SlideShare a Scribd company logo
1 of 35
June 22, 2022
Towards better performance of prediction models:
updating and extension with markers
: updating and marker
Ewout W. Steyerberg, PhD
Professor of Clinical Biostatistics and
Medical Decision Making
Dept of Biomedical Data Sciences
Leiden University Medical Center
Thanks to many, including Ben van Calster, Leuven
Key question: how to improve prediction models?
1. Better development + validation
a) Sample size
b) Methods
2. Updating of existing models
a) Local settings
b) Continuous learning
3. Extension with markers
4. Machine learning
22-Jun-22
2 Insert > Header & footer
Subquestion 1: How to assess model improvement?
1. Calibration (A + B, intercept + slope)
2. Discrimination (C, concordance)
3. Clinical usefulness (D, decision-analytic)
22-Jun-22
3 Insert > Header & footer
Multiple reviews
22-Jun-22
4 Insert > Header & footer
Calibration
22-Jun-22
5 Insert > Header & footer
BMJ style, understandable calibration plots
22-Jun-22
6 Insert > Header & footer
BMJ. 2012 Jun 21;344:e4181. doi: 10.1136/bmj.e4181.
With uncertainty
22-Jun-22
7 Insert > Header & footer
val.prob.ci.2() in R, based on Harrell’s val.prob() function
Decision-analytic perspectives
If we are serious about “using different thresholds that allow the
operator of the model to trade-off concerns in the errors made by
the model” we need a decision-analytic perspective
1. Define threshold
2. Evaluate quality of classification
 Decision Curve
22-Jun-22
8 Insert > Header & footer
BMJ style, understandable decision curves
22-Jun-22
9 Insert > Header & footer
22-Jun-22
10 Insert > Header & footer
Med Decis Making 2006;26:565–574
3 statements on Decision Curve Analysis (DCA)
1. A classic idea (1884 or older)
2. A good link with clinical context:
benefit of treatment vs harm by overtreatment to define thresholds
3. A good graphic because thresholds are ‘subjective’
22-Jun-22
11 Insert > Header & footer
Youden index and Net Benefit; Peirce, Science 1884
Event
Test: answer + –
+ aa ab
– ba bb
TP FP
sens spec
Youden index: sens + spec – 1
Vickers & Elkin, MDM 2006
22-Jun-22
13 Insert > Header & footer
Benefit: a – c
Harm: d – b
Odds of threshold:
Harm / Benefit
Net Benefit
Net Benefit = (TP – w FP)/N
w = harm / benefit ratio = threshold/ (1 – threshold)
• e.g.: threshold 50%: w = .5/.5=1;
threshold 20%: w=.2/.8=1/4
“Fraction of true-positive classifications,
penalized for false-positive classifications”
BMJ 2016;352:i6 doi: 10.1136/bmj.i6.
Illustration for CVD risk
22-Jun-22
15 Insert > Header & footer
BMJ style, understandable text
22-Jun-22
16 Insert > Header & footer
BMJ style, understandable decision curves
22-Jun-22
17 Insert > Header & footer
Key question: how to improve prediction models?
1. Better development + validation
a) Sample size
b) Methods
2. Updating of existing models
a) Local settings
b) Continuous learning
3. Extension with markers
22-Jun-22
18 Insert > Header & footer
Subquestion 2: How to balance global vs local models?
Prediction models need updating to local settings;
can we entertain the idea of a ‘global model’?
1. Global: baseline risk + predictor effects
2. Recalibrated: local baseline risk + global predictor effects
3. Refitted: local baseline risk + local predictor effects
22-Jun-22
19 Insert > Header & footer
Examples on updating
Single validation set
Robust approach + closed testing
22-Jun-22
20 Insert > Header & footer
Start “off the shelf”, update continuously
22-Jun-22
21 Insert > Header & footer
Examples on updating
Single validation set
Classic: approach SiM 2004; closed testing
Dynamic in calendar time
Multiple validation sets
Assess heterogeneity
a) Global model?
b) Fair representation of uncertainty?
22-Jun-22
22 Insert > Header & footer
22-Jun-22
23 Insert > Header & footer
Another advertisement: internal-external validation
22-Jun-22
24 Insert > Header & footer
Key question: how to improve prediction models?
1. Better development + validation
a) Sample size
b) Methods
2. Updating of existing models
a) Local settings
b) Continuous learning
3. Extension with markers
22-Jun-22
25 Insert > Header & footer
Cardiovascular risk without / with HDL
22-Jun-22
26 Insert > Header & footer
Incremental value of marker
• Define a reference model, add marker to evaluate incremental value
• Regression coefficient problematic (scaling); p-value assumed to be low
• Increase in AUC / c statistic usually small (typically: +0.01)
 Push to look beyond AUC: reclassification
22-Jun-22
27 Insert > Header & footer
29
7
173
174
22/183=12%
1/3081=.03%
NRI and delta(AUC) for binary classification
NRI = delta(sens) + delta(spec)
AUC for binary classification = (sens + spec) / 2
delta(AUC) = (delta(sens) + delta(spec)) / 2
NRI = 2 x delta(AUC)
NRI has ‘absurd’ weighting?
Decision-analytic variants
Weighted NRI
Delta NB (Vickers)
Delta Relative Utility (Baker) / standardized NB (Pepe / Janes)
22-Jun-22
33 Insert > Header & footer
Marker evaluation
NRI was a historical mistake?
Net benefit to the rescue?
22-Jun-22
34 Insert > Header & footer
Summary 20 June 2022
1. Prediction modeling research challenging
2. Performance assessment: calibration and Net Benefit
3. Improving performance:
a) Updating
b) Markers
c) Machine learning
22-Jun-22
35 Insert > Header & footer

