Clinical Prediction Models: A Practical Approach to Development, Validation, and Updating / Edition 2 available in Hardcover, eBook
![Clinical Prediction Models: A Practical Approach to Development, Validation, and Updating / Edition 2](http://img.images-bn.com/static/redesign/srcs/images/grey-box.png?v11.11.4)
Clinical Prediction Models: A Practical Approach to Development, Validation, and Updating / Edition 2
- ISBN-10:
- 3030163989
- ISBN-13:
- 9783030163983
- Pub. Date:
- 07/23/2019
- Publisher:
- Springer International Publishing
- ISBN-10:
- 3030163989
- ISBN-13:
- 9783030163983
- Pub. Date:
- 07/23/2019
- Publisher:
- Springer International Publishing
![Clinical Prediction Models: A Practical Approach to Development, Validation, and Updating / Edition 2](http://img.images-bn.com/static/redesign/srcs/images/grey-box.png?v11.11.4)
Clinical Prediction Models: A Practical Approach to Development, Validation, and Updating / Edition 2
Buy New
$99.99Buy Used
$54.29-
-
SHIP THIS ITEM
Temporarily Out of Stock Online
Please check back later for updated availability.
-
Overview
The second edition of this volume provides insight and practical illustrations on how modern statistical concepts and regression methods can be applied in medical prediction problems, including diagnostic and prognostic outcomes. Many advances have been made in statistical approaches towards outcome prediction, but a sensible strategy is needed for model development, validation, and updating, such that prediction models can better support medical practice.
There is an increasing need for personalized evidence-based medicine that uses an individualized approach to medical decision-making. In this Big Data era, there is expanded access to large volumes of routinely collected data and an increased number of applications for prediction models, such as targeted early detection of disease and individualized approaches to diagnostic testing and treatment. Clinical Prediction Models presents a practical checklist that needs to be considered for development of a valid prediction model. Steps include preliminary considerations such as dealing with missing values; coding of predictors; selection of main effects and interactions for a multivariable model; estimation of model parameters with shrinkage methods and incorporation of external data; evaluation of performance and usefulness; internal validation; and presentation formatting. The text also addresses common issues that make prediction models suboptimal, such as small sample sizes, exaggerated claims, and poor generalizability.
The text is primarily intended for clinical epidemiologists and biostatisticians. Including many case studies and publicly available R code and data sets, the book is also appropriate as a textbook for a graduate course on predictive modeling in diagnosis and prognosis. While practical in nature, the book also provides a philosophical perspective on data analysis in medicine that goes beyond predictive modeling.
Updates to this new and expanded edition include:
• A discussion of Big Data and its implications for the design of prediction models
• Machine learning issues
• More simulations with missing ‘y’ values
• Extended discussion on between-cohort heterogeneity
• Description of ShinyApp
• Updated LASSO illustration
• New case studies
Product Details
ISBN-13: | 9783030163983 |
---|---|
Publisher: | Springer International Publishing |
Publication date: | 07/23/2019 |
Series: | Statistics for Biology and Health |
Edition description: | 2nd ed. 2019 |
Pages: | 558 |
