Individual Participant Data Meta-Analysis: A Handbook for Healthcare Research

Content overview

Split into five parts (containing 18 chapters & over 500 pages), the book takes the reader through the journey from initiating and planning IPD projects to obtaining, checking, and meta-analysing IPD, and appraising and reporting findings.

The book initially focuses on the synthesis of IPD from randomised trials to evaluate treatment effects, including the evaluation of participant-level effect modifiers (treatment-covariate interactions). Detailed extension is then made to specialist topics such as diagnostic test accuracy, prognostic factors, risk prediction models, and advanced statistical topics such as multivariate and network meta-analysis, power calculations, and missing data.

Available to purchase from about £60.00 at Amazon & Wiley

Table of contents

Acknowledgements xxiii

1 Individual Participant Data Meta-analysis for Healthcare Research 1

Richard D. Riley, Lesley A. Stewart, and Jayne F. Tierney

1.1 Introduction 1

1.2 What Is IPD and How Does It Differ from Aggregate Data? 1

1.3 IPD Meta-analysis: A New Era for Evidence Synthesis 2

1.4 Scope of This Book and Intended Audience 2

Part I Rationale, Planning, and Conduct 7

2 Rationale for Embarking on an IPD Meta-analysis Project 9

Jayne F. Tierney, Richard D. Riley, Catrin Tudur Smith, Mike Clarke, and Lesley A. Stewart

2.1 Introduction 9

2.2 How Does the Research Process Differ for IPD and Aggregate Data Meta-analysis Projects? 10

2.2.1 The Research Aims 10

2.2.2 The process and methods 10

2.3 What Are the Potential Advantages of an IPD Meta-analysis Project? 11

2.4 What Are the Potential Challenges of an IPD Meta-Analysis Project? 14

2.5 Empirical Evidence of Differences between Results of IPD and Aggregate Data Meta-analysis Projects 14

2.6 Guidance for Deciding When IPD Meta-analysis Projects Are Needed to Evaluate Treatment Effects from Randomised Trials 15

