Effect and cardiovascular safety of adding rosiglitazone to insulin therapy in type 2 diabetes: A meta-analysis.

AIMS/
INTRODUCTION: Recently, the use of rosiglitazone has been limited or withdrawn from the market as a result of cardiovascular risk. However, theoretically adding rosiglitazone to insulin could help insulin to decrease the glucose level. The present meta-analysis was designed to investigate the effect and safety of adding rosiglitazone to insulin therapy in type 2 diabetes.
MATERIALS AND METHODS: We searched published and unpublished databases through to March 2012. Randomized controlled trials (RCTs) comparing rosiglitazone in combination with insulin (RSG + INS) vs insulin alone (INS) in type 2 diabetes with outcomes including glycated hemoglobin levels, insulin dose, lipid parameters, blood pressure, edema and cardiovascular adverse events were selected.
RESULTS: Nine RCTs with durations of 24-26 weeks involving 1,916 patients were included. The RSG + INS group showed significantly decreased glycated hemoglobin levels by 0.89% (P < 0.00001) with an 8.48-U reduction in daily insulin dose (P <0.00001). However, the risks of hypoglycemia and edema were more frequent in the RSG+INS group (P < 0.0001; P = 0.03, respectively). Total cholesterol level was significantly increased in the RSG+INS group (P < 0.00001), but none of the high-density lipoprotein cholesterol, low-density lipoprotein cholesterol, very low-density lipoprotein cholesterol or triglyceride levels were significantly different between groups. There were no significant differences between groups with regard to the risks of myocardial infarction, heart failure, cardiovascular death or all-cause death.
CONCLUSIONS: Rosiglitazone could help type 2 diabetes patients with poorly controlled glucose with insulin therapy to decrease glucose levels and reduce their daily insulin dose, but at the cost of increased total cholesterol level, hypoglycemia and edema risk. Compared with insulin therapy, adding rosiglitazone to insulin did not increase the risks of myocardial infarction, heart failure, cardiovascular death or all-cause death.

Introduction

There are 346 million people worldwide with diabetes mellitus. Of those, 90% have type 2 diabetes1. Type 2 diabetes is a chronic metabolic disorder resulting from a progressive insulin secretary defect in the background of insulin resistance2. For patients diagnosed with type 2 diabetes, lifestyle modifications (including diet, exercise and weight loss) are recommended first; if patients have failed to adequately improve hyperglycemia, monotherapy with metformin as initial pharmacological therapy is recommended; if hyperglycemia persistently fails to be controlled, a second-line hyperglycemic drug is added to metformin (which drug is added to metformin firstly is not clearly recommended), insulin treatment should be started when it is necessary3.

Rosiglitazone, an oral antidiabetic drug, was initially approved in 1999 in the USA, and was widely used for its great improvement of glycemic control; it was even once the world's best-selling antidiabetic drug, with a $3 billion bill in annual revenue. It is in a class of insulin sensitizers known as thiazolidinediones (TZDs), which reduce plasma glucose by mainly decreasing insulin resistance, and increasing insulin sensitivity in peripheral tissue (muscle and adipose tissue) and liver by activating the peroxisome proliferator-activated receptor-γ (PPAR-γ), and helps to preserve pancreatic β-cell function4. However, rosiglitazone is associated with weight gain, edema, heart failure, bone fractures and increased cardiovascular events. In September 2010, the US Food and Drug Administration (FDA) declared restrictions for rosiglitazone, whereas the European Medicines Agency (EMA) withdrew the drug from the market based on meta-analyses of mostly short-term randomized controlled trials (RCTs) suggesting that it could increase myocardial infarction (MI) risk5,6. In November 2013, the FDA removed the restrictions for rosiglitazone that were put in place in 2010 as a result of a recent re-evaluation of the data from a large, long-term clinical trial carried out by the Duke Clinical Research Institute7.

