoglycemia, prior severe events, and higher comorbidity burden parallels the multidimensional nature of risk identified by Freeman and Lara‐Abelenda. Moreover, the independent contribution of TBR and TIR in our analysis aligns with the emphasis placed on CGM‐derived variability metrics in those studies, underscoring the central role of continuous glucose monitoring in characterising hypoglycemia risk profiles [39].
Several limitations should be acknowledged. First, the retrospective case–control design may be subject to residual confounding, despite the strict matching criteria applied. Although cases and controls were selected within the same healthcare system using uniform inclusion and exclusion criteria, some potentially relevant factors—such as detailed insulin dosing strategies, carbohydrate counting accuracy, or other lifestyle behaviours—were not systematically recorded and may have influenced the observed associations.
Second, certain clinically relevant variables, including physical activity, dietary intake, use of rescue carbohydrates or glucagon, and the number of daily rapid‐acting insulin injections, were not consistently available in electronic records and therefore could not be evaluated. Similarly, the frequency of self‐monitoring of blood glucose in participants not using isCGM was not systematically documented.
Third, the models were internally validated using a split‐sample approach; however, external validation in independent cohorts and different healthcare settings will be required to confirm their generalisability and clinical applicability.
Finally, although isCGM use was recorded, detailed information regarding the specific device generation and the activation status of hypoglycemia alarms was not systematically available. As alarm functionality may influence the risk of severe hypoglycemia, this factor could not be evaluated in the present analysis [40].
Despite these limitations, the study has several strengths, including a large and well‐characterised regional sample, the use of a real‐world population encompassing emergency medical data, and the integration of continuous glucose metrics. The combination of robust internal validation, good model calibration, and practical clinical interpretability supports the reliability of the findings and their potential value for future implementation in diabetes care.
In conclusion, our study identifies key clinical and behavioural determinants of severe hypoglycemia in adults with T1D treated with MDI and demonstrates the feasibility of a nomogram‐based approach for individualised risk prediction. The findings highlight the protective role of isCGM, the impact of comorbid conditions, and the relevance of continuous glucose metrics such as TIR. Incorporating these elements into clinical assessment could enhance the precision of preventive strategies and reduce the burden of severe hypoglycemia in daily practice.
Author Contributions
Pablo Rodríguez de Vera Gómez: conceptualization, methodology, formal analysis, data curation, writing – original draft, visualization, funding acquisition. Virginia Bellido: data curation, investigation, writing – review and editing. Miguel Damas Fuentes: data curation and investigation. María del Carmen Serrano Laguna: data curation and investigation. María Victoria Cózar León: data curation and investigation. Marta Domínguez López: data curation and investigation. Carolina Sánchez Malo: data curation, investigation. Almudena Lara Barea: data curation and investigation. Víctor Siles‐Guerrero: data curation and investigation. María Laínez López: data curation and investigation. María del Carmen Ayala Ortega: data curation and investigation. Enrique Redondo Torres: data curation and investigation. María Dolores Alcántara Laguna: data curation and investigation. Nancy Sánchez: data curation and investigation. Pedro Mezquita Raya: data curation, investigation, writing – review and editing. Noelia Gros Herguido: data curation and investigation. Sandra Amuedo: data curation and investigation. Ángel Rebollo Román: data curation and investigation. Manuel de la Cal: data curation and investigation. Rafael Palomares Ortega: methodology, supervision, writing – review and editing. María Asunción Martínez‐Brocca: conceptualization, supervision, project administration, writing – review and editing.
Funding
This study was supported by the Andalusian Society of Endocrinology, Diabetes and Nutrition (SAEDYN) through the 2023 Diabetes Research Award.
Conflicts of Interest
The authors declare no conflicts of interest.