2.6.1 Are IPD Needed to Tackle the Research Question? 15

2.6.2 Are IPD Needed to Improve the Completeness and Uniformity of Outcomes and Participant-level Covariates? 17

2.6.3 Are IPD Needed to Improve the Information Size? 17

2.6.4 Are IPD Needed to Improve the Quality of Analysis? 18

2.7 Concluding Remarks 19

3 Planning and Initiating an IPD Meta-analysis Project 21

Lesley A. Stewart, Richard D. Riley, and Jayne F. Tierney

3.1 Introduction 22

3.2 Organisational Approach 22

3.2.1 Collaborative IPD Meta-analysis Project 22

3.2.2 IPD Meta-analysis Projects Using Data Repositories or Data-sharing Platforms 24

3.3 Developing a Project Scope 26

3.4 Assessing Feasibility and ‘In Principle’ Support and Collaboration 26

3.5 Establishing a Team with the Right Skills 29

3.6 Advisory and Governance Functions 30

3.7 Estimating How Long the Project Will Take 31

3.8 Estimating the Resources Required 33

3.9 Obtaining Funding 38

3.10 Obtaining Ethical Approval 39

3.11 Data-sharing Agreement 41

3.12 Additional Planning for Prospective Meta-analysis Projects 41

3.13 Concluding Remarks 43

4 Running an IPD Meta-analysis Project: From Developing the Protocol to Preparing Data for Meta-analysis 45

Jayne F. Tierney, Richard D. Riley, Larysa H.M. Rydzewska, and Lesley A. Stewart

4.1 Introduction 46

4.2 Preparing to Collect IPD 46

4.2.1 Defining the Objectives and Eligibility Criteria 46

4.2.2 Developing the Protocol for an IPD Meta-analysis Project 49

4.2.3 Identifying and Screening Potentially Eligible Trials 51

4.2.4 Deciding Which Information Is Needed to Summarise Trial Characteristics 51

4.2.5 Deciding How Much IPD Are Needed 52

4.2.6 Deciding Which Variables Are Needed in the IPD 52

4.2.7 Developing a Data Dictionary for the IPD 55

4.3 Initiating and Maintaining Collaboration 57

4.4 Obtaining IPD 59

4.4.1 Ensuring That IPD Are De-identified 59

4.4.2 Providing Data Transfer Guidance 60

4.4.3 Transferring trial IPD securely 61

4.4.4 Storing Trial IPD Securely 61

4.4.5 Making Best Use of IPD from Repositories 61

4.5 Checking and Harmonising Incoming IPD 62

4.5.1 The Process and Principles 63

4.5.2 Initial Checking of IPD for Each Trial 63

4.5.3 Harmonising IPD across Trials 64

4.5.4 Checking the Validity, Range and Consistency of Variables 65

4.6 Checking the IPD to Inform Risk of Bias Assessments 66

4.6.1 The Randomisation Process 68

4.6.2 Deviations from the Intended Interventions 71

4.6.3 Missing Outcome Data 73

4.6.4 Measurement of the Outcome 74

4.7 Assessing and Presenting the Overall Quality of a Trial 76

4.8 Verification of Finalised Trial IPD 77

4.9 Merging IPD Ready for Meta-analysis 77

4.10 Concluding Remarks 80

Part I References 81

Part II Fundamental Statistical Methods and Principles 87

5 The Two-stage Approach to IPD Meta-analysis 89

Richard D. Riley, Thomas P.A. Debray, Tim P. Morris, and Dan Jackson

5.1 Introduction 90

5.2 First Stage of a Two-stage IPD Meta-analysis 90

5.2.1 General Format of Regression Models to Use in the First Stage 92

5.2.2 Estimation of Regression Models Applied in the First Stage 92

5.2.3 Regression for Different Outcome Types 94

5.2.3.1 Continuous Outcomes 94

5.2.3.2 Binary Outcomes 98

5.2.3.3 Ordinal and Multinomial Outcomes 99

5.2.3.4 Count and Incidence Rate Outcomes 100

5.2.3.5 Time-to-Event Outcomes 101

5.2.4 Adjustment for Prognostic Factors 102

5.2.5 Dealing with Other Trial Designs and Missing Data 103

5.3 Second Stage of a Two-stage IPD Meta-analysis 106

5.3.1 Meta-analysis Assuming a Common Treatment Effect 106

5.3.2 Meta-analysis Assuming Random Treatment Effects 107

5.3.3 Forest Plots and Percentage Trial Weights 110

5.3.4 Heterogeneity Measures and Statistics 110

5.3.5 Alternative Weighting Schemes 112

5.3.6 Frequentist Estimation of the Between-Trial Variance of Treatment Effect 113

5.3.7 Deriving Confidence Intervals for the Summary Treatment Effect 113

5.3.8 Bayesian Estimation Approaches 115

5.3.8.1 An Introduction to Bayes’ Theorem and Bayesian Inference 115

5.3.8.2 Using a Bayesian Meta-Analysis Model in the Second Stage 115

5.3.8.3 Applied Example 117

5.3.9 Interpretation of Summary Effects from Meta-analysis 118

5.3.10 Prediction Interval for the Treatment Effect in a New Trial 118

5.4 Meta-regression and Subgroup Analyses 120

5.5 The ipdmetan Software Package 121

5.6 Combining IPD with Aggregate Data from non-IPD Trials 124

5.7 Concluding Remarks 125

6 The One-stage Approach to IPD Meta-analysis 127

Richard D. Riley and Thomas P.A. Debray 127

6.1 Introduction 128

6.2 One-stage IPD Meta-analysis Models Using Generalised Linear Mixed Models 129

6.2.1 Basic Statistical Framework of One-stage Models Using GLMMs 129

6.2.1.1 Continuous Outcomes 130

6.2.1.2 Binary Outcomes 130

6.2.1.3 Ordinal and Multinomial Outcomes 135

6.2.1.4 Count and Incidence Rate Outcomes 136

6.2.2 Specifying Parameters as Either Common, Stratified, or Random 136

6.2.3 Accounting for Clustering of Participants within Trials 139

6.2.3.1 Examples 141

6.2.4 Choice of Stratified Intercept or Random Intercepts 141

6.2.4.1 Findings from Simulation Studies 142

6.2.4.2 Our Preference for Using a Stratified Intercept 142

6.2.4.3 Allowing for Correlation between Random Effects on Intercept and Treatment Effect 143

6.2.5 Stratified or Common Residual Variances 144

6.2.6 Adjustment for Prognostic Factors 145

6.2.7 Inclusion of Trial-level Covariates 145

6.2.8 Estimation of One-stage IPD Meta-analysis Models Using GLMMs 146

6.2.8.1 Software for Fitting One-stage Models 146

6.2.8.2 ML Estimation and Downward Bias in Between-trial Variance Estimates 146

6.2.8.3 Trial-specific Centering of Variables to Improve ML Estimation of One-stage Models with a Stratified Intercept 147

