| |
| |
| Notation | |
| |
| |
| |
| Introduction | |
| |
| |
| |
| The medical test | |
| |
| |
| |
| Tests, classification and the broader context | |
| |
| |
| |
| Disease screening versus diagnosis | |
| |
| |
| |
| Criteria for a useful medical test | |
| |
| |
| |
| Elements of study design | |
| |
| |
| |
| Scale for the test result | |
| |
| |
| |
| Selection of study subjects | |
| |
| |
| |
| Comparing tests | |
| |
| |
| |
| Test integrity | |
| |
| |
| |
| Sources of bias | |
| |
| |
| |
| Examples and datasets | |
| |
| |
| |
| Overview | |
| |
| |
| |
| The CASS dataset | |
| |
| |
| |
| Pancreatic cancer serum biomarkers study | |
| |
| |
| |
| Hepatitis metastasis ultrasound study | |
| |
| |
| |
| CARET PSA biomarker study | |
| |
| |
| |
| Ovarian cancer gene expression study | |
| |
| |
| |
| Neonatal audiology data | |
| |
| |
| |
| St Louis prostate cancer screening study | |
| |
| |
| |
| Topics and organization | |
| |
| |
| |
| Exercises | |
| |
| |
| |
| Measures of accuracy for binary tests | |
| |
| |
| |
| Measures of accuracy | |
| |
| |
| |
| Notation | |
| |
| |
| |
| Disease-specific classification probabilities | |
| |
| |
| |
| Predictive values | |
| |
| |
| |
| Diagnostic likelihood ratios | |
| |
| |
| |
| Estimating accuracy with data | |
| |
| |
| |
| Data from a cohort study | |
| |
| |
| |
| Proportions: (FPF, TPF) and (PPV, NPV) | |
| |
| |
| |
| Ratios of proportions: DLRs | |
| |
| |
| |
| Estimation from a case-control study | |
| |
| |
| |
| Merits of case-control versus cohort studies | |
| |
| |
| |
| Quantifying the relative accuracy of tests | |
| |
| |
| |
| Comparing classification probabilities | |
| |
| |
| |
| Comparing predictive values | |
| |
| |
| |
| Comparing diagnostic likelihood ratios | |
| |
| |
| |
| Which test is better? | |
| |
| |
| |
| Concluding remarks | |
| |
| |
| |
| Exercises | |
| |
| |
| |
| Comparing binary tests and regression analysis | |
| |
| |
| |
| Study designs for comparing tests | |
| |
| |
| |
| Unpaired designs | |
| |
| |
| |
| Paired designs | |
| |
| |
| |
| Comparing accuracy with unpaired data | |
| |
| |
| |
| Empirical estimators of comparative measures | |
| |
| |
| |
| Large sample inference | |
| |
| |
| |
| Comparing accuracy with paired data | |
| |
| |
| |
| Sources of correlation | |
| |
| |
| |
| Estimation of comparative measures | |
| |
| |
| |
| Wide or long data representations | |
| |
| |
| |
| Large sample inference | |
| |
| |
| |
| Efficiency of paired versus unpaired designs | |
| |
| |
| |
| Small sample properties | |
| |
| |
| |
| The CASS study | |
| |
| |
| |
| The regression modeling framework | |
| |
| |
| |
| Factors potentially affecting test performance | |
| |
| |
| |
| Questions addressed by regression modeling | |
| |
| |
| |
| Notation and general set-up | |
| |
| |
| |
| Regression for true and false positive fractions | |
| |
| |
| |
| Binary marginal GLM models | |
| |
| |
| |
| Fitting marginal models to data | |
| |
| |
| |
| Illustration: factors affecting test accuracy | |
| |
| |
| |
| Comparing tests with regression analysis | |
| |
| |
| |
| Regression modeling of predictive values | |
| |
| |
| |
| Model formulation and fitting | |
| |
| |
| |
| Comparing tests | |
| |
| |
| |
| The incremental value of a test for prediction | |
| |
| |
| |
| Regression models for DLRs | |
| |
| |
| |
| The model form | |
| |
| |
| |
| Fitting the DLR model | |
| |
| |
| |
| Comparing DLRs of two tests | |
| |
| |
| |
| Relationships with other regression models | |
| |
| |
| |
| Concluding remarks | |
| |
| |
| |
| Exercises | |
| |
| |
| |
| The receiver operating characteristic curve | |
| |
| |
| |
| The context | |
| |
| |
| |
| Examples of non-binary tests | |
| |
| |
| |
| Dichotomizing the test result | |
