15th June 2009
The authors of this non-inferiority study claim non-inferiority of dual therapy with rosiglitazone plus metformin or a sulfonylurea compared to dual therapy with metformin plus a sulfonylurea with regard to major CV outcomes. However, questions over the analysis limit the degree of confidence which can be placed in this conclusion. The study confirms that therapy including rosiglitazone substantially increases the risk of heart failure as well as increasing the risk of fractures, and overall provides additional support for a strategy of reviewing patients taking rosiglitazone and considering their hypoglycaemic medication in the light of recently updated NICE guidance and MHRA advice.
Level of evidence:
Level 2 (limited-quality patient-oriented evidence) according to the SORT criteria.
Health professionals should continue to follow MHRA advice. Rosiglitazone (and pioglitazone▼) should not be used in people with heart failure or history of heart failure. Rosiglitazone should be used in patients with previous or current ischaemic heart disease only after a careful evaluation of the individual patient’s risk.
What is the background to this?
Concerns have arisen about the cardiovascular (CV) safety of the ‘glitazones’. Heart failure is a recognised risk, and analyses have suggested that rosiglitazone might also be associated with a small increased risk of myocardial infarction (MI). These concerns led to international reviews of available data for the safety and efficacy of glitazones, including CV safety, resulting in changes to prescribing information and advice (see blog). Glitazones may also increase the risk of fractures.
This study (n=4,447) was a multicentre, open label, randomised trial to test the non-inferiority of rosiglitazone added to metformin or a sulfonylurea, compared to metformin plus a sulfonylurea. At entry to the study, patients had an HbA1c greater than 7% (53mmol/mol) on monotherapy with either metformin or a sulfonylurea. The primary endpoint was CV hospitalisation or CV death, with a hazard ratio (HR) non-inferiority margin of 1·20 (upper 95% confidence interval [CI] 1.20 or less). Non-inferiority study design and results are complex – see the “So what” section below.
What does this study claim?
During a mean 5.5 year follow up, and according to an intention to treat (ITT) analysis, the HR for the primary outcome was 0.99, 95% CI 0.85 to 1.16, P=0.93. The HRs were 0·84 (95%CI 0·59 to 1·18, P=0.32) for CV death, 1·14 (95%CI 0·80 to 1·63, P=0.47) for MI and 0·72 (95%CI 0·49 to 1·06, P=0.1) for stroke. Although there appears to be no statistically significant increase in the risk of these outcomes, because this was a non-inferiority study and because of some potentially important differences in other treatments used by participants in this study, these results need further discussion (see the “So what” section below).
The HR for heart failure causing admission to hospital or death was 2·10 (95%CI 1·35 to 3·27, P=0.001). In addition, the risk of participant-reported bone fractures was greater in the rosiglitazone group: relative risk [RR] 1·57, 95%CI 1·26 to 1·97, P<0·0001). The risk of both of these outcomes was statistically significantly higher for the rosiglitazone-based regimen.
Non-inferiority studies such as this need careful critical appraisal. For some readers, it may be enough to know that there are some question marks over the analysis here, and that this study does not seem to give any reason to depart from MHRA guidance. For those who wish to know more, we examine some of those issues below.
Unlike the more commonly encountered superiority studies, the design of non-inferiority studies involves setting an acceptable upper threshold for the confidence interval for the HR for the harm, before conducting the study. In this case the upper 95%CI was set at 1.20. In other words, the authors decided that dual therapy including rosiglitazone would be judged no more harmful than NICE’s recommended standard dual therapy of metformin plus a sulfonylurea with regard to this composite outcome, if there was less than a 2.5% probability that it did not increase the risk (strictly, the HR) of the composite outcome by 20% or more.
Setting this upper threshold is a matter of clinical judgement. To set this in context, in superiority trials the observed beneficial effect is often a relative risk reduction (RRR) of 20-30%. When evaluating the results of the RECORD study we must ask ourselves whether or not it is clinically acceptable if rosiglitazone were to increase the risk of the composite outcome by this much. This has to be considered in the context of the indications for which rosiglitazone is used, and the balance of this possible harm against the likely benefits to patients.
Secondly, we must think about the degree of confidence we want to have in the results. Setting the criterion for non-inferiority using 95%CI means that the authors could not exclude a 2.5% chance that the “true” HR is greater than 1.20. To set this in context, the probability of throwing a double 6 with two standard dice is 1 in 36 or around 2.8%. If a criterion using 99%CI had been set, if that criterion was satisfied there would have been only a 0.5% probability that the “true” HR is greater than 1.20. To set this in context, the probability of throwing three standard dice and them all coming up 6 is 1 in 216 or around 0.46%. These two decisions – the upper acceptable CI threshold and whether the CIs are 95%, 99% or any other value – affect the size of the study needed to achieve acceptable power.