Insulin therapy is a traditional and classical treatment that can be used in different types of diabetes. Almost two-third of individuals with type 2 diabetes receiving oral antidiabetic therapy only cannot achieve the glycemic target glycated hemoglobin (HbA1c) of 7% and require insulin therapy8,9. However, insulin therapy requires injection, which reduces patients' compliance, and is associated with increased incidence of hypoglycemia, weight gain, edema and insulin resistance, which would lead to an increased insulin dose.

Theoretically, adding rosiglitazone to insulin could help insulin decrease the glucose level by reducing insulin resistance. However, some studies suggested that rosiglitazone increases the risk of MI, especially in patients treated with insulin5,10. To date, there has been no large trial or meta-analysis on the combination therapy of rosiglitazone and insulin. The aim of the present meta-analysis was to investigate the effect and safety of adding rosiglitazone to insulin therapy in patients with type 2 diabetes, and to evaluate whether we should/should not use rosiglitazone anymore.

Materials and Methods

Eligibility Criteria

We selected RCTs comparing the combination therapy of rosiglitazone and insulin with the same insulin therapy. The duration of trials was at least 24 weeks. Both therapies can be with or without the same additional intervention, such as other classes of oral medications/lifestyle programs. Participants were adults with type 2 diabetes of any sex or race. Trials had to report at least one of the following outcomes: HbA1c levels, daily insulin dose, lipid parameters (total cholesterol, high-density lipoprotein cholesterol [HDL-c], low-density lipoprotein cholesterol [LDL-c], very low-density lipoprotein cholesterol cholesterol [VLDL-c] and triglyceride), blood pressure (systolic blood pressure [SBP] and diastolic blood pressure [DBP]), fluid retention events (weight gain and edema) and cardiovascular adverse events (MI, heart failure [HF], cardiovascular death [CV death] and all-cause death).

Search Strategy

We searched the Pubmed, Embase and Cochrane Central Register of Controlled Trials (CENTRAL) for studies published in English up to March 2012. The following search terms were used: “rosiglitazone”, “insulin”, “randomized controlled trial” or “clinical trial”. To identify unpublished data, we reviewed the websites of the drug manufacturer GlaxoSmithKline, FDA and ClinicalTrials.gov.

In addition, bibliographies of included studies, meta-analyses and recent reviews were checked.

Study Selection and Data Extraction

Two researchers independently checked titles and abstracts for studies that could potentially meet the inclusion criteria. Full-text articles of these studies were retrieved and reviewed for detailed assessment. If there were both published and unpublished data for the same trial, the unpublished data was considered to be superior after comparison. From the included studies, we extracted types of interventions, baseline characteristics of participants and relevant outcomes on to a preformatted spreadsheet. Any uncertainties or disagreements between the two researchers were resolved through discussion or consultation with a third person by checking the original articles.

Assessing Risk of Bias

In accordance with the recommendations of the Cochrane Handbook for Systematic Reviews of Interventions, we used the following features to assess risk of bias for included studies: sequence generation, allocation concealment, blinding of participants and outcome assessors, withdrawal rate, and intention-to-treat (ITT) analysis11. The features were graded as adequate (or yes), unclear or inadequate (or no). If all features were adequate (or yes), the information from the study was at low risk of bias. If one or more features were unclear, it was at unclear risk of bias. If one or more features were inadequate (or no), it was at high risk of bias12. Funnel plots were used to evaluate publication bias only if there were at least 10 studies included for some outcome and no significant between-study heterogeneity13.

Statistical Analysis

Statistical analysis was carried out using RevMan 5.0 software (Nordic Cochrane Centre, The Cochrane Collaboration, Copenhagen, Danish) provided by the Cochrane Collaboration. For continuous data, we calculated the weighted mean difference (WMD) or the standard mean difference (SMD) with 95% confidence intervals (95% CI) for the change of parameters from baseline to the end of the study. WMD was used when the data were measured by the same unit, and SMD was used when the data were measured by different units. For dichotomous data, we calculated the risk ratio (RR) with 95% CI. P-values <0.05 were considered to be statistically significant. The pooled results of studies included in the meta-analysis and statistical homogeneity between trials were assessed using the fixed-effects model where a χ2-test P > 0.1 indicated no heterogeneity. The random-effects model of meta-analysis was used when statistical heterogeneity exited (χ2-test P ≤ 0.1)14. If there was considerable variation in results (I2 > 75%), we only used descriptive analysis15.

