Methods for Meta-Analysis in Medical Research

by Sutton, Alexander J.; Abrams, Keith R.; Jones, David R; Sheldon, Trevor A.; Song, Fujian

• ISBN13: 9780471490661
• ISBN10: 0471490660
• Edition: 1st
• Format: Hardcover
• Copyright: 2000-12-19
• Publisher: WILEY

Summary

With meta-analysis methods playing a crucial role in health research in recent years, this important and clearly-written book provides a much-needed survey of the field. Meta-analysis provides a framework for combining the results of several clinical trials and drawing inferences about the effectiveness of medical treatments. The move towards evidence-based health care and practice is underpinned by the use of meta-analysis. This book: * Provides a thorough criticism and an up-to-date survey of meta-analysis methods * Emphasises the practical approach, and illustrates the methods by numerous examples * Describes the use of Bayesian methods in meta-analysis * Includes discussion of appropriate software for each analysis * Includes numerous references to more advanced treatment of specialist topics * Refers to software code used in the examples available on the authors' Web site Practising statisticians, statistically-minded clinicians and health research professionals will benefit greatly from the clear presentation and numerous examples. Medical researchers will grasp the basic principles of meta-analysis, and learn how to apply the various methods.

Author Biography

Alex Sutton has published extensively on meta-analysis methodology generally, and on publication bias specifically in recent years, including a major systematic review on the topic of the methodology that has been developed for meta-analysis. He currently has an active interest in the area of partially reported study information, which is currently under-researched. Dr. Sutton is co-author of a textbook on metaanalysis (Methods for Meta Analysis in Medical Research), which was published by Wiley in 2000.

