Available:*
Library | Item Barcode | Call Number | Material Type | Item Category 1 | Status |
---|---|---|---|---|---|
Searching... | 30000004734939 | R853.C55 F34 2002 | Open Access Book | Book | Searching... |
On Order
Summary
Summary
More and more frequently, clinical trials include the evaluation of Health-Related Quality of Life (HRQoL), yet many investigators remain unaware of the unique measurement and analysis issues associated with the assessment of HRQoL. At the end of a study, clinicians and statisticians often face challenging and sometimes insurmountable analytic problems.
Design and Analysis of Quality of Life Studies in Clinical Trials details these issues and presents a range of solutions. Written from the author's extensive experience in the field, it focuses on the very specific features of QoL data: its longitudinal nature, multidimensionality, and the problem of missing data. The author uses three real clinical trials throughout her discussions to illustrate practical implementation of the strategies and analytic methods presented.
As Quality of Life becomes an increasingly important aspect of clinical trials, it becomes essential for clinicians, statisticians, and designers of these studies to understand and meet the challenges this kind of data present. In this book, SAS and S-PLUS programs, checklists, numerous figures, and a clear, concise presentation combine to provide readers with the tools and skills they need to successfully design, conduct, analyze, and report their own studies.
Author Notes
Diane L. Fairclough is an Associate Professor at the Colorado Health Outcomes Program and Department of Preventive Medicine and Biometry, University of Colorado Health Sciences Center, Denver, Colorado, USA
Table of Contents
Preface | p. xv |
Acknowledgments | p. xix |
1 Introduction | p. 1 |
1.1 Health-related quality of life | p. 1 |
1.2 Measuring health-related quality of life | p. 2 |
Characteristics of various measures | p. 2 |
Health status vs. patient preferences | p. 2 |
Objective vs. subjective | p. 4 |
Generic vs. disease-specific instruments | p. 4 |
Global index vs. profile of domain-specific measures | p. 4 |
Response format | p. 6 |
Period of recall | p. 6 |
1.3 Example 1: Adjuvant breast cancer trial | p. 7 |
Patient selection | p. 7 |
Treatment | p. 7 |
Quality of life measure | p. 8 |
Timing of HRQoL assessments | p. 8 |
Questionnaire completion/missing data | p. 10 |
1.4 Example 2: Advanced non-small-cell lung cancer (NSCLC) | p. 10 |
Treatment | p. 12 |
Quality of life measure | p. 12 |
Timing of assessments | p. 13 |
Questionnaire completion/missing data | p. 13 |
1.5 Example 3: Renal cell carcinoma trial | p. 15 |
Patient selection | p. 15 |
Treatment | p. 15 |
Quality of life measure | p. 15 |
Timing of HRQoL assessments | p. 16 |
Questionnaire completion/missing data | p. 17 |
1.6 Summary | p. 18 |
2 Study Design and Protocol Development | p. 19 |
2.1 Introduction | p. 19 |
2.2 Background and rationale | p. 19 |
2.3 Research objectives | p. 20 |
Domains of interest | p. 21 |
Pragmatic vs. explanatory inference | p. 21 |
2.4 Selection of subjects | p. 22 |
2.5 Longitudinal designs | p. 23 |
Event- or condition-driven designs | p. 23 |
Time-driven designs | p. 24 |
Timing of the initial HRQoL assessment | p. 24 |
Timing of the follow-up HRQoL assessments | p. 24 |
Timing of HRQoL assessments when therapy is cyclic | p. 25 |
Trials with different schedules of therapy | p. 25 |
Frequency of evaluations | p. 25 |
Duration of HRQoL assessment | p. 26 |
Assessment after discontinuation of therapy | p. 27 |
2.6 Selection of a quality of life measure | p. 27 |
Trial objectives | p. 28 |
Validity and reliability | p. 29 |
Appropriateness | p. 30 |
2.7 Conduct | p. 31 |
Mode of administration | p. 31 |
Data collection and management | p. 32 |
Avoiding missing data | p. 32 |
Education | p. 33 |
Forms | p. 34 |
Explicit procedures for follow-up | p. 34 |