More Related Content

What's hot

Measuring clinical utility: uncertainty in Net Benefit
Measuring clinical utility: uncertainty in Net BenefitMeasuring clinical utility: uncertainty in Net Benefit
Measuring clinical utility: uncertainty in Net Benefit
Laure Wynants
 

What's hot (20)

Shrinkage in medical prediction: the poor man’s solution for an inadequate sa...
Shrinkage in medical prediction: the poor man’s solution for an inadequate sa...Shrinkage in medical prediction: the poor man’s solution for an inadequate sa...
Shrinkage in medical prediction: the poor man’s solution for an inadequate sa...
 
P-values in crisis
P-values in crisisP-values in crisis
P-values in crisis
 
Development and evaluation of prediction models: pitfalls and solutions
Development and evaluation of prediction models: pitfalls and solutionsDevelopment and evaluation of prediction models: pitfalls and solutions
Development and evaluation of prediction models: pitfalls and solutions
 
Measuring clinical utility: uncertainty in Net Benefit
Measuring clinical utility: uncertainty in Net BenefitMeasuring clinical utility: uncertainty in Net Benefit
Measuring clinical utility: uncertainty in Net Benefit
 
Prognosis-based medicine: merits and pitfalls of forecasting patient health
Prognosis-based medicine: merits and pitfalls of forecasting patient healthPrognosis-based medicine: merits and pitfalls of forecasting patient health
Prognosis-based medicine: merits and pitfalls of forecasting patient health
 
Correcting for missing data, measurement error and confounding
Correcting for missing data, measurement error and confoundingCorrecting for missing data, measurement error and confounding
Correcting for missing data, measurement error and confounding
 
ML and AI: a blessing and curse for statisticians and medical doctors
ML and AI: a blessing and curse forstatisticians and medical doctorsML and AI: a blessing and curse forstatisticians and medical doctors
ML and AI: a blessing and curse for statisticians and medical doctors
 
Rage against the machine learning 2023
Rage against the machine learning 2023Rage against the machine learning 2023
Rage against the machine learning 2023
 
How to calculate sample size for different study
How to calculate sample size for different studyHow to calculate sample size for different study
How to calculate sample size for different study
 
Predictimands
PredictimandsPredictimands
Predictimands
 
26 shristi k.c. journal club presentation
26 shristi k.c. journal club presentation26 shristi k.c. journal club presentation
26 shristi k.c. journal club presentation
 
