(IPD) and how it differs from aggregate data
Illustrative example of 10 randomised trials examining the effect of anti‐hypertensive treatment
(a) IPD
The following table shows hypothetical IPD collected, checked and harmonised from 10 randomised trials examining the effect of anti‐hypertensive treatment versus control in participants with hypertension.
Each row provides the information for each participant in each trial, and each column provides participant‐level information such as baseline characteristics and outcome values.
Only a subset of the IPD is shown for brevity, as in reality many more rows and columns will be needed for each trial, to include all available participants and variables.Trial IDParticipant IDTreatment group,1 = treatment0 = controlAge(years)SBP before treatment(mmHg)SBP at 1 year(mmHg)1114613711112135143133(other rows for trial 1 omitted for brevity)114540622092192105517015522138144139(other rows for trial 2 omitted for brevity)2337144153129(rows for trials 3 to 9 omitted for brevity)101071149128102159168169(other rows for trial 10 omitted for brevity)104695063174128
This IPD can be used to produce aggregate data for each trial, as shown in the table on the following page.(b) Aggregate data
Now each row corresponds to a particular trial, and each column is a trial‐level variable containing aggregated data values such as the total number of particulars and the mean age in each group.Trial IDNumber of participantsMean age(years)Mean SBP before treatment(mmHg)Mean SBP at 1 year(mmHg)Treatment effect on SBP at 1 year adjusted for baseline(treatment minus control)ControlTreatmentControlTreatmentControlTreatmentControlTreatmentEstimate (variance)175070442.3642.17153.05153.88139.75132.54–6.53 (0.75)219913869.5769.71191.55188.30179.89164.67–13.81 (4.95)(rows for trials 3 to 9 omitted for brevity)102297239870.2170.26173.94173.75165.24154.87–10.26 (0.20)
Source: Richard Riley.
Figure 1.1 Number of published IPD meta‐analysis articles over time, based on a crude search* in PubMed
Source: Richard Riley.* from searching for the following keywords in the Title or Abstract of the article: (meta-analysis AND individual patient data) OR (meta-analysis AND individual participant data) OR (meta-analysis AND IPD).
IPD meta‐analysis projects require a multi‐disciplinary research team, including clinicians and healthcare professionals, statisticians, evidence synthesis experts, search and information specialists, database managers, trialists, and patient and public advisory groups, amongst others. Therefore, this book is aimed at a broad audience, and guides the reader through the journey from initiating and planning IPD projects to obtaining, checking, and meta‐analysing IPD, and appraising and reporting findings. Very little prior knowledge is required. We assume readers are aware of the importance of systematic reviews and meta‐analysis in general, and are reading this book to help guide their decisions as to whether to take the IPD approach; to learn what an IPD project entails (from start to finish); and to understand appropriate methodology and best practice, for example to inform protocols, data retrieval plans, statistical analyses, bias assessments, reporting standards, and critical appraisal.
Our book is split into five parts. Parts 1 to 3 focus on the synthesis of IPD from randomised trials to examine treatment effects. Parts 4 and 5 branch out to cover special topics and applications, including diagnosis, prognosis and prediction. Part 1 includes chapters 2 to 4, and covers practical guidance for initiating, planning and conducting IPD meta‐analysis projects. Part 2 includes chapters 5 to 8, and covers fundamental two‐stage and one‐stage statistical methods for conducting an IPD meta‐analysis of randomised trials to examine a treatment effect. These chapters are more technical than others, but should still be broadly accessible, as recommendations and illustrated examples are given throughout to reinforce the key messages. Part 3 includes Chapters 9 to 11, and focuses on the critical appraisal and dissemination of IPD projects. Part 4 includes Chapters 12 to 14, and covers special topics in statistics, including calculating power (in advance of IPD collection) and analysing multiple outcomes and multiple treatments. Part 5 concludes with Chapters 15 to 18, which broaden application of IPD projects to the evaluation of diagnostic tests, prognostic factors, and clinical prediction models.
This book is the first to be devoted entirely to IPD meta‐analysis projects, and complements other textbooks on systematic reviews and meta‐analysis that focus mainly on the aggregate data approach, such as the following.1,34–37 A general statistical textbook would also provide complementary reading to Part 2 of this book.38–41 Relevant methods for IPD meta‐analysis of prognosis studies are introduced in Prognosis Research in Healthcare: Concepts, Methods and Impact,32 and Part 5 builds extensively on this work. Detailed information is also available on our companion website for this book: www.ipdma.co.uk. Introductory videos are included, alongside links to relevant publications, talks, training courses, and workshops. Statistical code is also provided for educational purposes, so that readers can replicate various examples given throughout the book and reinforce their learning.
2 Rationale for Embarking on an IPD Meta‐Analysis Project
Jayne F. Tierney, Richard D. Riley, Catrin Tudur Smith, Mike Clarke, and Lesley A. Stewart
Summary Points
Many of the principles, methods and processes of IPD meta‐analysis projects are similar to those of a conventional systematic review and meta‐analysis of aggregate data. The most substantial differences relate to the collection, checking and analysis of data at the participant level, and collaboration with the investigators responsible for the existing trials.
Compared to using aggregate data, IPD projects can potentially provide substantial improvements to the extent and quality of data available, and give greater scope and flexibility in the analyses, for example to examine participant‐level associations.
Important differences can occur between IPD and aggregate data meta‐analysis results. This depends on many aspects including the availability of IPD, whether IPD leads to improvements in the completeness and quantity of information,