Hypoglycemia and Risk of Cardiovascular Disease and All-Cause Mortality in Insulin-Treated People With Type 1 and Type 2 Diabetes: A Cohort Study

Hypoglycemia is associated with a number of blood glucose–lowering therapies, but is a particular problem for patients receiving insulin therapy. As the most effective method of lowering blood glucose levels, insulin therapy is an essential element of treatment for people with type 1 diabetes and longer-duration type 2 diabetes. However, hypoglycemia is a major obstacle to achieving optimal glycemic control (1–4). A large prospective study by the UK Hypoglycaemia Study Group (5) demonstrated that severe hypoglycemia is a common problem in people with insulin-treated type 2 diabetes and that the incidence increases with the duration of insulin therapy. The acute effects of hypoglycemia are well documented, yet the relationship between hypoglycemia and long-term cardiovascular (CV) risk has been harder to elucidate, largely because of the complex nature of diabetes and the presence of comorbidities.

Large-scale diabetes outcomes trials (2,4,6–8) have presented conflicting results regarding the impact of intensive blood glucose–lowering therapy on mortality and CV risk. Notwithstanding, there is a growing consensus that hypoglycemia is a risk factor for CV events, particularly among high-risk patients (6,8–10).

In the ADVANCE, ACCORD, and VADT trials, severe hypoglycemia was significantly higher in the intensive glucose-lowering arms compared with the standard arms, as follows: ACCORD 16.2% vs. 5.1% (6); VADT 21.2% vs. 9.9% (9); and ADVANCE 2.7% vs. 1.5% (7). All these trials recruited participants who were at high risk for CV disease (CVD). The much lower risk for severe hypoglycemia in ADVANCE may be due to the fact that the patients in that trial had a shorter duration of diabetes, by 2–3 years, and lower HbA_1c level at study entry, despite very little use of insulin at baseline (11). A substudy from the treat-to-target ADVANCE trial investigated the impact of severe hypoglycemia (defined as requiring third-party assistance) on long-term CV risk and found that it was associated with a significant increase in the adjusted risk of major macrovascular events (hazard ratio [HR] 2.88 [95% CI 2.01, 4.12]), major microvascular events (1.81 [1.19, 2.74]), death from CV cause (2.68 [1.72, 4.19]), and death from any cause (2.69 [1.97, 3.67]) (12). A recent systematic review (13) collated data from cohort studies to evaluate the association between severe hypoglycemia and CV events in patients with type 2 diabetes. The study, which screened 34,443 citations (excluding studies from short-term hospital settings), reported a twofold increased risk of CV events among patients who experienced severe hypoglycemia (rate ratio 2.05 [95% CI 1.74, 2.42]) (13). Interestingly, the bias analysis suggested that comorbidity alone could not explain this association.

In patients with type 1 diabetes, the DCCT study (14) did not show any significant association between more frequent severe hypoglycemia and increased CV mortality among patients in the intensive treatment group. The Epidemiology of Diabetes Interventions and Complications (EDIC) study showed a reduction in CVD during follow-up; however, the risk of CV events among the people who experienced severe hypoglycemia was not reported, and this issue has scarcely been explored in clinical and epidemiological studies. The EURODIAB Prospective Complications Study (15) of 2,181 patients with type 1 diabetes, who were observed for 7 years, reported no association of baseline hypoglycemia with the risk of CVD. However, a retrospective analysis (16) of a large cohort of patients with type 1 diabetes on continuous subcutaneous insulin infusion pointed to a higher prevalence of CVD in patients with a history of repeated episodes of hypoglycemia. Further exploration of any potential relationship between severe hypoglycemia and CV events is vitally important because the large number of patients requiring insulin therapy means that even a small increase in risk could have major clinical and public health implications. In addition, a link between hypoglycemia and CV risk would be of particular consequence for those insulin-treated patients who require treatment intensification, especially when they have pre-existing CVDs or elevated CV risk factors. The current study explored whether there is an association between severe hypoglycemia and CV risk and mortality in a contemporary, real-world setting of people with type 1 or type 2 diabetes who were treated with insulin. In addition, we wanted to quantify the risk separately for those with and without pre-existing CVD.