Sales rank: | 225,074 |
Product dimensions: | 6.10(w) x 9.25(h) x (d) |
About the Author
Table of Contents
Preface viiAcknowledgements xi
Chapter 1 Introduction 1
1.1 Diagnosis, prognosis and therapy choice in medicine 1
1.1.1 Predictions for personalized evidence-based medicine 1
1.2 Statistical modeling for prediction 5
1.2.1 Model assumptions 5
1.2.2 Reliability of predictions: aleatory and epistemic uncertainty 6
1.2.3 Sample size 61.3 Structure of the book 8
1.3.1 Part I: Prediction models in medicine 8
1.3.2 Part II: Developing internally valid prediction models 8
1.3.3 Part III: Generalizability of prediction models 9
1.3.4 Part IV: Applications 9
Part I: Prediction models in medicine 11
Chapter 2 Applications of prediction models 13
2.1 Applications: medical practice and research 13
2.2 Prediction models for Public Health 142.2.1 Targeting of preventive interventions 14
*2.2.2 Example: prediction for breast cancer 14
2.3 Prediction models for clinical practice 17
2.3.1 Decision support on test ordering 17
*2.3.2 Example: predicting renal artery stenosis 17
2.3.3 Starting treatment: the treatment threshold 20
*2.3.4 Example: probability of deep venous thrombosis 20
2.3.5 Intensity of treatment 21
*2.3.6 Example: defining a poor prognosis subgroup in cancer 22
2.3.7 Cost-effectiveness of treatment 23
2.3.8 Delaying treatment 23
*2.3.9 Example: spontaneous pregnancy chances 24
2.3.10 Surgical decision-making 26
*2.3.11 Example: replacement of risky heart valves 27
2.4 Prediction models for medical research 28
2.4.1 Inclusion and stratification in a RCT 28
*2.4.2 Example: selection for TBI trials 29
2.4.3 Covariate adjustment in a RCT 30
2.4.4 Gain in power by covariate adjustment 31
*2.4.5 Example: analysis of the GUSTO-III trial 32
2.4.6 Prediction models and observational studies 32
2.4.7 Propensity scores 33*2.4.8 Example: statin treatment effects 34
2.4.9 Provider comparisons 35
*2.4.10 Example: ranking cardiac outcome 35
2.5 Concluding remarks 35
Chapter 3 Study design for prediction modeling 37
3.1 Studies for prognosis 37
3.1.1 Retrospective designs 37
*3.1.2 Example: predicting early mortality in esophageal cancer 37
3.1.3 Prospective designs 38*3.1.4 Example: predicting long-term mortality in esophageal cancer 39
3.1.5 Registry data 39
*3.1.6 Example: surgical mortality in esophageal cancer 39
3.1.7 Nested case-control studies 40
*3.1.8 Example: perioperative mortality in major vascular surgery 40
3.2 Studies for diagnosis 41
3.2.1 Cross-sectional study design and multivariable modeling 41
*3.2.2 Example: diagnosing renal artery stenosis 41
3.2.3 Case-control studies 41
*3.2.4 Example: diagnosing acute appendicitis 42
3.3 Predictors and outcome 42
3.3.1 Strength of predictors 42
3.3.2 Categories of predictors 42
3.3.3 Costs of predictors 43
3.3.4 Determinants of prognosis 44
3.3.5 Prognosis in oncology 44
3.4 Reliability of predictors 45
3.4.1 Observer variability 45
*3.4.2 Example: histology in Barrett’s esophagus 45
3.4.3 Biological variability 46
3.4.4 Regression dilution bias 46
*3.4.5 Example: simulation study on reliability of a binary predictor 46
3.4.6 Choice of predictors 47
3.5 Outcome 47
3.5.1 Types of outcome 47
3.5.2 Survival endpoints 48
*3.5.3 Examples: 5-year relative survival in cancer registries 48
3.5.4 Composite endpoints 49
*3.5.5 Example: composite endpoints in cardiology 49
3.5.6 Choice of prognostic outcome 49
3.5.7 Diagnostic endpoints 49
*3.5.8 Example: PET scans in esophageal cancer 50
3.6 Phases of biomarker development 50
3.7 Statistical power and reliable estimation 51