| |
| |
| |
| The ROC curve for continuous tests | |
| |
| |
| |
| Definition of the ROC | |
| |
| |
| |
| Mathematical properties of the ROC curve | |
| |
| |
| |
| Attributes of and uses for the ROC curve | |
| |
| |
| |
| Restrictions and alternatives to the ROC curve | |
| |
| |
| |
| Summary indices | |
| |
| |
| |
| The area under the ROC curve (AUC) | |
| |
| |
| |
| The ROC(t[subscript 0]) and partial AUC | |
| |
| |
| |
| Other summary indices | |
| |
| |
| |
| Measures of distance between distributions | |
| |
| |
| |
| The binormal ROC curve | |
| |
| |
| |
| Functional form | |
| |
| |
| |
| The binormal AUC | |
| |
| |
| |
| The binormal assumption | |
| |
| |
| |
| The ROC for ordinal tests | |
| |
| |
| |
| Tests with ordered discrete results | |
| |
| |
| |
| The latent decision variable model | |
| |
| |
| |
| Identification of the latent variable ROC | |
| |
| |
| |
| Changes in accuracy versus thresholds | |
| |
| |
| |
| The discrete ROC curve | |
| |
| |
| |
| Summary measures for the discrete ROC curve | |
| |
| |
| |
| Concluding remarks | |
| |
| |
| |
| Exercises | |
| |
| |
| |
| Estimating the ROC curve | |
| |
| |
| |
| Introduction | |
| |
| |
| |
| Approaches | |
| |
| |
| |
| Notation and assumptions | |
| |
| |
| |
| Empirical estimation | |
| |
| |
| |
| The empirical estimator | |
| |
| |
| |
| Sampling variability at a threshold | |
| |
| |
| |
| Sampling variability of ROC[subscript e](t) | |
| |
| |
| |
| The empirical AUC and other indices | |
| |
| |
| |
| Variability in the empirical AUC | |
| |
| |
| |
| Comparing empirical ROC curves | |
| |
| |
| |
| Illustration: pancreatic cancer biomarkers | |
| |
| |
| |
| Discrete ordinal data ROC curves | |
| |
| |
| |
| Modeling the test result distributions | |
| |
| |
| |
| Fully parametric modeling | |
| |
| |
| |
| Semiparametric location-scale models | |
| |
| |
| |
| Arguments against modeling test results | |
| |
| |
| |
| Parametric distribution-free methods: ordinal tests | |
| |
| |
| |
| The binormal latent variable framework | |
| |
| |
| |
| Fitting the discrete binormal ROC function | |
| |
| |
| |
| Generalizations and comparisons | |
| |
| |
| |
| Parametric distribution-free methods: continuous tests | |
| |
| |
| |
| LABROC | |
| |
| |
| |
| The ROC-GLM estimator | |
| |
| |
| |
| Inference with parametric distribution-free methods | |
| |
| |
| |
| Concluding remarks | |
| |
| |
| |
| Exercises | |
| |
| |
| |
| Proofs of theoretical results | |
| |
| |
| |
| Covariate effects on continuous and ordinal tests | |
| |
| |
| |
| How and why? | |
| |
| |
| |
| Notation | |
| |
| |
| |
| Aspects to model | |
| |
| |
| |
| Omitting covariates/pooling data | |
| |
| |
| |
| Reference distributions | |
| |
| |
| |
| Non-diseased as the reference population | |
| |
| |
| |
| The homogenous population | |
| |
| |
| |
| Nonparametric regression quantiles | |
| |
| |
| |
| Parametric estimation of S[subscript D,Z] | |
| |
| |
| |
| Semiparametric models | |
| |
| |
| |
| Application | |
| |
| |
| |
| Ordinal test results | |
| |
| |
| |
| Modeling covariate effects on test results | |
| |
| |
| |
| The basic idea | |
| |
| |
| |
| Induced ROC curves for continuous tests | |
| |
| |
| |
| Semiparametric location-scale families | |
| |
| |
| |
| Induced ROC curves for ordinal tests | |
| |
| |
| |
| Random effect models for test results | |
| |
| |
| |
| Modeling covariate effects on ROC curves | |
| |
| |
| |
| The ROC-GLM regression model | |
| |
| |
| |
| Fitting the model to data | |
| |
| |
| |
| Comparing ROC curves | |
| |
| |
| |
| Three examples | |
| |
| |
| |
| Approaches to ROC regression | |
| |
| |
| |
| Modeling ROC summary indices | |
| |
| |
| |
| A qualitative comparison | |
| |
| |
| |
| Concluding remarks | |
| |
| |
| |
| Exercises | |