This retrospective cohort study used data from the Clinical Practice Research Datalink (CPRD) database and included all insulin-treated patients (≥30 years of age) with a diagnosis of diabetes (Read/OXMIS code C10+) in the period between 1 January 2001 and 31 December 2007. The CPRD holds anonymized, longitudinal, primary care records of ∼5% of the population registered with a general practice in the U.K. and is widely used for epidemiological research. Patients registered with general practices participating in the database represent a population that is similar to, and therefore representative of, that of the U.K. (http://www.cprd.com/). Participating practices follow an agreed protocol for the collection of demographic, clinical, laboratory, and prescription data, and regularly submit anonymized records to the database. The patient-level primary care records in the CPRD are now linked to Hospital Episode Statistics (HES) data. The HES database contains details of all hospital admissions, outpatient appointments, and emergency attendances at National Health Service hospitals in England (not Northern Ireland, Scotland, or Wales). Only the subset of people in the CPRD from England who were eligible for linkage to HES was included in our study. All recordings of hypoglycemia in the HES database were severe (defined as requiring hospital admission), whereas those in the CPRD may be severe or nonsevere. Data on hypoglycemia recorded in CPRD plus HES and CPRD alone were analyzed. The CPRD-alone analysis excluded HES recordings of hypoglycemia: both analyses were performed in the same cohort of patients. CPRD plus HES data yield estimates for all recordings of hypoglycemia, both in general practice and on admission to hospital, whereas CPRD alone is more representative of those events reported only in general practice.

The last follow-up date was set on 31 December 2010. The index date for patients was the start of insulin treatment (identified by the second insulin prescription) after 1 January 2001. Data on hypoglycemic episodes were obtained from the HES via ICD-10 codes (E16.0, E16.1, and E16.2). Patients with a record (one or more episode) of severe hypoglycemia before the initiation of insulin therapy were excluded from the analysis. Patients were classified as exposed from the time of the first hypoglycemic episode or were otherwise considered to be unexposed, thus avoiding immortal time bias. Separate cohorts for people with type 1 and type 2 diabetes were created. Patients were classified into diabetes groups following the Royal College of General Practitioners guidelines (17), as follows: classification into diabetes type relies on a combination of Read codes, therapy, and age at diagnosis (18). Measurements of HbA_1c and BMI were recorded in a window extending to 3 months on either side of the index date. Patients with no data on HbA_1c levels, smoking, and/or BMI were categorized separately to be able to characterize these groups.

CV event was defined as a composite of myocardial infarction, stroke, or CV death (cause of death obtained through linkage to Office for National Statistics mortality data). The numbers of CV events were counted for each exposure group, and were only considered as happening in exposed risk time if patients had experienced a hypoglycemic episode before the CV event, thus retaining the time relationship between exposure and outcome. Patients with a registered code for any CV event in the HES or CPRD, which took place before the index date, were classified in a group with a history of CVD. (Read/OXMIS codes for diabetes, hypoglycemia, CV events, and smoking are provided in Supplementary Table 1.)

Basic statistics on the study population were presented as n (%), mean ± SD, or median (interquartile range [IQR]), as appropriate.

The rates of events were calculated as the number of events per 1,000 person-years along with the 95% CI using a standard life table analysis technique. For all patients, the exposure time started from the initiation of insulin treatment. To evaluate the risk of events associated with hypoglycemia, the “time to event” for those who experienced at least one hypoglycemic event was calculated from the time of the first episode of hypoglycemia. Stratified univariate and multivariate Cox regression models were fitted to estimate the risk of vascular events and mortality associated with hypoglycemia, with practice-level deprivation score (derived from the Index of Multiple Deprivation [19]) as the stratification factor. The covariates with complete data in the multivariate models were as follows: age on index date, sex, smoking status, geographical region, history of CV events before index date, use of oral antidiabetic medications, Charlson comorbidity index (20), BMI, and HbA_1c level on the index date. A sensitivity regression analysis was performed including only persons with known values in all covariates.

This protocol (13_027RMn) was approved by the Independent Scientific Advisory Committee.

The primary data set included a total of 265,868 individuals who had received a diagnosis of diabetes. After exclusion criteria were applied for the combined CPRD plus HES population, 3,260 patients with type 1 diabetes and 10,422 patients with type 2 diabetes were included in the regression analyses. (The results of the exclusion criteria are presented in Supplementary Table 2.) The baseline characteristics of people with type 1 and type 2 diabetes are shown in Table 1. Missing values for HbA_1c level, BMI, and smoking status did not affect the direction of the estimates. The number of patients in the type 1 and type 2 diabetes cohorts with data for blood pressure (systolic and diastolic) was 2,555 and 8,087, respectively, and for lipid parameters the number of patients in each diabetes cohort ranged between 805 and 1,800, and 2,636 and 5,569, respectively.