Results

Trial Flow

The process of study selection is shown in Figure1.

Figure 1

Flow chart of included studies.

A total of 39 articles were identified as potentially relevant RCTs, and full-texts were reviewed. Of these, nine RCTs16–19 that fulfilled the inclusion criteria were included in the present meta-analysis. The remaining trials were excluded as a result of different reasons listed in Figure1.

Study Characteristics

The characteristics of the included studies are summarized in Table1.

Table 1

Characteristics of included studies

Study	Intervations		Participants
	Study arms	Duration (weeks)	Population	n	Age (years)	Male sex (%)	Country
49653/082	RSG (4 mg/day) + INS	26	T2DM poorly controlled on insulin monotherapy	107	57.1	56.6	USA
	RSG (8 mg/day) + INS			105	57.7	54.4
	PBO + INS			107	55.6	55.8
49653/085	RSG (4–8 mg/day) + INS	26	T2DM poorly controlled on insulin monotherapy	138	61.3	54.0	Europe
	PBO + INS			139	61.5	46.8
49653/095	RSG (4 mg/day) + INS	26	T2DM poorly controlled on insulin monotherapy	99	57.8	63.9	USA
	RSG (8 mg/day) + INS			97	57.4	58.9
	PBO + INS			96	58.9	45.3
49653/347	RSG (2 mg/day) + INS	24	T2DM poorly controlled on insulin monotherapy	209	52.7	57.0	USA
	RSG (2–4 mg/day) + INS			209	52.6	48.1
	PBO + INS			212	53.8	46.2
AVD102209	RSG (4 mg/day) + INS	24	T2DM poorly controlled on insulin monotherapy	132	56.8	48.0	China
	PBO + INS			131	55.9	52.4
Naka 2011	RSG (4 mg/day) + INS	24	T2DM poorly controlled on insulin monotherapy	17	64.7	17.6	Greece
	INS			14	67.3	28.6
Reynolds 2002	RSG (4 mg/day) + INS + LP	24	Obese T2DM poorly controlled on insulin theropy with/without oral antidiabetic medications	8	NR	NR	USA
	PBO + INS + LP			10	NR	NR
Wong 2005	RSG (4 mg/day) + INS	24	Insulin treated T2DM with stable glycemic control on peritoneal dialysis therapy	26	62.9	NR	China (Hong Kong)
	INS			26	61.6	NR
Yilmaz 2007	RSG (8 mg/day) + INS	24	T2DM poorly controlled on insulin monotherapy	15	57.6	46.7	Turkey
	INS			19	61.5	63.2

INS, Insulin; LP,lifestyle program; MET, metformin; NR, not reported; PBO, placebo; RSG, rosiglitazone; T2DM, type 2 diabetes mellitus.

Nine RCTs with a duration of 24–26 weeks involving 1,916 patients were included. Studies were from all over the world with different races and reported in English. Four studies were published16–19, whereas the other five were unpublished and available from the GlaxoSmithKline website. The number of participants among the studies varied from 18 to 630. Most participants were middle to old-aged type 2 diabetes patients with inadequate glucose control (HbA1c > 7.0%) on previous insulin treatment with/without oral antidiabetic medications. Rosiglitazone doses varied from 2 to 8 mg/day. Eight studies used large doses of rosiglitazone (4–8 mg/day), except one with <4 mg/day. Only one trial used additional lifestyle intervention (diet change and physical activity) besides rosiglitazone17. Insulin doses could be regulated as a result of the blood glucose level or HbA1c level, or if hypoglycemia occurred.

