Rosiglitazone evaluated for cardiovascular outcomes in oral agent combination therapy for type 2 diabetes (RECORD): a multicentre, randomised, open-label trial
The Lancet, Early Online Publication, 5 June 2009
doi:10.1016/S0140-6736(09)60953-3Cite or Link Using DOI
Editors' note: There has been some concern over the safety of thiazolidinediones. This paper reports the results of the post-marketing surveillance RECORD trial. With more than 4000 participants with type 2 diabetes, it provides useful evidence to inform on the frequency of adverse events with rosiglitazone compared with metformin and sulphonyureas, and will be useful in clinical practice.
Prof Philip D Home DM a, Prof Stuart J Pocock PhD b, Prof Henning Beck-Nielsen DMSC c, Paula S Curtis PhD d, Prof Ramon Gomis MD e, Prof Markolf Hanefeld MD f, Nigel P Jones MA g, Prof Michel Komajda MD h, Prof John JV McMurray MD i, for the RECORD Study Team
a Newcastle Diabetes Centre and Newcastle University, Newcastle upon Tyne, UK
b Medical Statistics Unit, London School of Hygiene and Tropical Medicine, London, UK
c Department of Endocrinology and Metabolism, Odense, Denmark
d GlaxoSmithKline Research and Development, Greenford, UK
e Hospital Clinic, University of Barcelona, Barcelona, Spain
f Zentrum fur Klinische Studien Forschungsbereich Endokrinologie und Stoffwechsel, Dresden, Germany
g GlaxoSmithKline Research and Development, Harlow, UK
h Universite Pierre et Marie Curie Paris 6; Hopital Pitie-Salpetriere, Departement de Cardiologie, Paris, France
i British Heart Foundation Cardiovascular Research Centre, University of Glasgow, Glasgow, UK
Rosiglitazone is an insulin sensitiser used in combination with metformin, a sulfonylurea, or both, for lowering blood glucose in people with type 2 diabetes. We assessed cardiovascular outcomes after addition of rosiglitazone to either metformin or sulfonylurea compared with the combination of the two over 5-7 years of follow-up. We also assessed comparative safety.
In a multicentre, open-label trial, 4447 patients with type 2 diabetes on metformin or sulfonylurea monotherapy with mean haemoglobin A1c (HbA1c) of 7·9% were randomly assigned to addition of rosiglitazone (n=2220) or to a combination of metformin and sulfonylurea (active control group, n=2227). The primary endpoint was cardiovascular hospitalisation or cardiovascular death, with a hazard ratio (HR) non-inferiority margin of 1·20. Analysis was by intention to treat. This study is registered with ClinicalTrials.gov, number NCT00379769.
321 people in the rosiglitazone group and 323 in the active control group experienced the primary outcome during a mean 5·5-year follow-up, meeting the criterion of non-inferiority (HR 0·99, 95% CI 0·85-1·16). HR was 0·84 (0·59-1·18) for cardiovascular death, 1·14 (0·80-1·63) for myocardial infarction, and 0·72 (0·49-1·06) for stroke. Heart failure causing admission to hospital or death occurred in 61 people in the rosiglitazone group and 29 in the active control group (HR 2·10, 1·35-3·27, risk difference per 1000 person-years 2·6, 1·1-4·1). Upper and distal lower limb fracture rates were increased mainly in women randomly assigned to rosiglitazone. Mean HbA1c was lower in the rosiglitazone group than in the control group at 5 years.
Addition of rosiglitazone to glucose-lowering therapy in people with type 2 diabetes is confirmed to increase the risk of heart failure and of some fractures, mainly in women. Although the data are inconclusive about any possible effect on myocardial infarction, rosiglitazone does not increase the risk of overall cardiovascular morbidity or mortality compared with standard glucose-lowering drugs.
GlaxoSmithKline plc, UK.
