Incidence Trends and Predictors of Hospitalization for Hypoglycemia in 17,230 Adult Patients With Type 1 Diabetes: A Danish Register Linkage Cohort Study

Hypoglycemia is considered a major challenge to glucose-lowering therapy among patients with type 1 diabetes, with 90% of all patients who receive insulin experiencing hypoglycemia (1). Hypoglycemia can range from asymptomatic episodes that remain unrecognized by the patient to mild symptomatic events managed by the patient alone and severe hypoglycemia with cognitive disruption and need for assistance from others. Ultimately, severe hypoglycemic episodes may require hospital treatment as a result of severity or lack of treatment options in the surroundings. Such episodes are feared by patients and relatives and generate considerable societal costs.

Numerous studies have measured the incidence rate of hypoglycemia and severe hypoglycemia in type 1 diabetes, but results have varied considerably depending on the definition of hypoglycemia, group selection (whether high-risk patients were included or not), and the type of study (2). Over the last decades, insulin treatment has changed with increased use of analog insulin, insulin pump treatment, and continuous glucose monitoring (3). However, it is not clear to what extent recent advances in type 1 diabetes care has reduced the incidence of severe hypoglycemia, and we are not aware of any study that examined the recent trends in hospitalization for hypoglycemia (HH) incidence in adults and older patients with type 1 diabetes in a large sample. This study aims to describe the time trends in HH in adults and older patients with type 1 diabetes in Denmark from 2006 to 2012 and to investigate predictors of incident HH.

The study base comprised all patients diagnosed with type 1 diabetes aged 16 years and older who have been registered in the Danish Adult Diabetes Database (DADD) during 2006 to the end of 2011. In total, data from 23,268 patients were extracted from the DADD register. DADD is a nationwide clinical register, established in 2005, operated and financed by the Danish Regions (4). The aim of this register is to monitor the quality of diabetes care in Denmark. Hospital outpatient clinics and general practitioners are obliged to report annual clinical data for all patients with diabetes aged ≥16 years, including information on date of birth, sex, weight, height, smoking, date of diabetes diagnosis, HbA_1c, blood pressure, antihypertensive drug use, dyslipidemia drug use, urinary albumin-to-creatinine ratio (ACR), and status date (the date of clinical information updated/reported in DADD). The data completeness in DADD has been estimated to be >95%. The DADD data were linked to the following registries using the unique personal identification number for all people in Denmark: 1) the Danish Register of Causes of Death (information on cause of death and date of death available since 1970 covering all deaths in Denmark) and 2) the Danish National Patient Register (all somatic hospital admissions since 1977, including information on HH). Since 1995, data on outpatients and emergency contacts are also included.

The diagnosis of type 1 diabetes in DADD is based on the clinical definition according to national guidelines provided by the reporting specialist clinic. HH was defined as the ICD-10 codes E100, E110, E120, E130, E140, and E162; this wide range of code for HH was considered to ensure coverage of all HH events. However, the codes E100 and E120 made up the largest part of HH diagnoses (87% and 7%).

Baseline characteristics were derived from DADD and measured based upon the patients’ date of entry in DADD, as shown in Table 1. BMI was categorized as underweight (<18.50 kg/m²), normal weight (≥18.50 to ≤24.99 kg/m²), overweight (≥25 to ≤29.99 kg/m²), and obese (≥30 kg/m²) adapted from the World Health Organization; albuminuria (ACR) was stratified into normoalbuminuria (<3 mg/mmol), microalbuminuria (3 to ≤30 mg/mmol), and macroalbuminuria (>30 mg/mmol); smoking was characterized into never smoker, ex-smoker, occasional smoker, and daily smoker; dichotomous variables were sex (male or female), antihypertensive drug use (yes or no), and lipid-lowering treatment (yes or no). Age (age measured at entry), age at diabetes onset (years), diabetes duration (years), HbA_1c (%/mmol/mol), and systolic blood pressure (mmHg) were included as continuous variables.

HH was defined as an event when it required an outpatient, inpatient, or emergency care visit with the previously mentioned corresponding diagnostic code to ICD-10 code for HH.

All patients with type 1 diabetes in DADD were followed from or earliest status date in DADD (May 2005 and onward) prospectively to the end of 2011 or death, whichever came first. We excluded episodes of HH that followed <1 week after a previous event, as these are considered to be double registrations. A group of patients had their first entry in DADD by 31 December 2011 and were excluded as their follow-up period was estimated to be <1 day. In addition, due to erroneous coding, individuals aged <16 years old at entry in DADD, diabetes duration longer than age, BMI <10 or >100, or HbA_1c <0 were also excluded (total n = 1,455). Finally, 17,230 individuals (9,873 men and 7,357 women) were included in the study.

