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

ISBN-13: 9781119333722 / Angielski / Twarda / 2021 / 560 str.

Richard Riley;Jayne Tierney;Lesley Stewart

Opis

This book is a pivotal textbook for those considering, undertaking or appraising an evidence synthesis based on IPD for clinical research, especially those interested in intervention effects, modifiers of treatment response, identification of risk or prognostic factors, and the development and validation of risk prediction models. It covers all the key concepts and stages of a systematic review and meta-analysis of IPD, focusing primarily on the synthesis of randomised trials, as well as specialist topics, such as risk prediction, observational studies and advanced statistical methods. The book offers non-technical and practical examples, summary and learning points, and guidance including reporting criteria, software demonstrations, and illustrated applications. Describing the key features of the approach, this book will enable the reader to: Understand when the IPD approach is needed How to undertake the systematic review and identify relevant evidence How to obtain, check and manage the IPD How to minimise potential biases and clearly report methods and results How to make the best use of IPD meta-analyses to inform policy, practice and research. Case-study chapters are also provided, where the trials and tribulations of undertaking IPD projects of randomised trials and observational studies are described by researchers in a range of clinical fields.

Spis treści

Acknowledgements xxiii1 Individual Participant Data Meta-analysis for Healthcare Research 1Richard D. Riley, Lesley A. Stewart, and Jayne F. Tierney1.1 Introduction 11.2 What Is IPD and How Does It Differ from Aggregate Data? 11.3 IPD Meta-analysis: A New Era for Evidence Synthesis 21.4 Scope of This Book and Intended Audience 2Part I Rationale, Planning, and Conduct 72 Rationale for Embarking on an IPD Meta-analysis Project 9Jayne F. Tierney, Richard D. Riley, Catrin Tudur Smith, Mike Clarke, and Lesley A. Stewart2.1 Introduction 92.2 How Does the Research Process Differ for IPD and Aggregate Data Meta-analysis Projects? 102.2.1 The Research Aims 102.2.2 The process and methods 102.3 What Are the Potential Advantages of an IPD Meta-analysis Project? 112.4 What Are the Potential Challenges of an IPD Meta-Analysis Project? 142.5 Empirical Evidence of Differences between Results of IPD and Aggregate Data Metaanalysis Projects 142.6 Guidance for Deciding When IPD Meta-analysis Projects Are Needed to Evaluate Treatment Effects from Randomised Trials 152.6.1 Are IPD Needed to Tackle the Research Question? 152.6.2 Are IPD Needed to Improve the Completeness and Uniformity of Outcomes and Participant-level Covariates? 172.6.3 Are IPD Needed to Improve the Information Size? 172.6.4 Are IPD Needed to Improve the Quality of Analysis? 182.7 Concluding Remarks 193 Planning and Initiating an IPD Meta-analysis Project 21Lesley A. Stewart, Richard D. Riley, and Jayne F. Tierney3.1 Introduction 223.2 Organisational Approach 223.2.1 Collaborative IPD Meta-analysis Project 223.2.2 IPD Meta-analysis Projects Using Data Repositories or Data-sharing Platforms 243.3 Developing a Project Scope 263.4 Assessing Feasibility and 'In Principle' Support and Collaboration 263.5 Establishing a Team with the Right Skills 293.6 Advisory and Governance Functions 303.7 Estimating How Long the Project Will Take 313.8 Estimating the Resources Required 333.9 Obtaining Funding 383.10 Obtaining Ethical Approval 393.11 Data-sharing Agreement 413.12 Additional Planning for Prospective Meta-analysis Projects 413.13 Concluding Remarks 434 Running an IPD Meta-analysis Project: From Developing the Protocol to Preparing Data for Meta-analysis 45Jayne F. Tierney, Richard D. Riley, Larysa H.M. Rydzewska, and Lesley A. Stewart4.1 Introduction 464.2 Preparing to Collect IPD 464.2.1 Defining the Objectives and Eligibility Criteria 464.2.2 Developing the Protocol for an IPD