Individual oral glucose-lowering medications have limited efficacy,1-3 and hence are commonly used in combination.4 In 2000, the thiazolidinediones rosiglitazone and pioglitazone received marketing authorisation for use in combination with metformin and sulfonylureas in Europe. These thiazolidinediones were known to cause fluid retention and possibly heart failure, and both manufacturers were requested to undertake a post-marketing cardiovascular outcome study.5, 6
From data reported in the UK Prospective Diabetes Study (UKPDS) in 1998, metformin seemed to protect against cardiovascular risk, but uncertainty remained for sulfonylureas.1, 7 After 1999, evidence showed that thiazolidinediones improved some cardiovascular risk markers associated with diabetes-including insulin sensitivity, blood pressure, and coagulation factors.2, 8, 9 However, higher LDL cholesterol concentrations, albeit in the context of improvement in LDL phenotype and unchanged LDL to HDL cholesterol ratio,10 raised concern about the overall cardiovascular effect of rosiglitazone.
Concerns increased because several peroxisome proliferator-activated receptor (PPAR) αγ agonists failed in development as a result of cardiovascular problems in humans or malignancy, in particular bladder tumours, in animals.11 The PROactive secondary prevention study of the PPARαγ agonist pioglitazone was inconclusive for its primary composite cardiovascular endpoint, but showed a reduction for the secondary composite of death, myocardial infarction, and stroke compared with placebo.12
For rosiglitazone, which is a PPARγ agonist, an active-comparator cardiovascular outcome study (RECORD) was designed from the time of marketing authorisation.13 In 2006, the manufacturer (GlaxoSmithKline) submitted to drug regulators a combined analysis of several studies which suggested that, despite large observational studies to the contrary,14, 15 rosiglitazone increased myocardial ischaemia.16 Nissen and Wolski, using similar data sources, reached similar conclusions.17 The RECORD steering committee published an unplanned interim analysis at that time.18
Here, we report the final planned analysis of the RECORD trial, with 7935 (48%) more person-years of follow-up than at interim analysis, and we focus mainly on protocol-defined cardiovascular outcomes. Furthermore, accumulation of around 25 000 person-years of follow-up provides the opportunity to address, through adverse-event reporting, concern over malignancy11 and the issue of limb fractures raised by the ADOPT study.19
Our aim was to assess non-inferiority of rosiglitazone in combination with metformin or sulfonylurea compared with metformin and sulfonylurea dual therapy for cardiovascular outcomes. The primary endpoint was the time to first cardiovascular hospitalisation or cardiovascular death.
The aim of the RECORD trial was to assess the overall cardiovascular safety of rosiglitazone, when added to metformin or sulfonylurea treatment, compared with that of an active control of standard dual therapy, metformin plus sulfonylurea.13 The primary endpoint included cardiovascular death or any cardiovascular hospitalisation during follow-up, which occurred in 321 people of the rosiglitazone group and 323 of the active control group. The trial had less statistical power than initially planned because the overall primary event rate in the 4447 patients followed for a mean of 5·5 years was substantially lower than that anticipated in the protocol power calculation, for reasons previously discussed.18 Nevertheless, the CI for the primary endpoint's HR excluded the predefined 20% excess risk, thereby satisfying the criterion of non-inferiority.
Two predefined sensitivity analyses-a per-protocol analysis on randomised dual therapy and one excluding certain primary endpoints not of atherosclerotic origin-gave similar findings. Within-strata analyses also gave similar findings when comparing rosiglitazone with metformin, and rosiglitazone with sulfonylurea, all in combination with the other background therapy.
Previously recognised increased risk of heart failure with thiazolidinediones is confirmed in our study.12, 21, 22 Here, the risk was doubled with rosiglitazone compared with active control (61 vs 29 cases, p=0·0010), indicating an excess of one heart failure event per 385 person-years. Because these events resulted in hospitalisation and because the number of deaths for heart failure increased, at least in some cases thiazolidinedione-associated heart failure is not benign.
