Misleading meta-analysis: a need to look beyond the headlines

Analysis of decisions made in meta-analyses of depression screening and the risk of confirmation bias: a case study

BACKGROUND

Depression is common in primary care and clinicians are encouraged to screen their patients. Meta-analyses have evaluated the effectiveness of screening, but two author groups consistently reached completely opposite conclusions.

METHODS

We identified five systematic reviews on depression screening conducted between 2001 and 2009, three by Gilbody and colleagues and two by the United States Preventive Task Force. The two author groups consistently reached completely opposite conclusions. We analyzed two contemporaneous systematic reviews, applying a stepwise approach to unravel their methods. Decision points were identified, and discrepancies between systematic reviews authors' justification of choices made were recorded.

RESULTS

Two systematic reviews each addressing three research questions included 26 randomized controlled trials with different combinations in each review. For the outcome depression screening resulting in treatment, both reviews undertook meta-analyses of imperfectly overlapping studies. Two in particular, pooled each by only one of the reviews, influenced the recommendations in opposite directions. Justification for inclusion or exclusion of studies was obtuse.

CONCLUSION

Systematic reviews may be less objective than assumed. Based on this analysis of two meta-analyses we hypothesise that strongly held prior beliefs (confirmation bias) may have influenced inclusion and exclusion criteria of studies, and their interpretation. Authors should be required to declare a priori any strongly held prior beliefs within their hypotheses, before embarking on systematic reviews.

Cardiovascular risk and thiazolidinediones--what do meta-analyses really tell us?

Meta-analyses of clinical trials suggest that the use of the thiazolidinedione (TZD), rosiglitazone, in patients with type 2 diabetes mellitus may increase the risk of myocardial ischaemic events by 30-40%. Although these controversial data must be interpreted with caution, in the absence of definitive prospective cardiovascular (CV) outcomes data, they represent a prominent source of evidence concerning the CV safety of rosiglitazone. The results of meta-analyses and a large randomized-controlled CV outcomes trial provide strong evidence that pioglitazone does not increase the risk of coronary events. This article clarifies the clinical significance of these meta-analytical findings alongside other sources of evidence and assesses their impact on evolving treatment guidelines and recommendations for the use of TZDs in patients with type 2 diabetes.

Proteinuria in diabetes: bystander or pathway to cardiorenal disease?

The development of albuminuria in diabetics is closely associated with an enhanced risk of renal and cardiovascular disease. However, the role of albuminuria in the pathogenesis of these clinical conditions remains controversial. Whether albuminuria is simply a biomarker or qualifies as a surrogate endpoint for cardiorenal disease has wide-ranging implications from the monitoring and treatment of patients to the design of clinical trials and drug development. We critically review available data to determine whether the association between albuminuria and cardiorenal disease is causative. Current evidence suggests the significance of albuminuria depends on its severity (degree or level) and on the specific clinical outcome under consideration. For diabetic kidney disease, there is convincing epidemiologic and experimental evidence to assign clinical albuminuria status as a surrogate endpoint, but for lower levels of albuminuria (microalbuminuria and normoalbuminuria), the evidence is inconclusive or not available. Albuminuria of any degree is unlikely to be causally related to diabetic cardiovascular disease, but its onset might be useful to identify those subjects at cardiovascular risk and to detect and treat other modifiable risk factors.

Pioglitazone and the risk of cardiovascular events in patients with Type 2 diabetes receiving concomitant treatment with nitrates, renin-angiotensin system blockers, or insulin: results from the PROactive study (PROactive 20)

BACKGROUND

Patients with Type 2 diabetes mellitus (T2DM) are often treated with multiple glucose-lowering and cardiovascular agents. The concomitant use of nitrates, renin-angiotensin system (RAS) blockers, or insulin has been linked to a potential increase in myocardial ischemic risk with rosiglitazone. The PROactive database provides an opportunity to investigate the effects of these medications on the potential macrovascular benefits reported with pioglitazone.

METHODS

The PROactive study was a randomized double-blind prospective trial that evaluated mortality and cardiovascular morbidity in 5238 patients with T2DM and macrovascular disease. Patients received pioglitazone or placebo in addition to their baseline glucose-lowering and cardiovascular medications. The effect of pioglitazone on composite endpoints was evaluated, including all-cause death, myocardial infarction (MI), and stroke, as well as safety events of edema and serious heart failure, in subgroups using nitrates, RAS blockers, or insulin at baseline.

RESULTS

The risk of all-cause death, MI, and stroke for pioglitazone versus placebo was similar regardless of the baseline use of nitrates, RAS blockers, or insulin, with hazard ratios ranging from 0.81 to 0.87. Similar results were obtained for the other composite endpoints analyzed. There were no significant interactions between baseline medication subgroups and treatment. The increased risk of edema and serious heart failure was consistent across the baseline medication subgroups.

CONCLUSIONS

This post hoc analysis did not reveal an increased risk of macrovascular events with pioglitazone in patients receiving nitrates, RAS blockers, or insulin. Rather, all patients realized the same trend towards benefit with pioglitazone, and adverse events of edema and heart failure were predictable.