Effect of Glucose Management on Coronary Heart Disease Risk in Patients with Diabetes

Changing Epidemiology of Diabetes and Coronary Heart Disease

The general incidence and prevalence of CHD have declined in the United States in the last several decades, and this decline has been accompanied by a decline in CHD-related mortality. These trends have been attributed to better cardiovascular risk factor control and treatment during and after acute coronary syndromes over time, primarily with the use of statin medications, blood pressure management, and anti-platelet therapies. In contrast to CHD trends, the incidence and prevalence of diabetes have been steadily increasing over time, with the disease now affecting close to a third of older U.S. adults (65 years or older) (see also Chapter 1 ). ^, In addition, adults with diabetes are living longer. As a result, the burden of CHD attributable to diabetes is increasing (see also Chapter 7 ). These changes in the epidemiology of diabetes and CHD have important implications. First, strategies that mitigate the risk of CHD in diabetes patients will be of growing importance because heart disease is increasingly a complication of diabetes. Second, these strategies will be applied to an aging population with a high comorbidity burden and at higher risk for adverse effects of therapy.

Epidemiologic Relationship of Glucose with Coronary Heart Disease

Multiple studies have assessed the relationship between various glucose parameters—fasting glucose, 2-hour glucose during an oral glucose tolerance test, or HbA1c levels—and the risk of CHD in populations with and without diabetes. Most of this work suggests a continuous relationship between measures of glycemia and CHD risk.

Several studies and a metaregression analysis have shown that in nondiabetic populations, there is a graded relationship between initial fasting and postprandial glucose levels and subsequent occurrence of cardiovascular events over 12 years of follow-up. The association is apparent even at levels below the diabetic thresholds. However, because HbA1c is the preferred test for monitoring blood glucose control during the chronic management of diabetes, data summarized here will be predominantly based on this glycemic parameter.

In the large prospective population study of Norfolk, in the United Kingdom, HbA1c and cardiovascular risk factors were assessed from 1995 to 1997, and cardiovascular disease events and mortality were examined during the next 6 to 8 years of follow-up. The relationship between HbA1c and cardiovascular disease and total mortality was continuous and apparent even among persons without diabetes. The risk was lowest among persons with HbA1c below 5% and increased thereafter throughout the range of nondiabetic HbA1c levels up to 6.9%. Each one percentage point increase in HbA1c above 5% was associated with a 20% to 25% increase in the relative risk for CHD among men and women in age- and risk-factor adjusted models. Moreover, when known diabetes status and HbA1c concentration were included in the same model, diabetes was no longer a significant independent predictor of CHD, suggesting that the increased risk of CHD in dysglycemic states is mediated through hyperglycemia itself.

The prognostic value of HbA1c was also assessed in the Atherosclerosis Risk in Communities (ARIC) study of U.S. adults without a prior history of diabetes or cardiovascular disease and with up to 15 years of follow-up. Similar to the observations from the Norfolk study, the risk for CHD increased with higher HbA1c values in a continuous fashion independent of classic cardiovascular risk factors. When compared with study participants with HbA1c of 5% to less than 5.5% (the reference range), the hazard ratio (HR) for CHD was increased 23% in those with HbA1c of 5.5% to less than 6%, 78% for HbA1c of 6% to less than 6.5%, and 95% for HbA1c of 6.5% or higher. Although, clearly, the causal role of glucose in the development of CHD could not be evaluated in this epidemiologic study, the findings suggest that HbA1c, even in the nondiabetic range, can be a useful independent marker of cardiovascular risk.