The prespecified secondary composite endpoint of cardiovascular death, myocardial infarction, or stroke was similar for patients on rosiglitazone or active control (154 vs 165 cases, table 4). However, debate on the possible effects of rosiglitazone on myocardial infarction has been extensive after the publication of the revised European licence in 2006 and the Nissen and Wolski report in 2007.16, 17, 23, 24 In the RECORD trial, we had an excess of eight cases of myocardial infarction (both fatal and non-fatal) in the rosiglitazone group, which gives rise to an HR of 1·14 with a wide CI, which is not statistically significant. Therefore, the evidence regarding a potential risk of myocardial infarction of patients on rosiglitazone compared with that of controls is inconclusive; however, if a risk exists, it is low and does not increase the fatality rate (table 4). This risk is in comparison with agents that gave reductions of myocardial infarction in the UKPDS.1, 7, 25
In the rosiglitazone group, there were 17 fewer strokes than in the active control group (46 vs 63 cases), which corresponds to a non-significant risk reduction of 28%. Stroke was significantly reduced by treatment with rosiglitazone by around 50% in a combined analysis of all early trials.16
Both all-cause deaths (136 vs 157) and cardiovascular deaths (60 vs 71) were somewhat fewer in the rosiglitazone group, but neither difference was statistically significant (table 4). The significant excess of deaths due to heart failure in the rosiglitazone group is compatible with overall increased risk of heart failure. It is offset by a similar observed reduction in deaths from stroke and other acute vascular events in the rosiglitazone group.
Findings from prespecified subgroup analyses of the primary endpoint were unremarkable, except for a possible, but not statistically significant, increased risk for cardiovascular events for patients on rosiglitazone with previous ischaemic heart disease (figure 4). The excess relative risk for heart failure was similar for people with and without previous ischaemic heart disease. In people with previous myocardial infarction or angina, there was no excess of cardiovascular events with rosiglitazone compared with control medications.
Changes in cardiovascular risk factors over 5 years show evidence of superiority of rosiglitazone with regard to blood glucose control compared with both metformin and sulfonylurea (table 2), which is in line with a previous long-term monotherapy trial.26 However, the rosiglitazone group had higher serum LDL cholesterol (but also higher HDL cholesterol) than had the active control group, and greater bodyweight gain at 5 years.
Data from UKPDS suggested that monotherapy with metformin in an obese subpopulation was beneficial compared with conservative management: RR for myocardial infarction was 0·61 (0·41-0·89) at planned study end.7, 25 However, in a smaller study, metformin added to sulfonylurea seemed to increase diabetes-related death, with an unchanged risk of myocardial infarction.7 In the ADOPT study, in which cardiovascular events were not adjudicated, sulfonylureas were associated with lower rates of cardiovascular outcomes, whereas metformin and rosiglitazone were similar.16, 26 Here, we found in the two strata separately that rosiglitazone is similar in combination therapy for risk of cardiovascular events to both metformin and sulfonylureas. Therefore, the implication is that all three medications affect overall cardiovascular outcomes similarly.
The ADOPT study raised the issue of distal fractures in women.19 Our study shows that the increased risk of fractures seems confined to upper limb and non-femur/hip lower limb fractures, and is mainly in women (table 7). Although our data might be reassuring regarding the issue of osteoporotic hip and spine fractures, longer follow-up in a higher-risk population, such as the elderly, is probably warranted before drawing any conclusion.
No overall difference in the incidence of malignancies was recorded, including for bladder, prostate, breast, or colon cancer. The group treated with rosiglitazone had a reduced incidence of pancreatic cancer, but this might have happened by chance. A suggestion of increased pneumonia with pioglitazone in the PROactive trial is not confirmed for rosiglitazone.12 Data are available from adverse-event reporting on macular oedema, with a possible increase of non-serious events in patients treated with rosiglitazone.
Some potential weaknesses of the study might have had an effect on the conclusions. The greater use of loop diuretics in the rosiglitazone group than in the control group might have decreased the rate of heart failure, although the increased use might have been because of heart failure. Similarly, the rate of arteriosclerotic events in the rosiglitazone group might have been reduced by greater statin use than in the control group (9·2% higher at 5 years), although in both groups statin use was below that recommended as part of good clinical management for diabetes.4, 27 However, because statins reduce arteriosclerotic events by 25%, the HR could have been reduced by no more than 2%, which would not alter the conclusion of non-inferiority.