3.7.1 Sample size to identify predictor effects 51
3.7.2 Sample size for reliable modeling 53
3.7.3 Sample size for reliable validation 55
3.8 Concluding remarks 55
Chapter 4 Statistical models for prediction 57
4.1 Continuous outcomes 57
*4.1.1 Examples of linear regression 58
4.1.2 Economic outcomes 58
*4.1.3 Example: prediction of costs 58
4.1.4 Transforming the outcome 58
4.1.5 Performance: explained variation 59
4.1.6 More flexible approaches 60
4.2 Binary outcomes 61
4.2.1 R2 in logistic regression analysis 62
4.2.2 Calculation of R2 on the log likelihood scale 63
4.2.3 Models related to logistic regression 65
4.2.4 Bayes rule 65
4.2.5 Prediction with Naïve Bayes 66
4.2.6 Calibration and Naïve Bayes 67
*4.2.7 Logistic regression and Bayes 67
4.2.8 Machine learning: more flexible approaches 68
4.2.9 Classification and regression trees 69
*4.2.10 Example: mortality in acute MI patients 69
4.2.11 Advantages and disadvantages of tree models 70
4.2.12 Trees versus logistic regression modeling 70
*4.2.13 Other methods for binary outcomes 71
4.2.14 Summary on binary outcomes 72
4.3 Categorical outcomes 73
4.3.1 Polytomous logistic regression 73
4.3.2 Example: histology of residual masses 73
*4.3.3 Alternative models 75
*4.3.4 Comparison of modeling approaches 76
4.4 Ordinal outcomes 77
4.4.1 Proportional odds logistic regression 77
* 4.4.2 Relevance of the proportional odds assumption in RCTs 78
4.5 Survival outcomes 80
4.5.1 Cox proportional hazards regression 80
4.5.2 Prediction with Cox models 81
4.5.3 Proportionality assumption 81
4.5.4 Kaplan-Meier analysis 81
*4.5.5 Example: impairment after treatment of leprosy 82
4.5.6 Parametric survival 82
*4.5.7 Example: replacement of risky heart valves 83
4.5.8 Summary on survival outcomes 83
4.6 Competing risks 84
4.6.1 Actuarial and actual risks 84
4.6.2 Absolute risk and the Fine&Gray model 84
4.6.3 Example: Prediction of coronary heart disease incidence 85
4.6.4 Multi-state modeling 86
4.7 Dynamic predictions 87
4.7.1 Multi-state models and landmarking 874.7.2 Joint models 87
4.8 Concluding remarks 88
Chapter 5 Overfitting and optimism in prediction models 91
5.1 Overfitting and optimism 91
5.1.1 Example: surgical mortality in esophagectomy 92
5.1.2 Variability within one center 92
5.1.3 Variability between centers: noise vs. true heterogeneity 93
5.1.4 Predicting mortality by center: shrinkage 94
5.2 Overfitting in regression models 95
5.2.1 Model uncertainty and testimation bias 95
5.2.2 Other modeling biases 97
5.2.3 Overfitting by parameter uncertainty 97
5.2.4 Optimism in model performance 98
5.2.5 Optimism-corrected performance 99
5.3 Bootstrap resampling 100
5.3.1 Applications of the bootstrap 101
5.3.2 Bootstrapping for regression coefficients 102
5.3.3 Bootstrapping for prediction: optimism correction 102
5.3.4 Calculation of optimism-corrected performance 103
*5.3.5 Example: Stepwise selection in 429 patients 104
5.4 Cost of data analysis 105
*5.4.1 Degrees of freedom of a model 105
5.4.2 Practical implications 105
5.5 Concluding remarks 106
Chapter 6 Choosing between alternative models 109
6.1 Prediction with statistical models 109
6.1.1 Testing of model assumptions and prediction 110
6.1.2 Choosing a type of model 110
6.2 Modeling age – outcome relations 111
*6.2.1 Age and mortality after acute MI 111
*6.2.2 Age and operative mortality 112
*6.2.3 Age – outcome relations in other diseases 115
6.3 Head-to-head comparisons 116
6.3.1 StatLog results 116
*6.3.2 Cardiovascular disease prediction comparisons 117
