Prolonged Prothrombotic Effects of Antecedent Hypoglycemia in Individuals With Type 2 Diabetes

Cardiovascular (CV) disease is the leading cause of death in type 2 diabetes. A number of large clinical trials have attempted to address the role of intensive glucose control on vascular events and have shown either no reduction (1,2) or an increase in mortality (3). Hypoglycemia is associated with increased CV mortality (2,4), but evidence establishing cause and effect is lacking. In interventional trials, the increased mortality associated with hypoglycemia extended well beyond the acute event, with elevated risk months later (2). Mechanisms by which hypoglycemia might lead to increased CV mortality in the short to medium term are unclear.

Platelet hyperreactivity, altered fibrin clot characteristics, and hypofibrinolysis have been linked to excess CV events, especially in type 2 diabetes (5). Experimental hypoglycemia in people with type 1 diabetes has increased platelet aggregation (6) and inflammatory markers, including CD40 expression and soluble CD40 ligand, platelet-monocyte aggregates, plasma levels of vascular adhesion molecules, and P-selectin (7,8). Repeated episodes of hypoglycemia impair nitric oxide–mediated endothelial function and increase thrombin/antithrombin complex (9). In clinical studies, hypoglycemia increased factor VIII and von Willebrand factor, which are procoagulant (10).

However, studies examining the longer-term effects of hypoglycemia on thrombotic and inflammatory markers remain scarce. In particular, no study to date has studied prothrombotic changes in individuals with type 2 diabetes in response to hypoglycemia, where insulin resistance and clustering of metabolic risk factors may differentiate them from those with type 1 diabetes, predisposing them to a state of depressed fibrinolysis. Antecedent hypoglycemia has been shown to alter autonomic responses and decrease vagal activity downstream of the episode (11,12). Suppression of cholinergic anti-inflammatory pathways (13) may lead to prothrombotic and proinflammatory responses after hypoglycemia.

We hypothesized that hypoglycemia might oppose the CV benefits of intensive glycemic control by inducing prothrombotic responses both during the episode and downstream. Thus, the aim of this study was to investigate the effects of hypoglycemia in patients with type 2 diabetes on key thrombotic mechanisms, including 1) fibrin clot structure and fibrinolysis, 2) platelet reactivity and activation, and 3) plasma levels of inflammatory markers. We studied prothrombotic changes during acute hypoglycemia and up to 7 days after experimentally induced moderate hypoglycemia in both individuals with type 2 diabetes and matched healthy control subjects.

Twelve participants with type 2 diabetes aged 18–65 years and with an HbA_1c of 6.5–10.5% (48–91 mmol/mol) were recruited from Sheffield Teaching Hospitals outpatient clinics between 2011 and 2014. Eleven age- and BMI-matched individuals without diabetes were recruited as control subjects from staff at Sheffield Teaching Hospitals and the University of Sheffield. The same participants also had cardiac electrophysiological measurements recorded during the morning hypersinsulinemic clamps in a study of the effect of hypoglycemia on autonomic function and repolarization (14). Participants with diabetes were taking oral hypoglycemic agents and/or glucagon-like peptide 1 analogs and/or insulin for ≤2 years. Exclusion criteria were previous myocardial infarction, ischemic heart disease, cardiac arrhythmias, stroke or peripheral vascular disease, or other known CV disease; epilepsy; untreated hyperthyroidism; pregnancy; and serious intercurrent illness. None were taking antiplatelet agents or anticoagulants apart from two participants with diabetes who were taking aspirin. All participants had normal full blood count, renal function (estimated glomerular filtration rate >60 mL/min/1.73 m²), electrocardiogram at baseline, and cardiac autonomic function on the basis of standard autonomic function tests (15). All participants gave written informed consent. The study was approved by the South Yorkshire local research ethics committee.

