Clinical Prediction Models
Preface
1
Introduction
I Part I: Prediction models in medicine
2
Applications of Prediction Models
3
Study Design for Prediction Modeling
4
Statistical Models for Prediction
5
Overfitting and Optimism in Prediction Models
Figures 5.2 to 5.5
Fig 5.2: Noise in estimating 10% mortality per center
Figs 5.3 and 5.4
n=20
n=200
6
Choosing Between Alternative Models
Non-linearity illustrations
Prepare GUSTO data
Anova results for the different fits
Plotting of age effects
Start surgical mortality by age in Medicare
Anova results for the fit of age, with interaction by type of surgery
Plotting of predicted age effects, with interaction by type of surgery; add 95% CI
Plotting of age effects with original data points
II Part II: Developing Valid Prediction Models
7
Missing values
8
Case Study on Dealing with Missing Values
9
Coding of Categorical and Continuous Predictors
Figues 9.1 to 9.6
Fig 9.1: Age linear; add square; rcs in GUSTO-I
Fig 9.2: Impact of number of ST elevations (STE)
Fig 9.3: Non-linearities in small sample (n=751); and full GUSTO-I (n=40,830)
Fig 9.4: Glucose and hb in TBI
Fig 9.5: Non-linear association of glucose
Fig 9.6: Systolic blood pressure in TBI
10
Restrictions on Candidate Predictors
Case-study:
Mortality after surgery for esophageal cancer
Testing the equal weights assumption in a simple sumscore
Discussion
Literature
11
Selection of Main Effects
12
Assumptions in Regression Models - Additivity and Linearity
GUSTO-I interaction analysis
Examine interactions
Fig 12.1
Fig 12.2
Smart coding illustration
Table 12.2 Better predictions?
MFP and other non-linear analyses in n544 data
13
Modern Estimation Methods
14
Estimation with External Information
Learning from external information
14.0.1
A local model with external information: Table 14.1
The data: TBI (n=11022) and AAA (n=238) data sets
The impact study
14.0.2
Descriptives
14.0.3
Analyses for Table 14.2
14.0.4
Estimation of coefficients: naive, stratified and IPD-MA
14.0.5
Estimation of standard errors
14.0.6
Estimation for a specific study, assuming a global model holds
The AAA study
14.0.7
Descriptives
14.0.8
Literature, univariate estimates: Table 14.4
14.0.9
A simple prediction model: Table 14.5
14.0.10
Adaptation of univariate coefficients: Table 14.6
Conclusions
15
Evaluation of Performance
16
Evaluation of Clinical Usefulness
17
Validation of Prediction Models
18
Presentation Formats
Fit logistic models in n544 data set; 6 predictors
Internal validation by bootstrapping
Shrunk model and penalization
Fit penalized model and compare fits
Fig 18.1
Make nomogram from penalized model
Score chart creation with categorized and continuous predictors
Consider making rounded scores from penalized coefs
Scores for continuous predictors
LDH
Study scores for postchemotherapy size
Study scores for reduction in size
Table 18.2
Make a score chart
Fig 18.2
Translate score in probability estimates with 95% CI
End Table 18.2 (score chart) and Fig 18.2 (graphic translation from score to probability)
Categorized coding
Table 18.3
Alternative: make a score from 0 - 5
Fig 18.3
Iso-probability lines
Fig 18.4
Meta-model with tree presentation
Apparent performance of penalized vs rescaled vs simplified models
End comparisons of c statistics
III Part III: Generalizability of Prediction Models
19
Patterns of External Validity
Fig 19.1: a missed predictor Z
Fig 19.2: a missed predictor Z with identical distribution
Fig 19.3: more or less severe cases selected
Fig 19.4: more or less heterogeneous cases selected
Fig 19.5: more or less severe cases selected by Z
Fig 19.6: more or less heterogeneous cases selected by Z
Fig 19.7: Case-control design disturbs calibration
Fig 19.8: Overfitting disturbs discrimination and calibration
Fig 19.9: Different coeficients (model misspecification) disturbs discrimination and calibration
Scenarios: Table 19.4
Uncertainty in validation simulations
20
Updating for a New Setting
21
Updating for Multiple Settings
IV Part IIII: Applications
22
Case Study on a Prediction of 30-Day Mortality
23
Case Study on Survival Analysis: Prediction of Cardiovascular Event
24
Overall Lessons and Data Sets
Published with bookdown
Clinical prediction models
17
Validation of Prediction Models
Chapter 17 additional material upcoming.