Although the association between HbA1c level and CHD may be prognostically important in nondiabetic individuals, to understand the effect of glucose lowering on CHD risk we must examine data in patients with diabetes. A prospective observational study of type 2 diabetes patients enrolled in the United Kingdom Prospective Diabetes Study (UKPDS) examined the relationship between HbA1c and cardiovascular complications. They found that each 1% increase in the updated HbA1c was associated with a 14% relative risk increase for myocardial infarction (MI; Fig. 13-2 ). A meta-analysis of 13 prospective cohort studies of HbA1c and cardiovascular disease in persons with diabetes (type 1 or 2) suggested that chronic hyperglycemia is associated with an increased risk for cardiovascular disease. The pooled relative risk for cardiovascular disease associated with a 1% increase in HbA1c was 1.18. In a subgroup of six studies conducted in patients with type 2 diabetes, a 1% increase in HbA1c was associated with a 13% increased relative risk for CHD. The inclusion criteria for the meta-analysis did not specify pharmacologic treatment for diabetes; rather, these were observational studies involving patients on both medication and diet therapy. These results suggest a moderate increase in cardiovascular risk with increasing HbA1c in diabetic adults. However, the meta-analysis relied on small studies with some suggestion of heterogeneity of effects that could not be explored in detail.

One large analysis examined data from the U.K. General Practice Research Database (GPRD) on 27,965 patients with type 2 diabetes whose oral monotherapy was intensified to oral combination therapy, and 20,005 whose oral therapy was intensified to include insulin. The primary and secondary outcomes for the two cohorts were all-cause mortality and major cardiovascular events, respectively, over the mean follow-up of 4.5 years. HbA1c in the study was based on the mean of any values recorded between the therapeutic switch and death or date of censor. In combined cohort analysis, the HbA1c decile with a median value of 7.5% (interquartile range 7.5% to 7.6%) was associated with the lowest mortality and the lowest progression to large-vessel disease among those without prior history of cardiovascular events. Higher and lower HbA1c values were associated with an increased risk, and the pattern of risk was U shaped. In the oral combination therapy group, a wider range of HbA1c values was safe with respect to mortality risk (median HbA1c 6.9% to 8.9%), whereas this range was narrower for patients on insulin (median HbA1c 7.5% to 8.1%). In addition, the use of insulin was associated with an approximately 50% higher hazard of mortality compared with the use of oral agents. Although no evidence supports a direct cardiotoxic effect of insulin in type 2 diabetes, it is certainly possible that age, comorbidities, and diabetes duration may be related to the decision to initiate insulin as well as to the higher mortality risk. The findings from this study differ significantly from the graded, continuous epidemiologic relationships between HbA1c and cardiovascular outcomes in individuals without diabetes. In nondiabetic populations, lower HbA1c values predict better outcomes without a clear threshold, but the data from treated patients with diabetes suggest that there may be a risk associated with achieving near-normal glycemia.

Another retrospective cohort study, this time performed in the United States, confirmed the results of the GPRD analysis. Here, data from 71,092 patients with type 2 diabetes age 60 years or older within the Kaiser Permanente Northern California system were analyzed to examine the association between baseline HbA1c level and subsequent nonfatal complications (metabolic, microvascular, and cardiovascular events) and mortality. The authors found a similar U-shaped relationship between HbA1c level and mortality, with higher risk in those with HbA1c below 6% and 10% or higher in the adjusted models. In contrast, however, the relationship between HbA1c and cardiovascular events was continuous with increasing risk above HbA1c of 6%. Integrating all of the outcomes together, the “optimal” HbA1c range identified by this study lay somewhere in the 6% to 7.9% range. As in the GPRD study, the analysis added important information about optimal glycemic targets in diabetes, suggesting that achievement of low glycemic levels may provide benefits (such as lower risk of CHD), but that very low levels of glycemia may be associated with harm (e.g., higher mortality risk). A third study, this involving all adults with type 2 diabetes drawn from the Kaiser Permanente Southern California system, showed a U-shaped relationship between HbA1c and cardiovascular events, with HbA1c levels of 6% or lower and greater than 8% associated with an increased risk of cardiovascular events. Whether or not low HbA1c levels are a marker of sicker patients or a mediator of harm remains highly debatable. Moreover, whether this phenomenon is actually directly attributable to lower than desirable glycemia or to adverse effects of the medications clinicians use to achieve this range is not clear. Randomized clinical trials can test the effects of interventions directly on patient outcomes and may be able to provide greater insight into the effect of glucose lowering on CHD events.