Supporting information
Supporting Information S1
Supporting Information S2
Supporting Information S3
Acknowledgements
The authors wish to express their sincere gratitude to Henry Antonio Andrade Ruiz, from the Methodological and Statistical Management Unit of FISEVI (Foundation for the Management of Biomedical Research of Seville), for his valuable support in data analysis and methodological guidance throughout the development of this study. The authors also wish to acknowledge Pedro Emilio Ferro Gallego, from the METAnetwork of Andalusia, for his coordination efforts and support in enabling the multicenter collaboration participating in this study.
Contributor Information
Pablo Rodríguez de Vera Gómez, Email: pablo.rodriguezvera.sspa@juntadeandalucia.es.
María Asunción Martínez‐Brocca, Email: masuncion.martinez.sspa@juntadeandalucia.es.
Data Availability Statement
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
- 1. Umpierrez G. and Korytkowski M., “Diabetic Emergencies ‐ Ketoacidosis, Hyperglycaemic Hyperosmolar State and Hypoglycaemia,” Nature Reviews Endocrinology 12, no. 4 (April 2016): 222–232, 10.1038/nrendo.2016.15. [DOI] [PubMed] [Google Scholar]
- 2. American Diabetes Association Professional Practice Committee , “6. Glycemic Goals and Hypoglycemia: Standards of Care in Diabetes‐2025,” Supplement, Diabetes Care 48, no. 1 (January 2025): S128–S145, 10.2337/dc25-S006. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3. Parsaik A. K., Carter R. E., Pattan V., et al., “Population‐Based Study of Severe Hypoglycemia Requiring Emergency Medical Service Assistance Reveals Unique Findings,” Journal of Diabetes Science and Technology 6, no. 1 (January 2012): 65–73, 10.1177/193229681200600109. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4. Geller A. I., Shehab N., Lovegrove M. C., et al., “National Estimates of Insulin‐Related Hypoglycemia and Errors Leading to Emergency Department Visits and Hospitalizations,” JAMA Internal Medicine 174, no. 5 (May 2014): 678–686, 10.1001/jamainternmed.2014.136. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5. Barranco R. J., Gomez‐Peralta F., Abreu C., et al., “Incidence and Care‐Related Costs of Severe Hypoglycaemia Requiring Emergency Treatment in Andalusia (Spain): The PAUEPAD Project,” Diabetic Medicine 32, no. 11 (November 2015): 1520–1526, 10.1111/dme.12843. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6. Gómez‐Peralta F., Menéndez E., Conde S., Conget I., and Novials A., “Investigators of the Studies SED and SED1. Clinical Characteristics and Management of Type 1 Diabetes in Spain. the SED1 Study,” Endocrinología, Diabetes y Nutrición (English ed.) 68, no. 9 (November 2021): 642–653, 10.1016/j.endien.2021.11.020. [DOI] [PubMed] [Google Scholar]
- 7. Pinés Corrales P. J., Arias Lozano C., Jiménez Martínez C., et al., “Prevalence of Severe Hypoglycemia in a Cohort of Patients With Type 1 Diabetes,” Endocrinología, Diabetes y Nutrición (English ed.) 68, no. 1 (January 2021): 47–52. English, Spanish, 10.1016/j.endinu.2020.01.002. [DOI] [PubMed] [Google Scholar]
- 8. Brito‐Sanfiel M., Diago‐Cabezudo J., and Calderon A., “Economic Impact of Hypoglycemia on Healthcare in Spain,” Expert Review of Pharmacoeconomics & Outcomes Research 10, no. 6 (December 2010): 649–660, 10.1586/erp.10.73. [DOI] [PubMed] [Google Scholar]