*6.3.3 Traumatic brain injury modeling results 119
6.4 Concluding remarks 120
Part II: Developing valid prediction models 123
Checklist for developing valid prediction models 124
Chapter 7 Missing values 125
7.1 Missing values and prediction research 125
7.1.1 Inefficiency of complete case analysis 126
7.1.2 Interpretation of CC Analyses 127
7.1.3 Missing data mechanisms 127
7.1.4 Missing outcome data 1287.1.5 Summary points 129
7.2 Prediction under MCAR, MAR and MNAR mechanisms 130
7.2.1 Missingness patterns 130
7.2.2 Missingness and estimated regression coefficients 132
7.2.4 Missingness and estimated performance 134
7.3 Dealing with missing values in regression analysis 135
7.3.1 Imputation principle 135
7.3.2 Simple and more advanced single imputation methods 136
7.3.3 Multiple imputation 137
7.4 Defining the imputation model 138
7.4.1 Types of variables in the imputation model 138
*7.4.2 Transformations of variables 139
7.4.3 Imputation models for SI 139
7.4.4 Summary points 139
7.5 Success of imputation under MCAR, MAR and MNAR 140
7.5.1 Imputation in a simple model 140
7.5.2 Other simulation results 140
* 7.5.3 Multiple predictors 140
7.6 Guidance to dealing with missing values in prediction research 142
7.6.1 Patterns of missingness 142
7.6.2 Simple approaches 143
7.6.3 More advanced approaches 143
7.6.4 Maximum fraction of missing values before omitting a predictor 143
7.6.5 Single or multiple imputation for predictor effects? 144
7.6.6 Single or multiple imputation for deriving predictions? 145
7.6.7 Missings and predictions for new patients 145
*7.6.8 Performance across multiple imputed data sets 146
7.6.9 Reporting of missing values in prediction research 146
7.7 Concluding remarks 148
7.7.1 Summary statements 148
*7.7.2 Available software and challenges 149
Chapter 8 Case study on dealing with missing values 151
8.1 Introduction 151
8.1.1 Aim of the IMPACT study 151
8.1.2 Patient selection 152
8.1.3 Potential predictors 152
8.1.4 Coding and time dependency of predictors 153
8.2 Missing values in the IMPACT study 153
8.2.1 Missing values in outcome 153
8.2.2 Quantification of missingness of predictors 154
8.2.3 Patterns of missingness 156
8.3 Imputation of missing predictor values 159
8.3.1 Correlations between predictors 159
8.3.2 Imputation model 160
8.3.3 Distributions of imputed values 160
*8.3.4 Multilevel imputation 161
8.4 Predictor effect: adjusted analyses 162
8.4.1 Adjusted analysis for complete predictors: age and motor score 163
8.4.2 Adjusted analysis for incomplete predictors: pupils 165
8.5 Predictions: multivariable analyses 165
*8.5.1 Multilevel analyses 166
8.6 Concluding remarks 166
Chapter 9 Coding of categorical and continuous predictors 1699.1 Categorical predictors 169
9.1.1 Examples of categorical coding 170
9.2 Continuous predictors 171
*9.2.1 Examples of continuous predictors 171
9.2.2 Categorization of continuous predictors 172
9.3 Non-linear functions for continuous predictors 173
9.3.1. Polynomials 173
9.3.2. Fractional polynomials (FP) 174
9.3.3 Splines 175
*9.3.4 Example: functional forms with RCS or FP 176
9.3.5 Extrapolation and robustness 176
9.3.5 Preference for FP or RCS? 176
9.4 Outliers and winsorizing 177
9.4.1 Example: glucose values and outcome of TBI 178
9.5 Interpretation of effects of continuous predictors 180
*9.5.1 Example: predictor effects in TBI 181
9.6 Concluding remarks 182
9.6.1 Software 183
Chapter 10 Restrictions on candidate predictors 185
10.1 Selection before studying the predictor – outcome relation 185
10.1.1 Selection based on subject knowledge 185