During follow-up, 573 patients (18%) with type 1 diabetes and 1,463 patients (14%) with type 2 diabetes experienced hypoglycemia. The proportions of patients experiencing CV events or death were broadly similar both in patients with type 1 diabetes and type 2 diabetes (Table 2). The median duration of follow-up among patients with type 1 diabetes was 5.0 years, while for patients with type 2 diabetes it was 4.8 years. The incidence rates (per 1,000 person-years) of CV outcomes among patients with a history of CVD were 3.5-fold and 2.8-fold higher, respectively, in the type 1 and type 2 diabetes cohorts, compared with those patients without a history of CVD. Death rates were twofold higher in both cohorts for patients with a history of CVD (Table 3).

During follow-up, patients who experienced hypoglycemia were at a greater risk of CV events. Unadjusted and covariate-adjusted CV risk HRs for exposed patients with type 1 and type 2 diabetes are presented in Table 3. Among patients without a history of CVD, patients who experienced at least one episode of hypoglycemia had a 92% and 50% significantly increased risk, respectively, of composite CV events in type 1 and type 2 diabetes cohorts (CPRD plus HES population, Table 3). Hypoglycemia was also significantly associated with the increased risk of CV events among patients with a history of CVD in the type 2 diabetes cohort. It is important to mention here that the adjusted and unadjusted risk estimates associated with hypoglycemia were similar in the CPRD plus HES data, separately for the two cohorts of patients with type 1 and type 2 diabetes. The assumption of proportional hazards was not violated in the Cox regression analyses.

In the type 1 diabetes cohort, hypoglycemia was associated with an approximately twofold increased risk of all-cause mortality for both patient groups with and without the history of CVD (Table 3, CPRD plus HES data). In the type 2 diabetes cohort, after adjusting for the effects of possible confounding factors, hypoglycemia was associated with HRs of 94% (95% CI 1.52, 2.47) and 139% (2.13, 2.67), respectively, increased mortality risk among patients with and without history of CVD. The significant association of hypoglycemia with the increased mortality risk was also observed in the CPRD group (with the exception of an insignificant 53% increased risk among patients with a history of CVD in the type 1 diabetes cohort). Among patients with type 2 diabetes, those prescribed metformin had approximately half the risk of death compared with patients receiving sulphonylureas, both in patients with a history of CVD (HR 0.49 [95% CI 0.28, 0.87]) and in those without such a history (0.54 [0.42, 0.72]; P < 0.05). There was no difference in HR for CV events between metformin-treated and sulphonylurea-treated patients (data not shown). The assumption of proportional hazards was not violated in the Cox regression analyses.

The median time (from CPRD plus HES data) from the first episode of hypoglycemia to the first CV event in patients with type 1 diabetes was 1.5 years (IQR 0.5, 3.5 years) (n = 38 patients with hypoglycemia who experienced at least one CV event). Among patients with type 2 diabetes, the median time from the first episode of hypoglycemia to the first CV event was 1.5 years (IQR 0.5, 3.0 years) (n = 97 patients with hypoglycemia experienced at least one CV event). The median time (from CPRD plus HES data) from the first episode of hypoglycemia to death in patients with type 1 diabetes was 1.1 years (IQR 0.3, 2.3 years) (n = 169 patients who experienced at least one episode of hypoglycemia died). Among patients with type 2 diabetes, the median time from the first episode of hypoglycemia to death was 0.8 years (IQR 0.3, 2.3 years) (n = 493 patients who experienced at least one episode of hypoglycemia died).

The results of this retrospective cohort analysis demonstrate that, in a nationally representative contemporary population, hypoglycemia is associated with an increased risk of CV events and all-cause mortality in insulin-treated patients with type 1 and type 2 diabetes. These results confirm the findings of the ACCORD study in people with type 2 diabetes, and add to the evidence for a link between hypoglycemia and CV events in people with type 1 diabetes. Likewise, our analysis also reconfirms that the increased risk of CV events after a hypoglycemic episode persists over a period of months and years.