Meta-analysis Project 494.2.3 Identifying and Screening Potentially Eligible Trials 514.2.4 Deciding Which Information Is Needed to Summarise Trial Characteristics 514.2.5 Deciding How Much IPD Are Needed 524.2.6 Deciding Which Variables Are Needed in the IPD 524.2.7 Developing a Data Dictionary for the IPD 554.3 Initiating and Maintaining Collaboration 574.4 Obtaining IPD 594.4.1 Ensuring That IPD Are De-identified 594.4.2 Providing Data Transfer Guidance 604.4.3 Transferring trial IPD securely 614.4.4 Storing Trial IPD Securely 614.4.5 Making Best Use of IPD from Repositories 614.5 Checking and Harmonising Incoming IPD 624.5.1 The Process and Principles 634.5.2 Initial Checking of IPD for Each Trial 634.5.3 Harmonising IPD across Trials 644.5.4 Checking the Validity, Range and Consistency of Variables 654.6 Checking the IPD to Inform Risk of Bias Assessments 664.6.1 The Randomisation Process 684.6.2 Deviations from the Intended Interventions 714.6.3 Missing Outcome Data 734.6.4 Measurement of the Outcome 744.7 Assessing and Presenting the Overall Quality of a Trial 764.8 Verification of Finalised Trial IPD 774.9 Merging IPD Ready for Meta-analysis 774.10 Concluding Remarks 80Part I References 81Part II Fundamental Statistical Methods and Principles 875 The Two-stage Approach to IPD Meta-analysis 89Richard D. Riley, Thomas P.A. Debray, Tim P. Morris, and Dan Jackson5.1 Introduction 905.2 First Stage of a Two-stage IPD Meta-analysis 905.2.1 General Format of Regression Models to Use in the First Stage 925.2.2 Estimation of Regression Models Applied in the First Stage 925.2.3 Regression for Different Outcome Types 945.2.3.1 Continuous Outcomes 945.2.3.2 Binary Outcomes 985.2.3.3 Ordinal and Multinomial Outcomes 995.2.3.4 Count and Incidence Rate Outcomes 1005.2.3.5 Time-to-Event Outcomes 1015.2.4 Adjustment for Prognostic Factors 1025.2.5 Dealing with Other Trial Designs and Missing Data 1035.3 Second Stage of a Two-stage IPD Meta-analysis 1065.3.1 Meta-analysis Assuming a Common Treatment Effect 1065.3.2 Meta-analysis Assuming Random Treatment Effects 1075.3.3 Forest Plots and Percentage Trial Weights 1105.3.4 Heterogeneity Measures and Statistics 1105.3.5 Alternative Weighting Schemes 1125.3.6 Frequentist Estimation of the Between-Trial Variance of Treatment Effect 1135.3.7 Deriving Confidence Intervals for the Summary Treatment Effect 1135.3.8 Bayesian Estimation Approaches 1155.3.8.1 An Introduction to Bayes' Theorem and Bayesian Inference 1155.3.8.2 Using a Bayesian Meta-Analysis Model in the Second Stage 1155.3.8.3 Applied Example 1175.3.9 Interpretation of Summary Effects from Meta-analysis 1185.3.10 Prediction Interval for the Treatment Effect in a New Trial 1185.4 Meta-regression and Subgroup Analyses 1205.5 The ipdmetan Software Package 1215.6 Combining IPD with Aggregate Data from non-IPD Trials 1245.7 Concluding Remarks 1256 The One-stage Approach to IPD Meta-analysis 127Richard D. Riley and Thomas P.A. Debray 1276.1 Introduction 1286.2 One-stage IPD Meta-analysis Models Using Generalised Linear Mixed Models 1296.2.1 Basic Statistical Framework of One-stage Models Using GLMMs 1296.2.1.1 Continuous Outcomes 1306.2.1.2 Binary Outcomes 1306.2.1.3 Ordinal and Multinomial Outcomes 1356.2.1.4 Count and Incidence Rate Outcomes 1366.2.2 Specifying Parameters as Either Common, Stratified, or Random 1366.2.3 Accounting for Clustering of Participants within Trials 1396.2.3.1 Examples 1416.2.4 Choice of Stratified Intercept or Random Intercepts 1416.2.4.1 Findings from Simulation Studies 1426.2.4.2 Our Preference for Using a Stratified Intercept 1426.2.4.3 Allowing for Correlation between Random Effects on Intercept and Treatment Effect 1436.2.5 Stratified or Common Residual Variances 1446.2.6 Adjustment for Prognostic Factors 1456.2.7 Inclusion of Trial-level Covariates 1456.2.8 Estimation of One-stage IPD Meta-analysis Models Using GLMMs 1466.2.8.1 Software for Fitting One-stage Models 1466.2.8.2 ML Estimation and Downward Bias in Between-trial Variance Estimates 