Trials of Glucose-Lowering Interventions

The landmark trial in type 2 diabetes that investigated the effect of intensive glucose lowering on microvascular and macrovascular outcomes was the UKPDS. The trial was begun in 1977 and the results were published in 1998. In this trial, 3867 patients with newly diagnosed type 2 diabetes (median age 54) were randomized to intensive treatment with sulfonylureas (chlorpropamide, glibenclamide (glyburide in the U.S.), or glipizide) or with insulin, versus conventional therapy with diet alone. The median HbA1c level in the intensive group during the course of the trial was 7%, versus 7.9% in the conventional arm. Three separate aggregate endpoints were studied over the 10 years of follow-up. The risk in the intensive group was 12% lower for any diabetes-related endpoint ( P = 0.03), which included both macrovascular and microvascular events as well as metabolic complications; not significantly lower for any diabetes-related death (− 10%, P = 0.34); and not significantly lower for mortality (− 6%, P = 0.44), compared with patients treated with diet only. The reduction in diabetes-related endpoints was driven by a 25% risk reduction in microvascular events, and the reduction in MI did not reach statistical significance (− 16%, P = 0.052). A subgroup of UKPDS patients who were overweight (> 120% ideal body weight) were randomized either to intensive therapy with metformin (n = 342, median HbA1c 7.4%) or conventional diet therapy (n = 411, median HbA1c 8%). In this subset of patients, treatment with metformin was associated with a 32% reduction in any diabetes-related endpoint ( P = 0.002), 42% reduction in diabetes-related death ( P = 0.017), and 36% reduction in mortality ( P = 0.011). In this cohort of patients treated with metformin, there was a significant 39% reduction in MI ( P = 0.01) ( Fig. 13-3 ). In summary, the UKPDS trial established that intensive glucose control reduces the risk of microvascular complications in patients with newly diagnosed type 2 diabetes, but suggested that macrovascular benefits may be confined to overweight patients treated with metformin therapy.

After the UKPDS trial was completed, study participants and their clinicians were advised to lower levels of blood glucose as much as possible, and patients returned to community or hospital-based diabetes care according to their clinical needs without any attempts to maintain previously randomized therapies. In the 10-year post-trial monitoring study of patients who survived to the end of the UKDPS trial, HbA1c levels were no longer different between the original intensive and conventional arms (approximately 8% at the end of the post-trial monitoring period). In the sulfonylurea-insulin group, relative risk reductions for diabetes-related endpoints persisted, whereas significant risk reductions for MI (15%, P = 0.010) and mortality (13%, P = 0.007) emerged over time. In the metformin group, relative risk reductions persisted for any diabetes-related endpoint, MI (33%, P = 0.005), and mortality (27%, P = 0.002). These observations suggest a modest but sustained effect of intensive glucose lowering on cardiovascular events, but only after many years of follow-up. Whether the effect is confined to patients with newly diagnosed type 2 diabetes or whether it reflects the long period of time required to significantly affect subsequent atherosclerotic outcomes is not entirely clear.

Even before the cardiovascular benefits of intensive glucose therapy emerged in the long-term follow-up of the UKPDS trial, guidelines recommended a target HbA1c level of 7% or less in most patients. This was primarily driven by the expectation of microvascular benefits, albeit with uncertainty over the effects on macrovascular events. To settle the questions about the role of intensive glucose therapy in type 2 diabetes, three randomized controlled trials were specifically designed to examine the impact of targeting near-normal glycemia on cardiovascular risk. The HbA1c targets were set low because of the continuous epidemiologic relationship of glucose with cardiovascular risk, suggesting that perhaps much lower glucose levels need to be achieved for a significant benefit to emerge. The three trials all recruited participants with type 2 diabetes who had either a history of or multiple risk factors for cardiovascular disease, thus ensuring adequate event rates to study the effects of the interventions. Participants were therefore quite distinct from patients in the UKPDS trial—they were older, had a longer duration of diabetes, and had a greater comorbidity burden.