The study also had limited statistical power for individual components of the primary endpoint because it was never intended to answer these questions. Furthermore, the overall cardiovascular event rate was lower than anticipated at the time of trial design, limiting the scope for reliable exploratory analyses. Nevertheless, non-inferiority for the primary endpoint was still achieved because the point estimate of the HR was close to 1·00. Whether myocardial infarction and stroke are worsened or improved by treatment with rosiglitazone compared with metformin or sulfonylurea is not clear, but mortality is not increased.
Although this study was open label, in-depth site and data inspections in some countries did not suggest any bias in reporting. Non-compliance and other biases in study conduct might have been an issue after the publication of the Nissen and Wolski report in 2007,17 but publication of the interim RECORD data analysis and the statements of the study's data safety and monitoring board convinced investigators and participants of the ethics and safety of continuing the study.18 Loss of participants during follow-up is inevitable in a large multicentre international long-term clinical trial. In the RECORD study, this loss was low for vital status (2·0% person-years of follow-up), but the number of people withdrawing from study visits was greater (7·2% person-years of follow-up).
Strengths of the study include the blinded endpoint adjudication and recruitment of a clinically representative population from various countries across Europe and Australasia. Internal data checks did not indicate geographical variation or that the background strata represent populations from different regions. Additionally, blood glucose control at baseline and during the study, although on average worse than desirable levels, was not unrepresentative of usual practice. The comparators used were glucose-lowering drugs that are cardiovascular protective1, 7, 26 and are alternative agents of choice when prescribing rosiglitazone,4, 27 providing a strong clinical context for interpreting the findings.
What are the clinical implications for the future use of rosiglitazone? Rosiglitazone is not recommended for people with a history of heart failure or with previous problems that might have led to myocardial dysfunction. Rosiglitazone should be used with caution in women at high risk of fractures. Although our evidence is insufficient to rule out a small increased risk of myocardial infarction caused by rosiglitazone when compared with other glucose-lowering agents, rosiglitazone does not increase overall cardiovascular morbidity or mortality.
Of 7428 patients screened, 4458 were randomly allocated to study groups (figure 1). 11 did not take study medication and were excluded from analysis. 2222 people on metformin were assigned to addition of rosiglitazone (1117) or sulfonylurea (1105), and 2225 patients on a sulfonylurea were assigned to addition of rosiglitazone (1103) or metformin (1122). Mean follow-up duration was 5·5 years, which corresponded to 12 338 person-years in the rosiglitazone group and 12 272 person-years in the comparator group. Vital status was unknown at study end in 127 people (2·9%, 60 rosiglitazone and 67 control), and a further 394 were alive but withdrew from some study visits, thereby missing complete cardiovascular endpoint information (8·9%, 189 rosiglitazone and 205 control). In the 12 months after May 20, 2007 (the time of publication of the report by Nissen and Wolski17), discontinuations from rosiglitazone therapy increased slightly compared with those in the active control group. This increase amounted to an excess of 32 people (1·4% of the randomised population) withdrawn from randomised treatment.
Baseline characteristics between randomisation groups were similar (table 1). In the background metformin stratum, people were younger, more obese, and had a shorter duration from diabetes diagnosis than those in the background sulfonylurea stratum (table 1).
Mean HbA1c at 5 years was lowered more in people randomly assigned to rosiglitazone, whereas bodyweight and HDL cholesterol increased more, and LDL cholesterol was reduced less, than in the groups randomly assigned to either sulfonylurea or metformin (table 2). In the rosiglitazone group, 75% of person-years' follow-up were on dual oral therapy and 13% on triple oral therapy. In the control group, 83% of person-years' follow-up were on dual oral therapy. Use of concomitant cardiovascular-related medications had risen greatly by 5 years for both rosiglitazone and active control groups (table 3). Use of statins was significantly greater (9·2%) at 5 years in the rosiglitazone group than in the control group; this was also the case for loop diuretics (4·9%).