1466.2.8.3 Trial-specific Centering of Variables to Improve ML Estimation of One-stage Models with a Stratified Intercept 1476.2.8.4 REML Estimation 1476.2.8.5 Deriving Confidence Intervals for ParametersPpost-estimation 1496.2.8.6 Prediction Intervals 1516.2.8.7 Derivation of Percentage Trial Weights 1516.2.8.8 Bayesian Estimation for One-stage Models 1516.2.9 A Summary of Recommendations 1526.3 One-stage Models for Time-to-event Outcomes 1526.3.1 Cox Proportional Hazard Framework 1526.3.1.1 Stratifying Using Proportional Baseline Hazards and Frailty Models 1526.3.1.2 Stratifying Baseline Hazards without Assuming Proportionality 1546.3.1.3 Comparison of Approaches 1546.3.1.4 Estimation Methods 1546.3.1.5 Example 1566.3.2 Fully Parametric Approaches 1576.3.3 Extension to Time-varying Hazard Ratios and Joint Models 1576.4 One-stage Models Combining Different Sources of Evidence 1596.4.1 Combining IPD Trials with Partially Reconstructed IPD from Non-IPD Trials 1596.4.2 Combining IPD and Aggregate Data Using Hierarchical Related Regression 1606.4.3 Combining IPD from Parallel Group, Cluster and Cross-over Trials 1616.5 Reporting of One-stage Models in Protocols and Publications 1626.6 Concluding Remarks 1627 Using IPD Meta-analysis to Examine Interactions between Treatment Effect and Participant-level Covariates 163Richard D. Riley and David J. Fisher7.1 Introduction 1647.2 Meta-regression and Its Limitations 1667.2.1 Meta-regression of Aggregated Participant-level Covariates 1667.2.2 Low Power and Aggregation Bias 1667.2.3 Empirical Evidence of the Difference Between Using Across-trial and Within-trial Information to Estimate Treatment-covariate Interactions 1677.3 Two-stage IPD Meta-analysis to Estimate Treatment-covariate Interactions 1687.3.1 The Two-stage Approach 1687.3.2 Applied Example: Is the Effect of Anti-hypertensive Treatment Different for Males and Females? 1707.3.3 Do Not Quantify Interactions by Comparing Meta-analysis Results for Subgroups 1717.4 The One-stage Approach 1747.4.1 Merging Within-trial and Across-trial Information 1747.4.2 Separating Within-trial and Across-trial Information 1757.4.2.1 Approach (i) for a One-stage Survival Model: Center the Covariate and Include the Covariate Mean 1757.4.2.2 Approach (ii) for a One-stage Survival Model: Stratify All Nuisance Parameters by Trial 1767.4.2.3 Approaches (i) and (ii) for Continuous and Binary Outcomes 1767.4.2.4 Comparison of Approaches (i) and (ii) 1777.4.3 Applied Examples 1777.4.3.1 Is Age an Effect Modifier for Epilepsy Treatment? 1777.4.3.2 Is the Effect of an Early Support Hospital Discharge Modified by Having a Carer Present? 1787.4.4 Coding of the Treatment Covariate and Adjustment for Other Covariates 1787.4.4.1 Example 1807.4.5 Estimating the Aggregation Bias Directly 1807.4.6 Reporting Summary Treatment Effects for Subgroups after Adjusting for Aggregation Bias 1807.5 Combining IPD and non-IPD Trials 1817.5.1 Can We Recover Interaction Estimates from non-IPD Trials? 1817.5.2 How to Incorporate Interaction Estimates from non-IPD Trials in an IPD Metaanalysis 1827.6 Handling of Continuous Covariates 1847.6.1 Do Not Categorise Continuous Covariates 1847.6.2 Interactions May Be Non-linear 1857.6.2.1 Rationale and an Example 1857.6.2.2 Two-stage Multivariate IPD Meta-analysis for Summarising Non-linear Interactions 1867.6.2.3 One-stage IPD Meta-analysis for Summarising Non-linear Interactions 1907.7 Handling of Categorical or Ordinal Covariates 1917.8 Misconceptions and Cautions 1917.8.1 Genuine Treatment-covariate Interactions Are Rare 1917.8.2 Interactions May Depend on the Scale of Analysis 1927.8.3 Measurement Error May Impact Treatment-covariate Interactions 1937.8.4 Even without Treatment-covariate Interactions, the Treatment Effect on Absolute Risk May Differ across Participants 1937.8.5 Between-trial Heterogeneity in Treatment Effect Should Not Be Used to Guide Whether Treatment-covariate Interactions Exist at the Participant Level 1947.9 Is My Identified Treatment-covariate Interaction Genuine? 