6.2.8.4 REML Estimation 147

6.2.8.5 Deriving Confidence Intervals for ParametersPpost-estimation 149

6.2.8.6 Prediction Intervals 151

6.2.8.7 Derivation of Percentage Trial Weights 151

6.2.8.8 Bayesian Estimation for One-stage Models 151

6.2.9 A Summary of Recommendations 152

6.3 One-stage Models for Time-to-event Outcomes 152

6.3.1 Cox Proportional Hazard Framework 152

6.3.1.1 Stratifying Using Proportional Baseline Hazards and Frailty Models 152

6.3.1.2 Stratifying Baseline Hazards without Assuming Proportionality 154

6.3.1.3 Comparison of Approaches 154

6.3.1.4 Estimation Methods 154

6.3.1.5 Example 156

6.3.2 Fully Parametric Approaches 157

6.3.3 Extension to Time-varying Hazard Ratios and Joint Models 157

6.4 One-stage Models Combining Different Sources of Evidence 159

6.4.1 Combining IPD Trials with Partially Reconstructed IPD from Non-IPD Trials 159

6.4.2 Combining IPD and Aggregate Data Using Hierarchical Related Regression 160

6.4.3 Combining IPD from Parallel Group, Cluster and Cross-over Trials 161

6.5 Reporting of One-stage Models in Protocols and Publications 162

6.6 Concluding Remarks 162

7 Using IPD Meta-analysis to Examine Interactions between Treatment Effect and Participant-level Covariates 163

Richard D. Riley and David J. Fisher

7.1 Introduction 164

7.2 Meta-regression and Its Limitations 166

7.2.1 Meta-regression of Aggregated Participant-level Covariates 166

7.2.2 Low Power and Aggregation Bias 166

7.2.3 Empirical Evidence of the Difference Between Using Across-trial and Within-trial Information to Estimate Treatment-covariate Interactions 167

7.3 Two-stage IPD Meta-analysis to Estimate Treatment-covariate Interactions 168

7.3.1 The Two-stage Approach 168

7.3.2 Applied Example: Is the Effect of Anti-hypertensive Treatment Different for Males and Females? 170

7.3.3 Do Not Quantify Interactions by Comparing Meta-analysis Results for Subgroups 171

7.4 The One-stage Approach 174

7.4.1 Merging Within-trial and Across-trial Information 174

7.4.2 Separating Within-trial and Across-trial Information 175

7.4.2.1 Approach (i) for a One-stage Survival Model: Center the Covariate and Include the Covariate Mean 175

7.4.2.2 Approach (ii) for a One-stage Survival Model: Stratify All Nuisance Parameters by Trial 176

7.4.2.3 Approaches (i) and (ii) for Continuous and Binary Outcomes 176

7.4.2.4 Comparison of Approaches (i) and (ii) 177

7.4.3 Applied Examples 177

7.4.3.1 Is Age an Effect Modifier for Epilepsy Treatment? 177

7.4.3.2 Is the Effect of an Early Support Hospital Discharge Modified by Having a Carer Present? 178

7.4.4 Coding of the Treatment Covariate and Adjustment for Other Covariates 178

7.4.4.1 Example 180

7.4.5 Estimating the Aggregation Bias Directly 180

7.4.6 Reporting Summary Treatment Effects for Subgroups after Adjusting for Aggregation Bias 180

7.5 Combining IPD and non-IPD Trials 181

7.5.1 Can We Recover Interaction Estimates from non-IPD Trials? 181

7.5.2 How to Incorporate Interaction Estimates from non-IPD Trials in an IPD Metaanalysis 182

7.6 Handling of Continuous Covariates 184

7.6.1 Do Not Categorise Continuous Covariates 184

7.6.2 Interactions May Be Non-linear 185

7.6.2.1 Rationale and an Example 185

7.6.2.2 Two-stage Multivariate IPD Meta-analysis for Summarising Non-linear Interactions 186

7.6.2.3 One-stage IPD Meta-analysis for Summarising Non-linear Interactions 190

7.7 Handling of Categorical or Ordinal Covariates 191

7.8 Misconceptions and Cautions 191

7.8.1 Genuine Treatment-covariate Interactions Are Rare 191

7.8.2 Interactions May Depend on the Scale of Analysis 192

7.8.3 Measurement Error May Impact Treatment-covariate Interactions 193

7.8.4 Even without Treatment-covariate Interactions, the Treatment Effect on Absolute Risk May Differ across Participants 193

7.8.5 Between-trial Heterogeneity in Treatment Effect Should Not Be Used to Guide Whether Treatment-covariate Interactions Exist at the Participant Level 194

