Shari Bolen, MD, MPH; Leonard Feldman, MD; Jason Vassy, MD, MPH; Lisa Wilson, BS, ScM; Hsin-Chieh Yeh, PhD; Spyridon Marinopoulos, MD, MBA; Crystal Wiley, MD, MPH; Elizabeth Selvin, PhD; Renee Wilson, MS; Eric B. Bass, MD, MPH; Frederick L. Brancati, MD, MHS
Disclaimer: The authors of this article are responsible for its contents, including any clinical or treatment recommendations. No statement in this article should be construed as an official position of the Agency for Healthcare Research and Quality or of the U.S. Department of Health and Human Services.
Acknowledgment: The authors thank Steven Fox for his help as the Task Order Officer.
Grant Support: This article is based on research conducted by the Johns Hopkins Evidence-based Practice Center under contract number 290-02-0018 with the Agency for Healthcare Research and Quality. Dr. Brancati was supported by a mid-career investigator award for patient-oriented research in diabetes from the National Institute of Diabetes and Digestive and Kidney Diseases (5 K24 DK062222-05).
Potential Financial Conflicts of Interest: None disclosed.
Requests for Single Reprints: Shari Bolen, MD, MPH, Welch Center for Prevention, Epidemiology, and Clinical Research, Johns Hopkins University, 2024 East Monument Street, Suite 2-600, Room 2-615, Baltimore, MD 21205; e-mail, firstname.lastname@example.org.
Current Author Addresses: Drs. Bolen, Yeh, Selvin, and Brancati: Welch Center for Prevention, Epidemiology, and Clinical Research, Johns Hopkins University, 2024 East Monument Street, Suite 2-600, Baltimore, MD 21205.
Dr. Feldman: Johns Hopkins University, Jefferson Building, 600 North Wolfe Street, Room 242, Baltimore, MD 21287.
Dr. Vassy: University of Pennsylvania Health System, 3400 Spruce Street, Philadelphia, PA 19104.
Ms. L. Wilson, Ms. R. Wilson, and Drs. Wiley and Bass: Johns Hopkins University, 1830 East Monument Street, Eighth Floor, Baltimore, MD 21287.
Dr. Marinopoulos: University Health Services, Johns Hopkins University, 401 North Caroline Street, Baltimore, MD 21231.
Bolen S, Feldman L, Vassy J, Wilson L, Yeh H, Marinopoulos S, et al. Systematic Review: Comparative Effectiveness and Safety of Oral Medications for Type 2 Diabetes Mellitus. Ann Intern Med. 2007;147:386-399. doi: 10.7326/0003-4819-147-6-200709180-00178
Download citation file:
Published: Ann Intern Med. 2007;147(6):386-399.
As newer oral diabetes agents continue to emerge on the market, comparative evidence is urgently required to guide appropriate therapy.
To summarize the English-language literature on the benefits and harms of oral agents (second-generation sulfonylureas, biguanides, thiazolidinediones, meglitinides, and Î±-glucosidase inhibitors) in the treatment of adults with type 2 diabetes mellitus.
The MEDLINE, EMBASE, and Cochrane Central Register of Controlled Trials databases were searched from inception through January 2006 for original articles and through November 2005 for systematic reviews. Unpublished U.S. Food and Drug Administration and industry data were also searched.
216 controlled trials and cohort studies and 2 systematic reviews that addressed benefits and harms of oral diabetes drug classes available in the United States.
Using standardized protocols, 2 reviewers serially abstracted data for each article.
Evidence from clinical trials was inconclusive on major clinical end points, such as cardiovascular mortality. Therefore, the review was limited mainly to studies of intermediate end points. Most oral agents (thiazolidinediones, metformin, and repaglinide) improved glycemic control to the same degree as sulfonylureas (absolute decrease in hemoglobin A1c level of about 1 percentage point). Nateglinide and Î±-glucosidase inhibitors may have slightly weaker effects, on the basis of indirect comparisons of placebo-controlled trials. Thiazolidinediones were the only class that had a beneficial effect on high-density lipoprotein cholesterol levels (mean relative increase, 0.08 to 0.13 mmol/L [3 to 5 mg/dL]) but a harmful effect on low-density lipoprotein (LDL) cholesterol levels (mean relative increase, 0.26 mmol/L [10 mg/dL]) compared with other oral agents. Metformin decreased LDL cholesterol levels by about 0.26 mmol/L (10 mg/dL), whereas other oral agents had no obvious effects on LDL cholesterol levels. Most agents other than metformin increased body weight by 1 to 5 kg. Sulfonylureas and repaglinide were associated with greater risk for hypoglycemia, thiazolidinediones with greater risk for heart failure, and metformin with greater risk for gastrointestinal problems compared with other oral agents. Lactic acidosis was no more common in metformin recipients without comorbid conditions than in recipients of other oral diabetes agents.