Introduction to prediction modelling - Berlin 2018 - Part II
Introduction to prediction modelling - Berlin 2018 - Part IIIntroduction to prediction modelling - Berlin 2018 - Part II
Introduction to prediction modelling - Berlin 2018 - Part II
 
Machine learning in medicine: calm down
Machine learning in medicine: calm downMachine learning in medicine: calm down
Machine learning in medicine: calm down
 
Introduction to prediction modelling - Berlin 2018 - Part I
Introduction to prediction modelling - Berlin 2018 - Part IIntroduction to prediction modelling - Berlin 2018 - Part I
Introduction to prediction modelling - Berlin 2018 - Part I
 
Clinical prediction models: development, validation and beyond
Clinical prediction models:development, validation and beyondClinical prediction models:development, validation and beyond
Clinical prediction models: development, validation and beyond
 
Research Ethics and Ethics for Health Informaticians (November 15, 2021)
Research Ethics and Ethics for Health Informaticians (November 15, 2021)Research Ethics and Ethics for Health Informaticians (November 15, 2021)
Research Ethics and Ethics for Health Informaticians (November 15, 2021)
 
Algorithm based medicine: old statistics wine in new machine learning bottles?
Algorithm based medicine: old statistics wine in new machine learning bottles?Algorithm based medicine: old statistics wine in new machine learning bottles?
Algorithm based medicine: old statistics wine in new machine learning bottles?
 
A gentle introduction to AI for medicine
A gentle introduction to AI for medicineA gentle introduction to AI for medicine
A gentle introduction to AI for medicine
 
Meta analysis: Made Easy with Example from RevMan
Meta analysis: Made Easy with Example from RevManMeta analysis: Made Easy with Example from RevMan
Meta analysis: Made Easy with Example from RevMan
 
Uncertainty in AI
Uncertainty in AIUncertainty in AI
Uncertainty in AI
 

Similar to Prediction research Twente 22June22 sel.pptx

Download-manuals-surface water-software-47basicstatistics
 Download-manuals-surface water-software-47basicstatistics Download-manuals-surface water-software-47basicstatistics
Download-manuals-surface water-software-47basicstatistics
hydrologyproject001
 
Download-manuals-surface water-software-47basicstatistics
 Download-manuals-surface water-software-47basicstatistics Download-manuals-surface water-software-47basicstatistics
Download-manuals-surface water-software-47basicstatistics
hydrologyproject0
 
Download-manuals-surface water-software-47basicstatistics
 Download-manuals-surface water-software-47basicstatistics Download-manuals-surface water-software-47basicstatistics
Download-manuals-surface water-software-47basicstatistics
hydrologyproject0
 
Download-manuals-surface water-waterlevel-30howtovalidateratingcurve
 Download-manuals-surface water-waterlevel-30howtovalidateratingcurve Download-manuals-surface water-waterlevel-30howtovalidateratingcurve
Download-manuals-surface water-waterlevel-30howtovalidateratingcurve
hydrologyproject001
 
Probability Forecasting - a Machine Learning Perspective
Probability Forecasting - a Machine Learning PerspectiveProbability Forecasting - a Machine Learning Perspective
Probability Forecasting - a Machine Learning Perspective
butest
 

Similar to Prediction research Twente 22June22 sel.pptx (20)

Prediction research: perspectives on performance Stanford 19May22.pptx
Prediction research: perspectives on performance Stanford 19May22.pptxPrediction research: perspectives on performance Stanford 19May22.pptx
Prediction research: perspectives on performance Stanford 19May22.pptx
 
Petx I-Corps@NIH 121014
Petx I-Corps@NIH 121014Petx I-Corps@NIH 121014
Petx I-Corps@NIH 121014
 
Reducing Radiation Dose in CT Chest for Pulmonary Embolism (P.E) Studies - (QIP)
Reducing Radiation Dose in CT Chest for Pulmonary Embolism (P.E) Studies - (QIP)Reducing Radiation Dose in CT Chest for Pulmonary Embolism (P.E) Studies - (QIP)
Reducing Radiation Dose in CT Chest for Pulmonary Embolism (P.E) Studies - (QIP)
 
IRJET- Detection of Chronic Kidney Disease using Machine Learning in the R-En...
IRJET- Detection of Chronic Kidney Disease using Machine Learning in the R-En...IRJET- Detection of Chronic Kidney Disease using Machine Learning in the R-En...
IRJET- Detection of Chronic Kidney Disease using Machine Learning in the R-En...
 