Risk of Bias

The details of the risk of bias are summarized in Table2.

Table 2

Bias of included studies

Study	Sequence generation	Allocation concealment	Double-blind	Withdrawal (%)	ITT analysis	Risk of bias
49653/082	Unclear	Unclear	Yes	21.0	Yes	Unclear
49653/085	Unclear	Unclear	Yes	14.8	Yes	Unclear
49653/095	Unclear	Unclear	Yes	21.6	Yes	Unclear
49653/347	Unclear	Unclear	Yes	27.8	Yes	Unclear
AVD102209	Unclear	Unclear	Yes	9.1	Yes	Unclear
Naka 2011	Unclear	Unclear	No	8.8	No	High
Reynolds 2002	Unclear	Unclear	Yes	14.3	No	High
Wong 2005	Yes	Unclear	No	9.6	Yes	High
Yilmaz 2007	Unclear	Unclear	Unclear	0.0	No	High

ITT, Intention-to-treat.

The randomization procedure was not reported or was unclear for eight studies, except one that was adequately randomized18. None of the studies reported on allocation concealment. A double-blinded method was used for clinical data in six studies, whereas two studies were open-labeled. The participant dropout rate ranged from 0% to 27.8%. ITT analysis was carried out in six studies. In summary, five of the nine studies were at unclear risk of bias, whereas the other four were at high risk.

Results of Meta-Analysis

The results of different outcomes are summarized in Table3.

Table 3

Results of meta-analysis: rosiglitazone vs insulin

Outcomes	No. studies	No. participants	Overall effect			Heterogeneity test
			Statistical method	Effect estimate [95% CI]	P	I2 (%)	P′
HbAc1	8	1159	WMD, random	−0.89 [−1.19, −0.58]	<0.00001	78	<0.0001
Insulin dose	7	1152	WMD, random	−8.48 [−11.68, −5.27]	<0.00001	62	0.01
Hypoglycemia	6	1815	RR, fixed	1.33 [1.16, 1.52]	<0.0001	0	0.48
Lipid parameters
Total cholesterol	6	525	SMD, fixed	0.44 [0.27, 0.61]	<0.00001	0	0.64
HDL-c	6	525	SMD, fixed	0.14 [−0.03, 0.31]	0.12	3	0.4
LDL-c	6	502	SMD, random	0.56 [−0.40, 1.52]	0.25	95	<0.00001
VLDL-c	2	389	WMD, fixed	3.55 [−1.04, 8.13]	0.13	27	0.24
Triglyceride	6	525	SMD, random	0.17 [−0.15, 0.48]	0.29	59	0.03
Blood pressure
SBP	3	101	WMD, fixed	−3.16 [−9.76, 3.43]	0.35	35	0.22
DBP	3	101	WMD, random	−0.39 [−13.34, 12.56]	0.95	87	0.0005
Fluid retention
Edema	7	1846	RR, fixed	1.44 [1.03, 2.01]	0.03	40	0.12
Weight gain	4	1504	RR, fixed	1.96 [0.88, 4.33]	0.10	0	0.48
CV adverse events
Myocardial infarction	3	1226	RR, fixed	2.46 [0.52, 11.70]	0.26	0	0.94
Heart failure	4	1518	RR, fixed	1.82 [0.49, 6.74]	0.37	0	0.99
Cardiovascular death	5	1203	RR, fixed	0.77 [0.22, 2.65]	0.68	0	0.72
All-cause death	6	1833	RR, fixed	0.70 [0.25, 1.95]	0.50	0	0.72

CV adverse events, Cardiovascular adverse events; CI, confidence interval; DBP, diastolic blood pressure; HbA1c, glycated hemoglobin; HDL-c, high-density lipoprotein cholesterol; LDL-c, low-density lipoprotein cholesterol; RR, risk ratio; SBP, systolic blood pressure; SMD, standardized mean differece; VLDL-c, very low-density lipoprotein cholesterol; WMD, weighted mean difference.