The follow-up of patients was subdivided at the dates of recorded HH, and the number of HH events prior to entry of each patient were counted, enabling us to classify each interval of follow-up by the number of previously seen HH events (0, 1,. . .5) (Supplementary Fig. 1). Each follow-up interval thus ended at death, next HH, or censoring (at the end of the study period). Finally, follow-up intervals were subdivided in 3-month intervals along the calendar timescale. For each interval we could then compute the current age, current date, current duration of diabetes, and, for follow-up of people with at least one prior HH, the time since last HH event, thus rendering four different timescales. HH occurrence rates were modeled in a multistate model by Poisson regression using the event indicator (next HH) as outcome and the log person-years as offset. The same person may contribute several events, but because we are modeling these as separate types of events (with different rates), different events within the same individual contribute to the underlying likelihood as were they independent Poisson observations. Mortality rates were not modeled.

We modeled the HH occurrence rates by cubic splines (three parameters each) for the four timescales, including an interaction between number of previous HH events and time since last HH event, by sex, smoking status, and use of antihypertensive medication and lipid-lowering medication and by linear terms of the clinical variables HbA_1c, BMI, and systolic blood pressure. Because a substantial fraction of patients had missing values of albumin, the albumin effect (using the log10 transform of it) was estimated in a secondary (presented) analysis, including albumin in addition to the other variables. We tested for nonlinearity of all included continuous variables by including a quadratic term and using likelihood ratio tests. A full description of the analysis can be seen at http://bendixcarstensen.com/SDC/KZIA/HH-DK.pdf.

Register-based studies do not require ethical approval in Denmark. The Danish Data Protection Agency approved both the access and the use of the described data in this study (2007-58-0015, SDC-2013-003, serial number 02090).

A total of 18,285 patients, of whom 42.8% (7,821) were female, contributed a total of 70,002 patient-years during the follow-up period of May 2005 to the end of 2011. The analysis was based on 17,230 patients with all clinical variables known and follow-up to the sixth HH; 65,476 patient-years and a total of 2,369 events of HH among 1,735 patients. Before December 2011, 3.5% (609 patients) died.

Baseline characteristics are presented in Table 1 for 17,230 patients with clinical information. Median age at entry was 45.6 years, the median diabetes duration was 16.3 years, and median HbA_1c in the patient population was 64 mmol/mol. Almost half of the population was overweight, one-third were daily smokers, and one-third were treated with lipid-lowering or antihypertensive drugs. Twenty-five percent had one or more previous HH events at entry. In addition, a nonsignificant increase in mean HbA_1c was observed from 2006 to 2012.

The overall crude incidence of HH during follow-up was 3.38 per 100 patient-years and decreased from 2006 to 2012. Total number of HH events and crude incidence rates according to calendar year and sex are presented in Table 2. There was a steady decline in crude incidence rates over the 7 years of follow-up, with a reduction from 2005 at 5.1 per 100 patient-years to 2.8 per 100 patient-years in 2011. We found no evidence of nonlinear time trends with HH (P = 0.826), and a linear model of calendar time showed an annual decrease in HH of 8.4% (4.9–11.7%), confirming the initially observed time trends. Predicted rates of first HH occurrence excluding those with previous HH at entry are shown in Supplementary Fig. 2 as a function of age for the years 2006 and 2011, and approximate median values of all other variables in the model.

There was a strong nonlinear effect of age on HH with a rate ratio of ∼1.6 for patients at age 20 and 80 years compared with middle-aged patients (Fig. 1B). We found a modest but clear increase in HH incidence with increasing diabetes duration (Fig. 1C). Previous HH was strongly associated with HH with more than threefold increased risk of HH for patients with four previous events of HH compared with patients with no events, gradually lower with fewer previous events (Fig. 1D). Furthermore, the highest relative risk was observed within the first years after the last HH event. We found no interaction between time since last HH and number of previous HHs (P = 0.239); thus we report estimates from a model with a (log-) linear effect of calendar time and proportional hazards along the time since HH event between different number of previous HH events.

HbA_1c and albuminuria were significantly associated with incident HH (P < 0.05), but sex, BMI, smoking, and antihypertensive and lipid-lowering treatment were not significantly associated with HH (Table 3). The association with HbA_1c was nonlinear, with a marked increase in HH risk at HbA_1c >60 mmol/mol (Fig. 1A). Categorization into macro-, micro-, and normoalbuminuria showed similar associations (hazard ratio [HR] 1.98 [95% CI 1.21–3.22] and HR 1.49 [95% CI 1.22–1.84] vs. normoalbuminuria).