Table of Contents

Preface	xv
Acknowledgements	xvii
Part A: Meta-Analysis Methodology: The Basics	1	(160)
Introduction - Meta-analysis: Its Development and Uses	3	(14)
Evidence-based health care	3	(1)
Evidence-based everything!	4	(1)
Pulling together the evidence - systematic reviews	5	(3)
Why meta-analysis?	8	(4)
Aim of this book	12	(1)
Concluding remarks	13	(4)
References	13	(4)
Defining Outcome Measures used for Combining via Meta-analysis	17	(20)
Introduction	17	(1)
Non-comparative binary outcomes	18	(2)
Odds	18	(1)
Incidence rates	19	(1)
Comparative binary outcomes	20	(8)
The Odds ratio	20	(3)
Relative risk (or rate ratio/relative rate)	23	(2)
Risk differences between proportions (or the absolute risk reduction)	25	(2)
The number needed to treat	27	(1)
Comparisons of rates	28	(1)
Other scales of measurement used in summarizing binary data	28	(1)
Which scale to use?	28	(1)
Continuous data	28	(5)
Outcomes defined on their original metric (mean difference)	29	(2)
Outcomes defined using standardized mean differences	31	(2)
Ordinal outcomes	33	(1)
Summary/Discussion	33	(4)
References	34	(3)
Assessing Between Study Heterogeneity	37	(20)
Introduction	37	(1)
Hypothesis tests for presence of heterogeneity	38	(3)
Standard X2 test	38	(1)
Extensions/alternative tests	39	(1)
Example: Testing for heterogeneity in the cholesterol lowering trial dataset	40	(1)
Graphical informal tests/explorations of heterogeneity	41	(7)
Plot of normalized (z) scores	41	(1)
Forest plot	42	(4)
Radial plot (Galbraith diagram)	46	(1)
L'Abbe plot	47	(1)
Possible causes of heterogeneity	48	(2)
Specific factors that may cause heterogeneity in RCTs	49	(1)
Methods for investigating and dealing with sources of heterogeneity	50	(3)
Change scale of outcome variable	51	(1)
Include covariates in a regression model (meta-regression)	51	(1)
Exclude studies	52	(1)
Analyse groups of studies separately	52	(1)
Use of random effects models	52	(1)
Use of mixed-effect models	53	(1)
The validity of pooling studies with heterogeneous outcomes	53	(1)
Summary/Discussion	53	(4)
References	54	(3)
Fixed Effects Methods for Combining Study Estimates	57	(16)
Introduction	57	(1)
General fixed effect model - the inverse variance-weighted method	58	(5)
Example: Combining odds ratios using the inverse variance-weighted method	59	(3)
Example: Combining standardized mean differences using a continuous outcome scale	62	(1)
Specific methods for combining odds ratios	63	(7)
Mantel-Haenszel method for combining odds ratios	64	(2)
Peto's method for combining odds ratios	66	(2)
Combining odds ratios via maximum-likelihood techniques	68	(1)
Exact methods of interval estimation	69	(1)
Discussion of the relative merits of each method	69	(1)
Summary/Discussion	70	(3)
References	71	(2)
Random Effects Models for Combining Study Estimates	73	(14)
Introduction	73	(1)
Algebraic derivation for random effects models by the weighted method	74	(1)
Maximum likelihood and restricted maximum likelihood estimate solutions	75	(1)
Comparison of estimation methods	76	(1)
Example: Combining the cholesterol lowering trials using a random effects model	76	(4)
Extensions to the random effects model	80	(3)
Including uncertainty induced by estimating the between study variance	80	(1)
Exact approach to random effects meta-analysis of binary data	81	(1)
Miscellaneous extensions to the random effects model	82	(1)
Comparison of random with fixed effect models	83	(1)
Summary/Discussion	84	(3)
References	84	(3)
Exploring Between Study Heterogeneity	87	(22)
Introduction	87	(1)
Subgroup analyses	88	(5)
Example: Stratification by study characteristics	89	(1)
Example: Stratification by patient characteristics	89	(4)
Regression models for meta-analysis	93	(11)
Meta-regression models (fixed-effects regression)	93	(2)
Meta-regression example: a meta-analysis of Bacillus Calmette-Guerin (BCG) vaccine for the prevention of tuberculosis (TB)	95	(2)
Mixed effect models (random-effects regression)	97	(2)
Mixed model example: A re-analysis of Bacillus Calmette-Guerin (BCG) vaccine for the prevention of tuberculosis (TB) trials	99	(1)
Mixed modelling extensions	99	(5)
Summary/Discussion	104	(5)
References	105	(4)
Publication Bias	109	(24)
Introduction	109	(1)
Evidence of publication and related bias	110	(2)
Survey of authors	110	(1)
Published versus registered trials in a meta-analysis	110	(1)
Follow-up of cohorts of registered studies	111	(1)
Non-empirical evidence	111	(1)
Evidence of language bias	111	(1)
The seriousness and consequences of publication bias for meta-analysis	112	(1)
Predictors of publication bias (factors effecting the probability a study will get published)	112	(1)
Identifying publication bias in a meta-analysis	112	(7)
The funnel plot	113	(3)
Rank correlation test	116	(1)
Linear regression test	117	(2)
Other methods to detect publication bias	119	(1)