*10.1.2 Examples: too many candidate predictors 185
10.1.3 Meta-analysis for candidate predictors 186
*10.1.4 Example: predictors in testicular cancer 186
10.1.5 Selection based on distributions 186
10.2 Combining similar variables 187
10.2.1 Subject knowledge for grouping 187
10.2.2 Assessing the equal weights assumption 188
10.2.3 Biologically motivated weighting schemes 189
10.2.4 Statistical combination 189
10.3 Averaging effects 190*10.3.1 Example: Chlamydia trachomatis infection risks 190
*10.3.2 Example: acute surgery risk relevant for elective patients? 190
*10.4 Case study: family history for prediction of a genetic mutation 191
10.4.1 Clinical background and patient data 19110.4.2 Similarity of effects 191
10.4.3 CRC and adenoma in a proband 194
10.4.5 Full prediction model for mutations 196
10.5 Concluding remarks 197
Chapter 11 Selection of main effects 199
11.1 Predictor selection 199
11.1.1 Reduction before modeling 199
11.1.2 Reduction while modeling 200
11.1.3 Collinearity 200
11.1.4 Parsimony 200
11.1.5 Non-significant candidate predictors 201
11.1.6 Summary points on predictor selection 201
11.2 Stepwise selection 202
11.2.1 Stepwise selection variants 202
11.2.2 Stopping rules in stepwise selection 202
11.3 Advantages of stepwise methods 203
11.4 Disadvantages of stepwise methods 204
11.4.1 Instability of selection 204
11.4.2 Testimation: Biased in selected coefficients 206
*11.4.3 Testimation: empirical illustrations 207
11.4.4 Misspecification of variability and p-values 208
11.5 Influence of noise variables 210
11.6 Univariate analyses and model specification 211
11.6.1 Pros and cons of univariate pre-selection 211
*11.6.2 Testing of predictors in a domain 212
11.7 Modern selection methods 212
*11.7.1 Bootstrapping for selection 212
*11.7.2 Bagging and boosting 212
*11.7.3 Bayesian model averaging (BMA) 213
11.7.4 Shrinkage of regression coefficients to zero 213
11.8 Concluding remarks 214
Chapter 12 Assumptions in regression models: Additivity and linearity 217
12.1 Additivity and interaction terms 217
12.1.1 Potential interaction terms to consider 218
12.1.2 Interactions with treatment 218
12.1.3 Other potential interactions 219
*12.1.4 Example: time and survival after valve replacement 220
12.2 Selection, estimation and performance with interaction terms 220
12.2.1 Example: age interactions in GUSTO-I 220
12.2.2 Estimation of interaction terms 221
12.2.3 Better prediction with interaction terms? 222
12.2.4 Summary points 223
12.3 Non-linearity in multivariable analysis 223
12.3.1 Multivariable restricted cubic splines (rcs) 224
12.3.2 Multivariable fractional polynomials (FP) 225
12.3.3 Multivariable splines in gam 225
12.4 Example: non-linearity in testicular cancer case study 226
*12.4.1 Details of multivariable FP and gam analyses 227
*12.4.2 GAM in univariate and multivariable analysis 228
*12.4.3 Predictive performance 229
*12.4.4 R code for non-linear modeling in testicular cancer example 230
12.5 Concluding remarks 230
12.5.1 Recommendations 231
Chapter 13 Modern estimation methods 233
13.1 Predictions from regression and other models 233
*13.1.1 Estimation with other modeling approaches 234
13.2 Shrinkage 234
13.2.1 Uniform shrinkage 235
13.2.2 Uniform shrinkage: illustration 236
13.3 Penalized estimation 236
*13.3.1 Penalized maximum likelihood estimation 237
13.3.2 Penalized ML: illustration 238
*13.3.3 Optimal penalty by bootstrapping 238
13.3.4 Firth regression 239
*13.3.5 Firth regression: illustration 239
*13.4.1 Estimation of a LASSO model 240
13.5 Elastic net 241