The Action to Control Cardiovascular Risk in Diabetes (ACCORD) trial enrolled 10,251 patients (mean age 62, median baseline HbA1c 8.1%, 35% with history of prior cardiovascular event) to intensive glucose therapy (targeting HbA1c < 6%, median achieved HbA1c 6.4%) versus conventional therapy (targeting HbA1c 7% to 7.9%, median achieved HbA1c 7.5%). This trial was stopped prematurely after a mean follow-up of 3.5 years because of a higher mortality rate in the intensive therapy group compared with the control arm (HR 1.22, P = 0.04). The primary endpoint of the trial, major cardiovascular events, was not significantly reduced (HR 0.90, P = 0.16), although the rate of nonfatal MI was lower in the intensive therapy group (HR 0.76, P = 0.004). To date, analyses have not identified any clear explanation for the higher mortality risk associated with the intensive glucose-lowering strategy. In the intensive therapy group, a median HbA1c level of 6.4% was rapidly achieved and maintained, but subsequent post hoc analyses implicated factors associated with persistently higher HbA1c, rather than low HbA1c, as likely contributors to the increased mortality risk. In addition, rates of serious hypoglycemia requiring medical assistance were threefold higher in the intensive group than during standard therapy (10.5% versus 3.5%, P < 0.001). Subsequent retrospective epidemiologic analyses of ACCORD have suggested, however, that severe hypoglycemia may not, in fact, account for the difference in mortality between the two study arms. Although hypoglycemia was associated with increased mortality within each randomized group, the risk of death was actually lower in participants experiencing hypoglycemia in the intensive arm than in participants with hypoglycemia in the standard arm. Other explanations, such as the particular medication combinations or undetected medication interactions, have also been proposed, but no particular medication class has been implicated thus far. In the end, the explanation for the increased mortality may never be known, but the findings have led to a growing recognition that intensive glucose lowering may be associated with some benefits but also important risks.

Subsequent follow-up of the participants from the ACCORD trial, up to the originally planned 5 years, showed persistently increased mortality rates in the intensive therapy group (HR 1.19, P = 0.02), but still lower rates of nonfatal MI (HR 0.82, P = 0.01). Given these findings, strategies used in the ACCORD study targeting an HbA1c below 6% are not recommended for patients with advanced type 2 diabetes and established macrovascular complications or multiple risk factors for cardiovascular events.

At the same time the ACCORD study was published, results from the Action in Diabetes and Vascular Disease: Preterax and Diamicron Modified Release Controlled Evaluation (ADVANCE) trial also became available. In this trial, 11,140 patients with type 2 diabetes (mean age 66, median baseline HbA1c 7.2%, 32% with history of major macrovascular disease) were randomized to intensive therapy (preferentially with the sulfonylurea gliclazide, targeting HbA1c ≤ 6.5%, with mean achieved HbA1c 6.5%) or to standard therapy (HbA1c goal according to local guidelines, mean achieved HbA1c 7.3%). After a median follow-up of 5 years, there was a reduction in the primary outcome of the study, which was a composite of microvascular and macrovascular events (HR 0.90, P = 0.01), and this was almost entirely driven by effects on intermediate markers of nephropathy. There was no significant effect with respect to major cardiovascular events (HR 0.94, P = 0.32) in ADVANCE, but also no increase in mortality (HR 0.93, P = 0.28) as in ACCORD. Investigators of ADVANCE specifically examined various subgroups at potentially increased risk of death, but none were identified. Overall, the trial findings suggest a modest improvement in markers of microvascular complications with intensive treatment, but no significant benefit gained with respect to CHD endpoints.