7.9 Is My Identified Treatment-covariate Interaction Genuine? 195

7.10 Reporting of Analyses of Treatment-covariate Interactions 196

7.11 Can We Predict a New Patient’s Treatment Effect? 196

7.11.1 Linking Predictions to Clinical Decision Making 198

7.12 Concluding Remarks 198

8 One-stage versus Two-stage Approach to IPD Meta-analysis: Differences and Recommendations 199

Richard D. Riley, Danielle L. Burke, and Tim Morris

8.1 Introduction 200

8.2 One-stage and Two-stage Approaches Usually Give Similar Results 200

8.2.1 Evidence to Support Similarity of One-stage and Two-stage IPD Meta-analysis Results 200

8.2.2 Examples 202

8.2.3 Some Claims in Favour of the One-stage Approach Are Misleading 203

8.3 Ten Key Reasons Why One-stage and Two-stage Approaches May Give Different Results 203

8.3.1 Reason I: Exact One-stage Likelihood When Most Trials Are Small 204

8.3.2 Reason II: How Clustering of Participants Within Trials Is Modelled 207

8.3.3 Reason III: Coding of the Treatment Variable in One-stage Models Fitting with ML Estimation 208

8.3.4 Reason IV: Different Estimation Methods for τ2 210

8.3.5 Reason V: Specification of Prognostic Factor and Adjustment Terms 210

8.3.6 Reason VI: Specification of the Residual Variances 212

8.3.7 Reason VI: Choice of Common Effect or Random Effects for the Parameter of Interest 213

8.3.8 Reason VIII: Derivation of Confidence Intervals 213

8.3.9 Reason IX: Accounting for Correlation Amongst Multiple Outcomes or Time-points 214

8.3.10 Reason X: Aggregation Bias for Treatment Covariate Interactions 215

8.3.11 Other Potential Causes 215

8.4 Recommendations and Guidance 216

8.5 Concluding Remarks 217

Part II References 219

Part III Critical Appraisal and Dissemination 237

9 Examining the Potential for Bias in IPD Meta-analysis Results 239

Richard D. Riley, Jayne F. Tierney, and Lesley A. Stewart

9.1 Introduction 240

9.2 Publication and Reporting Biases of Trials 240

9.2.1 Impact on IPD Meta-analysis Results 240

9.2.2 Examining Small-study Effects Using Funnel Plots 241

9.2.3 Small-study Effects May Arise Due to the Factors Causing Heterogeneity 243

9.3 Biased Availability of the IPD from Trials 244

9.3.1 Examining the Impact of Availability Bias 245

9.3.2 Example: IPD Meta-analysis Examining High-dose Chemotherapy for the Treatment of Non-Hodgkin Lymphoma 246

9.4 Trial Quality (risk of bias) 247

9.5 Other Potential Biases Affecting IPD Meta-analysis Results 248

9.5.1 Trial Selection Bias 248

9.5.2 Selective Outcome Availability 250

9.5.3 Use of Inappropriate Methods by the IPD Meta-analysis Research Team 250

9.6 Concluding Remarks 251

10 Reporting and Dissemination of IPD Meta-analyses 253

Lesley A. Stewart, Richard D. Riley, and Jayne F. Tierney

10.1 Introduction 253

10.2 Reporting IPD Meta-analysis Projects in Academic Reports 254

10.2.1 PRISMA-IPD Title and Abstract Sections 255

10.2.2 PRISMA-IPD Introduction Section 259

10.2.3 PRISMA-IPD Methods Section 259

10.2.4 PRISMA-IPD Results Section 262

10.2.5 PRISMA-IPD Discussion and Funding Sections 266

10.3 Additional Means of Disseminating Findings 266

10.3.1 Key Audiences 267

10.3.1.1 The IPD Collaborative Group 267

10.3.1.2 Patient and Public Audiences 267

10.3.1.3 Guideline Developers 268

10.3.2 Communication Channels 268

10.3.2.1 Evidence Summaries and Policy Briefings 268

10.3.2.2 Press Releases 268

10.3.2.3 Social Media 270

10.4 Concluding Remarks 270

11 A Tool for the Critical Appraisal of IPD Meta-analysis Projects (CheckMAP) 271

Jayne F. Tierney, Lesley A. Stewart, Claire L. Vale, and Richard D. Riley

11.1 Introduction 271

11.2 The CheckMAP Tool 272

11.3 Was the IPD Meta-analysis Project Done within a Systematic Review Framework? 272

11.4 Were the IPD Meta-analysis Project Methods Pre-specified in a Publicly Available Protocol? 274

11.5 Did the IPD Meta-analysis Project Have a Clear Research Question Qualified by Explicit Eligibility Criteria? 276

11.6 Did the IPD Meta-analysis Project Have a Systematic and Comprehensive Search Strategy? 276

11.7 Was the Approach to Data Collection Consistent and Thorough? 277

11.8 Were IPD Obtained from Most Eligible Trials and Their Participants? 277

11.9 Was the Validity of the IPD Checked for Each Trial? 278

11.10 Was the Risk of Bias Assessed for Each Trial and Its Associated IPD? 27811.10.1 Was the Randomisation Process Checked Based on IPD? 278

