Modification of the Association Between Severe Hypoglycemia and Ischemic Heart Disease by Surrogates of Vascular Damage Severity in Type 1 Diabetes During ∼30 Years of Follow-up in the DCCT/EDIC Study

Severe hypoglycemia (SH) is one major barrier to achieving good glycemic control (1,2). Although these events are still quite frequent (3,4), the role of SH as a predictor for cardiovascular disease (CVD) is not clearly understood. Just recently, the International Hypoglycaemia Study Group emphasized the need to understand consequences of hypoglycemia, especially for the cardiovascular system (5).

Particularly for type 2 diabetes (T2D), several studies have linked effects of SH on CVD/mortality to older patients with high cardiovascular risk score (CV-score) (6,7). Literature on SH as a cardiovascular risk factor in type 1 diabetes (T1D), mostly for younger populations, is inconsistent. In the EURODIAB Prospective Complications Study, no significant association between baseline (B-)SH and B-CVD was observed (8). In the Diabetes Control and Complications Trial (DCCT)/Epidemiology of Diabetes Interventions and Complications study (EDIC), it has been found that updated current (C-)SH had no effect on CVD (no short-term effect of SH on CVD) during a mean follow-up of 27 years (maximum 30 years) (9). However, epidemiological studies found that any-SH was significantly associated with CVD (10) and that a dose-gradient relationship between SH frequency and CVD existed (11). None of the aforementioned T1D studies investigated interactions between vascular complications severity and SH.

Theoretically, in the long-term, SH effects on the cardiovascular system could be related to the induction and progression of atherosclerosis (12). Conceptually, while a healthy circulation might be able to compensate for SH insults on the cardiovascular system, SH might have potentially deleterious effects on a vasculature that has become compromised by micro- or macroangiopathies (13). If this holds true, then diabetes duration, which correlates with increased prevalence and diabetes complications severity (13), microvascular complications severity, and CV-score, should modify an association of SH with CVD (significant tests of interaction).

It is known that adapted Diabetes Complications Severity Index and diabetes duration are important CVD risk factors in T1D (9–11). However, it is unknown whether these surrogates of vascular complications severity impact the relationship between SH and CVD.

Analysis of the DCCT/EDIC cohort provides an opportunity to gain important insight into the role of SH in CVD by investigating whether an association between CVD and any-SH exists in a cohort with low B-CV-score and whether such relationship is conditional. At baseline, the DCCT cohort was a young cohort of generally good health; volunteers were without apparent macrovascular complications. While at baseline a relatively low percentage of the enrolled population had very mild neuropathy or albumin excretion rates (AERs), ≥30 mg/24 h but <200 mg/24 h, roughly 50% had retinopathy of variable severity (14,15).

As for CVD, evidence for an association between SH and ischemic heart disease (IHD) seems more consistent than for an association between SH and stroke (5). Therefore, we focus on IHD as the cardiovascular end point.

We now investigate the hypothesis that baseline microvascular damage severity (microvascular damage) surrogates, namely, diabetes duration, steps on DCCT Early Treatment Diabetic Retinopathy Study severity scale (DCCT-ETDRS) (15,16), and Diabetes Complications Severity Index (DCSI) (17), affect a prospective association between IHD and any-SH.

To gain a better understanding of what determines an individual’s vulnerability to adverse effects of SH on the cardiovascular system, in exploratory analyses, we evaluate the interaction between CV-score and any-SH and assess IHD predictors among the subset of participants with at least one SH episode during ∼30 years of follow-up in DCCT/EDIC.

We analyzed publicly available DCCT/EDIC data of 1,441 volunteers with T1D (aged 13–39 years) who were enrolled in DCCT between 1983 and 1989 (14,15). Data from DCCT entry until 31 December 2013 were provided by the National Institute of Diabetes and Digestive and Kidney Diseases (NIDDK). Marshall University Institutional Review Board approved the current analysis.

DCCT/EDIC has been well described in the past (14,15,18,19). Briefly, DCCT participants were randomly assigned to two treatment therapies (intensive and conventional). Generally, participants were of good health. None of the volunteers had any apparent CVD signs. Individuals with neuropathy who required treatment were excluded from the study. There was a primary cohort including volunteers with 1–5 years of diabetes duration, no retinopathy, and AER <40 mg/24 h and a secondary intervention cohort including participants with 1–15 years of diabetes duration, at least one microaneurysm, and AER <200 mg/24 h. The DCCT lasted until 1993, after which the observational EDIC (still ongoing) commenced (18,19).

