Atnaujinkite slapukų nuostatas

Medical Risk Prediction Models: With Ties to Machine Learning [Kietas viršelis]

,
  • Formatas: Hardback, 312 pages, aukštis x plotis: 234x156 mm, weight: 589 g
  • Serija: Chapman & Hall/CRC Biostatistics Series
  • Išleidimo metai: 01-Feb-2021
  • Leidėjas: CRC Press
  • ISBN-10: 113838447X
  • ISBN-13: 9781138384477
Kitos knygos pagal šią temą:
Medical Risk Prediction Models: With Ties to Machine LearningDidesnis paveikslėlis
  • Formatas: Hardback, 312 pages, aukštis x plotis: 234x156 mm, weight: 589 g
  • Serija: Chapman & Hall/CRC Biostatistics Series
  • Išleidimo metai: 01-Feb-2021
  • Leidėjas: CRC Press
  • ISBN-10: 113838447X
  • ISBN-13: 9781138384477
Kitos knygos pagal šią temą:
Pastovi nuoroda: https://www.kriso.lt/db/9781138384477.html
Raktažodžiai:
Medical Risk Prediction Models: With Ties to Machine Learning is a hands-on book for clinicians, epidemiologists, and professional statisticians who need to make or evaluate a statistical prediction model based on data. The subject of the book is the patients individualized probability of a medical event within a given time horizon. Gerds and Kattan describe the mathematical details of making and evaluating a statistical prediction model in a highly pedagogical manner while avoiding mathematical notation. Read this book when you are in doubt about whether a Cox regression model predicts better than a random survival forest.

Features:









All you need to know to correctly make an online risk calculator from scratch.













Discrimination, calibration, and predictive performance with censored data and competing risks.













R-code and illustrative examples.













Interpretation of prediction performance via benchmarks.













Comparison and combination of rival modeling strategies via cross-validation.

Recenzijos

"Two of the top researchers in the field of clinical prediction models have produced a highly innovative book that brings a very technical topic to public grasp by throwing out the formulas and just talking straight from the heart of practical experience. While clinicians and medical residents can now learn how to build, diagnose and validate risk models themselves, all public health researchers, old and new, will reap the benefits and enjoyment from reading this book." ~Donna Ankerst, Technical University of Munich "Two of the top researchers in the field of clinical prediction models have produced a highly innovative book that brings a very technical topic to public grasp by throwing out the formulas and just talking straight from the heart of practical experience. While clinicians and medical residents can now learn how to build, diagnose and validate risk models themselves, all public health researchers, old and new, will reap the benefits and enjoyment from reading this book." ~Donna Ankerst, Technical University of Munich

"Overall, the book offers a well-written, complete and illustrative overview of clinical prediction models with clear stances and directions on the modelling methods, choices and strategies. I find this a very welcome and much needed addition to the literature because prediction is the backbone of medical decision-making; few books are dedicated to modelling strategies and artificial intelligence is ascending in medical research. I thereby highly recommend this book for anyone who would be interested in performing predictive modelling for prognostic or diagnostic research."

-Evangelos I. Kritsotakis, International Society for Clinical Biostatistics, 72, 2021