11.10.2 Were the IPD Checked to Ensure That All (or Most) Randomised Participants Were Included? 279

11.10.3 Were All Important Outcomes Included in the IPD? 279

11.10.4 Were the Outcomes Measured/Defined Appropriately? 279

11.10.5 Was the Quality of Outcome Data Checked? 280

11.11 Were the Methods of Meta-analysis Appropriate? 280

11.11.1 Were the Analyses Pre-specified in Detail and the Key Estimands Defined? 280

11.11.2 Were the Methods of Summarising the Overall Effects of Treatments Appropriate? 281

11.11.3 Were the Methods of Assessing whether Effects of Treatments Varied by Trial-level Characteristics Appropriate? 281

11.11.4 Were the Methods of Assessing whether Effects of Treatments Varied by Participant-level Characteristics Appropriate? 282

11.11.5 Was the Robustness of Conclusions Checked Using Relevant Sensitivity or Other Analyses? 282

11.11.6 Did the IPD Meta-analysis Project’s Report Cover the Items Described in PRISMAIPD? 282

11.12 Concluding Remarks 283

Part III References 285

Part IV Special Topics in Statistics 291

12 Power Calculations for Planning an IPD Meta-analysis 293

Richard D. Riley and Joie Ensor

12.1 Introduction 294

12.1.1 Rationale for Power Calculations in an IPD Meta-analysis 294

12.1.2 Premise for This Chapter 294

12.2 Motivating Example: Power of a Planned IPD Meta-analysis of Trials of Interventions to Reduce Weight Gain in Pregnant Women 295