Table 4 shows the numbers of people affected, HR, risk difference, CIs, and p values for the main cardiovascular outcomes. Kaplan-Meier plots are in figures 2 and 3. The event rate was 28 per 1000 person-years for the primary endpoint. The primary endpoint (cardiovascular hospitalisation or cardiovascular death) occurred in 321 and 323 participants assigned to the rosiglitazone and active control groups, respectively (HR 0·99, 95% CI 0·85-1·16). The prespecified criterion of non-inferiority was met (table 4). A prespecified sensitivity analysis excluding some cardiovascular events not of atherosclerotic origin gave similar results (HR 0·97, 0·82-1·14). A per-protocol analysis excluding any participant 30 days after transfer from dual therapy gave similar results (HR 1·02, 0·85-1·21): cardiovascular deaths or hospitalisations were 240 on rosiglitazone and 260 on active control, with exposures of 8905 and 9818 person-years, respectively. Within-strata comparisons of those randomly assigned to rosiglitazone with those randomly assigned to metformin (HR 0·98, 0·79-1·21) or to sulfonylureas (HR 1·01, 0·81-1·26) gave similar results, with no evidence of between-strata interaction (p=0·84).
Cardiovascular deaths and all-cause deaths were fewer in the rosiglitazone group than in the control group, but not significantly different (table 4). HRs for myocardial infarction and stroke were slightly but not significantly different. The predefined composite secondary endpoint of cardiovascular death, myocardial infarction, and stroke gave an HR of 0·93 (0·74-1·15) for rosiglitazone against active comparator. A broader composite also including heart failure and unstable angina gave an HR of 0·99 (0·81-1·20). The rate of heart failure approximately doubled in the rosiglitazone group compared with that in the control group (table 4).
Table 5 shows the types of cardiovascular death and hospitalisation by treatment group. For cardiovascular mortality, the rosiglitazone group had more heart-failure-related deaths but fewer stroke-related deaths and other vascular deaths than had the active control group. A sensitivity analysis using investigator opinion to attribute cause to deaths adjudicated as of unknown aetiology did not affect the findings. Total cardiovascular hospitalisations were almost equal in the two groups: the rosiglitazone group had more heart-failure hospitalisations, fewer stroke hospitalisations, and fewer cardiovascular procedures than had the active control group.
Prespecified subgroup analyses on the primary endpoint showed consistent findings for the randomised comparison for participants subdivided by baseline characteristics (figure 4). Only for patients with previous history of ischaemic heart disease (yes or no) was there a suggestion of heterogeneity (interaction p=0·055), whereby more primary endpoints seem to be present for those with previous ischaemic heart disease in the rosiglitazone group than in the active comparator group (figure 4).
Although the incidence of heart failure was higher in the rosiglitazone group for people both with and without previous ischaemic heart disease, other reasons for cardiovascular hospitalisation or death showed no evidence of such excess. Specifically, the number of cases with heart failure with and without previous ischaemic heart disease were 17 (rosiglitazone) versus 8 (control) and 44 (rosiglitazone) versus 21 (control), giving similar relative risks (2·16 and 2·10) in these two subgroups. For people allocated to rosiglitazone, the presence of baseline stable angina or previous myocardial infarction carried no increased risk for a primary endpoint event (rosiglitazone 59 vs control 56 and 37 vs 35 events).
Table 6 summarises the rate of investigator-reported serious adverse events in more than 20 people or of events of particular interest in the context of PPAR agonists. Heart failure was significantly higher in people treated with rosiglitazone than in those treated with the active control. However, myocardial infarction did not differ between groups, whereas stroke might have been less frequent in people treated with rosiglitazone than in those treated with the active control (table 6). Overall incidence of malignancy or of some tumour types including bladder and colon cancer did not differ in the two groups, but pancreatic cancer was less frequently reported in the rosiglitazone group than in the active control group (table 6).