Foreword xiii
Preface xv
Terminology xvii
Software xxi
1 Why should I care about statistical prediction models?
1(7)
1.1 The many uses of prediction models in medicine
3(1)
1.2 The unique messages of this book
4(2)
1.3 Prognostic factor modeling philosophy
6(1)
1.4 The rest of this book
7(1)
2 I am going to make a prediction model. What do I need to know?
8(39)
2.1 Prediction model framework
8(4)
2.1.1 Target population
8(1)
2.1.2 The time origin
8(1)
2.1.3 The event of interest
8(1)
2.1.4 The prediction time horizon and follow-up
9(1)
2.1.5 Landmarking
10(1)
2.1.6 Risks and risk predictions
10(1)
2.1.7 Classification of risk
11(1)
2.1.8 Predictor variables
11(1)
2.1.9 Checklist
12(1)
2.2 Prediction performance
12(5)
2.2.1 Proper scoring rules
13(1)
2.2.2 Calibration
14(1)
2.2.3 Discrimination
14(1)
2.2.4 Explained variation
15(1)
2.2.5 Variability and uncertainty
15(1)
2.2.6 The interpretation is relative
16(1)
2.2.7 Utility
16(1)
2.2.8 Average versus subgroups
17(1)
2.3 Study design
17(3)
2.3.1 Study design and sources of information
17(1)
2.3.2 Cohort
18(1)
2.3.3 Multi-center study
18(1)
2.3.4 Randomized clinical trial
18(1)
2.3.5 Case-control
19(1)
2.3.6 Given treatment and treatment options
19(1)
2.3.7 Sample size calculation
19(1)
2.4 Data
20(5)
2.4.1 Purpose dataset
20(1)
2.4.2 Data dictionary
21(1)
2.4.3 Measurement error
21(1)
2.4.4 Missing values
21(1)
2.4.5 Censored data
22(1)
2.4.6 Competing risks
22(3)
2.5 Modeling
25(8)
2.5.1 Risk prediction model
25(2)
2.5.2 Risk classifier
27(1)
2.5.3 How is prediction modeling different from statistical inference?
27(2)
2.5.4 Regression model
29(1)
2.5.5 Linear predictor
29(1)
2.5.6 Expert selects the candidate predictors
29(1)
2.5.7 How to select variables for inclusion in the final model
30(1)
2.5.8 All possible interactions
31(1)
2.5.9 Checklist
32(1)
2.5.10 Machine learning
32(1)
2.6 Validation
33(6)
2.6.1 The conventional model
33(1)
2.6.2 Internal and external validation
33(1)
2.6.3 Conditional versus expected performance
34(1)
2.6.4 Cross-validation
34(1)
2.6.5 Data splitting
35(1)
2.6.6 Bootstrap
35(2)
2.6.7 Model checking and goodness of fit
37(1)
2.6.8 Reproducibility
38(1)
2.7 Pitfalls
39(8)
2.7.1 Age as time scale
39(1)
2.7.2 Odds ratios and hazard ratios are not predictions of risks
40(1)
2.7.3 Do not blame the metric
40(2)
2.7.4 Censored data versus competing risks
42(2)
2.7.5 Disease-specific survival
44(1)
2.7.6 Overfitting
44(1)
2.7.7 Data-dependent decisions
44(1)
2.7.8 Balancing data
45(1)
2.7.9 Independent predictor
45(1)
2.7.10 Automated variable selection
45(2)
3 How should I prepare for modeling?
47(15)
3.1 Definition of subjects
47(2)
3.2 Choice of time scale
49(1)
3.3 Pre-selection of predictor variables
50(2)
3.4 Preparation of predictor variables
52(6)
3.4.1 Categorical variables
53(1)
3.4.2 Continuous variables
53(1)
3.4.3 Derived predictor variables
54(1)
3.4.4 Repeated measurements
55(1)
3.4.5 Measurement error
56(1)
3.4.6 Missing values
57(1)
3.5 Preparation of event time outcome
58(4)
3.5.1 Illustration without competing risks
58(1)
3.5.2 Illustration with competing risks
59(1)
3.5.3 Artificial censoring at the prediction time horizon
60(2)
4 I am ready to build a prediction model
62(41)
4.1 Specifying the model type
63(3)
4.1.1 Uncensored binary outcome
63(1)
4.1.2 Right-censored time-to-event outcome (no competing risks)
64(1)
4.1.3 Right-censored time-to-event outcome with competing risks
65(1)
4.2 Benchmark model
66(4)
4.2.1 Uncensored binary outcome
66(1)
4.2.2 Right-censored time-to-event outcome (without competing risks)
67(1)
4.2.3 Right-censored time-to-event with competing risks
68(2)
4.3 Including predictor variables
70(14)
4.3.1 Categorical predictor variables
71(5)
4.3.2 Continuous predictor variables
76(5)
4.3.3 Interaction effects
81(3)
4.4 Modeling strategy
84(3)
4.4.1 Variable selection
84(1)
4.4.2 Conventional model strategy
85(1)
4.4.3 Whether to use a standard regression model or something else
86(1)
4.5 Advanced topics
87(2)
4.5.1 How to prevent overfitting the data
87(1)
4.5.2 How to deal with missing values
88(1)
4.5.3 How to deal with non-converging models
89(1)
4.6 What you should put in your manuscript
89(11)
4.6.1 Baseline tables
89(1)
4.6.2 Follow-up tables
90(1)