12.2.1 Background 295

12.2.2 What Is the Power to Detect a Treatment-BMI Interaction? 295

12.2.3 Power of an IPD Meta-analysis to Detect a Treatment-covariate Interaction for a Continuous Outcome 295

12.2.4 Closed-form Solutions 296

12.2.4.1 Application to the i-WIP Example 298

12.2.5 Simulation-based Power Calculations for a Two-stage IPD Meta-analysis 299

12.2.5.1 Application to the i-WIP Example 300

12.2.6 Power Results Naively Assuming the IPD All Come from a Single Trial 301

12.3 The Contribution of Individual Trials Toward Power 301

12.3.1 Contribution According to Sample Size 301

12.3.2 Contribution According to Covariate and Outcome Variability 302

12.4 The Impact of Model Assumptions on Power 302

12.4.1 Impact of Allowing for Heterogeneity in the Interaction 302

12.4.2 Impact of Wrongly Modelling BMI as a Binary Variable 304

12.4.3 Impact of Adjusting for Additional Covariates 304

12.5 Extensions 305

12.5.1 Power Calculations for Binary and Time-to-event Outcomes 305

12.5.2 Simulation Using a One-stage IPD Meta-analysis Approach 306

12.5.3 Examining the Potential Precision of IPD Meta-analysis Results 307

12.5.4 Estimating the Power of a New Trial Conditional on IPD Meta-analysis Results 307

12.6 Concluding Remarks 309

13 Multivariate Meta-analysis Using IPD 311

Richard D. Riley, Dan Jackson, and Ian R. White

13.1 Introduction 312

13.2 General Two-stage Approach for Multivariate IPD Meta-analysis 314

13.2.1 First-stage Analyses 315

13.2.1.1 Obtaining Treatment Effect Estimates and Their Variances for Continuous Outcomes 315

13.2.1.2 Obtaining Within-trial Correlations Directly or via Bootstrapping for Continuous Outcomes 316

13.2.1.3 Extension to Binary, Time-to-event and Mixed Outcomes 317

13.2.2 Second-stage Analysis: Multivariate Meta-analysis Model 319

13.2.2.1 Multivariate Model Structure 320

13.2.2.2 Dealing with Missing Outcomes 320

13.2.2.3 Frequentist Estimation of the Multivariate Model 321

13.2.2.4 Bayesian Estimation of the Multivariate Model 322

13.2.2.5 Joint Inferences and Predictions 322

13.2.2.6 Alternative Specifications for the Between-trial Variance Matrix with Missing Outcomes 323

13.2.2.7 Combining IPD and non-IPD Trials 323

13.2.3 Useful Measures to Accompany Multivariate Meta-analysis Results 324

13.2.3.1 Heterogeneity Measures 324

13.2.3.2 Percentage Trial Weights 325

13.2.3.3 The Efficiency (E) and Borrowing of Strength (BoS) Statistics 325

13.2.4 Understanding the Impact of Correlation and Borrowing of Strength 326

13.2.4.1 Anticipating the Value of BoS When Assuming Common Treatment Effects 326

13.2.4.2 BoS When Assuming Random Treatment Effects 327

13.2.4.3 How the Borrowing of Strength Impacts upon the Summary Meta-analysis Estimates 327

13.2.4.4 How the Correlation Impacts upon Joint Inferences across Outcomes 328

13.2.5 Software 328

13.3 Application to an IPD Meta-analysis of Anti-hypertensive Trials 329

13.3.1 Bivariate Meta-analysis of SBP and DBP 329

13.3.1.1 First-stage Results 329

13.3.1.2 Second-stage Results 329

13.3.1.3 Predictive Inferences 331

13.3.2 Bivariate Meta-analysis of CVD and Stroke 332

13.3.3 Multivariate Meta-analysis of SBP, DBP, CVD and Stroke 332

13.4 Extension to Multivariate Meta-regression 333

13.5 Potential Limitations of Multivariate Meta-analysis 334

13.5.1 The Benefits of a Multivariate Meta-analysis for Each Outcome Are Often Small 335

13.5.2 Model Specification and Estimation Is Non-trivial 335

13.5.3 Benefits Arise under Assumptions 335

13.6 One-stage Multivariate IPD Meta-analysis Applications 337

13.6.1 Summary Treatment Effects 337

13.6.1.1 Applied Example 337

13.6.2 Multiple Treatment-covariate Interactions 337

13.6.2.1 Applied Example 339

13.6.3 Multinomial Outcomes 339

13.7 Special Applications of Multivariate Meta-analysis 340

13.7.1 Longitudinal Data and Multiple Time-points 340

13.7.1.1 Applied Example 341

13.7.1.2 Extensions 342

13.7.2 Surrogate Outcomes 342

13.7.3 Development of Multi-parameter Models for Dose Response and Prediction 344

13.7.4 Test Accuracy 345

13.7.5 Treatment-covariate Interactions 345

13.7.5.1 Non-linear Trends 345

13.7.5.2 Multiple Treatment-covariate Interactions 345

13.8 Concluding Remarks 346

14 Network Meta-analysis Using IPD 347

Richard D. Riley, David M. Phillippo, and Sofia Dias

14.1 Introduction 348

14.2 Rationale and Assumptions for Network Meta-analysis 348

14.3 Network Meta-analysis Models Assuming Consistency 350

14.3.1 A Two-stage Approach 350

14.3.2 A One-stage Approach 351

14.3.3 Summary Results after a Network Meta-analysis 352

14.3.4 Example: Comparison of Eight Thrombolytic Treatments after Acute Myocardial Infarction 352

14.3.4.1 Two-stage Approach 353