One additional trial confirmed the lack of significant benefit of intensive glucose lowering on major cardiovascular events, the Veterans Affairs Diabetes Trial (VADT). In this study of 1791 U.S. military veterans with type 2 diabetes (mean age 60, median baseline HbA1c 9.4%, 40% with prior history of cardiovascular events), participants were randomized to intensive glucose or standard therapy using a combination of agents, with a goal of achieving an absolute reduction in HbA1c of 1.5% in the intensive versus the standard arm. Median HbA1c levels achieved were 6.9% versus 8.4% in the intensive and standard groups, respectively, over a median follow-up of 5.6 years. There was no significant benefit with respect to the primary composite outcome, of cardiovascular events (HR 0.88, P = 0.14), the individual outcome of death from any cause, nor any difference with respect to most microvascular complications (no changes in retinopathy, new neuropathy, or doubling of creatinine, but reduction in some albuminuria-based endpoints). In this study of patients with advanced type 2 diabetes, other cardiovascular risk factors were well controlled, and differences in HbA1c levels between the two groups were maintained. However, overall, the benefit of decreasing HbA1c from 8.4% to 6.9% was minimal, except in the progression of albuminuria, an intermediate marker with uncertain implications for long-term renal risk.

Multiple meta-analyses have followed the three randomized controlled trials described earlier to determine whether pooling results of existing studies will illuminate our understanding of these relationships ( Fig. 13-4 ). Although these analyses sometimes included studies of different intent (not necessarily glucose-lowering per se) and with variable patient characteristics (newly diagnosed versus advanced diabetic patients), they consistently show a modest, although significant, reduction of approximately 15% in the risk for nonfatal MI, but no impact on mortality or cardiovascular death. Moreover, all of these studies show that the risk for severe hypoglycemia with intensive glucose therapy is more than doubled.

Why have these large randomized controlled trials failed to show that intensive glucose lowering improves cardiovascular outcomes when the epidemiologic relationship between glycemia and cardiovascular events is so convincing? Many of the expectations for reduction in risk may have arisen from the effects of statin medications on major adverse cardiovascular events (MACEs) in early trials. A strong epidemiologic association between cholesterol and CHD exists, and interventional studies show a 23% relative reduction in risk achieved for every 1 mmol/L (38 mg/dL) of cholesterol lowering. Glycemia is a much weaker risk factor for CHD than cholesterol, but the assumption has been that some degree of risk reduction should result from glucose lowering. However, it is clear that the simple arithmetic (lower the level of a risk factor and cardiovascular events will naturally follow) does not apply in the case of glycemia. There may be a modest reduction in nonfatal MI, but overall disappointing results with respect to mortality and the composite of major cardiovascular events. Several potential explanations can be proposed: significant adverse effects of glucose therapy may counterbalance possible benefits, effects of glucose lowering applied in advanced diabetes may be too late to prevent atherosclerotic events, or the impact of glucose lowering may take a longer time to materialize than the 5 to 7 years assessed in most clinical trials. Regardless of the reasons, currently available therapies tested in these studies do not appear to constitute a “magic bullet” for the increased cardiovascular morbidity of diabetes.

Trials of GLUCOSE-LOWERING Interventions in Prediabetes and Early Diabetes

Because one potential reason for the lack of benefit of intensive glucose lowering on cardiovascular disease is that it is applied too late to prevent atherosclerosis, it is worthwhile to examine studies that have investigated glucose lowering before diabetes actually develops or very early in the disease course.