During DCCT/EDIC, cardiovascular outcome was assessed with standardized measures. Cardiovascular events (CVEs) were classified by an independent committee masked to DCCT treatment and HbA_1c levels. Supporting measures included medical records and annual centrally graded electrocardiograms. Additional details have previously been published (9,19).

The outcome of interest of this analysis was the time to the first occurrence of IHD, including cardiovascular (CV)-death, nonfatal/silent myocardial infarct (MI), revascularization procedures, or confirmed angina. We performed sensitivity analyses by assuming that: 1) no CV-deaths were of IHD origin; and 2) CV-deaths of participants with measured coronary calcification score >0 Agatston units were of IHD origin. We applied the midinterval time as time of IHD for three participants for whom the IHD event (IHE) was known only to have occurred within a specific time interval.

SH was defined as hypoglycemia requiring assistance of another person (20,21). Participants were considered to be exposed to SH from the time they experienced the first SH event (SHE) but otherwise considered unexposed. During DCCT, SHEs were reported at time of event or at quarterly visit. In EDIC, SH was reported for the 3 months prior to the annual visit (18,19).

B-diabetes duration, B-DCCT-ETDRS, B-DCSI, B-CV-score, and C-CV-score were considered potential modifiers of the association between SH and IHD. At baseline, diabetes duration serves as surrogate of microvascular complications severity. DCCT-ETDRS is one of the classification systems for diabetic retinopathy (16). The original DCCT-ETDRS ranged from 1 to 9 (15). Since only three participants were diagnosed with scale 9, scale 8 and 9 were combined. DCSI is an indicator capturing severity and type of complications at any point in time to predict risks of adverse outcomes (17). Complications include nephropathy, neuropathy, retinopathy, cerebrovascular, cardiovascular, peripheral vascular disease, and metabolic complications. Therefore, B-DCSI represents microvascular complications severity and ranges from 0 to 3. Based on DCSI, an AER ≥30 mg/24 h was considered an abnormality and given 1 point. A sensitivity analysis was performed based on contemporary guidelines (22), with nephropathy defined as AER ≥30 mg/24 h and persistence based on the first two DCCT checks.

Stroke and congestive heart failure (CHF) events were counted as any-stroke/CHF history if they occurred prior to an IHE or censored date. Few risk engines for assessing overall CVD risks in T1D exist. Since male sex was not a CVD risk factor in a recent DCCT/EDIC analysis (9), we used the validated Swedish National Diabetes Register T1D-risk score (Swedish risk engine), because it includes CVD history but not sex (23).

The following clinical factors were considered potential confounders (Textbox-1 Appendix): B-sex, B-age, B-family history of MI, past SH history, and B-neuropathy (sensory peripheral and/or autonomic), as well as time-dependent variables (mean and current [M,C]) M,C-HbA_1c, M,C-pulse, M,C-LDL, M,C-systolic blood pressure (SBP), M,C-insulin dose, M,C-BMI, C-triglycerides, C-use of ACE inhibitor, C-smoking, C-insulin regimen, C-AER, C-glomerular filtration rate, C-physical activity level, and any-stroke/CHF history. The most significant time-dependent factor among variables listed with M,C (updated mean level, current) was used.

Detailed information on listed covariates has previously been published (9,19). Participants were followed from randomization until time of IHE or censored date.

Analyses were performed with R, version 3.6.3 (https://www.R-project.org) and SAS, version 9.4 (SAS Institute, Cary, NC). Categorical variables are described as percentages and quantitative variables as mean and SD or median (interquartile range [IQR]). Separate semiparametric Cox proportional hazards models were used to assess whether any of the following three slightly different measures related to B-microvascular complications severity, namely, diabetes duration, DCCT-ETDRS, and DCSI, affect the relationship between any-SH and IHD (tests of interaction) during DCCT/EDIC follow-up time.