Individuals participated in paired hyperinsulinemic-euglycemic and hypoglycemic clamp studies separated by at least 4 but no more than 8 weeks to minimize carryover effects. We used a crossover design such that each participant served as his or her own control subject to reduce variability. Euglycemic studies preceded hypoglycemic studies because hypoglycemia may have persistent effects of the CV system. In previous studies, two periods of hypoglycemia resulted in altered autonomic and vagal function that could lead to proinflammatory and prothrombotic effects in the days after hypoglycemia (11,12). In one study, these changes were observed up to 6 days later (11). Thus, in the hypoglycemic arm, arterialized blood glucose was maintained at 2.5 mmol/L for two 60-min periods in the morning and afternoon with rapid-acting intravenous insulin at 120 and 240 mU/m²/min, respectively, alongside a variable infusion of 20% dextrose. Participants were blinded to blood glucose levels. Blood glucose was raised to euglycemic levels between morning and afternoon hypoglycemic clamps. In the euglycemic arm, arterialized blood glucose was maintained at 6 mmol/L for 60 min during both morning and afternoon, using similar rates of insulin infusion as during hypoglycemia. Insulin was administered at the same rates in both participant groups. Prothrombotic and inflammatory markers were measured at the end of the morning clamp and at recovery (30 min after end of morning clamp) but not during the afternoon clamp. To investigate downstream effects, prothrombotic and inflammatory markers were measured in the morning at 1 and 7 days after both euglycemic and hypoglycemic studies. The study design is outlined in Fig. 1, and details of the hyperinsulinemic clamp protocol are described in Supplementary Methods 1.

For epinephrine and norepinephrine, whole blood (6 mL) was collected into chilled lithium-heparin tubes containing 50 μL EGTA/glutathione as a preservative. These samples were assayed by high-performance liquid chromatography (interassay coefficient of variation for norepinephrine 6.03% and epinephrine 15.9%). For free insulin, whole blood (3 mL) was collected into a 6-mL lithium-heparin tube and immediately subjected to centrifugation. The resulting plasma (0.5 mL) was added to a chilled plastic tube containing 0.5 mL polyethylene glycol for precipitation of immune complexes and mixed. The product was analyzed by an immunometric assay (Insulin ELISA [interassay coefficient of variation 7.1%]; Invitron, Monmouth, U.K.). Serum cortisol analyzed from 4 mL of venous blood collected in a serum separator tube was measured using a commercial immunoassay (Roche e602 serum cortisol assay [interassay coefficient of variation 3.2%]; Roche Diagnostics, West Sussex, U.K.). Biochemical parameters were measured at baseline and at end of euglycemic and hypoglycemic arms in both groups.

Venous blood was collected into tubes containing 3.2% sodium citrate (BD Vacutainer Glass Citrate Tube) on ice. Ex vivo fibrin polymerization characteristics of plasma samples were investigated by a validated turbidimetric clotting assay (16) described in detail in Supplementary Methods 2.

Fibrinogen and plasminogen activator inhibitor 1 (PAI-1) assays were performed on venous blood (4.5 mL) collected into tubes containing 3.2% buffered sodium citrate solute (BD Vacutainer Glass Citrate Tube). hsCRP was analyzed using an immunoturbidmetric assay (Cardiac C-Reactive Protein [Latex] High Sensitive; Roche Diagnostics, Indianapolis, IN). Interleukin-6 (IL-6) was determined using cytometric bead array (Human IL6 Flex Set; BD Biosciences, Oxford, U.K.). Plasma levels of complement C3, a protein that is incorporated into the fibrin clot and modulates fibrinolysis (17), were determined using ELISA (Biosources, San Diego, CA). Additional details are described in Supplementary Methods 2.

Pooled samples of plasma were analyzed from 10 participants with type 2 diabetes and 10 control subjects. Fibrin clots were made as previously reported (18) and described in detail in Supplementary Methods 2.

Platelet aggregation was analyzed using multiple electrode impedance aggregometry (Multiplate; Roche Diagnostics, Basel, Switzerland) and described in detail in Supplementary Methods 2.