Scoring instruments | p. 35 |
Reverse coding | p. 35 |
Scoring multi-item scales | p. 35 |
Item nonresponse | p. 37 |
SAS example | p. 38 |
2.8 Summary | p. 39 |
3 Models for Longitudinal Studies | p. 41 |
3.1 Introduction | p. 41 |
Repeated measures models | p. 41 |
Growth curve models | p. 42 |
Selection between models | p. 42 |
Adjuvant breast cancer study (Example 1) | p. 42 |
NSCLC study (Example 2) | p. 43 |
Renal cell carcinoma study (Example 3) | p. 43 |
3.2 Building the analytic models | p. 43 |
Statistics guiding model reduction | p. 44 |
Likelihood ratio tests | p. 44 |
Other statistics | p. 46 |
3.3 Building repeated measures models | p. 46 |
Mean structure | p. 46 |
SAS example of a cell means model | p. 47 |
Covariance structures | p. 48 |
Unstructured covariance | p. 48 |
Structured covariance | p. 49 |
SAS example of a repeated measures model | p. 50 |
Hypothesis testing | p. 52 |
3.4 Building growth curve models | p. 54 |
Model for means | p. 54 |
Polynomial models | p. 54 |
Piecewise linear regression | p. 54 |
Covariance structure | p. 57 |
Variance of random effects | p. 57 |
Variance of residual errors | p. 58 |
SAS example of a polynomial growth curve model | p. 59 |
Fully parameterized model for the means | p. 59 |
Covariance structure | p. 59 |
Model reduction | p. 63 |
Estimation | p. 63 |
Hypothesis testing | p. 64 |
SAS example of a piecewise linear regression model | p. 65 |
Fully parameterized model for the means | p. 65 |
Covariance structure | p. 65 |
Model reduction | p. 66 |
Estimation | p. 66 |
Testing | p. 67 |
3.5 Summary | p. 68 |
4 Missing Data | p. 69 |
4.1 Introduction | p. 69 |
Terminology | p. 69 |
Why are missing data a problem? | p. 69 |
How much data can be missing? | p. 70 |
Similar patterns of dropout among intervention arms | p. 71 |
Prevention | p. 71 |
4.2 Patterns of missing data | p. 71 |
Example: NSCLC study | p. 72 |
4.3 Mechanisms of missing data | p. 72 |
Notation | p. 72 |
Example | p. 73 |
The concept | p. 74 |
MCAR: Missing completely at random | p. 75 |
The concept | p. 75 |
Covariate-dependent dropout | p. 75 |
Identifying covariate-dependent missingness | p. 76 |
Example: NSCLC study | p. 76 |
Analytic methods | p. 77 |
MAR: Missing at random | p. 77 |
The concept | p. 77 |
Identification of dependence on observed data (Y[subscript i superscript obs]) | p. 78 |
Analytic methods | p. 80 |
A test of MCAR vs. MAR for multivariate normal data | p. 81 |
Notation | p. 81 |
Test statistic | p. 81 |
NSCLC example | p. 81 |
Implementing in SAS | p. 82 |
MNAR: Missing not at random | p. 84 |
The concept | p. 84 |
Analytic methods | p. 84 |
Identification of dependence on unobserved data, Y[subscript i superscript mis] | p. 85 |
Example: NSCLC study | p. 85 |
4.4 Renal cell carcinoma study | p. 87 |
Plotting outcome by dropout | p. 89 |
4.5 Summary | p. 90 |
5 Analytic Methods for Ignorable Missing Data | p. 93 |
5.1 Introduction | p. 93 |
Hypothetical example | p. 93 |
5.2 Repeated univariate analyses | p. 94 |
NSCLC example | p. 96 |
5.3 Multivariate methods | p. 96 |
Complete case analysis (MANOVA) | p. 97 |
NSCLC example | p. 98 |
Maximum likelihood estimation with all available data | p. 101 |
NSCLC example | p. 102 |
Further comments | p. 104 |
Exclusion of subjects | p. 104 |
Exclusion of observations | p. 105 |
5.4 Baseline assessment as a covariate | p. 105 |
NSCLC example | p. 107 |
5.5 Change from baseline | p. 108 |
NSCLC example | p. 109 |
5.6 Adding other baseline covariates | p. 110 |
NSCLC example | p. 111 |
5.7 Empirical Bayes estimates | p. 112 |
5.8 Summary | p. 114 |
6 Simple Imputation | p. 115 |
6.1 Introduction | p. 115 |
Limitations of simple imputation | p. 116 |
NSCLC example | p. 116 |
6.2 Mean value substitution | p. 117 |
6.3 Explicit regression models | p. 118 |
Identification of the imputation model | p. 119 |
Simple univariate regression | p. 120 |
Conditional predicted values | p. 123 |