Glycemic Control (HbA1c) and Insulin Dose Reduction

The pooled data from eight studies reporting on HbA1c values showed that the combination therapy of rosiglitazone and insulin (RSG + INS group) significantly reduced HbA1c values compared with insulin therapy (INS group; WMD −0.89, 95% CI −1.19 to −0.58, P < 0.00001; Appendix S1). Meanwhile, the daily insulin dose was significantly reduced in the RSG + INS group than in INS group from seven studies (WMD −8.48, 95% CI −11.68 to −5.27, P < 0.00001; Appendix S2. However, there was significant heterogeneity among studies (I2 = 78%, P < 0.0001; I2 = 62%, P = 0.01).

Hypoglycemia

Compared with the INS group, the RSG + INS group was associated with a significantly increased risk of hypoglycemia from six studies (RR 1.33, 95% CI 1.16–1.52, P < 0.0001; Appendix S3). We found no evidence of statistical heterogeneity for this outcome (I2 = 0%, P = 0.48).

Lipid Parameters

Based on the pooled results of six studies, there was a significant increase in total cholesterol level in the RSG + INS group compared with the INS group (SMD 0.44, 95% CI 0.27–0.61, P < 0.00001; Appendix S4). There were no significant differences between two groups in the levels of HDL-c, LDL-c, VLDL-c or triglyceride (Appendix S5–S8). However, significant heterogeneity existed among studies for LDL-c and triglyceride (I2 = 95%, P < 0.00001; I2 = 59%, P = 0.03, respectively).

Blood Pressure

The pooled data from three studies showed there were no significant differences between the RSG + INS group and the INS group in improvement in SBP or DBP (WMD −3.16, 95% CI −9.76 to 3.43, P = 0.35; −0.39, 95% CI −13.34 to 12.56, P = 0.95; Appendix S9–S10). However, there was significant heterogeneity among studies for diastolic blood pressure (I2 = 87%, P = 0.0005).

Fluid Retention

Compared with the INS group, the RSG + INS group significantly increased the risk of edema (RR 1.44, 95% CI 1.03–2.01, P = 0.03; Appendix S11). Among 1,846 patients from seven studies that reported edema, there was only one patient that experienced serious edema. There was no significant increase in the risk of weight gain associated with rosiglitazone (RR 1.96, 95% CI 0.88–4.33, P = 0.10; Appendix S12). No heterogeneity existed among studies for these outcomes.

Cardiovascular Adverse Events

Based on the pooled results, the risks of MI, heart failure, CV death and all-cause death between the RSG + INS group and the INS group were not significantly different (RR 2.46, 95% CI 0.52–11.70, P = 0.26; RR 1.82, 95% CI 0.49–6.74, P = 0.37; RR 0.77, 95% CI 0.22–2.65, P = 0.68; 0.70, 95% CI 0.25–1.95, P = 0.50, respectively; Appendix S13–S16). No heterogeneity existed among studies for these outcomes.

Discussion

Nine RCTs involving 1,916 patients with type 2 diabetes were included in the present meta-analysis comparing adding rosiglitazone to insulin with insulin therapy. Five studies were at unclear risk of bias, whereas the other four were at high risk of bias. The results suggested that compared with the INS group, the RSG + INS group could effectively help insulin decrease HbA1c values by a mean of 0.89%; furthermore, with a mean reduction of 8.48 U daily insulin dose. Nevertheless, the RSG+INS group was associated with a 33% increased incidence of hypoglycemia. Total cholesterol level was increased in the RSG + INS group, but none of the levels of HDL-c, LDL-c, VLDL-c or triglyceride were significantly different. There were no significant changes of blood pressure (SBP or DBP) between the two groups. The RSG + INS group caused a 44% greater incidence of edema. The risks of MI, heart failure, CV death and all-cause death between the RSG + INS group and INS group were similar.