4.6.3 Regression tables
91(3)
4.6.4 Risk plots
94(2)
4.6.5 Nomograms
96(4)
4.7 Deployment
100(3)
4.7.1 Risk charts
100(1)
4.7.2 Internet calculator
100(1)
4.7.3 Cost-benefit analysis (waiting lists)
100(3)
5 Does my model predict accurately?
103(47)
5.1 Model assessment roadmap
104(3)
5.1.1 Visualization of the predictions
104(1)
5.1.2 Calculation of model performance
105(1)
5.1.3 Visualization of model performance
106(1)
5.2 Uncensored binary outcome
107(18)
5.2.1 Distribution of the predicted risks
107(6)
5.2.2 Brier score
113(3)
5.2.3 AUC
116(2)
5.2.4 Calibration curves
118(7)
5.3 Right-censored time-to-event outcome (without competing risks)
125(11)
5.3.1 Distribution of the predicted risks
126(2)
5.3.2 Brier score with censored data
128(3)
5.3.3 Time-dependent AUC for censored data
131(3)
5.3.4 Calibration curve for censored data
134(2)
5.4 Competing risks
136(11)
5.4.1 Distribution of the predicted risks
136(2)
5.4.2 Brier score with competing risks
138(4)
5.4.3 Time-dependent AUC for competing risks
142(1)
5.4.4 Calibration curve for competing risks
143(4)
5.5 The Index of Prediction Accuracy (IPA)
147(1)
5.6 Choice of prediction time horizon
148(1)
5.7 Time-dependent prediction performance
149(1)
6 How do I decide between rival models?
150(30)
6.1 Model comparison roadmap
151(1)
6.2 Analysis of rival prediction models
151(18)
6.2.1 Uncensored binary outcome
152(4)
6.2.2 Right-censored time-to-event outcome (without competing risks)
156(9)
6.2.3 Competing risks
165(4)
6.3 Clinically relevant change of prediction
169(6)
6.4 Does a new marker improve prediction?
175(5)
6.4.1 Many new predictors
179(1)
6.4.2 Updating a subject's prediction
179(1)
7 What would make me an expert?
180(44)
7.1 Multiple cohorts / Multi-center studies
180(2)
7.2 The role of treatment for making a prediction model
182(4)
7.2.1 Modeling treatment
183(2)
7.2.2 Comparative effectiveness tables
185(1)
7.3 Learning curve paradigm
186(1)
7.4 Internal validation (data splitting)
187(18)
7.4.1 Single split
187(6)
7.4.2 Calendar split
193(1)
7.4.3 Multiple splits (cross-validation)
194(7)
7.4.4 Dilemma of internal validation
201(1)
7.4.5 The apparent and the 632+ estimator
202(1)
7.4.6 Tips and tricks
202(3)
7.5 Missing values
205(14)
7.5.1 Missing values in the learning data
207(8)
7.5.2 Missing values in the validation data
215(4)
7.6 Time-varying coefficient models
219(1)
7.7 Time-varying predictor variables
220(4)
8 Can't the computer just take care of all of this?
224(33)
8.1 Zero layers of cross-validation
225(7)
8.1.1 What may happen if you do not look at the data
225(2)
8.1.2 Unsupervised modeling steps
227(5)
8.1.3 Final model
232(1)
8.2 One layer of cross-validation
232(8)
8.2.1 Penalized regression
233(7)
8.2.2 Supervised spline selection
240(1)
8.3 Machine learning (two levels of cross-validation)
240(10)
8.3.1 Random forest
243(5)
8.3.2 Deep learning and artificial neural networks
248(2)
8.4 The super learner
250(7)
9 Things you might have expected in our book
257(11)
9.1 Threshold selection for decision making
257(1)
9.2 Number of events per variable
258(1)
9.3 Confidence intervals for predicted probabilities
258(1)
9.4 Models developed from case-control data
259(1)
9.5 Hosmer-Lemeshow test
259(1)
9.6 Backward elimination and stepwise selection
260(1)
9.7 Rank correlation (c-index) for survival outcome
260(1)
9.8 Integrated Brier score
261(1)
9.9 Net reclassification index and the integrated discrimination improvement
261(1)
9.10 Re-classification tables
262(4)
9.11 Boxplots of rival models conditional on the outcome
266(2)
Bibliography 268(16)
Index 284
Thomas A. Gerds is professor at the biostatistics unit at the University of Copenhagen. He is affiliated with the Danish Heart Foundation. He is author of several R-packages on CRAN and has taught statistics courses to non-statisticians for many years.

Michael Kattan is a highly cited author and Chair of the Department of Quantitative Health Sciences at Cleveland Clinic. He is a Fellow of the American Statistical Association and has received two awards from the Society for Medical Decision Making: the Eugene L. Saenger Award for Distinguished Service, and the John M. Eisenberg Award for Practical Application of Medical Decision Making Research.