6.4 Last value carried forward | p. 125 |
[delta]-Adjustments | p. 126 |
Arbitrary high or low value | p. 127 |
6.5 Underestimation of variance | p. 128 |
6.6 Sensitivity analysis | p. 130 |
6.7 Summary | p. 130 |
7 Multiple Imputation | p. 131 |
7.1 Introduction | p. 131 |
7.2 Overview of multiple imputation | p. 131 |
Step 1 Selection of the imputation procedure | p. 131 |
Step 2 Generation of M imputed data sets | p. 132 |
Step 3 Analysis of M data sets | p. 132 |
Step 4 Combining results of M analyses | p. 132 |
7.3 Explicit univariate regression | p. 133 |
Identification of the imputation model | p. 133 |
Computation of imputed values | p. 134 |
Practical considerations | p. 135 |
Extensions to longitudinal studies | p. 135 |
Assumptions | p. 135 |
NSCLC example | p. 136 |
7.4 Closest neighbor and predictive mean matching | p. 140 |
Closest neighbor | p. 142 |
Predictive mean matching | p. 142 |
7.5 Approximate Bayesian bootstrap | p. 142 |
The basic procedure | p. 143 |
Extensions to longitudinal studies | p. 144 |
Propensity scores | p. 144 |
Practical issues | p. 144 |
The assumptions | p. 145 |
Nonignorable missing data | p. 145 |
7.6 Multivariate procedures for nonmonotone missing data | p. 146 |
NSCLC example | p. 146 |
7.7 Combining the M analyses | p. 147 |
SAS example | p. 148 |
7.8 Sensitivity analyses | p. 150 |
7.9 Imputation vs. analytic models | p. 151 |
7.10 Implications for design | p. 152 |
7.11 Summary | p. 152 |
8 Pattern Mixture Models | p. 153 |
8.1 Introduction | p. 153 |
NSCLC example | p. 154 |
8.2 Bivariate data (two repeated measures) | p. 155 |
NSCLC example | p. 156 |
Complete-case missing variable (CCMV) restriction | p. 156 |
NSCLC example | p. 158 |
Brown's protective restrictions | p. 158 |
NSCLC example | p. 159 |
Sensitivity analyses with intermediate restrictions | p. 160 |
Large-sample inferences for [mu subscript 2] | p. 160 |
NSCLC example | p. 161 |
8.3 Monotone dropout | p. 161 |
NSCLC study | p. 162 |
Complete-case missing value restriction | p. 162 |
Available case missing value restriction | p. 164 |
Neighboring case missing value restriction | p. 164 |
Comparison of CCMV, ACMV, and NCMV estimates | p. 167 |
8.4 Parametric models | p. 167 |
Linear trends | p. 168 |
Variance estimation | p. 172 |
SAS example of a pattern mixture model | p. 174 |
Step 1 Estimates of [pi superscript P subscript h] | p. 174 |
Step 2 Estimates of [beta superscript P subscript h] | p. 174 |
Step 3 Pooling estimates and computing variance | p. 175 |
Step 4 Hypothesis testing | p. 177 |
8.5 Additional reading | p. 178 |
Extensions of bivariate case | p. 178 |
Extensions of the sensitivity analysis | p. 178 |
Nonparametric analyses | p. 178 |
8.6 Algebraic details | p. 178 |
Simple linear regression of Y on X | p. 178 |
Complete-case missing variable restriction | p. 179 |
Equation 8.9 | p. 179 |
Equation 8.11 | p. 179 |
Brown's protective restriction | p. 179 |
Equation 8.17 | p. 179 |
Equation 8.19 | p. 180 |
Other | p. 180 |
8.7 Summary | p. 181 |
9 Random-Effects Mixture, Shared-Parameter, and Selection Models | p. 183 |
9.1 Introduction | p. 183 |
Mixture models | p. 183 |
Selection models | p. 184 |
Overview | p. 184 |
9.2 Conditional linear model | p. 185 |
Testing MAR vs. MNAR under assumptions of conditional linear model | p. 187 |
NSCLC example | p. 187 |
Estimation of the standard errors | p. 192 |
Assumptions | p. 192 |
Random-coefficient mixture model | p. 192 |
9.3 Joint mixed-effects and time to dropout | p. 193 |
Testing MAR vs. MNAR under the assumptions of the joint model | p. 194 |
Selection or mixture model? | p. 195 |
NSCLC example | p. 195 |
Initial estimates | p. 197 |
Extension to more complex mixed-effects models | p. 198 |
Renal cell carcinoma example | p. 198 |
9.4 Selection model for monotone dropout | p. 198 |
Outcome-dependent selection model | p. 201 |
NSCLC example | p. 201 |
Oswald program | p. 206 |