CVD remains the primary cause of death among insulin-treated people with diabetes (21,22). This is largely due to the effects of hyperglycemia and pre-existing comorbidities, but the role of hypoglycemia in contributing to CV events or mortality, even if modest, is still important. Since the ACCORD and VADT results were published, attempts have been made to assess whether a relationship exists between hypoglycemia and CV risk or mortality, but only recently has a consensus begun to emerge suggesting that there is a link between the two (6,8,9).

Several potential mechanisms linking hypoglycemia and CV risk have been proposed, including QT prolongation, hemodynamic changes arising from catecholamine release, inflammation, and endothelial dysfunction (23–25). Although real, these effects are more likely to explain a transient increase in CV risk during the acute phase of hypoglycemia rather than the longer-term relationship observed in the current study. Our investigation has not attempted to make a causal link between these observations, but we have highlighted the importance of CV risk in patients who experienced hypoglycemia by showing that hypoglycemia is associated with very early CV events and mortality. Indeed, there are common characteristics among patients with diabetes who are at high risk of experiencing hypoglycemia or CV events, notably an increased duration of insulin treatment and age (5,26). The relationship between severe hypoglycemic events and CV outcomes may be due to a patient’s health deteriorating much more at the time of their event than is the case for nonsevere hypoglycemic episodes. Conversely, our study, which includes both nonsevere and severe events, suggests that the relationship between hypoglycemia, and CV outcomes and all-cause mortality exists over a long period, and therefore possibly before any worsening of a patient’s health. Patients with type 2 diabetes treated solely with oral hypoglycemic agents were not included in our analysis. This is because these patients typically have a shorter duration of diabetes and variable levels of clinical inertia (27)—confounders that would be difficult to adjust for. In our study, CV risk and all-cause mortality (not HR) were higher among patients with a history of CVD compared with those without. When interpreting HRs, we are looking at the association between hypoglycemia and CV risk/all-cause mortality. This association is often stronger in those patients without a history of CV, and hence the HR is often higher in this subgroup. Possible explanations are that the association between hypoglycemia and CV risk/all-cause mortality is masked in those patients who have previously experienced a CV event, or that hypoglycemia is not as strong a predictor in this case.

The current study has several limitations. First, as there is no randomization of patients in exposure groups in observational studies, it is important to correct for bias and confounding. Not all relevant confounding factors are captured in the CPRD database; thus, predicting causality was not possible. Under-reporting of hypoglycemia in the CPRD is another issue potentially affecting our analysis because patients may often self-manage an episode and not present to a physician. This may be compounded by the exclusion of non-HES–eligible patients. Furthermore, we expect that reported hypoglycemic events will be more severe than those not registered, but we do not have any data available to confirm this. Patients’ second prescription of insulin was used as the index date to exclude erroneous registrations. Some patients may have started their insulin therapy in a hospital clinic and not received their insulin prescription from their general practitioner until later in the course of their illness.

Although variables like BMI, smoking status, and HbA_1c level at index are well captured with the introduction of the Quality and Outcomes Framework (QOF), they were not available for all patients during the 6-month window around the index date. It was not possible to accurately calculate the duration of diabetes in our analysis because patients may have received a diagnosis many years previously or in practices not included in the CPRD, but we believe that age is a good proxy for the duration of diabetes. Likewise, recordings of blood pressure and lipid levels in the CPRD are sparse, and were not incorporated in the main analysis. However, we do not believe that this invalidates the findings, as the measures would not independently alter the outcomes. Supporting our approach, a sensitivity analysis, carried out using the available lipid and blood pressure data in a subset of patients with known values for all covariates, did not change the size or direction of the estimates significantly (data not shown in the 5Results section). Finally, our study only included patients with Read codes in C10+. We believe that this is a minor limitation as this study focuses on insulin-treated patients only, and the QOF has resulted in more standardized use of Read codes by general practitioners as correct coding of patients leads to a higher prevalence of diabetes, and hence to higher QOF payments for general practices.

