An estimate of analytical bias in meta-analysis

Abstract

Objective: To systematically assess the benefits and harms of intravenous thrombolysis for patients with acute ischaemic stroke and to use this analysis to explore the influence of analytical bias on meta-analysis outcomes. Methods: After first cataloguing published concerns regarding the evidence supporting the use of intravenous thrombolysis in stroke, we conducted a systematic review and meta-analysis of randomised clinical studies of intravenous thrombolysis compared with control in patients with presumed acute ischaemic stroke. Analyses were then repeated with various limitations in order to explore the potential effect of analytical bias. Results: 26 studies that randomised 10,431 participants were included. In the primary analysis, the use of thrombolysis was associated with an increased odds of good functional outcome, OR 1.14 (95% CI 1.04-1.25, p=0.004), but also an increased risk of intracranial haemorrhage, OR 4.28 (95% CI 3.34-5.48, p<0.0005) and an increased risk of early mortality, OR 1.51 (95% CI 1.27-1.78, p<0.0005). In the secondary analysis, the odds of improved outcome associated with thrombolysis varied depending on the definition of ‘good outcome’ used, OR 1.45 (95% CI 1.23-1.70) to 0.99 (95% CI 0.89-1.11). Restricting analysis to rt-PA increased the odds of good functional outcome (1.20 (95%CI 1.08-1.33), while analysis of Desmoteplase in isolation did not suggest a significant benefit (OR 1.09, 95% CI 0.75-1.59). The use of data collected ‘per-protocol’ increased the odds of death, OR 1.88 (95% CI 1.36-2.59), but reduced the odds of intra-cerebral haemorrhage, OR 1.75 (95% CI 1.29-2.37). Conclusions: The primary analysis demonstrated evidence of increased early mortality and symptomatic intracranial haemorrhage but also of improved functional outcomes for patients treated with thrombolysis. Meta-analysis results were shown to be sensitive to potential analytical bias, reflecting the need for prospective and transparent meta-analysis design.