| |
| |
| |
| Incomplete data and imperfect reference tests | |
| |
| |
| |
| Verification biased sampling | |
| |
| |
| |
| Context and definition | |
| |
| |
| |
| The missing at random assumption | |
| |
| |
| |
| Correcting for bias with Bayes' theorem | |
| |
| |
| |
| Inverse probability weighting/imputation | |
| |
| |
| |
| Sampling variability of corrected estimates | |
| |
| |
| |
| Adjustments for other biasing factors | |
| |
| |
| |
| A broader context | |
| |
| |
| |
| Non-binary tests | |
| |
| |
| |
| Verification restricted to screen positives | |
| |
| |
| |
| Extreme verification bias | |
| |
| |
| |
| Identificable parameters for a single test | |
| |
| |
| |
| Comparing tests | |
| |
| |
| |
| Evaluating covariate effects on (DP, FP) | |
| |
| |
| |
| Evaluating covariate effects on (TPF, FPF) and on prevalence | |
| |
| |
| |
| Evaluating covariate effects on (rTPF, rFPF) | |
| |
| |
| |
| Alternative strategies | |
| |
| |
| |
| Imperfect reference tests | |
| |
| |
| |
| Examples | |
| |
| |
| |
| Effects on accuracy parameters | |
| |
| |
| |
| Classic latent class analysis | |
| |
| |
| |
| Relaxing the conditional independence assumption | |
| |
| |
| |
| A critique of latent class analysis | |
| |
| |
| |
| Discrepant resolution | |
| |
| |
| |
| Composite reference standards | |
| |
| |
| |
| Concluding remarks | |
| |
| |
| |
| Exercises | |
| |
| |
| |
| Proofs of theoretical results | |
| |
| |
| |
| Study design and hypothesis testing | |
| |
| |
| |
| The phases of medical test development | |
| |
| |
| |
| Research as a process | |
| |
| |
| |
| Five phases for the development of a medical test | |
| |
| |
| |
| Sample sizes for phase 2 studies | |
| |
| |
| |
| Retrospective validation of a binary test | |
| |
| |
| |
| Retrospective validation of a continuous test | |
| |
| |
| |
| Sample size based on the AUC | |
| |
| |
| |
| Ordinal tests | |
| |
| |
| |
| Sample sizes for phase 3 studies | |
| |
| |
| |
| Comparing two binary tests--paired data | |
| |
| |
| |
| Comparing two binary tests--unpaired data | |
| |
| |
| |
| Evaluating population effects on test performance | |
| |
| |
| |
| Comparisons with continuous test results | |
| |
| |
| |
| Estimating the threshold for screen positivity | |
| |
| |
| |
| Remarks on phase 3 analyses | |
| |
| |
| |
| Sample sizes for phase 4 studies | |
| |
| |
| |
| Designs for inference about (FPF, TPF) | |
| |
| |
| |
| Designs for predictive values | |
| |
| |
| |
| Designs for (FP, DP) | |
| |
| |
| |
| Selected verification of screen negatives | |
| |
| |
| |
| Phase 5 | |
| |
| |
| |
| Matching and stratification | |
| |
| |
| |
| Stratification | |
| |
| |
| |
| Matching | |
| |
| |
| |
| Concluding remarks | |
| |
| |
| |
| Exercises | |
| |
| |
| |
| More topics and conclusions | |
| |
| |
| |
| Meta-analysis | |
| |
| |
| |
| Goals of meta-analysis | |
| |
| |
| |
| Design of a meta-analysis study | |
| |
| |
| |
| The summary ROC curve | |
| |
| |
| |
| Binomial regression models | |
| |
| |
| |
| Incorporating the time dimension | |
| |
| |
| |
| The context | |
| |
| |
| |
| Incident cases and long-term controls | |
| |
| |
| |
| Interval cases and controls | |
| |
| |
| |
| Predictive values | |
| |
| |
| |
| Longitudinal measurements | |
| |
| |
| |
| Combining multiple test results | |
| |
| |
| |
| Boolean combinations | |
| |
| |
| |
| The likelihood ratio principle | |
| |
| |
| |
| Optimality of the risk score | |
| |
| |
| |
| Estimating the risk score | |
| |
| |
| |
| Development and assessment of the combination score | |
| |
| |
| |
| Concluding remarks | |
| |
| |
| |
| Topics we only mention | |
| |
| |
| |
| New applications and new technologies | |
| |
| |
| |
| Exercises | |
| |
| |
| Bibliography | |
| |
| |
| Index | |