Serious adverse events due to hyperglycaemia were less frequent, and those due to hypoglycaemia more frequent, in the rosiglitazone group than in the control group (table 6). The incidence of all adverse events due to hypoglycaemia (serious and non-serious) was higher in the sulfonylurea-containing groups (sulfonylurea added to metformin [197 patients, 18%], metformin added to sulfonylurea [148, 13%], and rosiglitazone added to sulfonylurea [175, 16%]) than when rosiglitazone was added to metformin (57 patients, 5%). There were no serious adverse-event reports of macular oedema. Non-serious adverse-event reports of macular oedema in people on rosiglitazone were seven, and three in people on the active control.
The overall incidence of participant-reported bone fractures was higher in the rosiglitazone group than in the active control group (risk ratio [RR] 1·57, 95% CI 1·26-1·97, p<0·0001) (table 7). The RR seemed to be higher in women than in men (1·82, 1·37-2·41 vs 1·23, 0·85-1·77; interaction p=0·10). The excess of fractures in people on rosiglitazone was mainly upper limb (RR 1·57, 1·12-2·19, p=0·0095) and distal lower limb (RR 2·60, 1·67-4·04, p<0·0001), and was mainly in women (RR 1·75 upper limb and 2·93 distal lower limb). Hip and femur fracture did not increase with rosiglitazone treatment; a non-significant excess in small numbers of spinal fractures was noted.
RECORD was a prospective, multicentre, randomised, open-label trial of dual therapy in patients with type 2 diabetes, comparing rosiglitazone plus either metformin or a sulfonylurea with an active control, metformin plus a sulfonylurea. The study design has been described in detail previously.13, 18 Patients with type 2 diabetes on monotherapy with either metformin or sulfonylurea and in less than optimal blood glucose control (haemoglobin A1c [HbA1c] >7·0-9·0%) were randomly assigned to addition of rosiglitazone or metformin (if already on sulfonylurea) or of rosiglitazone or sulfonylurea (if already on metformin). Data for glucose control and ambulatory blood pressure have already been published,2, 9 as was an interim analysis forced by controversy over cardiovascular risk.18
The study was done in 364 centres in 25 countries in Europe and Australasia. Randomisation was by telephone from a dedicated centre, using random-permuted blocks stratified by background medication. The study was open label because of planned differences in the strategy for rescue therapy and the need to allow different types and doses of comparator sulfonylurea therapy. Choice of sulfonylurea (glimepiride, gliclazide, or glibenclamide [glyburide]) was according to local investigator practice. Other glucose-lowering therapies were not permitted.
The study was monitored by a clinical trials organisation (Quintiles, Bracknell, UK), which also coordinated data collection. Biochemical measurements were done by a central laboratory (Quest Diagnostics, Heston, UK). The study was overseen by a steering committee. An independent data safety and monitoring board reviewed conduct of the study and unblinded data at about 6-month intervals. Interim analyses were not made available to the steering committee, except for those required for the interim publication.18 Adverse-event data were made available to the pharmacovigilance department of the sponsor, but not to staff involved with the study.
Eligibility criteria have been detailed previously.13 Requirements were: age between 40 and 75 years, body-mass index of more than 25·0 kg/m2, and being on maximum tolerated doses of metformin or a sulfonylurea monotherapy. Exclusions included hospitalisation for a major cardiovascular event in the 3 months before the trial, planned cardiovascular intervention, and presence, history, or treatment for heart failure. Written informed consent was obtained from each participant.13 Recruitment was from April, 2001, to April, 2003, and last study visits were from August to December, 2008.
Oral glucose-lowering therapies were managed throughout to a target HbA1c of 7·0% or less. Rosiglitazone (Avandia, GlaxoSmithKline, UK) was begun at 4 mg per day, and titrated to 8 mg any time after 8 weeks of therapy, if not to target. The starting dose of metformin and sulfonylurea varied by local practice, with dose increases permitted from 8 weeks after beginning of therapy. Maximum daily doses of 2550 mg metformin, 15 mg glibenclamide (or equivalent for different preparations), 240 mg gliclazide, or 4 mg glimepiride were stipulated.20 The criterion for rescue therapy by addition of a third oral agent (if in the rosiglitazone group) or transfer to insulin (in the metformin plus sulfonylurea group) was a confirmed HbA1c of 8·5% or more. Subsequently, if participants taking triple therapy (with rosiglitazone) had a confirmed HbA1c of 8·5% or more, rosiglitazone was to be stopped and insulin therapy substituted.