Hirudin-anticoagulated blood was added to tubes containing phycoerythrin Cy5 mouse anti-CD62P antibody (BD Biosciences) and platelet marker CD41a (BD Biosciences) for measurement of platelet P-selectin expression and is described in detail in Supplementary Methods 2.

Baseline demographic data and prothrombotic and inflammatory markers are summarized as mean ± SD for parametric data, unless otherwise stated, or median (interquartile range) for nonparametric data. hsCRP and IL-6 were logarithmically transformed because of a skewed distribution. End-of-clamp blood glucose and hormone concentrations were compared between euglycemic and hypoglycemic time points using two-way ANOVA.

We used a linear mixed model with repeated measures to compare 1) whether there was significant change in prothrombotic markers over time within glycemic arms in each participant group and 2) the effect of the glycemic arm on changes in prothrombotic markers in each participant group. For analysis 1, time point was specified as a fixed effect with multiple comparisons against baseline adjusted by the Sidak correction. In mixed-model analysis 2, glycemic arm, time, and interaction between glycemic arm and time were specified as fixed effects and the participant as random effect. Glycemic arm and time point were specified as repeated measures within each participant. Repeated measures were fitted with an unstructured, compound symmetry, or autoregressive 1 covariance structure, and the model with the best fit (lowest Akaike information criterion) was selected. P values were obtained by restricted maximum likelihood estimation. The effect of glycemic arm and interaction between time and glycemic arm on fiber network density and fibrin thickness were analyzed using two-way repeated-measures ANOVA. We further calculated the within-individual correlations between clot maximum absorbance (MA) or clot lysis time with fibrinogen, PAI-1, and C3, taking into account repeated measures, with ANCOVA (19). P < 0.05 was considered significant. Results were analyzed using SPSS version 20 software (IBM Corporation).

On the basis of our initial pilot data, a sample size of 10 participants per group had a power of 81% to detect a clinically relevant 15% difference in clot lysis time, assuming an SD of 100 s. We aimed to recruit 12–13 participants for each group, allowing for a 20–30% dropout rate.

Twelve patients with type 2 diabetes (nine male) and 11 individuals without diabetes (five male) participated in the study. Participants with diabetes were similar in age (54 years [50–58 years]) compared to control subjects (52 years [47–59 years]; P = 0.90). Mean BMI was comparable (34 ± 5 vs. 31 ± 8 kg/m² in the diabetes vs. nondiabetes group, respectively; P = 0.18). The median duration of diabetes was 10 years (8–12 years), and mean HbA_1c was 7.8 ± 1.3% (62 ± 14 mmol/mol). Among patients with diabetes, five were taking oral hypoglycemic agents only, five were taking oral hypoglycemic agents and a glucagon-like peptide 1 analog, and two were taking oral hypoglycemic agents and basal insulin. Baseline prothrombotic and inflammatory markers are shown in Supplementary Table 1. Participants with diabetes had higher platelet reactivity to ADP, but no significant differences were detected in fibrinogen, PAI-1, C3, and hsCRP plasma levels (Supplementary Table 1).

Blood glucose levels were similar at end-of-morning euglycemic and hypoglycemic clamps in both groups (Supplementary Table 2). Glucose targets were reached during afternoon clamps within each arm. During day 1, blood glucose was higher after hypoglycemic clamps compared with euglycemia in the diabetes group only. Blood glucose levels were similar between glycemic arms at day 7. Participants with diabetes did not report symptomatic hypoglycemia or capillary glucose values <3 mmol/L in the week after each clamp.

Median (interquartile range) free insulin levels at 120 min were 576 pmol/L (468–627 pmol/L) during euglycemia and 689 pmol/L (477–1,076 pmol/L) during hypoglycemia in the diabetes group. In the nondiabetes group, these values were 865 pmol/L (509–952 pmol/L) in the euglycemic arm and 665 pmol/L (468–967 pmol/L) in the hypoglycemic arm. Insulin levels between groups during both euglycemia and hypoglycemia were not different (P = 0.23).