Data on major clinical end points were limited. Studies inconsistently reported adverse events other than hypoglycemia, and definitions of adverse events varied across studies. Some harms not assessed in trials or observational studies may have been overlooked.
Compared with newer, more expensive agents (thiazolidinediones, Î±-glucosidase inhibitors, and meglitinides), older agents (second-generation sulfonylureas and metformin) have similar or superior effects on glycemic control, lipids, and other intermediate end points. Large, long-term comparative studies are needed to determine the comparative effects of oral diabetes agents on hard clinical end points.
The prevalence and morbidity associated with type 2 diabetes mellitus continue to increase in the United States and elsewhere (1, 2). Several studies of the treatment of type 2 diabetes suggest that improved glycemic control reduces microvascular risks (3–7). In contrast, the effects of treatment on macrovascular risk are more controversial (3, 4, 8, 9), and the comparative effects of oral diabetes agents on clinical outcomes are even less certain. As newer oral agents, such as thiazolidinediones and meglitinides, are increasingly marketed, clinicians and patients must decide whether they prefer these generally more costly medications over older agents, such as sulfonylureas and metformin.
Systematic reviews and meta-analyses of oral diabetes agents have attempted to fill this gap (10–19), but few have compared all agents with one another (18, 19). The few investigations that have compared all oral agents focused narrowly on individual outcomes, such as hemoglobin A1c level (18) or serum lipid levels (19). No systematic review has summarized all available head-to-head comparisons with regard to the full range of intermediate end points (including hemoglobin A1c level, lipid levels, and body weight) and other clinically important outcomes, such as adverse effects and macrovascular risks. Therefore, the Agency for Healthcare Research and Quality commissioned a systematic review to summarize the comparative benefits and harms of oral agents that are used to treat type 2 diabetes.
We searched MEDLINE, EMBASE, and the Cochrane Central Register of Controlled Trials from inception to January 2006 for original articles. We also searched these databases until November 2005 for systematic reviews. We reviewed reference lists of related reviews and original data articles, hand-searched recent issues of 15 medical journals, invited experts to provide additional citations, reviewed selected medications from the U.S. Food and Drug Administration (FDA) Web site, and reviewed unpublished data from several pharmaceutical companies and public registries of clinical trials. Our search strategy for the bibliographic databases combined terms for type 2 diabetes and oral diabetes agents and was limited to English-language articles and studies in adults. The search for systematic reviews was similar but included terms for study design as well.
We selected studies that included original data on adults with type 2 diabetes and assessed benefits or harms of FDA-approved oral diabetes agents that were available in the United States as of January 2006. To facilitate head-to-head comparisons of drug classes, we included drugs not on the U.S. market if members of their class were in use and had not been banned (voglibose, gliclazide, and glibenclamide). We also included studies of combinations of therapies that are commonly used, such as combinations of metformin, second-generation sulfonylureas, and thiazolidinediones. We excluded studies that evaluated combinations of 3 oral diabetes agents, and we also excluded first-generation sulfonylureas, because few clinicians prescribe these medications.
We sought studies that reported on major clinical outcomes (for example, all-cause mortality, cardiovascular morbidity and mortality, and microvascular outcomes) or any of the following intermediate end points or adverse events: hemoglobin A1c level, body weight, systolic and diastolic blood pressure, high-density lipoprotein (HDL) cholesterol level, low-density lipoprotein (LDL) cholesterol level, triglyceride level, hypoglycemia, gastrointestinal problems, congestive heart failure, edema or hypervolemia, lactic acidosis, elevated aminotransferase levels, liver failure, anemia, leukopenia, thrombocytopenia, allergic reactions requiring hospitalization or causing death, and other serious adverse events. For intermediate end points, we included only randomized, controlled trials, which were abundant. For major clinical end points and adverse events, we considered observational studies as well as trials, because fewer randomized trials assessed these end points. We excluded studies that followed patients for less than 3 months (the conventional threshold for determining effects on hemoglobin A1c) or had fewer than 40 patients. Figure 1 shows the search and selection process, and the full technical report (available at http://effectivehealthcare.ahrq.gov/repFiles/OralFullReport.pdf) provides a more detailed description of the study methods (20).
*Numbers add up to more than the number of abstracts or articles excluded because there may have been more than 1 reason for exclusion. †More than two thirds of the articles that were excluded for having fewer than 40 participants would have been excluded for other reasons as well. ‡The numbers of articles for intermediate outcomes, adverse events, microvascular and macrovascular outcomes, and mortality are not mutually exclusive.