Practical advice on methods for detecting publication bias	119	(1)
Taking into account publication bias or adjusting the results of a meta-analysis in the presence of publication bias	119	(7)
Analysing only the largest studies	120	(1)
Rosenthal's `file drawer' method	120	(2)
Models which estimate the number of unpublished studies, but do not adjust	122	(1)
Selection models using weighted distribution theory	123	(1)
The `Trim and Fill' method	123	(2)
The sensitivity approach of Copas	125	(1)
Broader perspective solutions to publication bias	126	(1)
Prospective registration of trials	126	(1)
Changes in publication process and journals	126	(1)
Including unpublished information	127	(1)
Summary/Discussion	127	(6)
References	128	(5)
Study Quality	133	(14)
Introduction	133	(1)
Methodological factors that may affect the quality of studies	134	(3)
Experimental studies	135	(1)
Observational Studies	136	(1)
Incorporating study quality into a meta-analysis	137	(6)
Graphical plot	137	(1)
Cumulative methods	138	(1)
Regression model	138	(2)
Weighting	140	(2)
Excluding studies	142	(1)
Sensitivity analysis	143	(1)
Practical implementation	143	(1)
Summary/Discussion	144	(3)
References	144	(3)
Sensitivity Analysis	147	(6)
Introduction	147	(1)
Sensitivity of results to inclusion criteria	147	(3)
Sensitivity of results to meta-analytic methods	150	(1)
Assessing the impact of choice of study weighting	150	(1)
Summary/Discussion	151	(2)
References	151	(2)
Reporting the Results of a Meta-analysis	153	(8)
Introduction	153	(1)
Overview and structure of a report	154	(1)
Graphical displays used for reporting the findings of a meta-analysis	155	(3)
Forest plots	155	(2)
Radial plots	157	(1)
Funnel plots	157	(1)
Displaying the distribution of effect size estimates	158	(1)
Graphs investigating length of follow-up	158	(1)
Summary/Discussion	158	(3)
References	158	(3)
Part B: Advanced and Specialized Meta-analysis Topics	161	(140)
Bayesian Methods in Meta-analysis	163	(28)
Introduction	163	(1)
Bayesian methods in health research	163	(6)
General introduction	163	(3)
General advantages/disadvantages of Bayesian methods	166	(1)
Example: Bayesian analysis of a single trial using a normal conjugate model	167	(2)
Bayesian meta-analysis of normally distributed data	169	(2)
Example: Combining trials with continuous outcome measures using Bayesian methods	171	(1)
Bayesian meta-analysis of binary data	171	(4)
Example: Combining binary outcome measures using Bayesian methods	173	(2)
Empirical Bayes methods in meta-analysis	175	(1)
Advantages/disadvantages of Bayesian methods in meta-analysis	176	(3)
Advantages	176	(2)
Disadvantages	178	(1)
Extensions and specific areas of application	179	(4)
Incorporating study quality	179	(1)
Inclusion of covariates	180	(1)
Model selection	180	(1)
Hierarchical models	181	(1)
Sensitivity analysis	181	(1)
Comprehensive modelling	182	(1)
Other developments	183	(1)
Summary/Discussion	183	(8)
References	183	(8)
Meta-analysis of Individual Patient Data	191	(8)
Introduction	191	(2)
Procedural methodology	193	(1)
Data collection	193	(1)
Checking data	193	(1)
Issues involved in carrying out IPD meta-analysis	193	(1)
Comparing meta-analysis using IPD or summary data?	194	(1)
Combining individual patient and summary data	195	(1)
Summary/Discussion	196	(3)
References	196	(3)
Missing Data	199	(6)
Introduction	199	(1)
Reasons for missing data	200	(1)
Categories of missing data at the study level	200	(1)
Analytic methods for dealing with missing data	201	(2)
General missing data methods which can be applied in the meta-analysis context	201	(1)
Missing data methods specific to meta-analysis	202	(1)
Example: Dealing with missing standard deviations of estimates in a meta-analysis	202	(1)
Bayesian methods for missing data	203	(1)
Summary/Discussion	203	(2)
References	204	(1)
Meta-analysis of Different Types of Data	205	(24)
Introduction	205	(1)
Combining ordinal data	205	(1)
Issues concerning scales of measurement when combining data	206	(3)
Transforming scales, maintaining same data type	207	(1)
Binary outcome data reported on different scales	207	(1)
Combining studies whose outcomes are reported using different data types	208	(1)
Combining summaries of binary outcomes with those of continuous outcomes	208	(1)
Non-parametric method of combining different data type effect measures	208	(1)
Meta-analysis of diagnostic test accuracy	209	(6)
Combining binary test results	209	(6)
Combining ordered categorical test results	215	(1)
Combining continuous test results	215	(1)
Meta-analysis using surrogate markers	215	(1)
Combining a number of cross-over trials using the patient preference outcome	216	(1)
Vote-counting methods	217	(1)
Combining p-values/significance levels	218	(5)
Minimum p method	219	(1)
Sum of z's method	220	(1)
Sum of logs method	220	(1)
Logit method	220	(1)