*13.5.1 Estimation of Elastic Net model 241
13.6 Performance after shrinkage 242
13.6.1 Shrinkage, penalization, and model selection 242
13.7 Concluding remarks 244
Chapter 14 Estimation with external information 247
Background 247
14.1 Combining literature and individual patient data (IPD) 247
14.1.1 A global prediction model 248
*14.1.2 A global model for traumatic brain injury 249
14.1.3 Developing a local prediction model 249
14.1.4 Adaptation of univariate coefficients 250
*14.1.5 Adaptation method 1 250
*14.1.6 Adaptation method 2 251
*14.1.7 Estimation of adaptation factors 251
*14.1.8 Simulation results 252
14.1.9 Performance of the adapted model 253
14.2 Case study: prediction model for AAA surgical mortality 254
14.2.1 Meta-analysis 254
14.2.2 Individual patient data analysis 255
14.2.3 Adaptation and clinical presentation 256
14.3 Alternative approaches 257
14.3.1 Overall calibration 257
14.3.2 Stacked regressions 257
14.3.3 Bayesian methods: using data priors to regression modeling 257
14.3.4 Example: predicting neonatal death 258
*14.3.5 Example: aneurysm study 258
14.4 Concluding remarks 258Chapter 15 Evaluation of performance 261
15.1 Overall performance measures 261
15.1.1 Explained variation: R2 261
15.1.2 Brier score 262
15.1.3 Performance of testicular cancer prediction model 263
15.3.4 Assessment of moderate calibration 283
15.3.5 Assessment of strong calibration 283
15.3.6 Calibration of survival predictions 284
15.3.7 Example: calibration in testicular cancer prediction model 285
*15.3.8 R code for assessing calibration 286
15.3.9 Calibration and discrimination 286
15.4 Concluding remarks 287
15.4.1 Bibliographic notes 287
Chapter 16 Evaluation of clinical usefulness 289
16.1 Clinical usefulness 289
16.1.1 Intuitive approach to the cutoff 290
16.1.2 Decision-analytic approach: benefit vs harm 290
16.1.3 Accuracy measures for clinical usefulness 291
16.1.4 Decision curve analysis 292
16.1.5 Interpreting net benefit in decision curves 293
16.1.6 Example: clinical usefulness of prediction in testicular cancer 295
16.1.7 Decision curves for testicular cancer example 296
16.1.8 Verification bias and clinical usefulness 297
*16.1.9 R code 298
16.2 Discrimination, calibration, and clinical usefulness 300
16.2.1 Discrimination, calibration, and Net Benefit in the testicular cancer case study 30016.2.2 Aims of prediction models and performance measures 301
16.2.2 Summary points 302
16.3 From prediction models to decision rules 303
16.3.1 Performance of decision rules 303
16.3.2 Treatment benefit in prognostic subgroups 305
16.3.3 Evaluation of classification systems 305
16.4 Concluding remarks 306
Chapter 17 Validation of prediction models 309
17.1 Internal versus external validation, and validity 309
17.1.1 Assessment of internal and external validity 310
17.2 Internal validation techniques 311
17.2.1 Apparent validation 311
17.2.3 Cross-validation 313
17.2.4 Bootstrap validation 314
17.2.5 Internal validation combined with imputation 315
17.3 External validation studies 315
17.3.1 Temporal validation 316
*17.3.2 Example: validation of a model for Lynch syndrome 316
17.3.3 Geographic validation 317
17.3.4 Fully independent validation 319
17.3.5 Reasons for poor validation 320
17.4 Concluding remarks 321
Chapter 18 Presentation formats 323
18.1 Prediction models versus decision rules 323
18.2 Clinical prediction models 325
18.2.1 Regression formulas 325
18.2.2 Confidence intervals for predictions 326
18.2.3 Nomograms 327
18.2.4 Score chart 329
18.2.5 Tables with predictions 330