1957.10 Reporting of Analyses of Treatment-covariate Interactions 1967.11 Can We Predict a New Patient's Treatment Effect? 1967.11.1 Linking Predictions to Clinical Decision Making 1987.12 Concluding Remarks 1988 One-stage versus Two-stage Approach to IPD Meta-analysis: Differences and Recommendations 199Richard D. Riley, Danielle L. Burke, and Tim Morris8.1 Introduction 2008.2 One-stage and Two-stage Approaches Usually Give Similar Results 2008.2.1 Evidence to Support Similarity of One-stage and Two-stage IPD Meta-analysis Results 2008.2.2 Examples 2028.2.3 Some Claims in Favour of the One-stage Approach Are Misleading 2038.3 Ten Key Reasons Why One-stage and Two-stage Approaches May Give Different Results 2038.3.1 Reason I: Exact One-stage Likelihood When Most Trials Are Small 2048.3.2 Reason II: How Clustering of Participants Within Trials Is Modelled 2078.3.3 Reason III: Coding of the Treatment Variable in One-stage Models Fitting with ML Estimation 2088.3.4 Reason IV: Different Estimation Methods for tau2 2108.3.5 Reason V: Specification of Prognostic Factor and Adjustment Terms 2108.3.6 Reason VI: Specification of the Residual Variances 2128.3.7 Reason VI: Choice of Common Effect or Random Effects for the Parameter of Interest 2138.3.8 Reason VIII: Derivation of Confidence Intervals 2138.3.9 Reason IX: Accounting for Correlation Amongst Multiple Outcomes or Time-points 2148.3.10 Reason X: Aggregation Bias for Treatment Covariate Interactions 2158.3.11 Other Potential Causes 2158.4 Recommendations and Guidance 2168.5 Concluding Remarks 217Part II References 219Part III Critical Appraisal and Dissemination 2379 Examining the Potential for Bias in IPD Meta-analysis Results 239Richard D. Riley, Jayne F. Tierney, and Lesley A. Stewart9.1 Introduction 2409.2 Publication and Reporting Biases of Trials 2409.2.1 Impact on IPD Meta-analysis Results 2409.2.2 Examining Small-study Effects Using Funnel Plots 2419.2.3 Small-study Effects May Arise Due to the Factors Causing Heterogeneity 2439.3 Biased Availability of the IPD from Trials 2449.3.1 Examining the Impact of Availability Bias 2459.3.2 Example: IPD Meta-analysis Examining High-dose Chemotherapy for the Treatment of Non-Hodgkin Lymphoma 2469.4 Trial Quality (risk of bias) 2479.5 Other Potential Biases Affecting IPD Meta-analysis Results 2489.5.1 Trial Selection Bias 2489.5.2 Selective Outcome Availability 2509.5.3 Use of Inappropriate Methods by the IPD Meta-analysis Research Team 2509.6 Concluding Remarks 25110 Reporting and Dissemination of IPD Meta-analyses 253Lesley A. Stewart, Richard D. Riley, and Jayne F. Tierney10.1 Introduction 25310.2 Reporting IPD Meta-analysis Projects in Academic Reports 25410.2.1 PRISMA-IPD Title and Abstract Sections 25510.2.2 PRISMA-IPD Introduction Section 25910.2.3 PRISMA-IPD Methods Section 25910.2.4 PRISMA-IPD Results Section 26210.2.5 PRISMA-IPD Discussion and Funding Sections 26610.3 Additional Means of Disseminating Findings 26610.3.1 Key Audiences 26710.3.1.1 The IPD Collaborative Group 26710.3.1.2 Patient and Public Audiences 26710.3.1.3 Guideline Developers 26810.3.2 Communication Channels 26810.3.2.1 Evidence Summaries and Policy Briefings 26810.3.2.2 Press Releases 26810.3.2.3 Social Media 27010.4 Concluding Remarks 27011 A Tool for the Critical Appraisal of IPD Meta-analysis Projects (CheckMAP) 271Jayne F. Tierney, Lesley A. Stewart, Claire L. Vale, and Richard D. Riley11.1 Introduction 27111.2 The CheckMAP Tool 27211.3 Was the IPD Meta-analysis Project Done within a Systematic Review Framework? 27211.4 Were the IPD Meta-analysis Project Methods Pre-specified in a Publicly Available Protocol? 27411.5 Did the IPD Meta-analysis Project Have a Clear Research Question Qualified by Explicit Eligibility Criteria? 27611.6 Did the IPD Meta-analysis Project Have a Systematic and Comprehensive Search Strategy? 27611.7 Was the Approach to Data Collection Consistent and Thorough? 27711.8 Were IPD Obtained from Most Eligible Trials and Their Participants? 