Counterregulatory hormones were unchanged during euglycemia (Supplementary Table 2). During acute hypoglycemia, epinephrine, norepinephrine, and cortisol increased significantly (all P < 0.05 vs. baseline) in both groups (Supplementary Table 2). Plasma epinephrine and cortisol returned to baseline at day 1 and day 7 after both arms in both participant groups. However, in the nondiabetes group, plasma norepinephrine levels increased significantly at day 1 after both clamps.

In the nondiabetes group, clot lysis times decreased after euglycemia (from 729 ± 216 to 611 ± 159 s, Δ −146 ± 110; P = 0.001 vs. baseline) but returned to baseline at day 1 (Fig. 2A). No changes in clot lysis times were found in the hypoglycemic arm in the nondiabetes group. In the diabetes group, clot lysis times decreased at the end of the euglycemic clamp (Δ −81 ± 86 s) but were prolonged at the end of hypoglycemia, with further increases at day 1 (Δ 71 ± 153 s) and day 7 (Δ 67 ± 107 s). Changes in clot lysis times were significantly different between the euglycemic and hypoglycemic arms in participants with diabetes (P = 0.001) (Fig. 2A).

In the nondiabetes group, clot MA decreased at the end of both clamps but to a lesser extent during hypoglycemia (Δ −0.02 ± 0.05 arbitrary units [AU]) compared with euglycemia (Δ −0.05 ± 0.05 AU; P = 0.02 for glycemic arm) (Fig. 2B). In the diabetes group, clot MA decreased during euglycemia and recovery. A nonsignificant increase was observed in clot MA at the end of hypoglycemia (Δ 0.02 ± 0.05 AU), which resolved at recovery, followed by gradual increases at day 1 and 7 (Fig. 2B). Significant differences were found in clot MA between glycemic arms in the diabetes group (P = 0.002) and interaction between time and glycemic arm (P = 0.02).

Fibrin diameter did not change significantly during euglycemia or hypoglycemia in the nondiabetes group (Fig. 3A). In the diabetes group, no changes were observed in fibrin fiber diameter during euglycemia, which increased at day 1 and 7 after hypoglycemia (P < 0.01 vs. euglycemia at equivalent time points) (Fig. 3A). A significant difference was found between glycemic arms (P < 0.0001) and in the interaction between time and glycemic arm (P < 0.0001).

In the nondiabetes group, fiber network density decreased at day 1 after euglycemia compared with no change after hypoglycemia (P < 0.001 for glycemic arm). The fibrin network density returned to baseline at day 7 after euglycemia but increased at day 7 after hypoglycemia (P < 0.01 for glycemic arm) (Fig. 3B).

In the diabetes group, fiber network density decreased at day 7 after euglycemia, whereas after hypoglycemia, an increase was found in fiber network density at day 1 and 7 (both P < 0.01 for glycemic arm) (Fig. 3B). Significant differences were found between the glycemic arms and the interaction between time and glycemic arm in both groups (both P < 0.001). Representative examples of scanning electron micrographs of pooled fibrin clots are shown in Fig. 3C.

In the nondiabetes group, fibrinogen tended to decrease at the end of the clamp and recover similarly during both euglycemia and hypoglycemia. However, in the diabetes group, fibrinogen levels did not change during euglycemia but increased: Δ 0.20 ± 0.10 mg/mL at day 1 and Δ 0.85 ± 0.69 mg/mL at day 7 after hypoglycemia (P = 0.05 for glycemic arm) (Supplementary Fig. 1A).

PAI-1 falls in both groups were similar during euglycemia and hypoglycemia (Supplementary Fig. 1B). PAI-1 decreased in the nondiabetes group during euglycemia and hypoglycemia (P = 0.15 and 0.005 for time, respectively), with no significant differences between arms (P = 0.56). In the diabetes group, PAI-1 decreased from baseline to Δ −811 ± 204 pg/mL at the end of euglycemia (P = 0.02 vs. baseline) and remained lower at day 7. PAI-1 decreased to a similar extent during hypoglycemia (Δ −888 ± 256; P = 0.02 vs. baseline), with no differences detected between glycemic arms (P = 0.85).