This present study has several strengths. First, the sample size was large and was derived from the CPRD, which is representative of patients across the UK. As the UK has a health system with universal coverage, this means that no section of the population is excluded from the CPRD (28). Additionally, our study takes place within the QOF period, beginning in 2004. QOF has greatly improved the completeness and accuracy of the data recorded in the CPRD (29,30). The study assessed type 1 and type 2 diabetes together, comparing the effect of hypoglycemia on CV risk and all-cause mortality in both populations. The high quality and nationally representative nature of the data used in the present analysis affirms its validity, especially in comparison with smaller studies, or with those using less robust or less well-curated indexes; however, we acknowledge that the selection criteria applied to our cohort could potentially affect its representative quality, particularly among the type 2 diabetes cohort. Despite this limitation, we believe that the results of the type 2 diabetes cohort play an important role in informing the debate concerning the growing body of evidence associating severe hypoglycemia with CV risk. Second, our study took a robust approach to the analysis, adjusting for multiple covariates, including the practice-level deprivation and Charlson comorbidity scores. Unlike the ACCORD study, which only analyzed severe hypoglycemia, our study captures all hypoglycemia (CPRD) and severe hypoglycemia (HES). This provides a broader picture of the potential impact of hypoglycemia on CV risk. Moreover, a recently published analysis (31) of national health record services has suggested that general practices submitting data to the CPRD are likely to be better at recording disease events compared with those practices not participating in the CPRD.

In conclusion, our study confirms a relationship between hypoglycemic events and an increased risk of CV outcomes and all-cause mortality in patients with both type 1 and type 2 diabetes. Furthermore, the association between these outcomes endures over a long period. The nature of our analysis precludes us from making a causal link; however, hypoglycemia has been proposed as a surrogate measure of greater disease burden and thus an indirect marker of CV risk (32). Based on the results of our analysis and previously published studies, we recommend that special attention should be paid to insulin-treated patients who experience a hypoglycemic event—especially those who are at elevated risk for CV events—and an effort should be made to reduce the incidence of future hypoglycemic events.

Acknowledgments. The authors thank Michael Soljak, PhD, Imperial College London, for his advice on analyzing smoking status in the current study.

Funding. This work was supported by both the East Midlands National Institute for Health Research (NIHR) Collaboration for Leadership in Applied Health Research and Care (CLAHRC) and the NIHR CLAHRC for Northwest London; the Leicester–Loughborough Diet, Lifestyle and Physical Activity Biomedical Research Unit; the Imperial NIHR Biomedical Research Centre; and the Imperial Centre for Patient Safety and Service Quality and received infrastructure support from Therapeutic Innovation Australia.The opinions expressed in this article are those of the authors and do not necessarily reflect the views of the NIHR, the Biomedical Research Centre, or the Imperial Centre for Patient Safety and Service Quality.

Duality of Interest. Medical writing assistance and editorial/manuscript submission support was funded by Novo Nordisk. The authors thank Paul Tisdale, PhD, and Daria Renshaw at Watermeadow Medical, U.K., for medical writing assistance and editorial support, respectively (sponsored by Novo Nordisk). K.K. has acted as a consultant, advisory board member, and speaker for and has received research grants from Novo Nordisk, Eli Lilly, Merck Sharp & Dohme, Bristol-Myers Squibb, AstraZeneca, Sanofi, Boehringer Ingelheim, and Roche. M.D. has acted as a consultant, advisory board member, and speaker for AstraZeneca, Novo Nordisk, Sanofi, Eli Lilly, Merck Sharp & Dohme, and Boehringer Ingelheim and has received grants in support of investigator and investigator-initiated trials from Novartis, Novo Nordisk, Sanofi, Eli Lilly, Pfizer, Merck Sharp & Dohme, and GlaxoSmithKline. B.L.T. and M.L.W. are employees and shareholders of Novo Nordisk A/S. S.K.P. has acted as a consultant and speaker for Novartis and Amylin Pharmaceuticals, LLC and has received grants in support of investigator and investigator-initiated clinical studies from Merck Sharp & Dohme, Novo Nordisk, Pfizer, Amylin Pharmaceuticals, Bristol-Myers Squibb, and Hospira. No other potential conflicts of interest relevant to this article were reported.

Author Contributions. K.K. and S.K.P. contributed equally to the conception and design of the article, the analysis plan, and the data interpretation. M.D. contributed to data interpretation and the writing and revision of the article. A.M. contributed to hypothesis generation and to the drafting of the manuscript. B.L.T. contributed to the study design and data analysis and interpretation. M.L.W. contributed to the study design and data interpretation. K.K. 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.