Longitudinal model | p. 206 |
Dropout model | p. 208 |
Oswald warnings | p. 210 |
9.5 Advanced readings | p. 210 |
Intermittent missing data | p. 210 |
More selection models | p. 210 |
Heckman probit stochastic dropout model | p. 210 |
Wu and Carroll | p. 210 |
Mori | p. 210 |
Nonparametric analyses | p. 211 |
9.6 Summary | p. 211 |
10 Summary Measures | p. 213 |
10.1 Introduction | p. 213 |
Addressing multiplicity of endpoints | p. 213 |
Summary measures vs. summary statistics | p. 213 |
Strengths and weaknesses | p. 215 |
Easier interpretation | p. 215 |
Increased power | p. 215 |
Weakness | p. 215 |
10.2 Choosing a summary measure | p. 215 |
10.3 Constructing summary measures | p. 217 |
Notation | p. 220 |
Missing data | p. 220 |
Average rate of change (slopes) | p. 222 |
NSCLC example | p. 222 |
Missing data | p. 222 |
Area under the curve | p. 225 |
Missing data | p. 225 |
Differences at baseline | p. 227 |
Average of ranks | p. 227 |
Missing data | p. 227 |
Univariate analysis of summary measures | p. 228 |
Stratified analysis of summary measures | p. 228 |
NSCLC example | p. 229 |
10.4 Summary statistics across time | p. 231 |
Notation | p. 231 |
Area under the curve | p. 231 |
Repeated measures | p. 231 |
Growth curve models | p. 232 |
10.5 Summarizing across HRQoL domains or subscales | p. 235 |
Summary measures | p. 235 |
Weighting proportional to the number of questions | p. 236 |
Factor analytic weights | p. 236 |
Patient weights | p. 237 |
Statistically derived weights: Inverse correlation | p. 238 |
Summary statistics | p. 239 |
10.6 Advanced notes | p. 240 |
Nonparametric procedures | p. 240 |
Combining HRQoL and time to event | p. 240 |
Area under the curve | p. 240 |
Latent variable models | p. 241 |
10.7 Summary | p. 241 |
11 Multiple Endpoints | p. 243 |
11.1 Introduction | p. 243 |
Limiting the number of confirmatory tests | p. 243 |
Summary measures and statistics | p. 244 |
Multiple comparison procedures | p. 244 |
11.2 Background concepts and definitions | p. 245 |
Univariate vs. multivariate test statistics | p. 245 |
Familywise and experimentwise error rates | p. 245 |
Global vs. individual tests | p. 246 |
11.3 Multivariate statistics | p. 246 |
Global tests | p. 246 |
Closed multivariate testing procedures | p. 246 |
Limitations | p. 248 |
11.4 Univariate statistics | p. 249 |
NSCLC example | p. 249 |
Alpha adjustments for K univariate tests | p. 249 |
Bonferroni adjustment | p. 249 |
Ruger's inequality | p. 252 |
Simes' global test | p. 252 |
Sequential rejective Bonferroni procedure | p. 253 |
p-value adjustments | p. 253 |
11.5 Resampling techniques | p. 254 |
11.6 Summary | p. 255 |
12 Design: Analysis Plans | p. 257 |
12.1 Introduction | p. 257 |
12.2 General analysis plan--Who is included? | p. 258 |
12.3 Models for longitudinal data | p. 259 |
Ignorable missing data | p. 259 |
Nonignorable missing data | p. 259 |
Event-driven designs and repeated measures models | p. 259 |
Time-driven designs and growth curve models | p. 260 |
Modification of analysis plan | p. 260 |
12.4 Multiplicity of endpoints | p. 260 |
Primary vs. secondary endpoints | p. 260 |
Summary measures | p. 261 |
Multiple comparison procedures | p. 261 |
12.5 Sample size and power | p. 262 |
Simple linear combinations of [beta] | p. 262 |
Basic assumptions | p. 264 |
Incomplete designs | p. 264 |
Example 1 Repeated measures | p. 266 |
Example 2 Growth curve model | p. 268 |
Other considerations | p. 270 |
Intermittent missing data patterns and time-varying covariates | p. 270 |
Unequal allocations of subjects to treatment groups | p. 270 |
Multivariate tests | p. 271 |
Small sample size approximations | p. 272 |
Restricted maximum likelihood estimation | p. 272 |
12.6 Reporting results | p. 272 |
12.7 Summary | p. 274 |
Appendix I Abbreviations | p. 275 |
Appendix II Notation | p. 277 |
Appendix III Formal Definitions for Missing Data | p. 281 |
References | p. 285 |
Index | p. 295 |