The Cox proportional hazards assumption was tested based on scaled Schoenfeld residuals and visual inspection. For accommodation of nonproportional hazards, time-by-covariate interaction using the time-transform functionality of coxph in R was added to models (24).

Minimal adjusted (basic) models included age, M-HbA_1c, and either SH or interaction between SH and diabetes duration (DCCT-ETDRS, DCSI, or CV-score). In fully adjusted (extended) models, in order to reduce risk of oversaturation, we employed the backward selection approach for model selection. The final model retained only covariates with P ≤ 0.1. If information on covariates was used to determine DCSI or CV-score, these covariates were excluded from models. For comparison with findings by Khunti et al. (10), median time from first SHE to first IHE was calculated.

A two-sided P value <0.05 was considered significant.

During DCCT/EDIC follow-up, of 1,441 DCCT participants, 168 had IHEs: 110 of 955 participants with and 58 of 486 participants without SH during follow-up.

At baseline, participants were young (mean age 26.8 years) and had low CV-scores (mean CV-score 0.65%); 47.2% of the participants were women. B-characteristics according to presence and absence of any-CVD (IHD, stroke, or CHF) and major adverse CVEs have previously been described in detail (9). We present means of selected baseline characteristics by IHE absence and presence in Table 1 Appendix. On average, participants with versus without IHE presented with significantly higher levels of vascular damage surrogates: longer diabetes duration, lower GFR, higher AER, and higher DCSI, DCCT-ETDRS, and CV-scores; a higher percentage of participants with IHD had neuropathy. Participants with versus without IHE were older, had higher SBP and pulse, had higher levels of triglycerides and LDL, had greater BMI, and higher HbA_1c levels. A higher percentage of participants with versus without an IHE smoked and reported family history of MI.

Spearman correlation coefficients were investigated to show that diabetes duration, retinopathy severity (DCCT-ETDRS), and DCSI represent similar measures related to B-microvascular damage. Age- and sex-adjusted Spearman coefficients for corresponding correlations were 0.77 (P < 0.0001) for diabetes duration and DCCT-ETDRS, 0.68 (P < 0.0001) for diabetes duration and DCSI, and 0.85 (P < 0.0001) for DCCT-ETDRS and DCSI.

Although the percentage of IHEs in participants with and without SH did not differ significantly (11.5% vs. 11.9%, respectively), in the basic model controlling for age and M-HbA_1c, the main SH effect was significantly associated with IHD (25). The hazard ratio (HR) for IHD based on SH was 1.67 (95% CI 1.19–2.33; P = 0.003). SH remained a significant IHD factor in extended models including diabetes duration, DCCT-ETDRS, or DCSI: HR 1.52 (95% CI 1.09–2.13), 1.53 (95% CI 1.10–2.14), and 1.55 (95% CI 1.11–2.16), respectively. All P values were <0.05. In this analysis, age and M-HbA_1c are negative confounders. In negative confounding, the observed unadjusted association is biased toward the null hypothesis. With adjustment for age and HbA_1c, bias due to age and HbA_1c is eliminated (Textbox-1 Appendix).

In basic and extended models, interactions between SH and microvascular damage were significant (Table 1). In extended models, HR based on SH increased 1.19-fold for each additional year of diabetes duration, 1.32-fold for each additional DCCT-ETDRS unit, and 2.21-fold for each additional DCSI unit. Figure 1 gives visual impressions on how measures related to microvascular damage modify the effect of SH on IHD. With increasing diabetes duration and DCCT-ETDRS and DCSI level, the SH effect on IHD increases. The effect of SH became significant at diabetes duration of 5.7 years, at DCCT-ETDRS level of 3, and at DCSI level of 1.

To gain additional insights, we explored the interaction between SH and B-nephropathy/neuropathy severity (maximum of 2 DCSI points). In basic and extended models, P = 0.12 and 0.23, respectively.

Neither B-CV-score nor C-CV-score modified the effect of SH on IHD in basic or extended models (nonsignificant interactions [Table 1]).

All sensitivity analyses were consistent with the main analysis (Supplementary Tables 2–4).

Median time from first SHE until first IHE was 14.0 years (IQR 9.5, 21.0). Median time from SHE immediately before the first IHE until this first IHE was 5.0 years (2.0, 11.0).