Insulin is a type of classical antidiabetic medication that has been used for almost 100 years. It can be used in every condition and different types of diabetes mellitus, but it also has some disadvantages. Hypoglycemia is a major treatment-associated complication of insulin, others include bruising, bleeding, lipohypertrophy, lipoatrophy, allergy, weight gain, edema, heart failure and potential cancer risk20. Besides, low compliance and a high withdrawal rate are common in insulin therapy. Some patients find daily insulin injection difficult and inconvenient, require training and are even fraught with some level of stigma. Type 2 diabetes is associated with insulin resistance, the long-term use of insulin might increase insulin resistance and a much higher dose of insulin injection is required, which would lead to elevated insulin adverse events. The synergistic effects of oral antidiabetic drugs with insulin might reduce the insulin dose, so insulin therapy is often accompanied with oral antidiabetic drugs. Common oral antidiabetic drugs and insulin combination therapy leads to a 0.3–0.6% decrease in HbA1c21, and is associated with a 43% reduction in total daily insulin requirement compared with insulin monotherapy22. The present results showed that rosiglitazone could decrease HbA1c by 0.89% accompanied with an 8.48-U reduction in daily insulin dose. Therefore rosiglitazone and insulin combination therapy has significant benefits for glycemic control over insulin monotherapy. Rosiglitazone is an insulin-sensitizing drug that can decrease insulin resistance, and improve insulin sensitivity in the liver and muscle. The synergistic effect of rosiglitazone and insulin could greatly help patients decrease glucose level and reduce daily insulin dose. However, the addition of rosiglitazone to insulin was associated with a high proportion of patients experiencing hypoglycemia. However, most hypoglycemic events were not serious.

Fluid retention is an adverse event of both insulin and rosiglitazone. The present results showed that combination therapy of rosiglitazone and insulin was associated with an increased incidence of edema. Though most edema was mild to moderate. The incidence of weight gain and heart failure were similar between the groups. A double-blind RCT showed that the incidence of heart failure of rosiglitazone was 1.5%, significantly higher than glyburide (0.6%), but similar to metformin (1.3%)23. A meta-analysis of seven RCTs showed that TZDs increased the risk of heart failure compared with controls in patients with prediabetes or type 2 diabetes by 72% (RR 1.72, 95% CI 1.21–2.42, P = 0.002) 24. Interestingly, TZDs-related heart failure would not increase the risk of CV death (RR 0.93, 95% CI 0.67–1.29, P = 0.68). Heart failure appeared to be a class effect, involving both rosiglitazone and pioglitazone, and was independent of dose or age or insulin use. TZDs-related heart failure was induced by fluid retention, but without the risk caused by progressive systolic or diastolic dysfunction of the left ventricle. In patients with type 2 diabetes and pre-existing chronic heart failure (New York Heart Association class I–II), 52 weeks with rosiglitazone therapy did not adversely affect left ventricular ejection fraction25. However, TZDs have been contraindicated in patients with all stages of heart failure (New York Heart Association class I–IV) in Europe and late stages (New York Heart Association III–IV) in the USA.

Over the past few years, the controversy about the cardiovascular safety of rosiglitazone has continued. In 2007, some meta-analyses ignited a firestorm about the ischemic cardiovascular risk of rosiglitazone. A meta-analysis by Nissen and Wolski including 42 trials with a duration of more than 24 weeks involving patients with type 2 diabetes, Alzheimer's disease and psoriasis reported a significant 43% increase risk of MI (odds ratio [OR] 1.43, 95% CI 1.03–1.98, P = 0.03) with rosiglitazone, with a borderline significant 64% increased risk in CV death (OR 1.64, 95% CI 0.98–2.74, P = 0.06)26. Similar results were seen in another meta-analysis including four long-term trials with a duration of at least 12 months. Among patients with impaired glucose tolerance or type 2 diabetes, rosiglitazone increased the risk of MI by 42%, but did not increase the risk of CV mortality6. Shortly thereafter, a patient-level meta-analysis carried out by the FDA showed that rosiglitazone was associated with a significant 40% increased risk of ischemic heart disease, but with no significant increase in the risks of MI or total mortality, or the composite end point of CV death, MI or stroke27.