One investigator used standardized forms to abstract data about study samples, interventions, designs, and outcomes, and a second investigator confirmed the abstracted data. Two investigators independently applied the Jadad scale to assess some aspects of the quality of randomized trials (21). We considered observational studies and nonrandomized trials to provide weaker evidence than randomized trials, and we did not use a standardized scoring system to assess their quality (22). We used the GRADE (Grading of Recommendations Assessment, Development and Evaluation) working group definitions to grade the overall strength of the evidence as high, moderate, low, very low, or insufficient (23).
We first performed a qualitative synthesis based on scientific rigor and type of end point. In general, we described the UKPDS (United Kingdom Prospective Diabetes Study) separately, because this large randomized, controlled trial differed from other trials in design, end points, and duration.
When data were sufficient (that is, obtained from at least 2 randomized, controlled trials) and studies were relatively homogeneous in sample characteristics, study duration, and drug dose, we conducted meta-analyses for the following intermediate outcomes and adverse effects: hemoglobin A1c level, weight, systolic blood pressure, LDL cholesterol level, HDL cholesterol level, triglyceride level, and hypoglycemia. For trials with more than 1 dosing group, we chose the dose that was most comparable with other trials and most clinically relevant. We combined drugs into drug classes only when similar results were found across individual drugs. We could not perform formal meta-analyses for microvascular or macrovascular outcomes, mortality, and adverse events other than hypoglycemia because of methodological diversity among the trials or insufficient numbers of trials.
We used a random-effects model with the DerSimonian and Laird formula to derive pooled estimates (posttreatment weighted mean differences for intermediate outcomes and posttreatment absolute risk differences for adverse events) (24). We tested for heterogeneity among the trials by using a chi-square test with α set to 0.10 or less and an I2 statistic greater than 50% (25). If heterogeneity was found, we conducted meta-regression analyses by using study-level characteristics of double-blinding, study duration, and dose ratio (calculated as the dose given in the study divided by the maximum approved dose of drug). The full report contains data on indirect comparisons, in which 2 interventions are compared through their relative effect against a common comparator (20). We tested for publication bias by using the tests of Begg and Mazumdar (26) and Egger and colleagues (27). All statistical analyses were done by using STATA Intercooled, version 8.0 (Stata, College Station, Texas).
The Agency for Healthcare Research and Quality suggested the initial questions and provided copyright release for this manuscript but did not participate in the literature search, data analysis, or interpretation of the results.
We found no definitive evidence about the comparative effectiveness of oral diabetes agents on all-cause mortality, cardiovascular mortality or morbidity, peripheral arterial disease, neuropathy, retinopathy, or nephropathy (Table 1). For each head-to-head comparison on specific outcomes, the number of randomized trials (≤3 trials) and the absolute number of events were small (20). The few observational studies were limited in quantity, consistency, and adjustment for key confounders.
Since our review, 2 high-profile comparative randomized trials with about 4 years of follow-up have been published, providing data on cardiovascular outcomes (28, 29). In ADOPT (A Diabetes Outcome Progression Trial) (28), the incidence of cardiovascular events was lower with glyburide than with rosiglitazone or metformin (1.8%, 3.4%, and 3.2%, respectively; P < 0.05). This effect was mainly driven by fewer congestive heart failure events and a lower rate of nonfatal myocardial infarction events in the glyburide group. Loss to follow-up was high (40%) and was disproportionate among the groups and therefore may account for some differences among groups.
The interim analysis of the RECORD (Rosiglitazone Evaluated for Cardiac Outcomes and Regulation of Glycaemia in Diabetes) study reported that rosiglitazone plus metformin or a sulfonylurea compared with metformin plus a sulfonylurea had a hazard ratio of 1.08 (95% CI, 0.89 to 1.31) for the primary end point of hospitalization or death from cardiovascular disease (29). The hazard ratio was driven by more congestive heart failure in the rosiglitazone plus metformin or sulfonylurea group than in the control group of metformin plus sulfonylurea (absolute risk, 1.7% vs. 0.8%, respectively). In Kaplan–Meier curves, the risk for hospitalization or death from myocardial infarction was slightly lower in the control group than in the rosiglitazone group, but the difference was not statistically significant. A limitation of this interim analysis was the lack of power to detect differences, owing to fewer cardiovascular events than initially predicted.
