Comparing Clinical Measurement Methods (eBook)

Bendix Carstensen, Senior Statistician at Steno Diabetes Center, Denmark; also of Department of Biostatistics, Copenhagen, Denmark Bendix Carstensen has been working as a biostatistician in research institutions since 1983. During the last 10 years he has taught courses for medical PhD-students at the Department of Biostatistics at the University of Copenhagen and has been involved in the delivery of overseas courses in epidemiology and the comparison of measurement methods.

Comparing Clinical Measurement Methods 5
Contents 7
Acknowledgments 13
1 Introduction 15
2 Method comparisons 19
2.1 One measurement by each method 19
2.1.1 Prediction of one method from another 22
2.1.2 Why not use the correlation? 22
2.1.3 A new method and a reference method 23
2.2 Replicate measurements by each method 24
2.2.1 Exchangeable replicates: fat data 24
2.2.2 Linked replicates: oximetry data 25
2.2.3 Why not use the averages of the replicates? 26
2.3 More than two methods 27
2.4 Terminology and notation 28
2.5 What it is all about 28
3 Howto. . . 31
3.1 . . . use this chapter 31
3.2 Two methods 32
3.2.1 Single measurements 32
3.2.2 Comparing with a gold standard 32
3.2.3 Replicate measurements 33
3.3 More than two methods 33
3.3.1 Single measurements 34
3.3.2 Replicate measurements 34
4 Two methods with a single measurement on each 35
4.1 Model for limits of agreement 36
4.1.1 Prediction between methods 38
4.1.2 The correlation of the difference and the average 40
4.2 Non-constant difference between methods 41
4.3 A worked example 44
4.4 What really goes on 45
4.4.1 Scaling 45
4.4.2 Independence 46
4.4.3 Actual behavior 46
4.5 Other regression methods for non-constant bias 47
4.5.1 Why ordinary regression fails 47
4.5.2 Deming regression 48
4.6 Comparison with a gold standard 49
4.7 Non-constant variance 49
4.7.1 Regression approach 50
4.7.2 A worked example 54
4.8 Transformations 59
4.8.1 Log transformation 59
4.9 Summary 61
5 Replicate measurements 63
5.1 Pairing of replicate measurements 63
5.1.1 Exchangeable replicates 64
5.1.2 Linked replicates 67
5.2 Plotting replicate measurements 69
5.3 Models for replicate measurements 69
5.3.1 Exchangeable replicates 69
5.3.2 Linked replicates 71
5.4 Interpretation of the random effects 73
5.5 Estimation 75
5.6 Getting it wrong and getting it almost right 75
5.6.1 Averaging over replicates 76
5.6.2 Replicates as items 77
5.7 Summary 78
6 Several methods of measurement 81
6.1 Model 81
6.2 Replicate measurements 82
6.3 Single measurement by each method 83
7 A general model for method comparisons 85
7.1 Scaling 86
7.2 Interpretation of the random effects 87
7.3 Parametrization of the mean 88
7.4 Prediction limits 89
7.4.1 Mean of replicates 91
7.4.2 Plotting predictions between methods 91
7.4.3 Reporting variance components 91
7.4.4 Comparison with a gold standard 93
7.5 Estimation 94
7.5.1 Alternating regressions 94
7.5.2 Estimation using BUGS 99
7.5.3 A worked example 101
7.6 Models with non-constant variance 106
7.6.1 Linear dependence of residual standard error 107
7.7 Summary 110
8 Transformation of measurements 113
8.1 Log transformation 114
8.2 Transformations of percentages 114
8.2.1 A worked example 115
8.2.2 Implementation in MethComp 118
8.3 Other transformations 119
8.4 Several methods 119
8.5 Variance components 119
8.6 Summary 120
9 Repeatability, reproducibility and coefficient of variation 121
9.1 Repeatability 122
9.2 Reproducibility 123
9.3 Coefficient of variation 124
9.3.1 Symmetric interval on the log scale 126
9.3.2 Computing the CV correctly 127
9.3.3 Transformations 127
10 Measures of association and agreement 129
10.1 Individual bioequivalence criterion 130
10.2 Agreement index 132
10.3 Relative variance index 133
10.4 Total deviation index 134
10.5 Correlation measures 135
10.5.1 Correlation coefficient 136
10.5.2 Intraclass correlation coefficient 136
10.5.3 Concordance correlation coefficient 138
10.6 Summary 140
11 Design of method comparison studies 141
11.1 Sample size 142
11.1.1 Mean parameters 142
11.1.2 Variance parameters 142
11.2 Repeated measures designs 144
11.3 Summary 145
12 Examples using standard software 147
12.1 SAS 148
12.1.1 Exchangeable replicates 148
12.1.2 Linked replicates 150
12.2 Stata 151
12.2.1 Exchangeable replicates 151
12.2.2 Linked replicates 153
12.3 R 155
12.3.1 Exchangeable replicates 155
12.3.2 Linked replicates 157
13 The MethComp package for R 163
13.1 Data structures 163
13.2 Function overview 164
13.2.1 Graphical functions 164
13.2.2 Data manipulating functions 165
13.2.3 Analysis functions 165
13.2.4 Reporting functions 166
References 167
Index 169

"This book presents useful information about the complexities of method comparison studies specific to clinical/biomedical research. . . I would consider using it in a course intended for students seeking advanced degrees in biostatistics and epidemiology." (Doody's, 16 September 2011)

"In conclusion, this book provides a statistical modeling approach to the comparison of clinical measurements. The modeling aspects will be particularly appreciated by researchers and others mathematically sophisticated, while the computer code at the end of the book will be useful for practitioners wishing to implement the methods." (Journal of Biopharmaceutical Statistics, January 2011)