One such study was the Diabetes Prevention Program that randomly assigned 3234 nondiabetic persons at high risk for diabetes (with elevated body mass index and fasting and postload glucose values) to intensive lifestyle therapy, metformin monotherapy, or placebo. Lifestyle intervention and metformin both significantly reduced the incidence of subsequent diabetes. In follow-up studies of the trial population, the lifestyle intervention also improved cardiovascular risk factors compared with metformin or placebo treatment, but the number of cardiovascular events was too small (n = 89 at 3 years) to allow any meaningful examination of the differences among groups. After 10 years of follow-up of the Diabetes Prevention Program, the cumulative incidence of diabetes was still lowest in the former lifestyle intervention group. Cardiovascular disease risk factors improved in all three treatment groups, but averaged over all follow-up, systolic and diastolic blood pressure and triglyceride levels were lower in the lifestyle than in the other groups (even though the use of antihypertensive medications was less frequent). However, the number of clinical events remained too small to determine the effect of diabetes prevention strategies on actual cardiovascular events.

In the Study to Prevent Non–Insulin-Dependent Diabetes Mellitus (STOP-NIDDM), investigators examined the effect of postprandial glucose lowering on the incidence of diabetes in 1429 participants with impaired glucose tolerance, elevated fasting glucose, and overweight or obesity. As a secondary outcome, investigators specifically examined the effect of intervention on major cardiovascular events and hypertension, although the trial was not adequately powered to answer that question. The participants were randomized to receive either acarbose or placebo and were followed for a mean of 3.3 years. In the course of the trial, almost one quarter of participants discontinued participation prematurely (including significantly greater numbers randomized to acarbose). Moreover, concerns about inconsistencies, failure to follow intention-to-treat analysis, and changes in the trial endpoints have been raised. Nevertheless, the trial reported an unanticipated 49% relative risk reduction ( P = 0.03) in major cardiovascular events (including revascularization procedures, congestive heart failure, and peripheral vascular disease in addition to the conventional major cardiovascular events) associated with acarbose therapy. This composite endpoint was primarily driven by an incredible 91% reduction in MI ( P = 0.02). Clearly, these trial findings will need to be confirmed in future studies before acarbose therapy can be recommended for cardiovascular risk reduction. One such trial, the Acarbose Cardiovascular Evaluation (ACE) study, is currently ongoing and testing the impact of acarbose on cardiovascular outcomes in persons with impaired glucose tolerance and established cardiovascular disease or acute coronary syndromes.

Encouraged by the STOP-NIDDM results, investigators of the Nateglinide and Valsartan in Impaired Glucose Tolerance Outcomes Research (NAVIGATOR) trial decided to test an alternative postprandial glucose-lowering approach with a short-acting insulin secretagogue, nateglinide, in addition to lifestyle modification. They randomized 9306 persons (mean age 64) with impaired glucose tolerance (baseline HbA1c 5.8%) at high risk for cardiovascular disease (24% had a prior history of cardiovascular events) to nateglinide or placebo and followed participants for a median of 5 years. Nateglinide did not reduce the occurrence of the three co-primary outcomes—incident diabetes (HR 1.07, P = 0.05), cardiovascular outcome (death from cardiovascular causes, nonfatal MI, nonfatal stroke, or hospitalization for heart failure, HR 0.94, P = 0.43), or the extended cardiovascular outcome (which in addition included unstable angina or arterial revascularization, HR 0.93, P = 0.16). The trial results contrast sharply with the findings of the STOP-NIDDM study and show that targeting postprandial hyperglycemia with nateglinide in participants with impaired glucose tolerance does not lead to cardiovascular benefits.