The strength of evidence was moderate to high that most oral agents (thiazolidinediones, metformin, and repaglinide) improved glycemic control to the same degree as sulfonylureas (decrease in hemoglobin A1c level, about 1 absolute percentage point). Nateglinide and α-glucosidase inhibitors may have slightly weaker effects on hemoglobin A1c levels on the basis of indirect comparisons of placebo-controlled trials (low strength of evidence). The strength of evidence was moderate that, compared with most other oral agents, thiazolidinediones had a beneficial effect on HDL cholesterol levels (relative mean increase, 0.08 to 0.13 mmol/L [3 to 5 mg/dL]) but a harmful effect on LDL cholesterol levels (relative mean increase, 0.26 mmol/L [10 mg/dL]). Metformin decreased LDL cholesterol levels by about 0.26 mmol/L (10 mg/dL), whereas other oral agents had no obvious effect on LDL cholesterol levels. The strength of evidence was moderate that thiazolidinediones, second-generation sulfonylureas, and metformin had similarly minimal effects on systolic blood pressure. There was moderate evidence that most agents other than metformin increased body weight by about 1 to 5 kg. Metformin had no effect on body weight in placebo-controlled trials.
Table 1 shows evidence grades and a summary of the comparative conclusions. These studies applied primarily to patients with type 2 diabetes and no major comorbid conditions.
The full report (20) provides a list of references and detailed evidence tables. We found 136 randomized trials that addressed intermediate outcomes and a systematic review on acarbose versus other oral diabetes agents (20). Study duration ranged from 12 weeks to 10 years, but most studies lasted 24 weeks or less. Participants were mainly middle-aged, overweight or obese adults of European ancestry who had had diabetes for more than 2 years and no major comorbid conditions. Mean baseline hemoglobin A1c levels ranged from 6% to 12% but were typically between 7% and 9%. About two thirds of studies received pharmaceutical industry support. Only 22 (16%) trials described their randomization techniques, and 83 (61%) reported double-blinding. In 33 (24%) studies, losses to follow-up and reasons for withdrawals were not described.
Figure 2 shows the comparative effects of oral diabetes agents on hemoglobin A1c. Thiazolidinediones, second-generation sulfonylureas, and metformin produced similar reductions in hemoglobin A1c levels when used as monotherapy (absolute reduction, about 1 percentage point). Repaglinide produced similar reductions in hemoglobin A1c levels compared with sulfonylureas. Combination therapies had additive effects, producing an absolute reduction in hemoglobin A1c levels of about 1 percentage point more than monotherapy.
Error bars represent 95% CIs. To convert cholesterol and triglyceride values to mmol/L, multiply by 0.0259 and 0.0113, respectively. Glyb = glyburide; HDL = high-density lipoprotein; LDL = low-density lipoprotein; Met = metformin; Pio = pioglitazone; RCT = randomized, controlled trial; Repag = repaglinide; Rosi = rosiglitazone; SU = sulfonylurea; TZD = thiazolidinedione.
The results of these meta-analyses were generally consistent with results of the UKPDS, a multicenter randomized trial starting in 1977 that had minimal loss to follow-up (3). After 3 months of dietary intervention, participants were stratified by ideal body weight and randomly assigned to receive insulin, chlorpropamide, glibenclamide, or dietary intervention alone. Overweight participants were also randomly allocated to metformin. All agents had similar effects on hemoglobin A1c levels. After 10 years, glibenclamide and metformin had a statistically insignificant between-group absolute difference of 0.3 percentage point (3, 30–32).
Few head-to-head comparisons involved repaglinide, nateglinide, or α-glucosidase inhibitors. To evaluate these agents, we therefore relied on indirect comparisons with placebo controls. Repaglinide produced similar reductions in hemoglobin A1c levels (about 1 absolute percentage point) when compared indirectly with thiazolidinediones and metformin. In contrast, nateglinide and α-glucosidase inhibitors produced weaker reductions in hemoglobin A1c levels (about 0.5 absolute percentage point). Appendix Table 1 shows findings for placebo-controlled trials and the full report on indirect comparisons (20).
Figure 2 shows the comparative effects of oral diabetes agents on blood pressure. Thiazolidinediones, second-generation sulfonylureas, and metformin had similarly minimal effects on systolic blood pressure (mean decrease <5 mm Hg). The greatest contrast was between thiazolidinediones and sulfonylureas—the former agent produced a 3–mm Hg greater reduction—but this difference was not statistically significant. Too few comparisons of meglitinides and acarbose with other oral diabetes agents in terms of blood pressure were available to draw firm conclusions. Results were similar for diastolic blood pressure (data not shown) (20).
Figure 2 shows the comparative effects of oral diabetes agents on plasma lipid levels. Metformin decreased LDL cholesterol levels by about 0.26 mmol/L (10 mg/dL), whereas thiazolidinediones consistently increased LDL cholesterol levels by a relative mean of 0.26 mmol/L (10 mg/dL). Sulfonylureas had similar minimal effects on LDL cholesterol compared with acarbose or repaglinide (33, 34).