Download-manuals-surface water-software-47basicstatistics
 Download-manuals-surface water-software-47basicstatistics Download-manuals-surface water-software-47basicstatistics
Download-manuals-surface water-software-47basicstatistics
 
Download-manuals-surface water-software-47basicstatistics
 Download-manuals-surface water-software-47basicstatistics Download-manuals-surface water-software-47basicstatistics
Download-manuals-surface water-software-47basicstatistics
 
Download-manuals-surface water-software-47basicstatistics
 Download-manuals-surface water-software-47basicstatistics Download-manuals-surface water-software-47basicstatistics
Download-manuals-surface water-software-47basicstatistics
 
Quality By Design
Quality By DesignQuality By Design
Quality By Design
 
Earned Value
Earned ValueEarned Value
Earned Value
 
Results-based payment (RBP): Who should be paid, and for what?
Results-based payment (RBP): Who should be paid, and for what?Results-based payment (RBP): Who should be paid, and for what?
Results-based payment (RBP): Who should be paid, and for what?
 
S-Curve Presentation- 20130529
S-Curve Presentation- 20130529S-Curve Presentation- 20130529
S-Curve Presentation- 20130529
 
Clinical Prediction Rules
Clinical Prediction RulesClinical Prediction Rules
Clinical Prediction Rules
 
Kommunikasjon: A tool for managing product quality
Kommunikasjon: A tool for managing product qualityKommunikasjon: A tool for managing product quality
Kommunikasjon: A tool for managing product quality
 
Machine learning Models for classification of cushing’s syndrome using retros...
Machine learning Models for classification of cushing’s syndrome using retros...Machine learning Models for classification of cushing’s syndrome using retros...
Machine learning Models for classification of cushing’s syndrome using retros...
 
A STUDY OF SUPPLY CHAIN MANAGEMENT IN FOOD INDUSTRY
A STUDY OF SUPPLY CHAIN MANAGEMENT IN FOOD INDUSTRYA STUDY OF SUPPLY CHAIN MANAGEMENT IN FOOD INDUSTRY
A STUDY OF SUPPLY CHAIN MANAGEMENT IN FOOD INDUSTRY
 
[2012] Empirical Evaluation on FBD Model-Based Test Coverage Criteria using M...
[2012] Empirical Evaluation on FBD Model-Based Test Coverage Criteria using M...[2012] Empirical Evaluation on FBD Model-Based Test Coverage Criteria using M...
[2012] Empirical Evaluation on FBD Model-Based Test Coverage Criteria using M...
 
gg_fall_15
gg_fall_15gg_fall_15
gg_fall_15
 
Download-manuals-surface water-waterlevel-30howtovalidateratingcurve
 Download-manuals-surface water-waterlevel-30howtovalidateratingcurve Download-manuals-surface water-waterlevel-30howtovalidateratingcurve
Download-manuals-surface water-waterlevel-30howtovalidateratingcurve
 
Diagnoza +
Diagnoza +Diagnoza +
Diagnoza +
 
Probability Forecasting - a Machine Learning Perspective
Probability Forecasting - a Machine Learning PerspectiveProbability Forecasting - a Machine Learning Perspective
Probability Forecasting - a Machine Learning Perspective
 

Recently uploaded

In Riyadh ((+919101817206)) Cytotec kit @ Abortion Pills Saudi Arabia
In Riyadh ((+919101817206)) Cytotec kit @ Abortion Pills Saudi ArabiaIn Riyadh ((+919101817206)) Cytotec kit @ Abortion Pills Saudi Arabia
In Riyadh ((+919101817206)) Cytotec kit @ Abortion Pills Saudi Arabia
ahmedjiabur940
 