27711.9 Was the Validity of the IPD Checked for Each Trial? 27811.10 Was the Risk of Bias Assessed for Each Trial and Its Associated IPD? 27811.10.1 Was the Randomisation Process Checked Based on IPD? 27811.10.2 Were the IPD Checked to Ensure That All (or Most) Randomised Participants Were Included? 27911.10.3 Were All Important Outcomes Included in the IPD? 27911.10.4 Were the Outcomes Measured/Defined Appropriately? 27911.10.5 Was the Quality of Outcome Data Checked? 28011.11 Were the Methods of Meta-analysis Appropriate? 28011.11.1 Were the Analyses Pre-specified in Detail and the Key Estimands Defined? 28011.11.2 Were the Methods of Summarising the Overall Effects of Treatments Appropriate? 28111.11.3 Were the Methods of Assessing whether Effects of Treatments Varied by Trial-level Characteristics Appropriate? 28111.11.4 Were the Methods of Assessing whether Effects of Treatments Varied by Participant-level Characteristics Appropriate? 28211.11.5 Was the Robustness of Conclusions Checked Using Relevant Sensitivity or Other Analyses? 28211.11.6 Did the IPD Meta-analysis Project's Report Cover the Items Described in PRISMAIPD? 28211.12 Concluding Remarks 283Part III References 285Part IV Special Topics in Statistics 29112 Power Calculations for Planning an IPD Meta-analysis 293Richard D. Riley and Joie Ensor12.1 Introduction 29412.1.1 Rationale for Power Calculations in an IPD Meta-analysis 29412.1.2 Premise for This Chapter 29412.2 Motivating Example: Power of a Planned IPD Meta-analysis of Trials of Interventions to Reduce Weight Gain in Pregnant Women 29512.2.1 Background 29512.2.2 What Is the Power to Detect a Treatment-BMI Interaction? 29512.2.3 Power of an IPD Meta-analysis to Detect a Treatment-covariate Interaction for a Continuous Outcome 29512.2.4 Closed-form Solutions 29612.2.4.1 Application to the i-WIP Example 29812.2.5 Simulation-based Power Calculations for a Two-stage IPD Meta-analysis 29912.2.5.1 Application to the i-WIP Example 30012.2.6 Power Results Naively Assuming the IPD All Come from a Single Trial 30112.3 The Contribution of Individual Trials Toward Power 30112.3.1 Contribution According to Sample Size 30112.3.2 Contribution According to Covariate and Outcome Variability 30212.4 The Impact of Model Assumptions on Power 30212.4.1 Impact of Allowing for Heterogeneity in the Interaction 30212.4.2 Impact of Wrongly Modelling BMI as a Binary Variable 30412.4.3 Impact of Adjusting for Additional Covariates 30412.5 Extensions 30512.5.1 Power Calculations for Binary and Time-to-event Outcomes 30512.5.2 Simulation Using a One-stage IPD Meta-analysis Approach 30612.5.3 Examining the Potential Precision of IPD Meta-analysis Results 30712.5.4 Estimating the Power of a New Trial Conditional on IPD Meta-analysis Results 30712.6 Concluding Remarks 30913 Multivariate Meta-analysis Using IPD 311Richard D. Riley, Dan Jackson, and Ian R. White13.1 Introduction 31213.2 General Two-stage Approach for Multivariate IPD Meta-analysis 31413.2.1 First-stage Analyses 31513.2.1.1 Obtaining Treatment Effect Estimates and Their Variances for Continuous Outcomes 31513.2.1.2 Obtaining Within-trial Correlations Directly or via Bootstrapping for Continuous Outcomes 31613.2.1.3 Extension to Binary, Time-to-event and Mixed Outcomes 31713.2.2 Second-stage Analysis: Multivariate Meta-analysis Model 31913.2.2.1 Multivariate Model Structure 32013.2.2.2 Dealing with Missing Outcomes 32013.2.2.3 Frequentist Estimation of the Multivariate Model 32113.2.2.4 Bayesian Estimation of the Multivariate Model 32213.2.2.5 Joint Inferences and Predictions 32213.2.2.6 Alternative Specifications for the Between-trial Variance Matrix with Missing Outcomes 32313.2.2.7 Combining IPD and non-IPD Trials 32313.2.3 Useful Measures to Accompany Multivariate Meta-analysis Results 32413.2.3.1 Heterogeneity Measures 32413.2.3.2 Percentage Trial Weights 32513.2.3.3 The Efficiency (E) and Borrowing of Strength (BoS) Statistics 32513.2.4 Understanding the Impact of Correlation and