Thiazolidinediones increased HDL cholesterol levels by a mean of 0.08 to 0.13 mmol/L (3 to 5 mg/dL) compared with metformin or second-generation sulfonylureas; these latter agents had little effect on HDL cholesterol. Combination therapy with thiazolidinediones increased HDL cholesterol levels similarly to monotherapy with thiazolidinediones. Repaglinide and acarbose had little effect on HDL cholesterol compared with second-generation sulfonylureas.
Only rosiglitazone increased triglyceride levels, by a mean of 0.11 mmol/L (10 mg/dL) in placebo-controlled trials (data not shown). Pioglitazone decreased triglyceride levels more than metformin, by a mean of 0.29 mmol/L (26 mg/dL), and metformin decreased triglyceride levels more than second-generation sulfonylureas, by a mean of 0.11 mmol/L (10 mg/dL). Repaglinide and acarbose produced similar reductions in triglyceride levels, by a mean of 0.11 to 0.34 mmol/L (10 to 30 mg/dL) compared with second-generation sulfonylureas.
Data on nateglinide were too sparse to draw conclusions about its comparative effects on lipid levels.
Compared with sulfonylureas, thiazolidinediones and repaglinide produced similar gains in body weight (1 to 5 kg). Metformin produced no weight gain compared with most other oral agents or placebo (Figure 2 and Appendix Table 2), and acarbose produced no weight gain compared with placebo (Appendix Table 2).
Three UKPDS articles reported weight changes that were consistent with these results favoring metformin over sulfonylurea (mean relative decrease, 2 kg at 10 years of follow-up) (3, 30, 32). Most of the weight gain in the glibenclamide group occurred in the first 2 years, whereas the metformin group maintained body weight in the first 2 years and then experienced weight gain (3).
Several randomized, controlled trials and some observational studies consistently demonstrate that minor and major hypoglycemic episodes are more frequent in adults receiving second-generation sulfonylureas (especially glyburide) than in those receiving metformin or thiazolidinediones. Repaglinide and second-generation sulfonylureas conferred similar risks for hypoglycemia.
In many trials and a few observational studies, metformin was almost always associated with more gastrointestinal problems (flatus, nausea, vomiting, and abdominal pain) than were most other oral diabetes agents. However, rates of lactic acidosis were similar between metformin and other oral diabetes agents, according to a systematic review of 176 comparative trials (35).
In many randomized trials, thiazolidinediones were associated with higher risk for edema than were sulfonylureas or metformin (absolute risk difference, 2% to 21%). Other than edema and hypoglycemia, we had difficulty assessing harms associated with thiazolidinediones because there were few trials and events. In addition, cohort studies often did not adjust for key confounders. Thiazolidinediones appeared to confer a higher risk for congestive heart failure (although absolute risks were small—generally 1% to 3%) and higher risk for mild anemia yet produced similarly low rates of elevated aminotransferase levels (<1%) compared with sulfonylureas and metformin.
Few studies compared the effect of meglitinides with that of other oral diabetes agents for outcomes other than hypoglycemia. Most studies on adverse effects were applicable to persons without major cardiovascular, renal, or hepatic comorbid conditions.
Overall, 167 original articles and 2 Cochrane systematic reviews evaluated adverse events (the full report provides a list of references and detailed evidence tables) (20). About two thirds of the studies were randomized, controlled trials, and the rest were observational. Most were based in the United States or Europe and had industry support. Study duration varied from 3 months to more than 10 years. Most participants were middle-aged to older adults of European ancestry who were overweight or obese. The duration of diabetes ranged from 1 year to 15 years, and mean baseline hemoglobin A1c levels were typically between 7% and 9%. Most randomized, controlled trials excluded people with major cardiovascular, hepatic, or renal disease.
Eighty-five percent (105 of 123) of the randomized, controlled trials with data relevant to adverse events did not describe the randomization technique in sufficient detail to determine whether the randomization was appropriate. About two thirds (66%) of these trials were reported as double-blind. However, 90 (73%) of these trials did not describe the masking procedure. Twenty-two (18%) trials did not report on withdrawals or losses to follow-up.
Minor and major hypoglycemic episodes were more frequent in patients receiving second-generation sulfonylureas (especially glyburide) than in those receiving metformin or thiazolidinediones. Absolute risk differences between groups ranged from 4% to 9% when sulfonylureas were compared with metformin or thiazolidinediones in short-term randomized trials, although reported levels of hypoglycemic risk ranged widely across studies: 0% to 36% for second-generation sulfonylureas, 0% to 21% for metformin, and 0% to 24% for thiazolidinediones.
