We identified 49 RCTs comparing the effects of 8 antihyperglycemic drug classes in patients with T2DM inadequately controlled with metformin monotherapy. To our knowledge, this analysis is the first to synthesize the available efficacy and safety data on all therapies for T2DM through Bayesian MTC meta-analysis. This approach combines direct and indirect evidence in a single analysis that enables simultaneous comparison of multiple treatment interventions in a clinically interpretable manner.18,79-81
Our results for HbA1c, hypoglycemia and body weight are generally consistent with other systematic reviews of oral antihyperglycemic drugs.10-14,82
All drug classes significantly reduced HbA1c relative to placebo to a similar degree. In some instances, our estimates of effect on HbA1c are somewhat lower than in other reviews. This may be due to our restricted focus on efficacy in the context of second-line therapy, because patients requiring second-line therapy may have more advanced diabetes and experience smaller treatment effects than treatment-naïve patients. However, our findings are similar to those reported by Phung and colleagues,82
who recently used MTC meta-analysis to assess the comparative efficacy of oral antihyperglycemic drugs added to metformin. Sulfonylureas, meglitinides, TZDs and insulins were associated with statistically significant increases in body weight ranging from approximately 2 kg to 3 kg relative to metformin alone. DPP-4 inhibitors and alpha-glucosidase inhibitors were found to not affect body weight, and GLP-1 analogues were associated with a statistically significant reduction in body weight of just under 2 kg. There are no well-accepted thresholds for the minimal weight change considered clinically significant, although weight reductions of 5%–10% (i.e., 3.5–7 kg for a 70-kg adult) are cited as such in the literature.83-88
In this context, the differences in body weight that we observed between classes are probably modest for most patients.
Both insulins and insulin secretagogues produced significantly increased hypoglycemia relative to placebo, whereas the TZDs, DPP-4 inhibitors, GLP-1 analogues and alpha-glucosidase inhibitors did not. Severe hypoglycemia events were rarely reported for all drug classes, including the insulins and insulin secretagogues. Large observational studies and long-term RCTs provide further insight into the risk of severe hypoglycemia among individuals with T2DM, although estimates vary considerably. Leese and colleagues reported 0.90 and 11.8 events that required emergency medical care per 100 patient-years with insulin secretagogues and insulin, respectively,89
whereas Bodmer and colleagues reported rates of 0.06 and 0.24 events that caused either hospitalization or death per 100 patient-years.90
In comparison, the ADVANCE trialists reported lower incidence rates than Leese and colleagues (0.7 per 100 patient-years in the intensive glycemic control arm versus 0.4 per 100 patient-years in the standard control arm), even though they defined severe hypoglycemia more liberally (i.e., medical resource use was not required).91
In the RECORD study, only 0.3% of subjects in the control arm (all of whom used metformin and a sulfonylurea) experienced a severe hypoglycemic event over the 5.5-year mean follow-up of the study.46
Overall, it appears that the risk of severe hypoglycemia with insulin secretagogues is quite low; therefore, any advantages of TZDs, GLP-1 analogues and DPP-4 inhibitors are probably modest in absolute terms. Further research is required to determine whether these agents provide greater benefits in patient groups at higher risk of severe hypoglycemia or its consequences.
Evidence regarding long-term diabetes-related complications and severe adverse events was inconclusive. The RECORD trial was the only included study powered to detect differences in long-term complications.46
Although we could not include these results in the review because of the lack of subgroup data for subjects initially taking metformin monotherapy, the overall results from RECORD are nevertheless noteworthy. Rosiglitazone was found to be non-inferior to the control treatment with respect to the primary macrovascular outcome of cardiovascular death or hospitalization, but the drug was associated with a significantly higher risk of heart failure and fractures. The data on fractures and heart failure were consistent with past studies,11,92,93
although controversy remains regarding the effects of TZDs on the risk of ischemic heart disease.94
The safety profile of the newest drug classes (i.e., DPP-4 inhibitors, GLP-1 analogues) requires further study in long-term observational studies and RCTs although there is evidence, albeit inconsistent, that they may be associated with pancreatitis.95,96
Advantages of older drug classes such as sulfonylureas and insulin are the availability of trial data regarding long-range safety97,98
and the extensive clinical experience with these agents.