Borrowing of Strength 32613.2.4.1 Anticipating the Value of BoS When Assuming Common Treatment Effects 32613.2.4.2 BoS When Assuming Random Treatment Effects 32713.2.4.3 How the Borrowing of Strength Impacts upon the Summary Meta-analysis Estimates 32713.2.4.4 How the Correlation Impacts upon Joint Inferences across Outcomes 32813.2.5 Software 32813.3 Application to an IPD Meta-analysis of Anti-hypertensive Trials 32913.3.1 Bivariate Meta-analysis of SBP and DBP 32913.3.1.1 First-stage Results 32913.3.1.2 Second-stage Results 32913.3.1.3 Predictive Inferences 33113.3.2 Bivariate Meta-analysis of CVD and Stroke 33213.3.3 Multivariate Meta-analysis of SBP, DBP, CVD and Stroke 33213.4 Extension to Multivariate Meta-regression 33313.5 Potential Limitations of Multivariate Meta-analysis 33413.5.1 The Benefits of a Multivariate Meta-analysis for Each Outcome Are Often Small 33513.5.2 Model Specification and Estimation Is Non-trivial 33513.5.3 Benefits Arise under Assumptions 33513.6 One-stage Multivariate IPD Meta-analysis Applications 33713.6.1 Summary Treatment Effects 33713.6.1.1 Applied Example 33713.6.2 Multiple Treatment-covariate Interactions 33713.6.2.1 Applied Example 33913.6.3 Multinomial Outcomes 33913.7 Special Applications of Multivariate Meta-analysis 34013.7.1 Longitudinal Data and Multiple Time-points 34013.7.1.1 Applied Example 34113.7.1.2 Extensions 34213.7.2 Surrogate Outcomes 34213.7.3 Development of Multi-parameter Models for Dose Response and Prediction 34413.7.4 Test Accuracy 34513.7.5 Treatment-covariate Interactions 34513.7.5.1 Non-linear Trends 34513.7.5.2 Multiple Treatment-covariate Interactions 34513.8 Concluding Remarks 34614 Network Meta-analysis Using IPD 347Richard D. Riley, David M. Phillippo, and Sofia Dias14.1 Introduction 34814.2 Rationale and Assumptions for Network Meta-analysis 34814.3 Network Meta-analysis Models Assuming Consistency 35014.3.1 A Two-stage Approach 35014.3.2 A One-stage Approach 35114.3.3 Summary Results after a Network Meta-analysis 35214.3.4 Example: Comparison of Eight Thrombolytic Treatments after Acute Myocardial Infarction 35214.3.4.1 Two-stage Approach 35314.3.4.2 One-stage Approach 35714.4 Ranking Treatments 35714.5 How Do We Examine Inconsistency between Direct and Indirect Evidence? 35914.6 Benefits of IPD for Network Meta-analysis 36114.6.1 Benefit 1: Examining and Plotting Distributions of Covariates across Trials Providing Different Comparisons 36114.6.2 Benefit 2: Adjusting for Prognostic Factors to Improve Consistency and Reduce Heterogeneity 36114.6.3 Benefit 3: Including Treatment-covariate Interactions 36214.6.4 Benefit 4: Multiple Outcomes 36514.7 Combining IPD and Aggregate Data in Network Meta-analysis 36514.7.1 Multilevel Network Meta-regression 36714.7.2 Example: Treatments to Reduce Plaque Psoriasis 36914.8 Further Topics 37014.8.1 Accounting for Dose and Class 37014.8.2 Inclusion of 'Real-world' Evidence 37214.8.3 Cumulative Network Meta-analysis 37214.8.4 Quality Assessment and Reporting 37214.9 Concluding Remarks 372Part IV References 375Part V Diagnosis, Prognosis and Prediction 38715 IPD Meta-analysis for Test Accuracy Research 389Richard D. Riley, Brooke Levis, and Yemisi Takwoingi 38915.1 Introduction 39015.1.1 Meta-analysis of Test Accuracy Studies 39015.1.2 The Need for IPD 39115.1.3 Scope of This Chapter 39415.2 Motivating Example: Diagnosis of Fever in Children Using Ear Temperature 39415.3 Key Steps Involved in an IPD Meta-analysis of Test Accuracy Studies 39715.3.1 Defining the Research Objectives 39715.3.2 Searching for Studies with Eligible IPD 39715.3.3 Extracting Key Study Characteristics and Information 39815.3.4 Evaluating Risk of Bias of Eligible Studies 39815.3.5 Obtaining, Cleaning and Harmonising IPD 40115.3.6 Undertaking IPD Meta-analysis to Summarise Test Accuracy at a Particular Threshold 40115.3.6.1 Bivariate IPD Meta-analysis to Summarise Sensitivity and Specificity 40115.3.6.2 Examining and Summarising Heterogeneity 40215.3.6.3 Combining IPD and non-IPD