Change in clot MA correlated with changes in fibrinogen (r = 0.98; P < 0.001), PAI-1 (r = 0.99; P < 0.001), and C3 (r = 0.99; P < 0.001) across both groups. Change in clot lysis time correlated with changes in fibrinogen (r = 0.98; P < 0.001), PAI-1 (r = 0.99; P < 0.001), and C3 (r = 0.98; P < 0.00). No correlation was found among epinephrine, norepinephrine, and clot MA or clot lysis times.

In the nondiabetes group, C3 levels were similar during euglycemia and hypoglycemia and did not change significantly (Supplementary Fig. 1C). C3 levels did not change during euglycemia in the diabetes group but tended to rise at day 7 after hypoglycemia (from 99 ± 5 mg/mL at hypoglycemic baseline to 108 ± 4 mg/mL at day 7).

In the nondiabetes group, hsCRP increased similarly at day 1 after euglycemia and hypoglycemia in both glycemic arms (Supplementary Fig. 1D). In the diabetes group, hsCRP decreased after euglycemia at day 7 (log hsCRP Δ −0.11 ± 0.01; P = 0.009 vs. baseline) compared with no change after hypoglycemia (log hsCRP Δ 0.06 ± 0.01 at day 7). A significant interaction was observed between glycemic arm and time (P = 0.04).

In the nondiabetes group, IL-6 was higher at day 1 after euglycemia but did not change after hypoglycemia (data not shown). In the diabetes group, IL-6 did not change during the euglycemic arm and trended toward an increase in IL-6 after hypoglycemia at day 7.

In the nondiabetes group, a nonsignificant change was found in collagen-induced platelet aggregation during euglycemia (Fig. 4A). During hypoglycemia, collagen-induced platelet aggregation increased significantly more than during euglycemia, which resolved at recovery. The overall difference in platelet aggregation response to collagen between arms was significant (P = 0.04).

In the diabetes group, collagen-induced platelet aggregation tended to decrease during acute euglycemia (Δ −15 ± 23 units; P = 0.11 vs. baseline) in contrast to an increase after hypoglycemia (P < 0.01 between glycemic arms), persisting to the recovery period but not beyond. A significant interaction was observed between time and glycemic arm (P = 0.03).

ADP-induced platelet aggregation decreased in the diabetes group during euglycemia compared with an increase at the end of hypoglycemia (P = 0.03 between arms). In the nondiabetes group, secondary rises in platelet aggregation induced by both collagen and ADP at day 1 occurred after both hypoglycemia and euglycemia (Fig. 4A and B).

No significant changes were found in platelet reactivity to 5-hydroxytryptamine during euglycemia in both groups (Fig. 4C). Trends toward increased platelet reactivity were seen after hypoglycemia in both groups, resolving at recovery. In the nondiabetes group, no significant changes in platelet activation occurred during euglycemia. Platelet activation tended to increase after hypoglycemia and at day 1, with no overall difference between arms (Fig. 4D). In the diabetes group, platelet activation, as measured by unstimulated P-selectin expression, decreased after euglycemia and maximally at day 1 but increased immediately after hypoglycemia (P = 0.01 for glycemic arm) (Fig. 4D).

To our knowledge, this study is the first to investigate the effects of hypoglycemia on both cellular and protein arms of the thrombosis pathway in the period beyond a hypoglycemic challenge. The novel findings are that 1) hypoglycemia was associated with early and late prothrombotic changes in the fibrin network in the participants with diabetes but was less evident in healthy control subjects, 2) a rise in fibrinogen and C3 levels may contribute to late prothrombotic changes in fibrin network properties, and 3) hypoglycemia enhanced platelet reactivity in participants with and without diabetes, lasting less than 24 h after the event.