The 10-year follow-up from UKPDS reported the annual rates of minor and major hypoglycemia as 17.5% and 2.5%, respectively, in the glibenclamide group (obese and nonobese persons) and 4.2% and 0%, respectively, in the metformin group (obese persons only). Results from observational studies were consistent with those of the UKPDS.
Glyburide and glibenclamide conferred a slightly higher risk for hypoglycemia compared with other second-generation sulfonylureas (absolute risk difference, about 2% in trials of short duration). Repaglinide and second-generation sulfonylureas conferred similar risks for hypoglycemia. Comparative data on acarbose and nateglinide were sparse. The incidence of minor and major hypoglycemia was higher with combinations that included sulfonylureas compared with metformin or sulfonylurea monotherapy (absolute risk differences of 8% to 14% for short-duration trials) (Figure 3).
Error bars represent 95% CIs. Glyb = glyburide; Met = metformin; Repag = repaglinide; SU = sulfonylurea; TZD = thiazolidinedione.
Metformin produced more gastrointestinal symptoms (range, 2% to 63%) than most other oral diabetes agents (range, 0% to 36% for thiazolidinediones, 0% to 32% for second-generation sulfonylureas, and 8% to 11% for repaglinide). The absolute risk differences among groups ranged from 0% to 31%, although most were between 5% and 15%. Acarbose produced percentages of gastrointestinal symptoms (range, 15% to 30%) similar to those with metformin and higher than those with thiazolidinediones and sulfonylureas in a few trials (<3 trials for each comparison). Too few comparative studies were available on nateglinide to draw firm conclusions (Table 2).
Currently marketed thiazolidinediones, second-generation sulfonylureas, and metformin had similarly low rates (generally <1%) of clinically significant elevated aminotransferase levels (>1.5 to 2 times the upper limit of normal). An insufficient number of studies evaluated or reported on the effects of meglitinides on serum aminotransferase levels, but they appeared to be similar to the effects of other oral diabetes agents (Table 2). Liver failure was so rare that agents could not be compared for this outcome by using these data.
Risk for congestive heart failure was greater with thiazolidinediones as monotherapy or combination therapy than with metformin or sulfonylureas (range of absolute risk differences, 0.7% to 2.2% in head-to-head, short-duration randomized trials). The absolute risk for congestive heart failure in the trials ranged from 0.8% to 3.6% for thiazolidinediones and 0% to 2.6% for nonthiazolidinediones. In contrast, neither metformin nor second-generation sulfonylureas were associated with congestive heart failure risk in 2 of 3 observational studies and 2 of 2 placebo-controlled trials. Congestive heart failure was reported mostly in cohort studies that did not adjust for key confounders, such as duration of diabetes, hemoglobin A1c level, blood pressure, and medication adherence. However, the cohort studies were consistent with one another and with limited data from randomized trials (Table 2).
Edema was more frequent in patients receiving thiazolidinediones as monotherapy or combination therapy (range, 0% to 26%) than in patients receiving second-generation sulfonylureas (range, 0% to 8%) or metformin (range, 0% to 4%). The absolute risk differences ranged from 2% to 21% in head-to-head randomized trials (Table 2).
We found a systematic review that reported similar rates of lactic acidosis between metformin and other oral diabetes agents (35). In this review, pooled data from 176 comparative trials and cohort studies totaling 35 619 patient-years revealed no cases of fatal or nonfatal lactic acidosis in any medication group. The estimated hypothetical upper limit of the underlying incidence of lactic acidosis was 8.4 cases per 100 000 patient-years in the metformin group and 9 cases per 100 000 patient-years in the nonmetformin group (35). We found 8 additional studies with data on lactic acidosis (3 randomized trials and 5 cohort studies). All showed little or no elevated risk for lactic acidosis in metformin recipients (Table 2).
Six head-to-head randomized trials, 7 placebo-controlled randomized trials, and 1 cohort study evaluated anemia as an outcome. Thiazolidinediones may be associated with an increased risk for anemia compared with other oral diabetes agents (posttreatment absolute risk differences, 1% to 5%). The mean decrease in hemoglobin level was small (≤1 g/dL). Only 1 study reported an adverse event of thrombocytopenia and leukopenia.
No study reported an allergic reaction to oral diabetes medications that led to hospitalization or death.
In addition to data published in peer-reviewed journals, we reviewed data from the FDA, unpublished trials conducted by industry, and clinical trial registries. The only new finding was that pioglitazone was associated with an increased risk for hospitalization for acute cholecystitis (12 patients) compared with placebo (1 patient) in a pooled analysis of 1526 patients (20). Otherwise, unpublished data were consistent with those from the published literature.