Another potential approach is to test whether early provision of basal insulin to normalize fasting plasma glucose may reduce cardiovascular outcomes in persons with mildly elevated glucose levels or early diabetes. This strategy does not specifically target postprandial hyperglycemia, but rather postulates that insulin, which has been demonstrated to have anti-inflammatory effects, may actually be cardioprotective and that it may also preserve pancreatic beta cell function over time. The Outcome Reduction with Initial Glargine Intervention (ORIGIN) trial randomized 12,537 people (mean age 64; 82% with early diabetes, 6% with new diabetes, and 12% with impaired fasting glucose or impaired glucose tolerance; median baseline HbA1c 6.4%; 59% with prior cardiovascular disease) to insulin glargine or standard care. The target in the insulin group was to achieve a fasting glucose level of 95 mg/dL or lower. After a median 6 years of follow-up, HbA1c levels were 6.2% versus 6.5% in the insulin and standard arms, respectively. The trial found no significant reduction in two co-primary outcomes, major cardiovascular events (HR 1.02, P = 0.63) and major cardiovascular events plus revascularization and heart failure (HR 1.04, P = 0.27). However, there was an increased risk of hypoglycemia with insulin therapy and some weight gain (+ 1.6 kg versus − 0.5 kg in the two groups). Although diabetes incidence was decreased (a finding of questionable clinical application), the trial findings were generally disappointing. The ORIGIN trial did not support the original hypothesis that normalizing glucose with early insulin therapy would lead to better cardiovascular outcomes.

These studies illustrate the impact of glucose lowering with a variety of different approaches in persons with prediabetes or early diabetes on cardiovascular outcomes. With the exception of the STOP-NIDDM study, which had important limitations, the studies to date have not supported the notion that glucose lowering is beneficial for cardiovascular outcomes at the prediabetic stage. Despite the epidemiologic association of glucose with cardiovascular events that extends well into the nondiabetic range, interventions targeting glycemia have thus far failed to deliver an effective strategy to reduce this risk.

The “How” of Glucose Lowering: the Evidence for Specific Medications and Medication Classes

Perhaps one of the most important lessons in cardiovascular risk reduction in type 2 diabetes has been the growing recognition that the exact strategy used to reduce glucose may actually matter with respect to outcomes. For several decades now, the thrust of clinical research has been to test approaches that target specific degrees of glucose lowering. Less attention was previously paid to the different ways in which glycemia is actually improved and how that may affect downstream events. The early UKPDS experience, albeit based on a small subgroup (n = 342) of patients, has given metformin a preferred place in the diabetic regimen for type 2 diabetes. In most subsequent trials (ACCORD, ADVANCE, VADT), combinations of various medications, including insulin use, had to be used to lower glucose levels, and there was no particular advantage to one strategy versus another. However, these glucose-lowering trials were not designed to test the effects of specific medications on outcomes.

Interest in the effects of specific antihyperglycemic medications on outcomes was boosted by the publication of a meta-analysis by Nissen and colleagues in 2007 that showed an adverse effect of the thiazolidinedione rosiglitazone on MI risk. In the analysis of 42 trials that was performed, there were 86 MI events in the rosiglitazone group compared with 71 in the comparator arm (including placebo, metformin, sulfonylurea, or insulin), resulting in an increased odds ratio of 1.43 ( P = 0.03). Although the methodology and the results of this meta-analysis have been debated, and some have noted no increase in risk associated with rosiglitazone use, ^, there is no doubt that the study provided a cautionary tale for glucose lowering. Most important, the study suggested that even though a medication may reduce glucose levels, and thus appear to treat diabetes effectively, it may in fact increase the risk of clinical events that are the target of glucose lowering in the first place. After the rosiglitazone experience, however, the U.S. Food and Drug Administration (FDA) mandated that glucose-lowering medications must have data that support their safety with respect to cardiovascular events before they are approved for use in diabetes.

Despite the intense interest in the specific effects of medication classes (and individual agents) that followed, there is a paucity of data to guide choice of therapy in diabetes with respect to long-term outcomes. Indeed, a recent comparative effectiveness analysis of 140 trials and 26 observational studies of medications for type 2 diabetes concluded that evidence on long-term clinical outcomes, including mortality, cardiovascular disease, nephropathy, and neuropathy, was of low strength and insufficient. Evidence supported metformin as a first-line agent in treatment of diabetes, but this evidence was primarily based on its efficacy to lower HbA1c levels, safety, adverse effect profile, and cost. We shall now review the available evidence for the major classes of antihyperglycemic therapy. These results are summarized in Table 13-1 .