Author: Dr Tom Moullaali, twitter: @tom_moullaali email: firstname.lastname@example.org
Affiliations: Centre for Clinical Brain Sciences, University of Edinburgh; George Institute for Global Health, Sydney.
Dr Xia Wang is a biostatistician and senior research fellow at the George Institute for Global Health, Australia. I’ve had the pleasure of working with Xia since 2014 when she supervised my elective research project at the George Institute. Since then, we’ve collaborated on several research papers, including two analyses of pooled patient-level data on the management of blood pressure after acute intracerebral haemorrhage.
What I’d do for a flat white and a proper catch up in the Sydney sunshine…
Instead, I asked Xia for her insights about the analysis of pooled patient-level data and she provided responses by email due to time differences. I hope the audience find these helpful; we welcome any questions through Twitter or email.
- What are the benefits of pooling patient-level data to answer important questions in stroke research?
You get more statistical power. For interventions with small effect sizes, many thousands of patients are needed to demonstrate an effect of treatment. This might not be possible to deliver with a single trial. Therefore, pooling data from several trials can harness larger sample sizes and detect effects that may have been missed by underpowered single studies. You can also assess whether or not the effects of the interventions vary in patient subgroups.
A good example of this is the meta-analysis of patient-level data from randomised controlled trials that tested the effects of intravenous thrombolysis for acute ischaemic stroke: the study was able to address uncertainties about the influence of the timing of treatment, and its efficacy in older patients and across different stroke severities.1
- What are the challenges of acquiring patient-level data and how can you overcome them?
Meta-analyses of patient-level data require collaboration. Prospective collaborations such as the Blood pressure in Acute Stroke Collaboration (BASC)2 and several others3,4 aim to involve investigators leading trials in their respective areas. There are political and legal barriers to sharing data. In our experience, these can be overcome with transparency and inclusiveness. This involves publishing a protocol with detailed information about how you are going to acquire, manage and analyse the data, and the publication strategy. Data transfer agreements provide legal assurance that data will be used according to the protocol. Authorship should be inclusive (for example, by forming a writing committee) and investigators who share data but do not meet international authorship criteria should be acknowledged as collaborators.
- What are the statistical pitfalls of analysing patient-level data
First, a great deal of care is needed to harmonise heterogeneous datasets. Again, a prospective plan helps here. I recommend prospectively agreeing nomenclature for all key variables in the database before data are transferred. Sometimes investigators sharing data do not have the time to harmonise their datasets according to the specifications; be prepared to do the hard work and check inconsistencies with the individual investigators.
In our experience analysing pooled data from INTERACT2 and ATACH-II,5 complex pre-specified analyses that involved multiple imputation of missing data were challenging to undertake and present to the audience. One essential consideration is that the data are likely to come from heterogeneous populations (baseline risks of the outcome may vary from one study to another) who were treated slightly (or very) differently (treatment effects may vary by study). The statistical models you use should account for these differences, also known as ‘clustering’. There are several ways to do this, and I would recommend seeking support from a statistician with experience of analysing pooled patient-level data.
1 Emberson J, Lees KR, Lyden P, et al. Effect of treatment delay, age, and stroke severity on the effects of intravenous thrombolysis with alteplase for acute ischaemic stroke: A meta-analysis of individual patient data from randomised trials. Lancet 2014; 384: 1929–35.
2 Bath FJ, Bath PMW. What is the correct management of blood pressure in acute stroke? The blood pressure in acute stroke collaboration. Cerebrovasc Dis 1997; 7: 205–13.
3 Goyal M, Menon BK, Van Zwam WH, et al. Endovascular thrombectomy after large-vessel ischaemic stroke: a meta-analysis of individual patient data from five randomised trials. Lancet 2016; 387: 1723–31.
4 Graham C, Lewis S, Forbes J, et al. The FOCUS, AFFINITY and EFFECTS trials studying the effect(s) of fluoxetine in patients with a recent stroke: Statistical and health economic analysis plan for the trials and for the individual patient data meta-analysis. Trials 2017; 18: 627.
5 Moullaali TJ, Wang X, Martin RH, et al. Blood pressure control and clinical outcomes in acute intracerebral haemorrhage: a preplanned pooled analysis of individual participant data. Lancet Neurol 2019; 18: 857–64.
Click here to add your own text