The antiaggregatory, anti-inflammatory, and profibrinolytic responses to euglycemic hyperinsulinemia are consistent with previous human studies (20,21). In the current work, a striking observation was the difference in clot lysis time between euglycemia and hypoglycemia in the diabetes group. Initially, the decrease in clot lysis time during euglycemia may be explained by a reduction in PAI-1. Insulin-mediated suppression of fibrinogen synthesis also may result in enhanced clot lysis but is unlikely to contribute to early changes given the long half-life of the protein, unlike PAI-1. On the other hand, early prolongation in clot lysis in the diabetes group did not appear to be PAI-1 mediated because protein levels were reduced after hypoglycemic clamps. In a previous study, PAI-1 was decreased during both hyperinsulinemic hypoglycemia and euglycemia in individuals with type 1 diabetes; however, the authors found increased PAI-1 after hypoglycemia in healthy control subjects (8).

A novel and intriguing observation is the persistent effect of hypoglycemia on clot density and impaired fibrinolysis at least 7 days after hypoglycemia in the diabetes group but not in the nondiabetes group. This finding may be important clinically, particularly in individuals at higher vascular risk. The mechanisms for the late impairment in lysis are not entirely clear, although the rise in fibrinogen and complement C3 plasma levels provide plausible explanations. Increased levels of fibrinogen are associated with denser clots, as we have previously shown (18), which may impair fibrinolysis. This finding is supported by an observational study in type 2 diabetes wherein low fasting glucose was associated with denser ex vivo fibrin clots compared with fasting glucose in the normal range (22). We have demonstrated that C3 modulates clot lysis independently of PAI-1, and this effect is particularly pronounced in individuals with diabetes (17,23). Taken together, repeated hypoglycemic episodes possibly lead to a state of chronic low-grade inflammation, resulting in elevated fibrinogen and C3 levels, which in turn compromise fibrinolysis. However, we cannot exclude the possibility that other plasma components also may have an effect. Joy et al. (9) demonstrated that nitric oxide–mediated endothelial dysfunction, which creates an inflammatory environment, is exaggerated by repeated experimental hypoglycemia 24 h later, but effects beyond 48 h were not reported.

We also show that hypoglycemia modulates platelet reactivity in type 2 diabetes, probably mediated by catecholamine release, but the effects are short-lived. Our results are consistent with previous reports of individuals with type 1 diabetes demonstrating increased platelet reactivity during acute hypoglycemia (24). The hypoglycemia-mediated increase in platelet reactivity can be abolished by α-receptor blockade, implicating α₂-adrenoreceptors in mediating this effect (6). In contrast to fibrin clot studies, we did not observe consistent differences between platelet responses to acute hypoglycemia in the groups studied.

The current data identify a mechanism whereby hypoglycemia could contribute to increased CV mortality by opposing the benefits of intensive glycemic control. In a post hoc analysis of the Normoglycemia in Intensive Care Evaluation-Survival Using Glucose Algorithm Regulation (NICE-SUGAR) study, which reported excess mortality in patients treated to tight glucose levels, the median time to death as a result of a hypoglycemic event was 7 days for moderate hypoglycemia and 8 days for severe hypoglycemia (25). The time course matches our findings of a worsening atherothrombotic risk profile 1 week after hypoglycemia. In the Action in Diabetes and Vascular Disease: Preterax and Diamicron MR Controlled Evaluation (ADVANCE) study and Veterans Affairs Diabetes Trial (VADT), the increased CV risk extended for many months after a severe hypoglycemic event, and it is unlikely that the changes we observed would persist over this period. On the other hand, less severe episodes (equivalent to those induced in this study) were not measured consistently in these trials. Arguably, they are occurring more frequently in those who experience severe events and, if recurrent, might contribute to thrombotic events, particularly in individuals at increased CV risk.