A high CV-score is defined as ≥7.5% (23). Of 110 participants with SH and IHE, roughly 85% had low and moderate CV-scores. Median CV-score at time of IHE for participants with and those without SH during DCCT/EDIC was 3.7% (IQR 2.3, 5.5) and 2.7% (1.8, 3.7), respectively. Information on CV-scores as of 31 December 2013 can be found in Textbox-3 Appendix.

Table 2 lists baseline characteristics by absence and presence of IHE with corresponding HRs for participants with SH during follow-up and general baseline information for the entire subpopulation (mean age 26.5 years and mean CV-score 0.65%; 47.7% of participants were female).

Significant B-IHD risk factors included vascular damage surrogates: longer diabetes duration; higher DCCT-ETDRS, DCSI, and CV-score (although low); neuropathy; and greater AER. Other risk factors were older age, higher LDL levels, smoking, family history of MI, higher SBP, and pulse. HbA_1c and triglycerides were not significant B-IHD predictors.

Table 3 presents the most important IHD predictors (P < 0.05) including B-diabetes duration as a surrogate of microvascular damage for participants with at least one SHE during ∼30 follow-up years. Predictors are listed in order of their z values. M-SBP and diabetes duration showed the strongest association with IHD (z = 4.76 and z = 4.08, respectively). Other important factors were any-stroke/CHF history, C-insulin regimen, female sex, C-triglycerides, B-age, M-insulin dose, C-LDL, B-family MI history, and B-neuropathy. Of note, sex was significant in the extended but not univariate model. Models containing B-DCCT-ETDRS (or B-DCSI) instead of B-diabetes duration were consistent (Supplementary Table 4). M-SBP and DCCT-ETDRS (or DCSI) showed the strongest association with IHD. Noteworthy, in age-only adjusted models, the average effect of C-HbA_1c and M-HbA_1c was significant (z = 4.39 and z = 4.19, respectively). However, in extended models, the average effect of C-HbA_1c and M-HbA_1c was not significant.

Several novel observations were made. For the first time, we now report that in a young T1D cohort without any apparent macrovascular complications at baseline, surrogates of baseline microvascular damage (diabetes duration, DCCT-ETDRS, and DCSI) modify the association between SH and IHD. With increasing microvascular complications severity, the adverse effect of SH on the cardiovascular system increases.

SH is often considered as a trigger for CVEs in vulnerable individuals with high CVD risk. However, we found a significant association between SH and IHD in a population with a low B-CV-score. Only 15% of participants with SH and IHD had high CV-scores at time of IHE. Noteworthy, even after ∼30 follow-up years, the median CV-score of the study participants is in the moderate range.

In this analysis, neither B-CV-score nor C-CV-score modified the association between SH and IHD. These findings contrast with those by Veterans Affairs Diabetes Trial (VADT) investigators (26), who found a significant interaction between CV-score level (assessed by the T2D-specific UK Prospective Diabetes Study risk engine [27]) and SH. Possible explanations for our observation of nonsignificant interaction between SH and CV-score include differences in pathophysiology of CVD in T1D and T2D. More likely, the current observation might be a consequence of the young age of the analyzed population and relatively low CV-score.

Our findings strongly indicate that high CV-score (6,7) per se is not mandatory for an association of SH with IHD. Rather, they present evidence that the association between SH and IHD is multifactorial, as previously hypothesized (5).

The association between the interaction of SH with surrogates of microvascular damage and IHD was present over roughly three decades of DCCT/EDIC follow-up. HRs based on SH or interaction of SH with diabetes duration, DCCT-ETDRS, or DCSI were reasonably constant during this follow-up. Without interactions, in the basic model controlling for negative confounding bias by age and HbA_1c, the effect of SH on IHD was significant (Results and Textbox-1 Appendix). The effect remained significant in fully adjusted models. Previously, an association between SH and CVD (10,11) was found. Due to relatively short follow-up times of 5 years in both studies, there were no indications that the effect of SH on CVD might exist over a decade or decades. While Khunti et al. (10), analyzing an older population (mean age 60 years), reported a median time between first exposure to SH and first CVE of 1.5 years (IQR 0.5, 3.5), we report a median time of 14.0 years (9.5, 21.0).