Studies 40315.3.6.4 Application to the Fever Example 40315.3.6.5 Bivariate Meta-analysis of PPV and NPV 40415.3.7 Examining Accuracy-covariate Associations 40615.3.7.1 Model Specification Using IPD Studies 40715.3.7.2 Combining IPD and Aggregate Data 40815.3.7.3 Application to the Fever Example 40815.3.8 Performing Sensitivity Analyses and Examining Small-study Effects 40915.3.9 Reporting and Interpreting Results 40915.4 IPD Meta-analysis of Test Accuracy at Multiple Thresholds 41015.4.1 Separate Meta-analysis at Each Threshold 41015.4.2 Joint Meta-analysis of All Thresholds 41015.4.2.1 Modelling Using the Multinomial Distribution 41115.4.2.2 Modelling the Underlying Distribution of the Continuous Test Values 41215.5 IPD Meta-analysis for Examining a Test's Clinical Utility 41415.5.1 Net Benefit and Decision Curves 41515.5.2 IPD Meta-analysis Models for Summarising Clinical Utility of a Test 41615.5.3 Application to the Fever Example 41715.6 Comparing Tests 41815.6.1 Comparative Test Accuracy Meta-analysis Models 41915.6.2 Applied Example 42015.7 Concluding Remarks 42016 IPD Meta-analysis for Prognostic Factor Research 421Richard D. Riley, Karel G.M. Moons, and Thomas P.A. Debray16.1 Introduction 42216.1.1 Problems with Meta-analyses Based on Published Aggregate Data 42216.1.2 Scope of This Chapter 42416.2 Potential Advantages of an IPD Meta-analysis 42416.2.1 Standardise Inclusion Criteria and Definitions 42416.2.2 Standardise Statistical Analyses 42516.2.3 Advanced Statistical Modelling 42616.3 Key Steps Involved in an IPD Meta-analysis of Prognostic Factor Studies 42716.3.1 Defining the Research Question 42716.3.1.1 Unadjusted or Adjusted Prognostic Factor Effects? 42916.3.2 Searching and Selecting Eligible Studies and Datasets 43016.3.3 Extracting Key Study Characteristics and Information 43316.3.4 Evaluating Risk of Bias of Eligible Studies 43316.3.5 Obtaining, Cleaning and Harmonising IPD 43316.3.6 Undertaking IPD Meta-analysis to Summarise Prognostic Effects 43416.3.6.1 A Two-stage Approach Assuming a Linear Prognostic Trend 43416.3.6.2 A Two-stage Approach with Non-linear Trends Using Splines or Polynomials 43516.3.6.3 Incorporating Measurement Error 43816.3.6.4 A One-stage Approach 44016.3.6.5 Checking the Proportional Hazards Assumption 44116.3.6.6 Dealing with Missing Data and Adjustment Factors 44116.3.7 Examining Heterogeneity and Performing Sensitivity Analyses 44216.3.8 Examining Small-study Effects 44216.3.9 Reporting and Interpreting Results 44316.4 Software 44416.5 Concluding Remarks 44417 IPD Meta-analysis for Clinical Prediction Model Research 447Richard D. Riley, Kym I.E. Snell, Laure Wynants, Valentijn M.T. de Jong, Karel G.M. Moons, and Thomas P.A. Debray17.1 Introduction 44817.2 IPD Meta-analysis for Prediction Model Research 44817.2.1 Types of Prediction Model Research 44817.2.2 Why IPD Meta-analyses Are Needed 45017.2.3 Key Steps Involved in an IPD Meta-analysis for Prediction Model Research 45217.2.3.1 Define the Research Question and PICOTS System 45217.2.3.2 Identify Relevant Existing Studies and Datasets 45217.2.3.3 Examine Eligibility and Risk of Bias of IPD 45217.2.3.4 Obtain, Harmonise and Summarise IPD 45417.2.3.5 Undertake Meta-analysis and Quantify Heterogeneity 45517.3 External Validation of an Existing Prediction Model Using IPD Meta-analysis 45517.3.1 Measures of Predictive Performance in a Single Study 45617.3.1.1 Overall Measures of Model Fit 45617.3.1.2 Calibration Plots and Measures 45617.3.1.3 Discrimination Measures 45617.3.2 Potential for Heterogeneity in a Model's Predictive Performance 45917.3.2.1 Causes of Heterogeneity in Model Performance 46017.3.2.2 Disentangling Sources of Heterogeneity 46117.3.3 Statistical Methods for IPD Meta-analysis of Predictive Performance 46117.3.3.1 Two-stage IPD Meta-analysis 46117.3.3.2 Example 1: Validation of Prediction Models for Cardiovascular Disease 46317.3.3.3 Example 2: Meta-analysis of Case-mix Standardised Estimates of Model Performance 46617.3.3.4 Example 3: Examining