Individual Participant Data Meta-Analysis. Группа авторов
compare with those done by original trial investigators.
Given the additional resource requirements, it is important to consider carefully whether an IPD project is needed instead of a conventional systematic review using aggregate data. The decision will depend on the particular research question, and whether IPD would produce a more reliable and comprehensive answer than using the aggregate data already available for eligible trials.
2.1 Introduction
In this chapter, we overview those elements of an IPD meta‐analysis project that differ from a conventional meta‐analysis of aggregate data (Section 2.2), describe the advantages (Section 2.3) and challenges of the IPD approach (Section 2.4), and summarise empirical evidence comparing results of IPD and aggregate data meta‐analyses (Section 2.5). Although IPD projects almost always provide advantages, sometimes a standard aggregate data meta‐analysis may be sufficient to answer a particular research question. Hence, researchers should only embark on an IPD project after careful consideration, especially as it requires additional time, resources and skills. We provide guidance to help researchers decide when the use of IPD is likely to provide more robust conclusions than using available aggregate data alone (Section 2.6). We focus on the synthesis of evidence from randomised trials evaluating treatment effects, but most of what is presented also applies to other study types and to other types of research questions, such as those for diagnosis and prognosis (Part 5).
2.2 How Does the Research Process Differ for IPD and Aggregate Data Meta‐Analysis Projects?
IPD meta‐analysis projects follow many of the same principles and research processes as conventional systematic reviews and meta‐analyses of aggregate data. However, there are also important differences, as now described.
2.2.1 The Research Aims
A first and fundamental step of all research projects is to define their aims. As for systematic reviews and meta‐analyses based on aggregate data, the aims of an IPD meta‐analysis project should be defined in relation to key components such as the participants, interventions, comparators or controls, outcomes and study designs of interest, aided by a framework such as PICOS (an example is given in Section 3.3).42 Most reviews based on aggregate data focus on summarising the overall treatment effect, and often IPD meta‐analysis projects also have this objective. However, IPD additionally allows participant‐level information to be examined and analysed, and so most IPD projects are specifically set up to utilise this. In particular, they may aim to summarise treatment effects conditional on prognostic factors (Chapters 5 and 6); to assess whether the treatment effect varies according to participant‐level characteristics (Chapter 7), or to evaluate treatment effects at multiple time‐points during follow‐up (Chapter 13). Indeed, the potential research questions that can be addressed by an IPD project are broad, and a wide variety of applications are demonstrated throughout this book.
2.2.2 The Process and Methods
Figure 2.1 provides key differences in the process of conducting IPD meta‐analysis projects compared to conventional reviews based on aggregate data.7,43 Best practice is to publish and adhere to a protocol, regardless of whether aggregate data or IPD are being used, although protocols for IPD projects will usually be more detailed (Section 4.2.2). Methods for identifying trials are very similar in the two approaches, but in an IPD project searches may be conducted prior to or in tandem with protocol development, in order to generate a preliminary list of trials (Section 4.2.3), and to identify the associated investigators from whom IPD will be sought (Section 3.2).
Prior to data collection, an IPD project may require ethical approval (Section 3.10) and development of formal data‐sharing agreements (Section 3.11), as well as the preparation of a detailed data dictionary (Section 4.2.7). These are rarely required for an aggregate data review. Furthermore, the subsequent data collection, checking and analytical aspects of an IPD project are much more exacting than those for aggregate data reviews. They may include data entry, data re‐coding and harmonisation, together with checking, querying and subsequent validation of IPD with original trial investigators (Chapter 4),7,43,44 as well as advanced statistical methods for meta‐analysis (Part 2).
Unlike aggregate data reviews, IPD projects usually involve and benefit from establishing partnerships with trial investigators who, in addition to providing their IPD, play an active role throughout the process, from identifying relevant trials through to helping interpret and disseminate IPD meta‐analysis results. This may include establishing a collaborative group that authors the main project publication, with all those involved being listed as co‐authors, and holding a meeting of this Group where preliminary results are presented and discussed (Section 3.8).7,43 Recently, a range of clinical study data repositories and platforms have been established, offering another source of IPD from existing trials, but there are both advantages and disadvantages of obtaining for IPD meta‐analysis projects in this way (Sections 3.2.2 and 4.4.5).45
Figure 2.1 Key differences between the process for a IPD meta‐analysis project and a conventional systematic review and meta‐analysis of aggregate data.
Source: Jayne Tierney.
Given these differences, IPD meta‐analysis projects require a greater range of skills (Section 3.5), generally take longer (Section 3.7), and need more resources (Section 3.8) than traditional systematic reviews and meta‐analyses based on aggregate data.
2.3 What Are the Potential Advantages of an IPD Meta‐Analysis Project?
Provided it is conducted