18.2.6 Specific formats 331
18.2.7 Black box presentations 331
18.3 Case study: clinical prediction model for testicular cancer model 333
18.3.1 Regression formula from logistic model 333
18.3.2 Nomogram 334
*18.3.3 Score chart 334
18.3.4 Summary points 335
18.4 Clinical decision rules 335
18.4.1 Regression tree 335
18.4.2 Score chart rule 335
18.4.3 Survival groups 336
18.4.4 Meta-model 337
18.5 Concluding remarks 338
Part III: Generalizability of prediction models 341
Chapter 19 Patterns of external validity 343
19.1 Determinants of external validity 343
19.1.1 Case-mix 343
19.1.2 Differences in case-mix 343
19.1.3 Differences in regression coefficients 344
19.2.1 Simulation set-up 345
19.2.2 Performance measures 347
19.3 Distribution of predictors 348
19.3.1 More or less severe case-mix according to X 348
*19.3.2 Interpretation of testicular cancer validation 349
19.3.3 More or less heterogeneous case-mix according to X 349
19.3.4 More or less severe case-mix according to Z 350
19.3.5 More or less heterogeneous case-mix according to Z 351
19.4 Distribution of observed outcome y 353
19.5 Coefficients β 354
19.5.1 Coefficient of linear predictor < 1 354
19.5.2 Coefficients β different 355
19.6 Summary of patterns of invalidity 356
19.6.1 Other scenarios of invalidity 35719.7 Reference values for performance 358
19.7.1 Model-based performance: performance if the model is valid 358
19.7.2 Performance with refitting 358
*19.7.3 Examples: testicular cancer and TBI 359
*19.7.4 R code 360
19.8 Limited validation sample size 361
19.8.1 Uncertainty in validation of performance 361
*19.8.2 Estimating standard errors in validation studies 363
19.8.3 Summary points 363
19.9 Design of external validation studies 363
19.9.1 Power of external validation studies 364
*19.9.2 Calculating sample sizes for validation studies 365
19.9.3 Rules for sample size of validation studies 366
19.9.4 Summary points 367
19.10 Concluding remarks 368
Chapter 20 Updating for a new setting 371
20.1 Updating only the intercept 372
20.1.1 Simple updating methods 372
20.2 Approaches to more extensive updating 372
20.2.1 Eight updating methods for predicting binary outcomes 373
20.3 Validation and updating in GUSTO-I 375
20.3.1 Validity of TIMI-II model for GUSTO-I 37620.3.2 Updating the TIMI-II model for GUSTO-I 377
20.3.3 Performance of updated models 378
*20.3.4 R code for updating methods 379
20.4 Shrinkage and updating 379
20.4.1 Shrinkage towards recalibrated values in GUSTO-I 380
*20.4.2 R code for shrinkage and penalization in updating 381
20.4.4 Bayesian updating 382
20.5 Sample size and updating strategy 383
*20.5.1 Simulations of sample size, shrinkage, and updating strategy 384
20.5.2 A closed test for the choice of updating strategy 386
20.6 Validation and updating of tree models 386
20.7 Validation and updating of survival models 388
*20.7.1 Validation of a simple index for non-Hodgkin's lymphoma 388
20.7.2 Updating the prognostic index 389
20.7.3 Recalibration for groups by time points 389
20.7.4 Recalibration with a Cox or Weibull regression model 390
20.7.6 Summary points 391
20.8 Continuous updating 392
*20.8.1 Precision and updating strategy 392
*20.8.2 Continuous updating in GUSTO-I 393
*20.8.3 Other dynamic modeling approaches 394
20.9 Concluding remarks 396
*20.9.1 Further illustrations of updating 397
Chapter 21 Updating for multiple settings 401
21.1 Differences in outcome 401
21.1.1 Testing for calibration-in-the large 401*21.1.2 Illustration of heterogeneity in GUSTO-I 402
21.1.3 Updating for better calibration-in-the large 403