We did not find strong evidence of possible publication bias. Only 2 drug comparisons, from studies of hypoglycemia, had statistically significant results for publication bias (P < 0.05) according to the less conservative test of Egger and colleagues (27): metformin versus second-generation sulfonylureas (8 studies; P = 0.04) and repaglinide versus placebo (3 studies; P = 0.035). The 3 largest studies in the comparison of metformin with sulfonylureas had smaller absolute risk differences than the smaller studies; however, all studies showed that metformin is associated with less hypoglycemia than sulfonylureas. There were too few studies in the comparison of repaglinide versus placebo to draw conclusions about publication bias. For all other comparisons, the funnel plots appeared to be roughly symmetrical, and results of the tests of Begg and Mazumdar (26) and Egger and colleagues (27) were not statistically significant.
Ideally, oral diabetes agents should improve microvascular and macrovascular outcomes and mortality. We found no definitive comparative evidence on these outcomes. Because of this uncertainty, we evaluated medication effects on intermediate outcomes and other adverse events. By these criteria, we found that metformin was similar to, or better than, other currently available oral agents. Second-generation sulfonylureas also fared well against other agents, apart from the increased risk for hypoglycemia. Compared with newer agents, metformin and second-generation sulfonylureas share 3 additional advantages: lower cost, longer use in practice, and more intensive scrutiny in long-term trials with clinically relevant end points. Thiazolidinediones, although they pose a lower risk for hypoglycemia and a slight beneficial effect on HDL cholesterol level, showed no advantage in glucose-lowering effect and were associated with adverse effects on LDL cholesterol level, body weight, and risk for congestive heart failure.
These findings support the current American Diabetes Association and International Diabetes Federation recommendations that favor metformin as initial pharmacotherapy for type 2 diabetes (36, 37). They are also consistent with the 2007 American College of Endocrinology guidelines that suggest choosing an oral diabetes agent on the basis of the individual patient's burden of comorbid conditions (38). Of course, optimal glycemic control often requires multidrug therapy. Our review confirms that a second agent is additive both in terms of improved glycemic control and increased risk for adverse events, unless both agents are used at lower doses. Although they are not clearly superior to newer agents, sulfonylureas remain a reasonable alternative as second-line therapy, especially if cost is an issue.
Our findings are generally consistent with those of previous reviews of the effects of oral diabetes agents on intermediate outcomes, such as hemoglobin A1c level, lipid levels, and body weight (10, 12, 14, 16, 18, 19, 39, 40). Inzucchi (18) conducted a systematic review of the effect of oral diabetes agents and placebo on hemoglobin A1c and drew conclusions similar to ours. Our study adds to this research by including more recent articles, comparisons involving meglitinides, and meta-analyses of head-to-head comparisons. In a 2002 systematic review (without quantitative meta-analyses) on the lipid effects of oral diabetes medications, Buse and coworkers (19) reported findings similar to ours. Our investigation updates their review and adds more detail on differences between drugs from formal meta-analyses. The main contribution of our review is its comprehensiveness: We included a broad range of clinically relevant outcomes and adverse effects across all available drug classes.
Nissen and Wolski (11) recently reported results of a meta-analysis suggesting a relationship between use of rosiglitazone and risk for myocardial infarction. When they analyzed specific drug–drug or drug–placebo comparisons, however, their results were not statistically significant. Likewise, we found no statistically significant differences between specific oral diabetes medications in terms of cardiovascular outcomes other than congestive heart failure. Limitations of Nissen and Wolski's study included the small number of largely unadjudicated events and the fact that cardiovascular events were not the primary outcome. An additional limitation that influenced their conclusions was the decision to include studies with 2 diverse patient samples: nondiabetic persons, in whom the risk-to-benefit ratio of an oral diabetes agent may differ greatly from that in their diabetic counterparts, and diabetic persons with congestive heart failure, for whom rosiglitazone is contraindicated. The decision to include these studies may have biased the meta-analysis toward showing harm. Finally, exclusion of studies with no cardiovascular events in either group introduced a small bias against finding no difference in cardiovascular risk. Given the limitations of Nissen and Wolski's analysis, the effects of rosiglitazone on cardiovascular mortality and myocardial infarction are still uncertain. A recently published interim analysis from the RECORD study showed no statistically significant elevation in cardiovascular risk (besides congestive heart failure) related to rosiglitazone compared with metformin and sulfonylureas (20). Overall, these recent findings are consistent with ours: We found no conclusive evidence of worse cardiovascular morbidity or mortality with oral diabetes agents, other than the higher risk for congestive heart failure with thiazolidinediones than with other oral medications.
