The days of a meta-analysis being the simple adding up of lots of studies, pretending that they are all just tiny pieces of the One Big Trial that was performed before the world was made are on their way out. Newer ways of using synthesised evidence – like meta-regression and individual patient data analysis – are coming up quickly.
The problem with ‘traditional’ meta-analysis is that you’re limited by what is published. If the studies publish that kids with asthma on steroids had 10% admission rate, and on placebo they had a 25% rate, then all your traditional review tells you is that steroids work – but nothing about which of the does, routes or exact type is best. What newer techniques can do is try to look at questions like ‘is it better in children with milder or more severe disease’ or ‘does the dose/intensity’ of treatment matter? Meta-regression is a technique that uses some of these variables (‘co-variates’) and looks at them in comparison with outcomes, and tries to estimate the strength of that association. Note – this is purely observational, even with RCT data – so it’s subject to biases like other observational studies (such as the trial with more severe children also having better high dependency care, and ‘loosing’ the real improvement with more severe disease, for example). Its strength is that is can be done with just the published data. If instead of just the published, summary data, you can get your hands on more detailed stuff – ideally down to the individual patient data (IPD) level – then you can ask these sorts of questions without struggling with these biases. Although it’s nice to fine tune our understanding of treatment, where these techniques are making a huge difference is in the analysis of diagnostic and prognostic information, allowing us to produce far more accurate measures from the usually tiny studies undertaken. These studies are already here and producing real results: IPD analysis of valproate and carbamazapine trials for epilepsy management shows what single RCTs have failed to do – that valproate is better for generalised epilepsies and carbamazapine is better for partial onset seizures. In glue ear management, individual RCTs and systematic reviews showed virtually no benefit for prompt insertion of grommits. The use of IPD data allows the interaction between surgery and day-care attendance to emerge (presumably proxy for high likelihood of recurrent infection), and demonstrates that if ventilation tubes are of any benefit in improving hearing, it is in the young who go to nursery. IPD analysis and meta-regression are extracting greater depth and quality from existing study data – surely, this, rather than more studies in new children, is the best way to advance paediatric health care !
New statistical method for analyzing time to first seizure: example using data comparing carbamazepine and valproate monotherapy. Epilepsia. 2007 Jun;48(6):1173-8
Grommets in otitis media with effusion: an individual patient data meta-analysis. Arch Dis Child. 2005 May;90(5):480-5. Review.