Top profile Call Girls In dimapur [ 7014168258 ] Call Me For Genuine Models W...
Top profile Call Girls In dimapur [ 7014168258 ] Call Me For Genuine Models W...Top profile Call Girls In dimapur [ 7014168258 ] Call Me For Genuine Models W...
Top profile Call Girls In dimapur [ 7014168258 ] Call Me For Genuine Models W...
gajnagarg
 
Top profile Call Girls In Bihar Sharif [ 7014168258 ] Call Me For Genuine Mod...
Top profile Call Girls In Bihar Sharif [ 7014168258 ] Call Me For Genuine Mod...Top profile Call Girls In Bihar Sharif [ 7014168258 ] Call Me For Genuine Mod...
Top profile Call Girls In Bihar Sharif [ 7014168258 ] Call Me For Genuine Mod...
nirzagarg
 
PLE-statistics document for primary schs
PLE-statistics document for primary schsPLE-statistics document for primary schs
PLE-statistics document for primary schs
cnajjemba
 
Top profile Call Girls In Chandrapur [ 7014168258 ] Call Me For Genuine Model...
Top profile Call Girls In Chandrapur [ 7014168258 ] Call Me For Genuine Model...Top profile Call Girls In Chandrapur [ 7014168258 ] Call Me For Genuine Model...
Top profile Call Girls In Chandrapur [ 7014168258 ] Call Me For Genuine Model...
gajnagarg
 
Top profile Call Girls In Purnia [ 7014168258 ] Call Me For Genuine Models We...
Top profile Call Girls In Purnia [ 7014168258 ] Call Me For Genuine Models We...Top profile Call Girls In Purnia [ 7014168258 ] Call Me For Genuine Models We...
Top profile Call Girls In Purnia [ 7014168258 ] Call Me For Genuine Models We...
nirzagarg
 
Top profile Call Girls In Satna [ 7014168258 ] Call Me For Genuine Models We ...
Top profile Call Girls In Satna [ 7014168258 ] Call Me For Genuine Models We ...Top profile Call Girls In Satna [ 7014168258 ] Call Me For Genuine Models We ...
Top profile Call Girls In Satna [ 7014168258 ] Call Me For Genuine Models We ...
nirzagarg
 
怎样办理伦敦大学城市学院毕业证(CITY毕业证书)成绩单学校原版复制
怎样办理伦敦大学城市学院毕业证(CITY毕业证书)成绩单学校原版复制怎样办理伦敦大学城市学院毕业证(CITY毕业证书)成绩单学校原版复制
怎样办理伦敦大学城市学院毕业证(CITY毕业证书)成绩单学校原版复制
vexqp
 
Top profile Call Girls In Hapur [ 7014168258 ] Call Me For Genuine Models We ...
Top profile Call Girls In Hapur [ 7014168258 ] Call Me For Genuine Models We ...Top profile Call Girls In Hapur [ 7014168258 ] Call Me For Genuine Models We ...
Top profile Call Girls In Hapur [ 7014168258 ] Call Me For Genuine Models We ...
nirzagarg
 
Top profile Call Girls In Tumkur [ 7014168258 ] Call Me For Genuine Models We...
Top profile Call Girls In Tumkur [ 7014168258 ] Call Me For Genuine Models We...Top profile Call Girls In Tumkur [ 7014168258 ] Call Me For Genuine Models We...
Top profile Call Girls In Tumkur [ 7014168258 ] Call Me For Genuine Models We...
nirzagarg
 
Gartner's Data Analytics Maturity Model.pptx
Gartner's Data Analytics Maturity Model.pptxGartner's Data Analytics Maturity Model.pptx
Gartner's Data Analytics Maturity Model.pptx
chadhar227
 
Top profile Call Girls In Begusarai [ 7014168258 ] Call Me For Genuine Models...
Top profile Call Girls In Begusarai [ 7014168258 ] Call Me For Genuine Models...Top profile Call Girls In Begusarai [ 7014168258 ] Call Me For Genuine Models...
Top profile Call Girls In Begusarai [ 7014168258 ] Call Me For Genuine Models...
nirzagarg
 
Lecture_2_Deep_Learning_Overview-newone1
Lecture_2_Deep_Learning_Overview-newone1Lecture_2_Deep_Learning_Overview-newone1
Lecture_2_Deep_Learning_Overview-newone1
ranjankumarbehera14
 

Recently uploaded (20)

Digital Advertising Lecture for Advanced Digital & Social Media Strategy at U...
Digital Advertising Lecture for Advanced Digital & Social Media Strategy at U...Digital Advertising Lecture for Advanced Digital & Social Media Strategy at U...
Digital Advertising Lecture for Advanced Digital & Social Media Strategy at U...
 