Other methods of combining significance levels	220	(1)
Appraisal of the methods	221	(1)
Example of combining p-values	221	(2)
Novel applications of meta-analysis using non-standard methods or data	223	(1)
Summary/Discussion	223	(6)
References	223	(6)
Meta-analysis of Multiple and Correlated Outcome Measures	229	(10)
Introduction	229	(1)
Combining multiple p-values	230	(1)
Method for reducing multiple outcomes to a single measure for each study	231	(1)
Development of a multivariate model	231	(5)
Model of Raudenbush et al.	231	(1)
Model of Gleser and Olkin	232	(1)
Multiple outcome model for clinical trials	232	(1)
Random effect multiple outcome regression model	232	(1)
DuMouchel's extended model for multiple outcomes	233	(1)
Illustration of the use of multiple outcome models	233	(3)
Summary/Discussion	236	(3)
References	236	(3)
Meta-analysis of Epidemiological and Other Observational Studies	239	(20)
Introduction	239	(1)
Extraction and derivation of study estimates	240	(6)
Scales of measurement used to report and combine observational studies	243	(1)
Data manipulation for data extraction	243	(1)
Methods for transforming and adjusting reported results	244	(2)
Analysis of summary data	246	(2)
Heterogeneity of observational studies	246	(1)
Fixed or random effects?	247	(1)
Weighting of observational studies	247	(1)
Methods for combining estimates of observational studies	247	(1)
Dealing with heterogeneity and combining the OC and breast cancer studies	248	(1)
Reporting the results of meta-analysis of observational studies	248	(1)
Use of sensitivity and influence analysis	248	(1)
Study quality considerations for observational studies	249	(1)
Other issues concerning meta-analysis of observational studies	250	(4)
Analysing individual patient data from observational studies	250	(1)
Combining dose-reponse data	251	(2)
Meta-analysis of single case research	253	(1)
Unresolved issues concerning the meta-analysis of observational studies	254	(1)
Summary/Discussion	255	(4)
References	255	(4)
Generalized Synthesis of Evidence-Combining Different Sources of Evidence	259	(18)
Introduction	259	(1)
Incorporating Single-arm studies: models for incorporating historical controls	259	(3)
Example	260	(2)
Combining matched and unmatched data	262	(1)
Approaches for combining studies containing multiple and/or different treatment arms	263	(2)
Approach of Gleser and Olkin	264	(1)
Models of Berkey et al.	264	(1)
Method of Higgins	264	(1)
Mixed model of DuMouchel	264	(1)
The confidence profile method	265	(1)
Cross-design synthesis	266	(7)
Beginnings	267	(1)
Bayesian hierarchical models	267	(4)
Grouped random effects models of Larose and Dey	271	(1)
Synthesizing studies with disparate designs to assess the exposure effects on the incidence of a rare adverse event	271	(1)
Combining the results of cancer studies in humans and other species	272	(1)
Combining biochemical and epidemiological evidence	272	(1)
Combining information from disparate toxicological studies using stratified ordinal regression	272	(1)
Summary/Discussion	273	(4)
References	273	(4)
Meta-analysis of Survival Data	277	(10)
Introduction	277	(1)
Inferring/estimating and combining (log) hazard ratios	278	(1)
Calculation of the `log-rank' odds ratio	278	(1)
Calculation of pooled survival rates	279	(1)
Method of Hunink and Wong	279	(1)
Iterative generalized least squares for meta-analysis of survival data at multiple times	280	(2)
Application of the model	281	(1)
Identifying prognostic factors using a log (relative risk) measure	282	(1)
Combining quality of life adjusted survival data	282	(1)
Meta-analysis of survival data using individual patient data	283	(1)
Pooling independent samples of survival data to form an estimator of the common survival function	283	(1)
Is obtaining and using survival data necessary?	283	(1)
Summary/Discussion	284	(3)
References	284	(3)
Cumulative Meta-analysis	287	(8)
Introduction	287	(1)
Example: Ordering by date of publication	288	(2)
Using study characteristics other than date of publication	290	(1)
Example: Ordering the cholesterol trials by baseline risk in the control group	290	(1)
Bayesian approaches	291	(1)
Issues regarding uses of cumulative meta-analysis	291	(1)
Summary/Discussion	292	(3)
References	292	(3)
Miscellaneous and Developing Areas of Application in Meta-analysis	295	(6)
Introduction	295	(1)
Alternatives to conventional meta-analysis	295	(2)
Estimating and extrapolating a response surface	295	(1)
Odd man out method	296	(1)
Best evidence synthesis	296	(1)
Developing areas	297	(4)
Prospective meta-analysis	297	(1)
Economic evaluation through meta-analysis	298	(1)
Combining meta-analysis and decision analysis	299	(1)
Net benefit model synthesizing disparate sources of information	299	(1)
References	299	(2)
Appendix I: Software Used for the Examples in this Book	301	(8)
Subject index	309