This longitudinal study provides the most recent data on time trends in HH, with information available until 2012. The study shows a steady decline of HH incidence rates during 2006–2012 in patients with type 1 diabetes in Denmark. The incidence rates observed were remarkably low compared with previous studies in patients with type 1 diabetes (5–8). This study also reports a number of risk factors of HH, including past history of HH, albuminuria, long diabetes duration, and high HbA_1c in adults with type 1 diabetes.

A novel finding of our study was the remarkably low rates of HH observed in patients with type 1 diabetes with a steady decline from 2006 to 2012. Moreover, this decrease in incidence rates was observed in the fully adjusted model. These low rates of incidence of HH are comparable with a recent study in the U.S. where the rate for an emergency department visit due to hypoglycemia reported was 1.8 in 2006 and 1.4 in 2011 per 100 patient-years; however, the study was unable to classify diabetes into type 1 and type 2 (5). Another recent U.S. study in a large group of patients with diabetes reported incidence rates of any hypoglycemia-related emergency care visits or inpatient encounter in the range of 1.4–1.6 per 100 patient-years during 2005–2011 (6). Nevertheless, other studies that estimated high incidence rates of HH differ from our study in several aspects. A Scottish study that reported a rate of 11.5% per year in patients with type 1 diabetes in the study period from 1997 to 1998 (7) defined hypoglycemia as an episode that required emergency treatment from primary care, ambulance, accident, emergency, or hospital services. Another study in Minnesota reported that 18% of patients with type 1 diabetes were admitted to the hospital due to hypoglycemia during 7 years of follow-up. In this study, hypoglycemia was defined as any ambulance calls and emergency medical services due to hypoglycemia in patients with type 1 diabetes (8). Most patients in the cohort of Minnesota were treated with multiple daily insulin injections and had a high degree of comorbidities. This may explain the high rate together with the broad definition of hypoglycemia (8). The steep decline in HH with calendar time in our study is in line with some recent studies; Wang et al. (5) in their nationwide study in the U.S. observed decreasing trends of emergency department visits due to hypoglycemia for all age and sex groups. Another recent U.S. study observed decreasing trends of HH in all age and sex groups in older Medicare beneficiaries all over the U.S. (9). However, none of these studies were able to scrutinize incidence trends of HH in adult patients with type 1 diabetes.

These encouraging findings are most likely due to several improvements in diabetes care. Changes in clinical practice by the implementation of the American Diabetes Association/International Society for Pediatric and Adolescent Diabetes guidelines in 2007 with glycemic target settings (HbA_1c 7.5% [58 mmol/mol]) and widespread use of analog insulin, improved diabetes education, and frequent use of insulin pumps with or without continuous glucose monitoring in recent years may have contributed to this improvement in severe hypoglycemia control in Denmark (10–12). Furthermore, it is relevant to mention that as part of the improvement in diabetes complication treatment, it is likely that more cases of severe hypoglycemia are now treated at home with glucagon injection by acute ambulance services or by educated family members. Finally, the reduction of HH observed during 2011 and 2012 may be explained by increased family treatment as a consequence of the implementation of stricter new EU driver licensing regulations disqualifying patients with recurrent severe hypoglycemia from maintaining their license (13). Unfortunately, no register information on prehospital treatment is available in Denmark.

Like most of the existing studies, we also found that previous HH was a strong predictor of HH in this patient group (14–16). Risk of HH was increased by ∼3 times, 2.6 times, ∼2 times, and 1.6 times, respectively, in the patient groups with four or more events, three events, two events, and one episode of HH before study entry compared with those who had no event. None of the earlier studies examined past experience of HH similarly (14–16). Previous HH may be a marker of several risk factors, including impaired hypoglycemia awareness and poor self-care or lack of family support during daily life. Further studies are needed to understand this mechanism.

Longer diabetes duration was identified as a risk marker for HH in our analyses. This finding is consistent with several other large studies (7,15) and may be driven by a growing proportion of subjects with impaired hypoglycemia awareness with increasing diabetes duration. Development of complications in the natural progression of diabetes at a later stage could increase the risk of HH in patients with type 1 diabetes (17), supported by the fact that albuminuria was also associated with increased risk of HH. However, the current study could not adjust for patient’s comorbidity status, especially long-term diabetes complications and hypoglycemia unawareness, which needs to be taken into account when interpreting our results.

In contrast to previous large studies, including the Diabetes Control and Complications Trial (DCCT) in patients with type 1 diabetes, the current study observed a positive association of HH with increasing HbA_1c >60 mmol/mol (15,18). Lipska et al. (19) reported a U-shaped relationship and increased risk of HH with HbA_1c levels lower or higher than the normal 6–6.9% (42–52 mmol/mol) level. However, a possible explanation for the observed association is that patients with a high level of HbA_1c may have poor self-care as a predisposing factor and may be less familiar with handling hypoglycemia (the same is probably true for the families). On the other hand, as already mentioned, the introduction of new treatment guidelines in Denmark with glycemic target settings (HbA_1c 7.5% [58 mmol/mol]) (10) may have prevented some cases of HH associated with strict glycemic control in type 1 diabetes over the last decade.