Predictive Performance of QRISK2 across Multiple Practices 46817.3.3.5 One-stage IPD Meta-analysis 46917.4 Updating and Tailoring of a Prediction Model Using IPD Meta-analysis 47017.4.1 Example 1: Updating of the Baseline Hazard in a Prognostic Prediction Model 47017.4.2 Example 2: Multivariate IPD Meta-analysis to Compare Different Model Updating Strategies 47117.5 Comparison of Multiple Existing Prediction Models Using IPD Meta-analysis 47217.5.1 Example 1: Comparison of QRISK2 and Framingham 47217.5.2 Example 2: Comparison of Prediction Models for Pre-eclampsia 47617.5.3 Comparing Models When Predictors Are Unavailable in Some Studies 47617.6 Using IPD Meta-analysis to Examine the Added Value of a New Predictor to an Existing Prediction Model 47817.7 Developing a New Prediction Model Using IPD Meta-analysis 47917.7.1 Model Development Issues 47917.7.1.1 Examining and Handling Between-study Heterogeneity in Case-mix Distributions 47917.7.1.2 One-stage or Two-stage IPD Meta-analysis Models 48217.7.1.3 Allowing for Between-study Heterogeneity and Inclusion of Study-specific Parameters 48317.7.1.4 Studies with Different Designs 48417.7.1.5 Predictor Selection Based on Statistical Significance 48417.7.1.6 Conditional and Marginal Apparent Performance 48517.7.1.7 Sample Size, Overfitting and Penalisation 48517.7.2 Internal-external Cross-validation to Examine Transportability 48717.7.2.1 Overview of the Method 48717.7.2.2 Example: Diagnostic Prediction Model for Deep Vein Thrombosis 48817.8 Examining the Utility of a Prediction Model Using IPD Meta-analysis 49117.8.1 Example: Net Benefit of a Diagnostic Prediction Model for Ovarian Cancer 49217.8.1.1 Summary and Predicted Net Benefit of the LR2 Model 49317.8.1.2 Comparison to Strategies of Treat All or Treat None 49317.8.2 Decision Curves 49317.9 Software 49417.10 Reporting 49517.11 Concluding Remarks 49518 Dealing with Missing Data in an IPD Meta-analysis 499Thomas Debray, Kym I.E. Snell, Matteo Quartagno, Shahab Jolani, Karel G.M. Moons, and Richard D. Riley 49918.1 Introduction 50018.2 Motivating Example: IPD Meta-analysis Validating Prediction Models for Risk of Preeclampsia in Pregnancy 50018.3 Types of Missing Data in an IPD Meta-analysis 50218.4 Recovering Actual Values of Missing Data within IPD 50218.5 Mechanisms and Patterns of Missing Data in an IPD Meta-analysis 50218.5.1 Mechanisms of Missing Data 50418.5.2 Patterns of Missing Data 50418.5.3 Example: Risk of Pre-eclampsia in Pregnancy 50518.6 Multiple Imputation to Deal with Missing Data in a Single Study 50618.6.1 Joint Modelling 50618.6.2 Fully Conditional Specification 50718.6.3 How Many Imputations Are Required? 50818.6.4 Combining Results Obtained from Each Imputed Dataset 50818.7 Ensuring Congeniality of Imputation and Analysis Models 50918.8 Dealing with Sporadically Missing Data in an IPD Meta-analysis by Applying Multiple Imputation for Each Study Separately 50918.8.1 Example: Risk of Pre-eclampsia in Pregnancy 51118.9 Dealing with Systematically Missing Data in an IPD Meta-analysis Using a Bivariate Metaanalysis of Partially and Fully Adjusted Results 51118.10 Dealing with Both Sporadically and Systematically Missing Data in an IPD Meta-analysis Using Multilevel Modelling 51418.10.1 Motivating Example: Prognostic Factors for Short-term Mortality in Acute Heart Failure 51518.10.2 Multilevel Joint Modelling 51618.10.3 Multilevel Fully Conditional Specification 51918.11 Comparison of Methods and Recommendations 52118.11.1 Multilevel FCS versus Joint Model Approaches 52118.11.2 Sensitivity Analyses and Reporting 52318.12 Software 52318.13 Concluding Remarks 524Part V References 525Index 000

O autorze

Richard D. Riley is Professor of Biostatistics in the School of Medicine, Keele University, UK.Jayne F. Tierney is Professor of Evidence Synthesis at the MRC Clinical Trials Unit, University College London, UK.Lesley A. Stewart is Professor of Evidence Synthesis and Director of the Centre for Reviews and Dissemination, University of York, UK.