In contrast to meta-analyses, large clinical trials had more favorable results on the cardiovascular safety of rosiglitazone. To date, there has been only one large randomized clinical trial prospectively designed to evaluate the effect of rosiglitazone on cardiovascular outcomes for type 2 diabetes, the Rosiglitazone Evaluated for Cardiac Outcomes and Regulation of Glycemia in Diabetes (RECORD) trial. The primary end-point was cardiovascular hospitalization or CV death, with a hazard ratio (HR) non-inferiority margin of 1.20. After a mean 5.5-year follow up, the results showed that rosiglitazone increased the risk of heart failure (HR 2.10, P = 0.01), but with no significant increased risk of MI (HR 1.14), CV death (HR 0.84), stroke (HR 0.72) or the composite of CV death, MI and stroke (HR 0.93)28. Large prospective randomized clinical trials designed to assess the macrovascular outcomes are the best approach to evaluate the relationship between rosiglitazone and cardiovascular events. Unfortunately, the weakness of the RECORD study was challenged. First, the open-label design introduced unavoidable biases that favored the rosiglitazone-treated group. Second, questions about the reliability of ascertainment and adjudication of the outcomes persisted. Therefore, the results from RECORD are inconclusive. Similar results were seen in another large RCT, the Diabetes Reduction Assessment with Ramipril and Rosiglitazone Medication (DREAM), which focused on patients with impaired glucose tolerance or impaired fasting glucose, and no previous cardiovascular disease. There was no significant evidence that rosiglitazone might increase the risk of MI, CV death, stroke or death compared with placebo with a 3-year follow up29. Data from A Diabetes Outcome Progression Trial (ADOPT) showed that rosiglitazone was associated with a higher cardiovascular risk than glyburide, but similar to metformin23.

In June 2010, Nissen and Wolski updated their meta-analysis by including 13 small trials and the recently completed RECORD trial in addition to their original 42 trials. Although the risk of MI remained statistically significant, it downgraded from 43% (P = 0.03) to 28% (P = 0.04). The previously borderline significant risk of CV death was no longer evident, from 64% (P = 0.06) to 3% (P = 0.86)5.

Most evidence on cardiovascular risk of rosiglitazone was derived from meta-analyses, but there were deficiencies in these meta-analyses. First, rosiglitazone was compared with active controls on blood glucose (life program intervention, metformin, sulfonylurea, pioglitazone, insulin) or placebo or other controls (donepezil). Second, except for the RECORD trial, most trials were designed to evaluate the hypoglycemic effect of drugs, not cardiovascular outcomes. Third, participants were greatly varied among trials, including participants with type 2 diabetes, prediabetes, psoriasis, Alzheimer's disease and rheumatoid arthritis. These would lead to great clinical differences. Fourth, the meta-analyses included many “zero-event” trials. Some researchers applied different methodological approaches to reanalyze the data of a meta-analysis, and found a non-significant or significant increased risk for MI30. Thus, the evidence regarding the cardiovascular risk of rosiglitazone is inconsistent, fragile and methodologically deficient.