Albuminuria appeared to be a strong independent marker of HH in our analysis; patients with macroalbuminuria (ACR >30 mg/mmol) had nearly a doubling of risk of HH compared with patients with normoalbuminuria (<3 mg/mmol). None of any previously mentioned studies have examined albuminuria as a predictor of HH in adult patients with type 1 diabetes. Yun et al. (20) in their 10-year follow-up study reported 2.5 times increased risk of HH in patients with type 2 diabetes. Patients with macroalbuminuria in our study might be under more aggressive treatment than their counterpart of patients with normoalbuminuria; and furthermore, macroalbuminuria is a marker of comorbidities, especially chronic kidney disease, which is widely considered as a risk marker for HH.

Being underweight increased the risk of HH compared with normal-weight patients in our study, although this finding was not statistically significant. These findings are in line with some previous studies (21,22). This study is unable to explain the mechanism behind this connection, although we assume poor nutritional status may potentially imply poor glycogen depots or eating disorder, which might explain this relation.

The main strength of the study lies in the large data set that covers the whole Danish population aged 16 years and older diagnosed with type 1 diabetes in recent years. The follow-up information on HH is individually linked and as such less susceptible to nominator-denominator bias compared with cross-sectional studies. The diagnosis of type 1 diabetes is clinically verified, which facilitated the inclusion of patients diagnosed during adulthood. Almost one in three patients in the population with type 1 diabetes is diagnosed after age 30 years, and information on this group of patients is particularly scarce in the literature. Other strengths are its prospective design with 7 years of follow-up, using register data with precise information about diabetes type, and unbiased hospital-registered HH event records, which indicates a precise estimation of incidence trends of HH in an adult population with type 1 diabetes. Moreover, the study was able to analyze a number of lifestyle-related factors and a wide variety of clinical risk factors of HH in patients with type 1 diabetes.

The study has some limitations. This study is incapable of revealing the total incidence and trends of severe hypoglycemia in type 1 diabetes in Denmark as a lot of severe hypoglycemia cases are treated at home by family members and acute ambulance services at prehospital triage (7,23–26). Furthermore, we were unable to distinguish whether the events of “hospitalized hypoglycemia” occurred as inpatient or outpatient or emergency care visits. The quality of the Danish National Patient Register data is often discussed and a few studies reported minor misclassification, especially where the coding practice is not clear (27–29). However, these misclassifications are nondifferential (27–29) and unlikely to influence the observed trends. Also, we did not have information on insulin type, ethnicity, socioeconomic status, lifestyle factors (including missing meals, alcohol intake, and exercise), hypoglycemia awareness, or diabetes education, which might influence the incidence of HH (30,31). As hypoglycemia unawareness may be a major risk factor for HH and may furthermore be an important factor underlying the association between diabetes duration and poor control with HH, systematic registration of hypoglycemia awareness in the quality registers would be useful for both clinical and research purposes.

In conclusion, this study of recent trends of HH in adult patients with type 1 diabetes revealed low incidence rates of HH, with a decreasing trend of incidence in Denmark from 2006 to 2012. However, due to increasing incidence of type 1 diabetes combined with a decline in mortality (32), the absolute number of cases of HH will probably continue to increase in the coming decades.

Acknowledgments. The authors thank Gregers Stig Andersen (Steno Diabetes Centre) for data managerial support.

Duality of Interest. B.C. and M.E.J. are employed by Steno Diabetes Centre A/S, a research hospital working in the Danish National Health Service and owned by Novo Nordisk A/S. B.C. and M.E.J. own shares in Novo Nordisk A/S. Steno Diabetes Centre receives part of its core funding from unrestricted grants from the Novo Foundation and Novo Nordisk A/S. U.P.-B. served on advisory boards for AstraZeneca/Bristol-Myers Squibb and Novo Nordisk and received lecture fees from AstraZeneca/Bristol-Myers Squibb, Sanofi, and Novo Nordisk. No other potential conflicts of interest relevant to this article were reported.

Author Contributions. K.I.-A. contributed to the study design, performed the data analyses, drafted the primary version of the manuscript, contributed to successive drafts of the manuscript, and approved the final version. B.C. contributed to the study design, performed the data analyses, contributed to successive drafts of the manuscript, and approved the final version. U.P.-B. contributed to successive drafts of the manuscript and approved the final version. M.E.J. contributed to the study design, contributed to successive drafts of the manuscript, and approved the final version. M.E.J. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.