- 9. Pedersen‐Bjergaard U. and Thorsteinsson B., “Reporting Severe Hypoglycemia in Type 1 Diabetes: Facts and Pitfalls,” Current Diabetes Reports 17, no. 12 (October 2017): 131, 10.1007/s11892-017-0965-1. [DOI] [PubMed] [Google Scholar]
- 10. Little S. A., Leelarathna L., Barendse S. M., et al., “Severe Hypoglycaemia in Type 1 Diabetes Mellitus: Underlying Drivers and Potential Strategies for Successful Prevention,” Diabetes/Metabolism Research and Reviews 30, no. 3 (March 2014): 175–190, 10.1002/dmrr.2492. [DOI] [PubMed] [Google Scholar]
- 11. Gubitosi‐Klug R. A., Braffett B. H., White N. H., et al., “Risk of Severe Hypoglycemia in Type 1 Diabetes Over 30 Years of Follow‐Up in the DCCT/EDIC Study,” Diabetes Care 40, no. 8 (August 2017): 1010–1016, 10.2337/dc16-2723. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12. Sherr J. L., Laffel L. M., Liu J., et al., “Severe Hypoglycemia and Impaired Awareness of Hypoglycemia Persist in People With Type 1 Diabetes Despite Use of Diabetes Technology: Results From a Cross‐Sectional Survey,” Diabetes Care 47, no. 6 (June 2024): 941–947, 10.2337/dc23-1765. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13. Barnard‐Kelly K. D., Martínez‐Brocca M. A., Glatzer T., and Oliver N., “Identifying the Deficiencies of Currently Available CGM to Improve Uptake and Benefit,” Diabetic Medicine 41, no. 8 (August 2024): e15338, 10.1111/dme.15338. [DOI] [PubMed] [Google Scholar]
- 14. McCoy R. G., Herrin J., Galindo R. J., et al., “All‐Cause Mortality After Hypoglycemic and Hyperglycemic Emergencies Among U.S. Adults With Diabetes, 2011‐2020,” Diabetes Research and Clinical Practice 197 (March 2023): 110263, 10.1016/j.diabres.2023.110263. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15. Canha D., Choudhary P., Cosson E., et al., “Time Below Range Alone Is Insufficient to Identify Severe Hypoglycaemia Risk in Type 1 Diabetes‐The Critical Role of Hypoglycaemia Awareness: Results From the SFDT1 Study,” Diabetologia 68, no. 12 (September 2025): 2719–2731, 10.1007/s00125-025-06536-x. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16. Wohland T., Holstein J. D., Patzer O. M., et al., “New Risk and Protective Factors for Severe Hypoglycaemia in People With Type 1 Diabetes,” Nutrition, Metabolism, and Cardiovascular Diseases 27, no. 5 (May 2017): 407–414, 10.1016/j.numecd.2016.12.006. [DOI] [PubMed] [Google Scholar]
- 17. Cook A. J., DuBose S. N., Foster N., et al., “Cognitions Associated With Hypoglycemia Awareness Status and Severe Hypoglycemia Experience in Adults With Type 1 Diabetes,” Diabetes Care 42, no. 10 (October 2019): 1854–1864, 10.2337/dc19-0002. [DOI] [PubMed] [Google Scholar]
- 18. Jensen M. H., Dethlefsen C., Hejlesen O., and Vestergaard P., “Association of Severe Hypoglycemia With Mortality for People With Diabetes Mellitus During a 20‐Year Follow‐Up in Denmark: A Cohort Study,” Acta Diabetologica 57, no. 5 (May 2020): 549–558, 10.1007/s00592-019-01447-x. [DOI] [PubMed] [Google Scholar]
- 19. Zhu B., Abu Irsheed G. M., Martyn‐Nemeth P., and Reutrakul S., “Type 1 Diabetes, Sleep, and Hypoglycemia,” Current Diabetes Reports 21, no. 12 (December 2021): 55, 10.1007/s11892-021-01424-1. [DOI] [PubMed] [Google Scholar]
- 20. Ratzki‐Leewing A. A., Black J. E., Ryan B. L., et al., “Development and Validation of a Real‐World Model to Predict 1‐Year Level 3 (Severe) Hypoglycaemia Risk in Adults With Diabetes (The iNPHORM Study, United States),” Diabetes, Obesity and Metabolism 25, no. 10 (October 2023): 2910–2927, 10.1111/dom.15186. [DOI] [PubMed] [Google Scholar]