Outcome measures and adverse events
The primary outcome was time to first occurrence of cardiovascular hospitalisation or cardiovascular death. The primary hypothesis to be tested was non-inferiority of rosiglitazone compared with active control with a prespecified non-inferiority margin of 1·20 for the hazard ratio (HR). Therefore, if the upper limit of the CI of the HR was less than 1·20, non-inferiority could be claimed.
Some sensitivity analyses on the primary endpoint were predefined, including one with non-atherosclerotic events removed (ie, pacemaker insertion, cardiac conducting tissue ablation, bradycardia, deep venous thrombosis, pericarditis, cardiac surgery complications, pulmonary embolism, syncope, valve surgery or valvular heart disease, hypertensive emergency, and subdural haematoma).
Deaths and investigator-diagnosed cardiovascular events were identified through adverse-event reporting, direct questioning, or both, at study visits with trial record forms. Data from all relevant clinical sources were obtained by Quintiles and provided to an independent clinical endpoints committee who were blind to treatment allocation. All deaths were adjudicated with predefined criteria. Cardiovascular deaths included those with a definite cardiovascular cause and those of unknown cause. Cardiovascular hospitalisations included any acute or unplanned admission for a cardiovascular reason-such as heart failure, transient ischaemic attack, thrombotic events, unplanned cardiovascular revascularisations, and amputation of extremities-in addition to traditional cardiovascular emergencies, such as acute myocardial infarction or stroke. In a sensitivity analysis, cardiovascular deaths adjudicated as of unknown cause were reattributed according to investigator opinion.
Adverse events and serious adverse events were obtained for all participants while on dual or triple oral therapy, and serious adverse events thereafter. After publication of the ADOPT study,19 fracture events were also obtained by direct questioning at study visits.
For the non-inferiority hypothesis, 4000 participants followed for a median time of 6 years were needed to give 99% power, provided that the active control group had an 11% event rate per year, allowing 2% annual loss to follow-up. Blinded overall event tracking showed the event rate during the study was well below this rate. Therefore, endpoint sweeps were implemented to identify any missed events. An in-depth review of a sample of individual records showed very few missed events.
The hypothesis of non-inferiority for the primary endpoint was tested on the randomised and treated population, using analysis by intention to treat. A sensitivity per-protocol analysis was restricted to each participant's time on dual combination therapy plus 30 days thereafter. Time from randomisation to first defined event was calculated for each relevant event. The comparison between the rosiglitazone group and the control group was estimated as an HR (with 95% CI) on the basis of Cox proportional hazards regression stratified for background medication. Two-sided p values were derived from the asymptotic Wald tests, unadjusted for multiple testing. Cumulative incidence was estimated with the Kaplan-Meier method. Event rates were calculated according to individual time in study by randomisation group, without censoring for rescue therapy, and enabled calculation of absolute risk differences.
Analyses of changes in HbA1c and other quantitative measures at 5 years used a repeated measures model (based on all available data and assuming data were missing at random), including terms for baseline and baseline by visit interaction, with an unstructured covariance matrix for the within-patient variability in each treatment group. Incidence of adverse-event data was compared using a stratified Fisher's exact test.
The statistical analysis was done by the sponsor's statisticians according to a detailed predefined statistical analysis plan agreed by the steering committee, with SAS software version 8.2 (Cary, NC, USA). Confirmatory analyses were done by an independent investigator (London School of Hygiene and Tropical Medicine, UK) with SAS software. This study is registered with ClinicalTrials.gov, number NCT00379769.
Role of the funding source
The sponsor of the study was GlaxoSmithKline (Brentford, UK). Sponsor statisticians were involved in the design, reporting plan, and data analysis. The steering committee had responsibility for study conduct, data collected, data analysis, the writing of this report, and the decision to publish.