In an exploratory analysis, we identified IHD predictors among the T1D subpopulation with SH during DCCT/EDIC follow-up. SBP and surrogates of microvascular damage were the most important IHD predictors. Other important IHD predictors included traditional factors. In T1D, studies on sex differences in CVD are scarce. While similar CVD rates have been seen in women and men (9), in this analysis female sex was a significant IHD risk factor for the entire cohort and participants with SH. Reasonable explanations include sexual dimorphism, genetic and epigenetics factors, and CV-health/CVD care disparities in women versus men (28–30). Furthermore, women are less susceptible to blunting effects of antecedent hypoglycemia (31). Noteworthy, sex was not a significant factor for participants without SH (not shown). The interaction between sex and SH was not significant, possibly due to small power.

HbA_1c had a less important role in IHD in the population with SH. Reasonable explanations include mediation of the HbA_1c effect on CVD by other CVD factors (32) and masking the beneficial effect of good glycemic control by SH. Whether this finding holds true in the future needs to be investigated.

Mechanism(s) by which hypoglycemia exerts its effect on the cardiovascular system are not entirely understood. It is thought that the long-term SH effect might be linked to subclinical myocardial damage and atherosclerosis with its thrombotic complications rather than to temporary arrhythmia (33). Hypoglycemia invokes a cascade of physiologic responses (34) including abnormalities in coagulation and fibrinolytic activities. It has been reported that hypoglycemia promotes monocyte-platelet aggregates formation within monocyte subsets (CD16⁺). Promoted by monocyte-platelet aggregates formation, 1) monocytes release proinflammatory cytokines such as TNFα and C-X-C motif chemokine ligand 8 and 2) adhesive properties of monocytes (35) are increased. Reports of increased levels of soluble ICAM-1, VCAM-1, and E-selectin and of impairment in flow-mediated dilation provide evidence of endothelial dysfunction. Inflammation and endothelial dysfunction are cornerstones of atherosclerosis (36), a slow and multifactorial process.

It seems plausible that SH (especially repeated SH) works as a catalyst and advances atherosclerosis more rapidly in an environment in which atherosclerosis is already accelerated due to vascular damage (37). Basic research suggests that hypoglycemia might affect signaling pathways, changing immune cell responses to other stimuli such as proatherogenic factors (38). Both our findings of a conditional relationship between SH and IHD and VADT findings (26) that the association between SH and CVE was greater for individuals with higher B-CV-score give additional support.

The strength of this analysis is due to the careful design and capturing and assessments of factors by the DCCT/EDIC investigators over roughly three decades (9). Our results are robust: all results were consistent, independent of the microvascular damage surrogate used. Subanalyses of cohorts stratified by lower/higher microvascular damage surrogate levels showed that SH only correlated with IHD for participants with higher but not lower severity levels. In our exploratory analysis, the finding that diabetes duration, DCCT-ETDRS, and DCSI are particularly important IHD predictors for individuals with SH strengthens the validity of our main results.

Limitations include the post hoc nature of this analysis. However, we have a priori hypothesized that the effect of SH on IHD is most likely modified by measures related to microvascular damage. Currently, no cardiovascular risk model exists including female sex as a risk factor. We concentrated on the Swedish risk engine, which is not validated in other populations. This engine predicts the 5-year CVD risk and was based on observations of patients aged 30–65 years. However, for the subpopulation without any prior stroke/CHF history before the IHE, CV-scores assessed by the Steno type 1 risk engine (39) and Swedish risk engine were consistent (not shown). Although we adjusted for covariates, there might still be residual bias. During EDIC, all variables were measured/reported at frequencies different from those of DCCT. Exploratory analyses suggest a conditional dose-response relationship. However, due to differences in SH ascertainment, results on SH rate in comparison with current level or history of SH must be interpreted more cautiously. Thresholds for diabetes duration, retinopathy severity, and DCSI at which the association between SH and IHD becomes significant might differ for another population.

In summary, since SH remains one of the obstacles in T1D management (1,2), the importance of understanding the role of SH in CVD is obvious. With this analysis, we are adding to a better understanding of the complicated nature of the SH role in CVD. Our analysis highlights the importance of both glycemic control and SH reduction in lowering cardiovascular risk in diabetes. Just recently, although regarding T2D, it was emphasized that special attention should be directed toward SH reduction in the older diabetes population with comorbidities (6). We now show for the first time that this special attention needs to include younger T1D individuals with DCCT-ETDRS levels ≥3 for reduction of IHD risk.