14.3.4.2 One-stage Approach 357

14.4 Ranking Treatments 357

14.5 How Do We Examine Inconsistency between Direct and Indirect Evidence? 359

14.6 Benefits of IPD for Network Meta-analysis 361

14.6.1 Benefit 1: Examining and Plotting Distributions of Covariates across Trials Providing Different Comparisons 361

14.6.2 Benefit 2: Adjusting for Prognostic Factors to Improve Consistency and Reduce Heterogeneity 361

14.6.3 Benefit 3: Including Treatment-covariate Interactions 362

14.6.4 Benefit 4: Multiple Outcomes 365

14.7 Combining IPD and Aggregate Data in Network Meta-analysis 365

14.7.1 Multilevel Network Meta-regression 367

14.7.2 Example: Treatments to Reduce Plaque Psoriasis 369

14.8 Further Topics 370

14.8.1 Accounting for Dose and Class 370

14.8.2 Inclusion of ‘Real-world’ Evidence 372

14.8.3 Cumulative Network Meta-analysis 372

14.8.4 Quality Assessment and Reporting 372

14.9 Concluding Remarks 372

Part IV References 375

Part V Diagnosis, Prognosis and Prediction 387

15 IPD Meta-analysis for Test Accuracy Research 389

Richard D. Riley, Brooke Levis, and Yemisi Takwoingi 389

15.1 Introduction 390

15.1.1 Meta-analysis of Test Accuracy Studies 390

15.1.2 The Need for IPD 391

15.1.3 Scope of This Chapter 394

15.2 Motivating Example: Diagnosis of Fever in Children Using Ear Temperature 394

15.3 Key Steps Involved in an IPD Meta-analysis of Test Accuracy Studies 397

15.3.1 Defining the Research Objectives 397

15.3.2 Searching for Studies with Eligible IPD 397

15.3.3 Extracting Key Study Characteristics and Information 398

15.3.4 Evaluating Risk of Bias of Eligible Studies 398

15.3.5 Obtaining, Cleaning and Harmonising IPD 401

15.3.6 Undertaking IPD Meta-analysis to Summarise Test Accuracy at a Particular Threshold 401

15.3.6.1 Bivariate IPD Meta-analysis to Summarise Sensitivity and Specificity 401

15.3.6.2 Examining and Summarising Heterogeneity 402

15.3.6.3 Combining IPD and non-IPD Studies 403

15.3.6.4 Application to the Fever Example 403

15.3.6.5 Bivariate Meta-analysis of PPV and NPV 404

15.3.7 Examining Accuracy-covariate Associations 406

15.3.7.1 Model Specification Using IPD Studies 407

15.3.7.2 Combining IPD and Aggregate Data 408

15.3.7.3 Application to the Fever Example 408

15.3.8 Performing Sensitivity Analyses and Examining Small-study Effects 409

15.3.9 Reporting and Interpreting Results 409

15.4 IPD Meta-analysis of Test Accuracy at Multiple Thresholds 410

15.4.1 Separate Meta-analysis at Each Threshold 410

15.4.2 Joint Meta-analysis of All Thresholds 410

15.4.2.1 Modelling Using the Multinomial Distribution 411

15.4.2.2 Modelling the Underlying Distribution of the Continuous Test Values 412

15.5 IPD Meta-analysis for Examining a Test’s Clinical Utility 414

15.5.1 Net Benefit and Decision Curves 415

15.5.2 IPD Meta-analysis Models for Summarising Clinical Utility of a Test 416

15.5.3 Application to the Fever Example 417

15.6 Comparing Tests 418

15.6.1 Comparative Test Accuracy Meta-analysis Models 419

15.6.2 Applied Example 420

15.7 Concluding Remarks 420

16 IPD Meta-analysis for Prognostic Factor Research 421

Richard D. Riley, Karel G.M. Moons, and Thomas P.A. Debray

16.1 Introduction 422

16.1.1 Problems with Meta-analyses Based on Published Aggregate Data 422

16.1.2 Scope of This Chapter 424

16.2 Potential Advantages of an IPD Meta-analysis 424

16.2.1 Standardise Inclusion Criteria and Definitions 424

16.2.2 Standardise Statistical Analyses 425

16.2.3 Advanced Statistical Modelling 426

16.3 Key Steps Involved in an IPD Meta-analysis of Prognostic Factor Studies 427

16.3.1 Defining the Research Question 427

16.3.1.1 Unadjusted or Adjusted Prognostic Factor Effects? 429

16.3.2 Searching and Selecting Eligible Studies and Datasets 430

16.3.3 Extracting Key Study Characteristics and Information 433

16.3.4 Evaluating Risk of Bias of Eligible Studies 433

16.3.5 Obtaining, Cleaning and Harmonising IPD 433

16.3.6 Undertaking IPD Meta-analysis to Summarise Prognostic Effects 434

16.3.6.1 A Two-stage Approach Assuming a Linear Prognostic Trend 434

16.3.6.2 A Two-stage Approach with Non-linear Trends Using Splines or Polynomials 435

16.3.6.3 Incorporating Measurement Error 438

16.3.6.4 A One-stage Approach 440

16.3.6.5 Checking the Proportional Hazards Assumption 441

16.3.6.6 Dealing with Missing Data and Adjustment Factors 441

16.3.7 Examining Heterogeneity and Performing Sensitivity Analyses 442

16.3.8 Examining Small-study Effects 442

16.3.9 Reporting and Interpreting Results 443

16.4 Software 444

16.5 Concluding Remarks 444

17 IPD Meta-analysis for Clinical Prediction Model Research 447

Richard D. Riley, Kym I.E. Snell, Laure Wynants, Valentijn M.T. de Jong, Karel G.M. Moons, and Thomas P.A. Debray