Several adverse events merit further discussion. First, because of concerns about lactic acidosis, metformin is contraindicated in patients with impaired renal function or congestive heart failure. However, neither our review nor that of Salpeter and colleagues (35) found evidence of an elevated risk for lactic acidosis in patients taking metformin compared with other oral diabetes agents. The evidence for metformin-induced lactic acidosis stems mainly from about 300 case reports. We did not consider case reports in our review because they pose problems in determining causality and provide no clear denominator for risk estimation. Underlying comorbid conditions, such as chronic kidney disease or myocardial infarction, are well-established risk factors for lactic acidosis; therefore, attributing lactic acidosis to metformin use versus an underlying comorbid condition is often difficult. Most reported cases of metformin-related lactic acidosis were associated with severe underlying illnesses (41, 42). Because of lingering fears about biguanides (phenformin was unequivocally related to risk for lactic acidosis), we suspect that apparent cases of “metformin-induced lactic acidosis” may have been overreported. However, we could not rule out the possibility that metformin conferred additional risk in the presence of severe underlying cardiac or renal disease, because these conditions were excluded in most randomized trials and were too uncommon in cohort studies to allow assessment.
Second, macular edema has been mentioned as an adverse event related to use of rosiglitazone only in case reports (43), which we excluded from our review. Third, the ADOPT study (published after our review was completed) reported an increase in fracture risk in women taking rosiglitazone compared with metformin or sulfonylureas (28). No cases were reported in the studies from our review, but this will need further investigation. Finally, repaglinide may be associated with less serious hypoglycemia compared with second-generation sulfonylureas, as was seen in 1 study of elderly persons (44), and in patients who skip meals, as was seen in 1 randomized trial not included in our review (because it was <3 months in duration) (45).
Our study has limitations. First, most of the trials, especially those of newer agents, were short-term trials, generally lasting less than 1 year. Ideally, therapeutic decision making should be based on long-term effectiveness. Second, head-to-head data were limited in many instances. This was especially true for multidrug regimens now in common use and for some of the newer agents, such as rosiglitazone, nateglinide, and miglitol. Third, although almost all studies reported the incidence of hypoglycemia, reporting of other adverse events was inconsistent, and the definitions of adverse events varied across studies. For instance, gastrointestinal events could include nausea, vomiting, abdominal pain, flatulence, or a combination of these events, making comparisons across studies difficult. Few trials reported data on elevated liver aminotransferase levels, congestive heart failure, anemia, and allergic reactions; therefore, we relied on cohort studies for many of these outcomes. The available cohort studies, however, were limited by their ability to adjust for key confounding factors, such as hemoglobin A1c level, blood pressure, duration of diabetes, adherence to medications, and medication dose. Finally, we focused on safety issues by making an a priori hypothesis of potential harm, and we may have missed harms reported only in case reports or those that were not assessed in trials or observational studies.
Compared with newer, more expensive agents (thiazolidinediones, α-glucosidase inhibitors, and meglitinides), older agents (second-generation sulfonylureas and metformin) have similar or superior effects on glycemic control and other cardiovascular risk factors (blood pressure, lipid levels, and body weight). Each oral diabetes agent is associated with adverse events that counterbalance its benefits. Overall, metformin seemed to have the best profile of benefit to risk. Large, long-term comparative studies on major clinical end points, such as myocardial infarction, chronic kidney disease, and cardiovascular mortality, are needed to determine definitively the comparative effects of the oral diabetes agents, especially in light of recent controversy regarding rosiglitazone.
Appendix Table 1.
Appendix Table 2.
The In the Clinic® slide sets are owned and copyrighted by the American College of Physicians (ACP). All text, graphics, trademarks, and other intellectual property incorporated into the slide sets remain the sole and exclusive property of the ACP. The slide sets may be used only by the person who downloads or purchases them and only for the purpose of presenting them during not-for-profit educational activities. Users may incorporate the entire slide set or selected individual slides into their own teaching presentations but may not alter the content of the slides in any way or remove the ACP copyright notice. Users may make print copies for use as hand-outs for the audience the user is personally addressing but may not otherwise reproduce or distribute the slides by any means or media, including but not limited to sending them as e-mail attachments, posting them on Internet or Intranet sites, publishing them in meeting proceedings, or making them available for sale or distribution in any unauthorized form, without the express written permission of the ACP. Unauthorized use of the In the Clinic slide sets will constitute copyright infringement.
Other Audio Options: Download MP3
Cardiology, Endocrine and Metabolism, Diabetes, High Value Care, Coronary Risk Factors.
Results provided by:
Copyright © 2016 American College of Physicians. All Rights Reserved.
Print ISSN: 0003-4819 | Online ISSN: 1539-3704
Conditions of Use
This PDF is available to Subscribers Only