The current analysis suggests that DCSI affects the impact of SH on IHD. While the interaction between retinopathy severity and SH was significant, the interaction between nephropathy/neuropathy and SH did not reach significance. Due to the study design by the DCCT investigators, the event rate and severity of nephropathy/neuropathy were low. The power of this analysis is not sufficient to clearly state whether the increased effect of SH on IHD is a consequence of any combination of retinopathy with nephropathy/neuropathy complications severity or due to increased retinopathy severity; DCCT-ETDRS and DCSI were highly correlated. Further research should examine whether neuropathy and/or nephropathy severity alters a relationship between SH and IHD.

In this analysis, we concentrated on IHD. Future studies should investigate whether potential conditional relationships between SH and cerebral vascular disease, and between SH and mortality, exist.

Albeit of exploratory nature, the observations here on what factors are involved that make some patients with SH more susceptible to adverse effects of SH than others should be considered for risk stratification. Our results indicate that the role of female sex in CVD, especially for women with T1D who were exposed to SH, requires further investigation. Supported by previous literature, the need for diversifying care and individualizing support provided to both men and women is apparent (29).

Our results provide important information for the implementation of the most effective management for patients with T1D. Both the physician and the younger patient with T1D need to recognize that a long-term effect of SH on the cardiovascular system exists that relates to preexisting microvascular disease severity, particularly retinopathy severity. Individuals whom we identified to be susceptible to adverse SH effects on the cardiovascular system might benefit from closer monitoring. Psychological and social assessments, considering initiating insulin pump treatment, and continuous glucose monitoring with an alarm for hypoglycemia might be helpful to reduce the risk of repeated SHEs (3,40,41).

The focus of the American Diabetes Association (ADA) recommendations on statin use is based on risk profile rather than LDL-cholesterol measurement (40,41). ADA endorses the consideration of statin therapy for all patients with T1D aged ≥40 years and younger patients with presence of cardiovascular risk factors. Although the findings here need to be confirmed, for now, SH should be considered a cardiovascular risk factor for younger T1D individuals with DCCT-ETDRS ≥3. If so, it might be worthwhile to consider statin treatment as a preventative measure for such individuals with T1D.

In conclusion, for the young DCCT/EDIC cohort here, microvascular complications severity but not CV-score affects the strength of association between SH and IHD. It needs to be determined whether in the next decade(s), with older age and increasing overall incidence of IHEs, CV-score becomes the dominant factor impacting the effect of SH on IHD.

Acknowledgments. The authors extend a special big thank you to the DCCT/EDIC investigators for the generosity and making the data publicly accessible. The data from DCCT/EDIC were supplied by the NIDDK Central Repository. The authors also thank the NIDDK, librarian Denise Smith, Marshall University, for providing all of the requested literature right away, and Dr. Todd Gress, Assistant Dean and Director of Clinical Research, Joan C. Edwards School of Medicine, Marshall University, and VA Medical Center, for the continued support.

Funding. DCCT and its follow-up, the EDIC study, were conducted by the DCCT/EDIC Research Group and supported by National Institutes of Health grants and contracts and the General Clinical Research Center Program, National Center for Research Resources.

This manuscript was not prepared under the The Diabetes Control and Complications Trial (DCCT) and its follow-up the Epidemiology of Diabetes Interventions and Complications (EDIC) study were conducted by the DCCT/EDIC Research Group and supported by National Institute of Health grants and contracts and by the General Clinical Research Center Program, NCRR. The data from the DCCT/EDIC study were supplied by the NIDDK Central Repositories. This manuscript was not prepared under the auspices of the DCCT/EDIC study and does not represent analyses or conclusions of the DCCT/EDIC study group, the NIDDK Central Repositories, or the NIH.

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. E.R.F. performed statistical analyses and wrote the manuscript. L.A. performed statistical analyses and reviewed and edited the manuscript. H.K.D. contributed to discussions and reviewed and edited the manuscript. E.R.F. and H.K.D. are the guarantors of this work and, as such, had full access to all of the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.