In the ITT analysis, the upper 95% CI for the primary outcome was 1.16. This is less than 1.20, suggesting that non-inferiority was demonstrated (within the parameters discussed above). However, the EMEA states that non-inferiority studies should also be analysed on a per-protocol basis, and this is also discussed in a CONSORT statement. This is because an ITT analysis includes patients who did not actually take the treatment as intended, so any differences between the two groups may be blurred and the chance of falsely finding non-inferiority is increased (type 1 error). A per-protocol analysis uses data only from patients who satisfied the entry criteria, actually received treatment and properly followed the protocol. Non-inferiority can be established only if both analyses support the finding. A limited per-protocol analysis (which excluded any participant 30 days after transfer from dual therapy because of inadequate HbA1c control) found an HR of 1·02, with 95% CI of 0·85 to 1·21. Although similar to the ITT result, the upper 95%CI breaches the predetermined criterion for non-inferiority. Further appraisal of this is not possible given the data presented.
This study confirms the recognised increased risk of heart failure associated with rosiglitazone (and pioglitazone▼), and supports previous evidence about an increased risk of fractures (although the study was not set up to address this latter point). Regarding other cardiovascular risks with rosiglitazone, the accompanying editorial states “Rosiglitazone was associated with higher LDL cholesterol levels, leading to an increased use of statins in the rosiglitazone group [55.2% versus 46.0%, P not stated but authors say difference was significant], which might have reduced the incidence of cardiovascular events. Thus, although the RECORD study has confirmed known risks associated with rosiglitazone (including increased rates of heart failure, fractures, and hyperlipidaemia), uncertainty remains regarding the effect of rosiglitazone on cardiovascular disease”.
The prescribing of rosiglitazone in primary care in England has recently fallen substantially. From a peak of approximately 310,000 items in the quarter to June 2007, prescribing of rosiglitazone tablets nearly halved to approximately 170,000 items in the quarter to December 2008 (the last quarter for which we have data). The number of items of the combination product with metformin has also declined in that period from nearly 170,000 items to approximately 110,000. This study provides additional support for a strategy of reviewing patients taking rosiglitazone, considering their hypoglycaemic medication and making adjustments where appropriate in the light of recently updated NICE guideline.
More information on the management of people with type 2 diabetes can be found on the relevant floors of NPC. NICE has also recently partially updated its guidance on the management of type 2 diabetes (including guidance on the use of glitazones) in NICE clinical guideline 87 (see our blog).
Home PD et al. Rosiglitazone evaluated for cardiovascular outcomes in oral agent combination therapy for type 2 diabetes (RECORD): a multicentre, randomised, open-label trial Lancet 2009; Published Online June 5, 2009 DOI:10.1016/S0140-6736(09)60953-3
Design: Multicentre open label randomised controlled trial
Patients: 4447 patients with type 2 diabetes on monotherapy with either metformin or sulfonylurea and in less than optimal blood glucose control (HbA1c >7·0% to 9·0% [53mmol/mol to 75mmol/mol])
Intervention and comparison: rosiglitazone (initially 4mg/day, titrated to 8mg any time after 8 weeks of therapy if not to target), or metformin (if previously taking sulfonylurea) or sulfonylurea (if previously taking metformin). Maximum daily doses of 2550mg metformin, 15mg glibenclamide (or equivalent for different preparations), 240mg gliclazide, or 4mg glimepiride were stipulated. 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% (69mmol/mol) or more. Subsequently, if participants taking triple therapy (with rosiglitazone) had a confirmed HbA1c of 8·5% (69mmol/mol) or more rosiglitazone was to be stopped and insulin therapy substituted.
Outcomes and results: The primary outcome was time to first occurrence of CV hospitalisation or CV 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 HR. Therefore, if the upper limit of the CI of the HR was less than 1·20, non-inferiority could be claimed. During a mean 5.5 year follow up, and according to an intention to treat analysis, the HR for the primary outcome was 0.99, 95% CI 0.85 to 1.16, P=0.93. A limited per-protocol analysis (excluding any participant 30 days after transfer from dual therapy because of inadequate HbA1c control) found an HR of 1·02, with 95% CI of 0·85 to 1·21.
The HRs were 0·84 (95% CI 0·59 to 1·18, P=0.32) for CV death, 1·14 (95% CI 0·80 to 1·63, P=0.47) for MI and 0·72 (95% CI 0·49 to 1·06, P=0.1) for stroke. The HR for heart failure causing admission to hospital or death was 2·10 (95%CI 1·35 to 3·27, P=0.001). In addition, the risk of participant-reported bone fractures was greater in the rosiglitazone group: RR 1·57, 95% CI 1·26 to 1·97, P<0·0001).
Sponsorship: GlaxoSmithKline plc. Two of the authors are employees of, and hold stocks in, GlaxoSmithKline plc
Make sure you are signed up to NPC Email updates — the free email alerting system that keeps you up to date with the NPC news and outputs relevant to you.