Our study had the following relevant limitations. First, the relatively small number of participants, although adequate to demonstrate changes in fibrin clot properties, may have left us short of statistical power to identify subtle differences between the type 2 diabetes and nondiabetes groups. However, recruiting individuals into experimental studies is challenging, and this study remains one of the largest to investigate both participants with type 2 diabetes and control subjects with combined platelet and fibrin network analyses. Second, the experimental model requires supraphysiological doses of insulin to maintain stable hypoglycemia higher than that observed during routine clinical care. This might artifactually diminish prothrombotic and proinflammatory changes during experimental hypoglycemia. Third, two participants were receiving aspirin treatment, which can affect collagen-induced platelet aggregation and clot lysis (26). However, these individuals remained on aspirin throughout the study and displayed patterns of collagen-induced platelet aggregation similar to those of other participants (data not shown). In these studies, we induced hypoglycemia on two occasions during day 0, and we cannot be certain that similar persistent prothrombotic changes would have been observed after a single hypoglycemic episode or shorter hypoglycemic episodes. Future studies should be designed to address this. For obvious ethical reasons, we did not induce hypoglycemia in individuals with known CV disease. It is unknown whether the deleterious atherothrombotic effects of hypoglycemia would be exaggerated in patients with a high CV risk and the protective effects of insulin attenuated compared with our study group.

In conclusion, we have shown that two episodes of moderate hypoglycemia have acute prothrombotic effects in individuals with and without type 2 diabetes. We also have demonstrated that the impact of moderate, short-lived hypoglycemia is maintained for at least 7 days after the event, with adverse effects on fibrin clot properties, fibrinolysis, and subclinical inflammation. These effects were more prominent among individuals with type 2 diabetes compared with control subjects without diabetes. We have identified potential mechanisms whereby hypoglycemia could increase the risk of CV events during and after an episode and so oppose the benefits of intensive glycemic control. The precise clinical relevance of these findings remains to be established by additional studies. Nevertheless, we believe that clinicians should consider these results when addressing vascular health in patients with type 2 diabetes and choose approaches that minimize hypoglycemia when optimizing glycemic management.

Acknowledgments. The authors thank the Diabetes Department and Clinical Research Facility in Sheffield Teaching Hospitals for assistance and all the patients who gave time to this study. The authors also thank Jenny Freeman (University of Leeds) and Eric Lau (Chinese University of Hong Kong) for statistical advice.

Funding. This article is a summary of independent research funded in part by the National Institute for Health Research (NIHR) Biomedical Research Fellowship, U.K. (BRF-2011-004) and carried out at the NIHR Sheffield Clinical Research Facility.

The views expressed are those of the authors and not necessarily those of the National Health Service, NIHR, or the Department of Health.

Duality of Interest. R.F.S. reports institutional research grants/support from AstraZeneca and PlaqueTec; consultant fees from Actelion, AstraZeneca, Avacta, Bayer, Bristol-Myers Squibb/Pfizer, Idorsia, Novartis, PlaqueTec, and Thromboserin; and honoraria from AstraZeneca and Bayer. R.A. received honoraria and educational and research support from Abbott Diabetes Care, AstraZeneca, Novo Nordisk, Eli Lilly, Bayer, Sanofi, MSD, and Boehringer Ingelheim. S.R.H. has served as an advisory board panel member for Eli Lilly, Novo Nordisk, Boehringer Ingelheim, Zealand Pharma, UNEEG medical, and Takeda Pharmaceuticals and is a member of speakers’ bureaus for AstraZeneca, Novo Nordisk, Eli Lilly, and MSD. No other potential conflicts of interest relevant to this article were reported.

Author Contributions. E.C. designed the study, collected and analyzed the data, and wrote the manuscript. A.I., E.W., and F.P. helped to collect the data, conducted clotting and platelet assays, and reviewed the manuscript. I.A.M. analyzed the catecholamine data and reviewed the manuscript. R.F.S., R.A., and S.R.H. designed the study, reviewed the data, and edited and redrafted the manuscript. E.C. 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.

Prior Presentation. Parts of this study were presented in abstract form at the 73rd Scientific Sessions of the American Diabetes Association, Chicago, IL, 21–25 June 2013.