Discover Why Less is More in B2B Research
Discover Why Less is More in B2B ResearchDiscover Why Less is More in B2B Research
Discover Why Less is More in B2B Research
 
Ranking and Scoring Exercises for Research
Ranking and Scoring Exercises for ResearchRanking and Scoring Exercises for Research
Ranking and Scoring Exercises for Research
 
In Riyadh ((+919101817206)) Cytotec kit @ Abortion Pills Saudi Arabia
In Riyadh ((+919101817206)) Cytotec kit @ Abortion Pills Saudi ArabiaIn Riyadh ((+919101817206)) Cytotec kit @ Abortion Pills Saudi Arabia
In Riyadh ((+919101817206)) Cytotec kit @ Abortion Pills Saudi Arabia
 
Top profile Call Girls In dimapur [ 7014168258 ] Call Me For Genuine Models W...
Top profile Call Girls In dimapur [ 7014168258 ] Call Me For Genuine Models W...Top profile Call Girls In dimapur [ 7014168258 ] Call Me For Genuine Models W...
Top profile Call Girls In dimapur [ 7014168258 ] Call Me For Genuine Models W...
 
5CL-ADBA,5cladba, Chinese supplier, safety is guaranteed
5CL-ADBA,5cladba, Chinese supplier, safety is guaranteed5CL-ADBA,5cladba, Chinese supplier, safety is guaranteed
5CL-ADBA,5cladba, Chinese supplier, safety is guaranteed
 
Top profile Call Girls In Bihar Sharif [ 7014168258 ] Call Me For Genuine Mod...
Top profile Call Girls In Bihar Sharif [ 7014168258 ] Call Me For Genuine Mod...Top profile Call Girls In Bihar Sharif [ 7014168258 ] Call Me For Genuine Mod...
Top profile Call Girls In Bihar Sharif [ 7014168258 ] Call Me For Genuine Mod...
 
PLE-statistics document for primary schs
PLE-statistics document for primary schsPLE-statistics document for primary schs
PLE-statistics document for primary schs
 
Top profile Call Girls In Chandrapur [ 7014168258 ] Call Me For Genuine Model...
Top profile Call Girls In Chandrapur [ 7014168258 ] Call Me For Genuine Model...Top profile Call Girls In Chandrapur [ 7014168258 ] Call Me For Genuine Model...
Top profile Call Girls In Chandrapur [ 7014168258 ] Call Me For Genuine Model...
 
Top profile Call Girls In Purnia [ 7014168258 ] Call Me For Genuine Models We...
Top profile Call Girls In Purnia [ 7014168258 ] Call Me For Genuine Models We...Top profile Call Girls In Purnia [ 7014168258 ] Call Me For Genuine Models We...
Top profile Call Girls In Purnia [ 7014168258 ] Call Me For Genuine Models We...
 
Top profile Call Girls In Satna [ 7014168258 ] Call Me For Genuine Models We ...
Top profile Call Girls In Satna [ 7014168258 ] Call Me For Genuine Models We ...Top profile Call Girls In Satna [ 7014168258 ] Call Me For Genuine Models We ...
Top profile Call Girls In Satna [ 7014168258 ] Call Me For Genuine Models We ...
 