- 21. McCoy R. G., Lipska K. J., Van Houten H. K., and Shah N. D., “Association of Cumulative Multimorbidity, Glycemic Control, and Medication Use With Hypoglycemia‐Related Emergency Department Visits and Hospitalizations Among Adults With Diabetes,” JAMA Network Open 3, no. 1 (January 2020): e1919099, 10.1001/jamanetworkopen.2019.19099. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22. International Hypoglycaemia Study Group , “Hypoglycaemia, Cardiovascular Disease, and Mortality in Diabetes: Epidemiology, Pathogenesis, and Management,” Lancet Diabetes & Endocrinology 7, no. 5 (May 2019): 385–396, 10.1016/S2213-8587(18)30315-2. [DOI] [PubMed] [Google Scholar]
- 23. Longendyke R., Grundman J. B., and Majidi S., “Acute and Chronic Adverse Outcomes of Type 1 Diabetes,” Endocrinology and Metabolism Clinics of North America 53, no. 1 (March 2024): 123–133, 10.1016/j.ecl.2023.09.004. [DOI] [PubMed] [Google Scholar]
- 24. Sepúlveda E., Poínhos R., Nata G., et al., “Relationship Between Severe Hypoglycemia or Impaired Awareness of Hypoglycemia and diabetes‐Related Health Status, Global Cognition and Executive Functions in Adults With Type 1 Diabetes Without Severe Anxiety or Depression,” Diabetes Research and Clinical Practice 221 (March 2025): 112004, 10.1016/j.diabres.2025.112004. [DOI] [PubMed] [Google Scholar]
- 25. Shigemoto S., Imbe H., Fujisawa R., et al., “Decreased Cognitive Function Is Associated With Preceding Severe Hypoglycemia and Impaired Blood Glucose Control in the Elderly Individuals With Type 1 Diabetes,” Diabetology International 13, no. 4 (May 2022): 679–686, 10.1007/s13340-022-00588-9. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26. Castellano‐Guerrero A. M., Guerrero R., Relimpio F., et al., “Prevalence and Predictors of Depression and Anxiety in Adult Patients With Type 1 Diabetes in Tertiary Care Setting,” Acta Diabetologica 55, no. 9 (September 2018): 943–953, 10.1007/s00592-018-1172-5. [DOI] [PubMed] [Google Scholar]
- 27. Ruiz‐Aranda D., Zysberg L., García‐Linares E., Castellano‐Guerrero A. M., Martínez‐Brocca M. A., and Gutiérrez‐Colosía M. R., “Emotional Abilities and HbA1c Levels in Patients With Type 1 Diabetes,” Psychoneuroendocrinology 93 (July 2018): 118–123, 10.1016/j.psyneuen.2018.04.015. [DOI] [PubMed] [Google Scholar]
- 28. Iqbal A. and Heller S. R., “The Role of Structured Education in the Management of Hypoglycaemia,” Diabetologia 61, no. 4 (April 2018): 751–760, 10.1007/s00125-017-4334-z. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29. Rondags S. M., de Wit M., and Snoek F. J., “HypoAware: Development and Pilot Study of a Brief and Partly Web‐Based Psychoeducational Group Intervention for Adults With Type 1 and Insulin‐Treated Type 2 Diabetes and Problematic Hypoglycaemia,” Diabetic Medicine 33, no. 2 (February 2016): 184–191, 10.1111/dme.12876. [DOI] [PubMed] [Google Scholar]
- 30. Rodríguez de Vera Gómez P., Mayoral Sánchez E., Vilches Arenas Á, et al., “Population‐Based Study on the Implementation of Flash Glucose Monitoring and Severe Hypoglycemia in Adults With Type 1 Diabetes,” Diabetes Technology & Therapeutics 26, no. 12 (December 2024): 951–959, 10.1089/dia.2024.0201. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 31. Hermanns N., Ehrmann D., Heinemann L., Freckmann G., Waldenmaier D., and Calhoun P., “Real‐Time Continuous Glucose Monitoring Can Predict Severe Hypoglycemia in People With Type 1 Diabetes: Combined Analysis of the HypoDE and DIAMOND Trials,” Diabetes Technology & Therapeutics 24, no. 9 (September 2022): 603–610, 10.1089/dia.2022.0130. [DOI] [PubMed] [Google Scholar]