17.1 Introduction 448

17.2 IPD Meta-analysis for Prediction Model Research 448

17.2.1 Types of Prediction Model Research 448

17.2.2 Why IPD Meta-analyses Are Needed 450

17.2.3 Key Steps Involved in an IPD Meta-analysis for Prediction Model Research 452

17.2.3.1 Define the Research Question and PICOTS System 452

17.2.3.2 Identify Relevant Existing Studies and Datasets 452

17.2.3.3 Examine Eligibility and Risk of Bias of IPD 452

17.2.3.4 Obtain, Harmonise and Summarise IPD 454

17.2.3.5 Undertake Meta-analysis and Quantify Heterogeneity 455

17.3 External Validation of an Existing Prediction Model Using IPD Meta-analysis 455

17.3.1 Measures of Predictive Performance in a Single Study 456

17.3.1.1 Overall Measures of Model Fit 456

17.3.1.2 Calibration Plots and Measures 456

17.3.1.3 Discrimination Measures 456

17.3.2 Potential for Heterogeneity in a Model’s Predictive Performance 459

17.3.2.1 Causes of Heterogeneity in Model Performance 460

17.3.2.2 Disentangling Sources of Heterogeneity 461

17.3.3 Statistical Methods for IPD Meta-analysis of Predictive Performance 461

17.3.3.1 Two-stage IPD Meta-analysis 461

17.3.3.2 Example 1: Validation of Prediction Models for Cardiovascular Disease 463

17.3.3.3 Example 2: Meta-analysis of Case-mix Standardised Estimates of Model Performance 466

17.3.3.4 Example 3: Examining Predictive Performance of QRISK2 across Multiple Practices 468

17.3.3.5 One-stage IPD Meta-analysis 469

17.4 Updating and Tailoring of a Prediction Model Using IPD Meta-analysis 470

17.4.1 Example 1: Updating of the Baseline Hazard in a Prognostic Prediction Model 470

17.4.2 Example 2: Multivariate IPD Meta-analysis to Compare Different Model Updating Strategies 471

17.5 Comparison of Multiple Existing Prediction Models Using IPD Meta-analysis 472

17.5.1 Example 1: Comparison of QRISK2 and Framingham 472

17.5.2 Example 2: Comparison of Prediction Models for Pre-eclampsia 476

17.5.3 Comparing Models When Predictors Are Unavailable in Some Studies 476

17.6 Using IPD Meta-analysis to Examine the Added Value of a New Predictor to an Existing Prediction Model 478

17.7 Developing a New Prediction Model Using IPD Meta-analysis 479

17.7.1 Model Development Issues 479

17.7.1.1 Examining and Handling Between-study Heterogeneity in Case-mix Distributions 479

17.7.1.2 One-stage or Two-stage IPD Meta-analysis Models 482

17.7.1.3 Allowing for Between-study Heterogeneity and Inclusion of Study-specific Parameters 483

17.7.1.4 Studies with Different Designs 484

17.7.1.5 Predictor Selection Based on Statistical Significance 484

17.7.1.6 Conditional and Marginal Apparent Performance 485

17.7.1.7 Sample Size, Overfitting and Penalisation 485

17.7.2 Internal-external Cross-validation to Examine Transportability 487

17.7.2.1 Overview of the Method 487

17.7.2.2 Example: Diagnostic Prediction Model for Deep Vein Thrombosis 488

17.8 Examining the Utility of a Prediction Model Using IPD Meta-analysis 491

17.8.1 Example: Net Benefit of a Diagnostic Prediction Model for Ovarian Cancer 492

17.8.1.1 Summary and Predicted Net Benefit of the LR2 Model 493

17.8.1.2 Comparison to Strategies of Treat All or Treat None 493

17.8.2 Decision Curves 493

17.9 Software 494

17.10 Reporting 495

17.11 Concluding Remarks 495

18 Dealing with Missing Data in an IPD Meta-analysis 499

Thomas Debray, Kym I.E. Snell, Matteo Quartagno, Shahab Jolani, Karel G.M. Moons, and Richard D. Riley 499

18.1 Introduction 500

18.2 Motivating Example: IPD Meta-analysis Validating Prediction Models for Risk of Preeclampsia in Pregnancy 500

18.3 Types of Missing Data in an IPD Meta-analysis 502

18.4 Recovering Actual Values of Missing Data within IPD 502

18.5 Mechanisms and Patterns of Missing Data in an IPD Meta-analysis 502

18.5.1 Mechanisms of Missing Data 504

18.5.2 Patterns of Missing Data 504

18.5.3 Example: Risk of Pre-eclampsia in Pregnancy 505

18.6 Multiple Imputation to Deal with Missing Data in a Single Study 506

18.6.1 Joint Modelling 506

18.6.2 Fully Conditional Specification 507

18.6.3 How Many Imputations Are Required? 508

18.6.4 Combining Results Obtained from Each Imputed Dataset 508

18.7 Ensuring Congeniality of Imputation and Analysis Models 509

18.8 Dealing with Sporadically Missing Data in an IPD Meta-analysis by Applying Multiple Imputation for Each Study Separately 509

18.8.1 Example: Risk of Pre-eclampsia in Pregnancy 511

18.9 Dealing with Systematically Missing Data in an IPD Meta-analysis Using a Bivariate Metaanalysis of Partially and Fully Adjusted Results 511

18.10 Dealing with Both Sporadically and Systematically Missing Data in an IPD Meta-analysis Using Multilevel Modelling 514

18.10.1 Motivating Example: Prognostic Factors for Short-term Mortality in Acute Heart Failure 515

18.10.2 Multilevel Joint Modelling 516

18.10.3 Multilevel Fully Conditional Specification 519

18.11 Comparison of Methods and Recommendations 521

18.11.1 Multilevel FCS versus Joint Model Approaches 521

18.11.2 Sensitivity Analyses and Reporting 523

18.12 Software 523

18.13 Concluding Remarks 524

Part V References 525