Evaluating, adapting and extending individual patient data (IPD) methods of evidence synthesis

Health and healthcare decisions should be informed by robust research evidence of the highest possible standard.

To reach reliable conclusions about the benefits and risks of a particular treatment, it is important to consider results from all relevant clinical trials and to use objective and transparent methods to summarise findings across trials. This is called systematic review. Systematic reviews that collect and re-analyse original ?raw? research data are called IPD (individual patient data) meta-analyses, and are considered to be a gold standard approach. These projects allow more detailed analysis than can be done in systematic reviews that rely on summary data extracted from published trial reports.

The methods used in IPD meta-analysis were developed around projects in cancer and heart disease and usually involve relatively straightforward methods of analysing and combining data. As IPD meta-analyses become used in new clinical areas, and as we need more sophisticated evaluation to guide complex decisions, we need to evaluate, extend and adapt these methods.

We will use (anonymous) data from 32,217 pregnant women enrolled in 31 randomised trials (that we previously collected in an international project exploring whether antiplatelet agents such as aspirin can prevent pre-eclampsia) to explore how we can improve the methods that researchers use in future. We will explore and develop both practical and statistical methods. These will relate to:

(1) issues specific to data from pregnancy and childbirth (e.g. how to best analyse data from twin or other multiple pregnancies)
(2) issues that are general to IPD meta-analysis (e.g. most appropriate statistical approaches to analysis, including how best to find out whether particular types of people do particularly well or badly from a treatment, and how to deal with situations where trials have data missing)
(3) issues relating all types of systematic review (e.g. trying to find out why the first trials of antiplatelet agents for prevention of pre-eclampsia reported the most beneficial results)

We will also try to find out which aspects of analysis are likely to change the findings of a systematic review and which are unlikely to change findings, so that we know where to focus efforts if time and resource are limited.

Findings of this project will be published and shared with the Cochrane Collaboration methods groups feeding into the guidance and advice provided by them. Ultimately, findings should help ensure that healthcare decisions are underpinned by the most reliable evidence synthesis methods.

Evidence synthesis underpinning health and healthcare decisions should be robust and of the highest possible standard. Individual patient data (IPD) meta-analyses which centrally collect, check, re-analyse and pool the original ?raw? data from relevant studies, are considered to be a gold standard of systematic review. Current approaches to IPD meta-analyses are largely those developed in cancer and cardiovascular disease and usually involve relatively straightforward methods of synthesis. As the IPD approach moves to new areas of healthcare, and we require more sophisticated evaluation to guide complex decisions, there is a need to evaluate, extend and adapt the methodology of IPD meta-analysis.

We will utilise previously collected data from 32,217 pregnant women enrolled in 31 randomised trials of pre-eclampsia primary prevention, to explore and develop the methods used in both IPD and aggregate data systematic reviews. We will explore and develop both practical and statistical methods relating to:

? issues specific to synthesis of data from pregnancy and childbirth research (linkage between mothers and babies and dealing with multiple pregnancies)
? issues that are generally applicable to the conduct of IPD meta-analysis (one or two stage analysis, fixed or random effects, subgroup or modelling based approach to exploring potential effect modifiers/treatment interactions, dealing with sparse and missing data)
? issues relating to all types of systematic review (selective reporting of outcomes, the utility of meta-regression).

Distinguishing between aspects of analysis that materially alter the results of systematic review from those that, whilst theoretically/ statistically valid, are unlikely to impact importantly on results, may also guide where best to focus efforts where time and resource are limited.

In addition to journal publications and conference presentations, findings will be shared with the Cochrane Collaboration IPD Meta-analysis Methods Group and Statistical Methods Group and will feed into the guidance and advice provided by these groups.

Given the likely increased need for, and potential of IPD meta-analysis to influence practice and future research, the proposed work could have far-reaching impact. Ultimately, findings should help to ensure that healthcare decisions are underpinned by the most reliable evidence synthesis methods.