21.1.4 Empirical Bayes estimates 403
*21.1.5 Illustration of updating in GUSTO-I 404
21.1.6 Testing and updating of predictor effects 405
*21.1.7 Heterogeneity of predictor effects in GUSTO-I 405
*21.1.8 R code for random effect analyses in GUSTO-I 405
21.2 Provider profiling 406
21.2.1 Ranking of centers: the expected rank 407
*21.2.2 Example: provider profiling in stroke 408
*21.2.4 Estimation and interpreting differences between centers 409
*21.2.5 Ranking of centers 410
*21.2.6 R code for provider profiling 41121.3 Concluding remarks 412
*21.3.1 Further literature 413
Part IV: Applications 415
Chapter 22 Case study on a prediction of 30-day mortality 417
22.1 GUSTO-I study 417
22.1.1 Acute myocardial infarction 417
*22.1.2 Treatment results from GUSTO-I 418
22.1.3 Prognostic modeling in GUSTO-I 418
22.2 General considerations of model development 421
22.2.1 Research question and intended application 421
22.2.2 Outcome and predictors 421
22.2.3 Study design and analysis 421
22.3 Seven modeling steps in GUSTO-I 423
22.3.1 Preliminary 423
22.3.2 Coding of predictors 423
22.3.3 Model specification 423
22.3.4 Model estimation 423
22.3.5 Model performance 424
22.3.6 Model validation 424
22.3.7 Presentation 425
22.3.8 Predictions 426
22.4 Validity 428
22.4.1 Internal validity: overfitting 42822.4.2 External validity: generalizability 428
22.4.3 Summary points 429
22.5 Translation into clinical practice 429
22.5.1 Score chart for choosing thrombolytic therapy 429
22.5.2 From predictions to decisions 430
22.6 Concluding remarks 432
Chapter 23 Case study on survival analysis: prediction of cardiovascular events 435
23.1 Prognosis in the SMART study 435
*23.1.1 Patients in SMART 436
23.2 General considerations in SMART 438
23.2.1 Research question and intended application 438
23.2.2 Outcome and predictors 438
23.2.3 Study design and analysis 438
23.3 Preliminary modeling steps in the SMART cohort 440
23.3.1 Patterns of missing values 440
23.3.2 Imputation of missing values 441
23.3.3 R code 442
23.4 Coding of predictors 443
23.4.1 Extreme values 443
23.4.2 Transforming continuous predictors 444
23.4.3 Combining predictors with similar effects 445
23.4.4 R code 446
23.5.1 A full model 447
23.5.2 Impact of imputation 449
23.5.3 R code for full model and imputation variants 449
23.6 Model selection and estimation 451
23.6.1 Stepwise selection 451
23.6.2 LASSO for selection with imputed data 452
23.7 Model performance and internal validation 453
23.7.1 Estimation of optimism in performance 453
23.7.2 Model presentation 456
23.7.3 R code for presentations 457
23.8 Concluding remarks 458
Chapter 24 Overall lessons and data sets 461
24.1 Sample size 461
24.1.1 Model selection, estimation, and sample size 462
24.1.2 Calibration improvement by penalization 463
24.1.3 Poorer performance with more predictors 464
24.1.4 Model selection with noise predictors 465
24.1.5 Potential solutions 466
24.1.6 R code for model selection and penalization 466
24.2 Validation 467
24.2.1 Examples of internal and external validation 467
24.3 Subject matter knowledge versus machine learning 468
24.3.1 Exploiting subject matter knowledge 468
24.3.2 Machine learning and Big Data 470
24.4 Reporting of prediction models and risk of bias assessments 470
24.4.1 Reporting guidelines 470
24.4.2 Risk of bias assessment 47224.5 Data sets 473
24.5.1 GUSTO-I prediction models 473
24.5.2 SMART case study 475
24.5.3 Testicular cancer case study 476
24.5.4 Abdominal aortic aneurysm case study 478
24.6 Concluding remarks 481
References 483