Pharmacologic reduction of angiographic vasospasm in experimental subarachnoid hemorrhage: systematic review and meta-analysis

Subarachnoid hemorrhage (SAH) is an important cause of morbidity and mortality in adult life, and as in so many areas of neuroscience we do not—yet—have much in the way of effective treatments. Subarachnoid hemorrhage is particularly interesting because in many cases neuronal injury occurs after hospital admission, due either to delayed ischemic neurological deficits or to rebleeding; this raises the possibility of initiating treatment before damage occurs, and therefore the opportunity to demonstrating proof of concept for neuroprotection.

The work of Zoerle et al (2012, this issue) on interventions to prevent vasospasm in experimental SAH is important because it seeks to understand how best to use animal data to inform human health. For seven interventions where efficacy in humans is known from systematic reviews (three effective, four not) they find that, taking all the animal data together, all drugs are effective. The sensitivity (100%) and specificity (0%) is similar to previous studies (Perel et al, 2007). In all, 55% of animal studies were concordant with the human data, not very much higher than would be expected by chance. However, they suggest that where outcome is measured beyond 3 days then concordance is higher; this interesting hypothesis requires further testing.

This work is also interesting because it broadens the application of meta-analysis in animal models of neurological disease, where translation is most challenging. This technique has now been used for stroke (Nava-Ocampo et al, 2000; Horn et al, 2001), intracerebral hemorrhage (Frantzias et al, 2011), multiple sclerosis (Vesterinen et al, 2010), Parkinson's disease (Rooke et al, 2011), glioma (Amarasingh et al, 2009), and spinal cord injury (Fiore et al, 2012).

The current work illustrates some emerging themes. First, not all experiments have the most appropriate experimental design, and specifically many do not report simple measures such as randomization or blinding. Second, sample sizes are generally small, and this raises at least the suspicion that in some hands sample size expands until statistical significance is achieved. Third, publication bias is an issue here as elsewhere (Sena et al, 2010), and we need to find effective ways to counter this. Together, these factors argue for a more robust approach to animal modeling with larger, better designed experiments; this is almost certain to require multicenter animal studies (Bath et al, 2009).

This work also illustrates some methodological issues. The statistical tools for meta-analysis were developed to summarize data from large clinical studies where a single summary estimate is meaningful. In contrast, animal studies describe efficacy across a range of circumstances and are usually orders of magnitude smaller. The statistical approaches to meta-analysis perform differently at small sample sizes, and the exploration of sources of heterogeneity in animal studies (testing the significance of observed differences between different groups of experiments) provides a further challenge; with the relative merits of partitioning of heterogeneity and meta-regression being, as yet, unclear. We are only now beginning to understand these methodological challenges, but their solutions will support the application of this important tool, and will help provide evidence to support successful translation.

The author declares no conflict of interest.