Table of Contents
Acknowledgments xi
1 Introduction 1
2 Method comparisons 5
2.1 One measurement by each method 5
2.1.1 Prediction of one method from another 8
2.1.2 Why not use the correlation? 8
2.1.3 A new method and a reference method 9
2.2 Replicate measurements by each method 10
2.2.1 Exchangeable replicates: fat data 10
2.2.2 Linked replicates: oximetry data 11
2.2.3 Why not use the averages of the replicates? 12
2.3 More than two methods 13
2.4 Terminology and notation 14
2.5 What it is all about 14
3 How to… 17
3.1 … use this chapter 17
3.2 Two methods 18
3.2.1 Single measurements 18
3.2.2 Comparing with a gold standard 18
3.2.3 Replicate measurements 19
3.3 More than two methods 19
3.3.1 Single measurements 20
3.3.2 Replicate measurements 20
4 Two methods with a single measurement on each 21
4.1 Model for limits of agreement 22
4.1.1 Prediction between methods 24
4.1.2 The correlation of the difference and the average 26
4.2 Non-constant difference between methods 27
4.3 A worked example 30
4.4 What really goes on 31
4.4.1 Scaling 31
4.4.2 Independence 32
4.4.3 Actual behavior 32
4.5 Other regression methods for non-constant bias 33
4.5.1 Why ordinary regression fails 33
4.5.2 Deming regression 34
4.6 Comparison with a gold standard 35
4.7 Non-constant variance 35
4.7.1 Regression approach 36
4.7.2 A worked example 40
4.8 Transformations 45
4.8.1 Log transformation 45
4.9 Summary 47
5 Replicate measurements 49
5.1 Pairing of replicate measurements 49
5.1.1 Exchangeable replicates 50
5.1.2 Linked replicates 53
5.2 Plotting replicate measurements 55
5.3 Models for replicate measurements 55
5.3.1 Exchangeable replicates 55
5.3.2 Linked replicates 57
5.4 Interpretation of the random effects 59
5.5 Estimation 61
5.6 Getting it wrong and getting it almost right 61
5.6.1 Averaging over replicates 62
5.6.2 Replicates as items 63
5.7 Summary 64
6 Several methods of measurement 67
6.1 Model 67
6.2 Replicate measurements 68
6.3 Single measurement by each method 69
7 A general model for method comparisons 71
7.1 Scaling 72
7.2 Interpretation of the random effects 73
7.3 Parametrization of the mean 74
7.4 Prediction limits 75
7.4.1 Mean of replicates 77
7.4.2 Plotting predictions between methods 77
7.4.3 Reporting variance components 77
7.4.4 Comparison with a gold standard 79
7.5 Estimation 80
7.5.1 Alternating regressions 80
7.5.2 Estimation using Bugs 85
7.5.3 A worked example 87
7.6 Models with non-constant variance 92
7.6.1 Linear dependence of residual standard error 93
7.7 Summary 96
8 Transformation of measurements 99
8.1 Log transformation 100
8.2 Transformations of percentages 100
8.2.1 A worked example 101
8.2.2 Implementation in MethComp 104
8.3 Other transformations 105
8.4 Several methods 105
8.5 Variance components 105
8.6 Summary 106
9 Repeatability, reproducibility and coefficient of variation 107
9.1 Repeatability 108
9.2 Reproducibility 109
9.3 Coefficient of variation 110
9.3.1 Symmetric interval on the log scale 112
9.3.2 Computing the CV correctly 113
9.3.3 Transformations 113
10 Measures of association and agreement 115
10.1 Individual bioequivalence criterion 116
10.2 Agreement index 118
10.3 Relative variance index 119
10.4 Total deviation index 120
10.5 Correlation measures 121
10.5.1 Correlation coefficient 122
10.5.2 Intraclass correlation coefficient 122
10.5.3 Concordance correlation coefficient 124
10.6 Summary 126
11 Design of method comparison studies 127
11.1 Sample size 128
11.1.1 Mean parameters 128
11.1.2 Variance parameters 128
11.2 Repeated measures designs 130
11.3 Summary 131
12 Examples using standard software 133
12.1 SAS 134
12.1.1 Exchangeable replicates 134
12.1.2 Linked replicates 136
12.2 Stata 137
12.2.1 Exchangeable replicates 137
12.2.2 Linked replicates 139
12.3 R 141
12.3.1 Exchangeable replicates 141
12.3.2 Linked replicates 143
13 The MethComp package for R 149
13.1 Data structures 149
13.2 Function overview 150
13.2.1 Graphical functions 150
13.2.2 Data manipulating functions 151
13.2.3 Analysis functions 151
13.2.4 Reporting functions 152
References 153
Index 155