Long-term studies such as the United Kingdom Prospective Diabetes Study (UKPDS) have convincingly demonstrated a progressive time-dependent increase in the HbA1c levels of patients with T2DM.99,100
This gradual loss of glycemic control is primarily attributable to a corresponding decrease in pancreatic beta-cell function. There is speculation that newer agents such as DPP-4 inhibitors, GLP-1 analogues and TZDs can offer the benefit of prolonged glycemic control by slowing the decline of beta-cell function; however, the evidence is limited and inconclusive. A recent systematic review of DPP-4 inhibitors reported that no definite conclusions can be made regarding their effects on beta-cell function.12
In contrast, A Diabetes Outcome Progression Trial (ADOPT) reported a statistically significant difference in the number of patients experiencing monotherapy failure, with a lower failure rate for TZDs than for sulfonylureas and metformin.101
The progressive nature of T2DM means that many patients will eventually require insulin therapy to maintain glycemic control. "In this context, oral agents that are capable of producing longer periods of sustained glycemic control could delay initiation of insulin initiation, which may be desirable for some patients and could result in cost savings, given the expense of insulin therapy. Given the expense of insulin therapy, delay in insulin initiation may also result in cost savings . We could find no conclusive evidence that TZDs and incretin mimetics have more durable effects on glycemia than sulfonylureas. Further long-term studies are needed to explore differences in glycemic durability between agents over time, especially for the newer, more expensive oral antidiabetes drugs.
Strengths and limitations
The strengths of our analysis were its comprehensiveness in terms of the drug classes considered, the number of outcomes assessed and the use of MTC meta-analyses incorporating both direct and indirect evidence in a clinically interpretable manner. However, certain limitations also deserve mention. First, potentially relevant non-English studies may have been excluded, although restriction to English-language studies has been reported to have minimal impact on systematic review results.102-105
Second, we did not assess non-serious adverse effects that can affect the tolerability of antihyperglycemic agents. For example, acarbose is commonly associated with gastrointestinal adverse effects that may limit its usefulness.61
Third, inclusion of insulin in the MTC meta-analysis may be viewed with scepticism because it is not commonly considered as second-line therapy after metformin in clinical practice and because trials of insulin may have enrolled patients with more advanced or severe disease than trials of oral agents. However, we believed it important to quantify the effects of insulin relative to other antihyperglycemic agents so that patients and clinicians can make informed choices regarding all available treatment options. Furthermore, scrutiny of subject characteristics revealed no major differences between the subjects enrolled in insulin trials and trials of other agents. Meta-regression analyses to adjust for differences in baseline HbA1c and duration of diabetes produced results that were similar to the reference case; therefore, any differences in these parameters between insulin and non-insulin studies were of little consequence.
Possible limitations concerning the internal validity and generalizability of the included studies should be noted. A majority of the RCTs in our analysis, including the largest trials, received a poor rating upon assessment for risk of bias. In addition, the majority of the trials failed to address 2 or more of the major sources of bias, that is, proper allocation concealment, use of intention-to-treat analysis and equal treatment of patients in each trial arm except for study medications. The clinical population of interest with respect to optimal second-line therapy consists of patients whose T2DM is inadequately controlled with metformin alone, the first-line treatment recommended by most guidelines. However, the available RCTs typically included patients with various treatment histories, such that metformin monotherapy failure did not necessarily occur in the context of first-line therapy. Nevertheless, we believe the relative treatment effects we report are transferable to patients treated with initial metformin monotherapy, because the reference case results were robust to adjustment (through meta-regression) for differences across studies in duration of T2DM and baseline HbA1c. These factors are probably more important predictors of treatment efficacy than treatment history per se. Although meta-regression analyses are limited by an inherent lack of statistical power, the fact that adjusted and unadjusted effect estimates were similar in nearly all analyses supports the generalizability of our results to a broad population of patients with T2DM inadequately controlled with metformin monotherapy.