Data Analyst Tasks to do the internship.pdf
Data Analyst Tasks to do the internship.pdfData Analyst Tasks to do the internship.pdf
Data Analyst Tasks to do the internship.pdf
 
怎样办理伦敦大学城市学院毕业证(CITY毕业证书)成绩单学校原版复制
怎样办理伦敦大学城市学院毕业证(CITY毕业证书)成绩单学校原版复制怎样办理伦敦大学城市学院毕业证(CITY毕业证书)成绩单学校原版复制
怎样办理伦敦大学城市学院毕业证(CITY毕业证书)成绩单学校原版复制
 
Aspirational Block Program Block Syaldey District - Almora
Aspirational Block Program Block Syaldey District - AlmoraAspirational Block Program Block Syaldey District - Almora
Aspirational Block Program Block Syaldey District - Almora
 
Top profile Call Girls In Hapur [ 7014168258 ] Call Me For Genuine Models We ...
Top profile Call Girls In Hapur [ 7014168258 ] Call Me For Genuine Models We ...Top profile Call Girls In Hapur [ 7014168258 ] Call Me For Genuine Models We ...
Top profile Call Girls In Hapur [ 7014168258 ] Call Me For Genuine Models We ...
 
Top profile Call Girls In Tumkur [ 7014168258 ] Call Me For Genuine Models We...
Top profile Call Girls In Tumkur [ 7014168258 ] Call Me For Genuine Models We...Top profile Call Girls In Tumkur [ 7014168258 ] Call Me For Genuine Models We...
Top profile Call Girls In Tumkur [ 7014168258 ] Call Me For Genuine Models We...
 
Gartner's Data Analytics Maturity Model.pptx
Gartner's Data Analytics Maturity Model.pptxGartner's Data Analytics Maturity Model.pptx
Gartner's Data Analytics Maturity Model.pptx
 
Vadodara 💋 Call Girl 7737669865 Call Girls in Vadodara Escort service book now
Vadodara 💋 Call Girl 7737669865 Call Girls in Vadodara Escort service book nowVadodara 💋 Call Girl 7737669865 Call Girls in Vadodara Escort service book now
Vadodara 💋 Call Girl 7737669865 Call Girls in Vadodara Escort service book now
 
Top profile Call Girls In Begusarai [ 7014168258 ] Call Me For Genuine Models...
Top profile Call Girls In Begusarai [ 7014168258 ] Call Me For Genuine Models...Top profile Call Girls In Begusarai [ 7014168258 ] Call Me For Genuine Models...
Top profile Call Girls In Begusarai [ 7014168258 ] Call Me For Genuine Models...
 
Lecture_2_Deep_Learning_Overview-newone1
Lecture_2_Deep_Learning_Overview-newone1Lecture_2_Deep_Learning_Overview-newone1
Lecture_2_Deep_Learning_Overview-newone1
 