On the contrary, pioglitazone has a more favorable effect on cardiovascular disease. A clinical trial designed to assess the effect of pioglitazone on macrovascular events, the Prospective Pioglitazone Clinical Trial in Macrovascular Events (PROactive) trial, showed that pioglitazone significantly reduced the composite of all-cause mortality, non-fatal MI and stroke (HR 0.84, 95% CI 0.72–0.98)31. A meta-analysis also showed the benefit of pioglitazone in regard to cardiovascular events, with an 18% reduced risk of the composite of death, MI and stroke32. An observational, retrospective, inception cohort study suggested that rosiglitazone was associated with increased risks of stroke, heart failure and the composite of MI, stroke, heart failure or death compared with pioglitazone33. Similar results for increased MI and death risks for rosiglitazone were obtained by a meta-analysis of 16 observational studies relative to pioglitazone34. The mechanism as to why the two TZDs have an opposite effect on cardiovascular events remains uncertain. The main potential factor might contribute to the different effect of the two TZDs on lipid metabolism. Some trials suggested pioglitazone had a more favorable effect on blood lipids. Pioglitazone was associated with a reduction in total cholesterol and triglycerides, and with an increase in HDL-c35. In contrast, rosiglitazone might increase total cholesterol and LDL-c compared with other therapies. In a comparative trial, pioglitazone produced less of an increase in LDL-c and triglyceride levels, a greater increase in HDL-c, reduced LDL particle concentration and increased LDL particle size compared with rosiglitazone36. The opposite effect between the two TZDs might be because pioglitazone is a PPAR-α/γ dual agonist, whereas rosiglitazone is a PPAR-γ agonist. The combination therapy of PPAR-α and PPAR-γ agonists led to a decrease in LDL-c levels, and might be of benefit in the treatment of type 2 diabetes patients with cardiovascular disease by downregulating cytokines37. PPAR-α agonist could improve cardiac dysfunction of adipose triglyceride lipase gene deletion mice38.

Despite the controversy about the cardiovascular safety of rosiglitazone being inconclusive, in 2010, the FDA announced restrictions of access to rosiglitazone and the EMA withdrew the drug from the market as a result of the cardiovascular risk. The FDA required the GSK to develop a restricted access program for rosiglitazone under a risk evaluation and mitigation strategy: rosiglitazone-containing medicines should only be used in patients already being treated with these medicines, and new patients only if they are unable to achieve glucose control on other medications and are unable to take pioglitazone.

However, in November 2013, the FDA removed the prescribing and dispensing restrictions for rosiglitazone that were put in place in 2010 based on a comprehensive, outside, expert re-evaluation of the data of the RECORD study by Duke Clinical Research Institute7. The HR for rosiglitazone vs metformin/sulfonylurea of CV death/MI/stroke was 0.95 (95% CI 0.78–1.17) compared with 0.93 (95% CI 0.74–1.15) for the original RECORD results. Treatment comparisons for MI (HR 1.13, 95% CI 0.80–1.59) and mortality (HR 0.86, 95% CI 0.68–1.08) were also the same compared with the original RECORD results (HR 1.14, 95% CI 0.80–1.63 for MI; 0.86, 95% CI 0.68–1.08 for mortality). The re-evaluation results showed that the risk of MI of rosiglitazone was similar to the standard type 2 diabetes medicines, metformin and sulfonylurea.

Because type 2 diabetes is a chronic metabolic disease, the risks of macrovascular events and mortality as a result of rosiglitazone were intended to be seen in long-term trials. The studies included in the present meta-analysis were too short (just 24–26 weeks), and fewer than six trials were used to evaluate the cardiovascular events and mortality. There was a high risk of bias in four studies. As sequence generation and allocation concealment were clearly described in only one study, measurement bias existed. Blinding was not used in two studies, so selective bias existed. Attrition bias cannot be ignored in two studies, with withdrawals where ITT analysis was not carried out.

In conclusion, in patients with type 2 diabetes who were poorly glucose controlled, adding rosiglitazone to insulin therapy showed an advantage in the reduction in HbA1c of 0.89% compared with insulin therapy, but at the cost of increased total cholesterol level, hypoglycemia and edema risk. As a result of the existing limitation (e.g. short study duration), the conclusion should be drawn cautiously that adding rosiglitazone to insulin does not increase the risks of MI, heart failure, cardiovascular or all-cause death. Clinical physicians should weigh the potential benefits and risks of rosiglitazone in different patients.