- 32. Deshmukh H., Wilmot E. G., Choudhary P., et al., “Time Below Range and Its Influence on Hypoglycemia Awareness and Severe Hypoglycemia: Insights From the Association of British Clinical Diabetologists Study,” Diabetes Care 48, no. 3 (March 2025): 437–443, PMID: 39746160; PMCID: PMC11870288, 10.2337/dc24-1833. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 33. Gabbay M. A. L., Rodacki M., Calliari L. E., et al., “Time in Range: A New Parameter to Evaluate Blood Glucose Control in Patients With Diabetes,” Diabetology & Metabolic Syndrome 12, no. 1 (March 2020): 22, 10.1186/s13098-020-00529-z. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 34. Lu J., Chen D., Lin B., et al., “A CGM‐Based Model for Predicting Hypoglycemia in Type 2 Diabetes Patients With TIR in Target,” Diabetology & Metabolic Syndrome 17, no. 1 (May 2025): 169, 10.1186/s13098-025-01713-9. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 35. Wu J., Wu J., Zhou Y., Lu X., Zhao W., and Xu F., “Nomogram for Predicting Hypoglycemia in Type 2 Diabetes Mellitus Patients Treated With Insulin Pump During Enteral Nutrition,” Diabetes, Metabolic Syndrome and Obesity 17 (May 2024): 2147–2154, 10.2147/DMSO.S436390. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 36. Han K., Yun J. S., Park Y. M., et al., “Development and Validation of a Risk Prediction Model for Severe Hypoglycemia in Adult Patients With Type 2 Diabetes: A Nationwide Population‐Based Cohort Study,” Clinical Epidemiology 10 (October 2018): 1545–1559, 10.2147/CLEP.S169835. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 37. Freeman N. L. B., Muthukkumar R., Weinstock R. S., Wickerhauser M. V., and Kahkoska A. R., “Use of Machine Learning to Identify Characteristics Associated With Severe Hypoglycemia in Older Adults With Type 1 Diabetes: A Post‐Hoc Analysis of a Case‐Control Study,” BMJ Open Diabetes Research & Care 12, no. 1 (February 2024): e003748, 10.1136/bmjdrc-2023-003748. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 38. Lara‐Abelenda F. J., Chushig‐Muzo D., Wägner A. M., Tayefi M., and Soguero‐Ruiz C., “Interpretable and Multimodal Fusion Methodology to Predict Severe Hypoglycemia in Adults With Type 1 Diabetes,” Engineering Applications of Artificial Intelligence 144 (2025): 110142, 10.1016/j.engappai.2025.110142. [DOI] [Google Scholar]
- 39. Harvengt A., Bastin M., Toussaint C., Beckers M., Helleputte T., and Lysy P., “Development of a Prediction Model for Severe Hypoglycemia in Children and Adolescents With Type 1 Diabetes: The Epi‐GLUREDIA Study,” Nutrients 17, no. 16 (August 2025): 2610, 10.3390/nu17162610. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 40. Oriot P. and Hermans M. P., “Intermittent‐Scanned Continuous Glucose Monitoring With Low Glucose Alarms Decreases Hypoglycemia Incidence in Middle‐Aged Adults With Type 1 Diabetes in Real‐Life Setting,” Journal of Diabetic Complications 37, no. 2 (February 2023): 108385, 10.1016/j.jdiacomp.2022.108385. [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Supporting Information S1
Supporting Information S2
Supporting Information S3
Data Availability Statement
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.