Prediction research Twente 22June22 sel.pptx

  • 1. June 22, 2022 Towards better performance of prediction models: updating and extension with markers : updating and marker Ewout W. Steyerberg, PhD Professor of Clinical Biostatistics and Medical Decision Making Dept of Biomedical Data Sciences Leiden University Medical Center Thanks to many, including Ben van Calster, Leuven
  • 2. Key question: how to improve prediction models? 1. Better development + validation a) Sample size b) Methods 2. Updating of existing models a) Local settings b) Continuous learning 3. Extension with markers 4. Machine learning 22-Jun-22 2 Insert > Header & footer
  • 3. Subquestion 1: How to assess model improvement? 1. Calibration (A + B, intercept + slope) 2. Discrimination (C, concordance) 3. Clinical usefulness (D, decision-analytic) 22-Jun-22 3 Insert > Header & footer
  • 6. BMJ style, understandable calibration plots 22-Jun-22 6 Insert > Header & footer BMJ. 2012 Jun 21;344:e4181. doi: 10.1136/bmj.e4181.
  • 7. With uncertainty 22-Jun-22 7 Insert > Header & footer val.prob.ci.2() in R, based on Harrell’s val.prob() function
  • 8. Decision-analytic perspectives If we are serious about “using different thresholds that allow the operator of the model to trade-off concerns in the errors made by the model” we need a decision-analytic perspective 1. Define threshold 2. Evaluate quality of classification  Decision Curve 22-Jun-22 8 Insert > Header & footer
  • 9. BMJ style, understandable decision curves 22-Jun-22 9 Insert > Header & footer
  • 10. 22-Jun-22 10 Insert > Header & footer Med Decis Making 2006;26:565–574
  • 11. 3 statements on Decision Curve Analysis (DCA) 1. A classic idea (1884 or older) 2. A good link with clinical context: benefit of treatment vs harm by overtreatment to define thresholds 3. A good graphic because thresholds are ‘subjective’ 22-Jun-22 11 Insert > Header & footer
  • 12. Youden index and Net Benefit; Peirce, Science 1884 Event Test: answer + – + aa ab – ba bb TP FP sens spec Youden index: sens + spec – 1
  • 13. Vickers & Elkin, MDM 2006 22-Jun-22 13 Insert > Header & footer Benefit: a – c Harm: d – b Odds of threshold: Harm / Benefit
  • 14. Net Benefit Net Benefit = (TP – w FP)/N w = harm / benefit ratio = threshold/ (1 – threshold) • e.g.: threshold 50%: w = .5/.5=1; threshold 20%: w=.2/.8=1/4 “Fraction of true-positive classifications, penalized for false-positive classifications” BMJ 2016;352:i6 doi: 10.1136/bmj.i6.
  • 15. Illustration for CVD risk 22-Jun-22 15 Insert > Header & footer
  • 16. BMJ style, understandable text 22-Jun-22 16 Insert > Header & footer
  • 17. BMJ style, understandable decision curves 22-Jun-22 17 Insert > Header & footer
  • 18. Key question: how to improve prediction models? 1. Better development + validation a) Sample size b) Methods 2. Updating of existing models a) Local settings b) Continuous learning 3. Extension with markers 22-Jun-22 18 Insert > Header & footer
  • 19. Subquestion 2: How to balance global vs local models? Prediction models need updating to local settings; can we entertain the idea of a ‘global model’? 1. Global: baseline risk + predictor effects 2. Recalibrated: local baseline risk + global predictor effects 3. Refitted: local baseline risk + local predictor effects 22-Jun-22 19 Insert > Header & footer
  • 20. Examples on updating Single validation set Robust approach + closed testing 22-Jun-22 20 Insert > Header & footer
  • 21. Start “off the shelf”, update continuously 22-Jun-22 21 Insert > Header & footer
  • 22. Examples on updating Single validation set Classic: approach SiM 2004; closed testing Dynamic in calendar time Multiple validation sets Assess heterogeneity a) Global model? b) Fair representation of uncertainty? 22-Jun-22 22 Insert > Header & footer
  • 23. 22-Jun-22 23 Insert > Header & footer
  • 24. Another advertisement: internal-external validation 22-Jun-22 24 Insert > Header & footer
  • 25. Key question: how to improve prediction models? 1. Better development + validation a) Sample size b) Methods 2. Updating of existing models a) Local settings b) Continuous learning 3. Extension with markers 22-Jun-22 25 Insert > Header & footer
  • 26. Cardiovascular risk without / with HDL 22-Jun-22 26 Insert > Header & footer
  • 27. Incremental value of marker • Define a reference model, add marker to evaluate incremental value • Regression coefficient problematic (scaling); p-value assumed to be low • Increase in AUC / c statistic usually small (typically: +0.01)  Push to look beyond AUC: reclassification 22-Jun-22 27 Insert > Header & footer
  • 28.
  • 30. NRI and delta(AUC) for binary classification NRI = delta(sens) + delta(spec) AUC for binary classification = (sens + spec) / 2 delta(AUC) = (delta(sens) + delta(spec)) / 2 NRI = 2 x delta(AUC)
  • 31.
  • 32. NRI has ‘absurd’ weighting?
  • 33. Decision-analytic variants Weighted NRI Delta NB (Vickers) Delta Relative Utility (Baker) / standardized NB (Pepe / Janes) 22-Jun-22 33 Insert > Header & footer
  • 34. Marker evaluation NRI was a historical mistake? Net benefit to the rescue? 22-Jun-22 34 Insert > Header & footer
  • 35. Summary 20 June 2022 1. Prediction modeling research challenging 2. Performance assessment: calibration and Net Benefit 3. Improving performance: a) Updating b) Markers c) Machine learning 22-Jun-22 35 Insert > Header & footer