Methoden zur medizinischen Datenmodellierung€¦ · Methoden zur medizinischen Datenmodellierung...

68
Methoden zur medizinischen Datenmodellierung Georg Dorffner Section for Artificial Intelligence, Center for Medical Statistics, Informatics and Intelligent Systems Medical University of Vienna [email protected] Gastpräsentation: Matthias Samwald

Transcript of Methoden zur medizinischen Datenmodellierung€¦ · Methoden zur medizinischen Datenmodellierung...

Page 1: Methoden zur medizinischen Datenmodellierung€¦ · Methoden zur medizinischen Datenmodellierung Georg Dorffner Section for Artificial Intelligence, Center for Medical Statistics,

Methoden zur medizinischen Datenmodellierung

Georg Dorffner

Section for Artificial Intelligence, Center for Medical Statistics, Informatics and Intelligent Systems

Medical University of Vienna

[email protected]

Gastpräsentation: Matthias Samwald

Page 2: Methoden zur medizinischen Datenmodellierung€¦ · Methoden zur medizinischen Datenmodellierung Georg Dorffner Section for Artificial Intelligence, Center for Medical Statistics,

Genomic CDS ontology and the

Medicine Safety CodeA technology stack for anchoring personalized 

medicine into clinical practice

Mag. Dr. Matthias Samwald Medical University of Vienna

Page 3: Methoden zur medizinischen Datenmodellierung€¦ · Methoden zur medizinischen Datenmodellierung Georg Dorffner Section for Artificial Intelligence, Center for Medical Statistics,

WHAT IS PERSONALIZED MEDICINE / STRATIFIED MEDICINE?

Background

Page 4: Methoden zur medizinischen Datenmodellierung€¦ · Methoden zur medizinischen Datenmodellierung Georg Dorffner Section for Artificial Intelligence, Center for Medical Statistics,

Drug efficacy and toxicity can vary drastically between patients with different genetic profiles

Significant cause of morbidity and mortality!One reason why many promising therapeutics in development fail to reach patients!

Page 5: Methoden zur medizinischen Datenmodellierung€¦ · Methoden zur medizinischen Datenmodellierung Georg Dorffner Section for Artificial Intelligence, Center for Medical Statistics,

Pharmacogenetic assays and treatment algorithms are becoming more and more numerous

Page 6: Methoden zur medizinischen Datenmodellierung€¦ · Methoden zur medizinischen Datenmodellierung Georg Dorffner Section for Artificial Intelligence, Center for Medical Statistics,

Pharmacogenetic assays and treatment algorithms are becoming more and more numerous

Page 7: Methoden zur medizinischen Datenmodellierung€¦ · Methoden zur medizinischen Datenmodellierung Georg Dorffner Section for Artificial Intelligence, Center for Medical Statistics,

Pharmacogenetic assays and treatment algorithms are becoming more and more numerous

Sequencing-based:PGRNseq

Microarray-based:23andMe

Affymetrix DMET chipFlorida/Stanford chip

Page 8: Methoden zur medizinischen Datenmodellierung€¦ · Methoden zur medizinischen Datenmodellierung Georg Dorffner Section for Artificial Intelligence, Center for Medical Statistics,

CAPTURING PHARMACOGENOMIC DOMAIN KNOWLEDGE IN COMPUTABLE FORM

The Genomic CDS ontology

Page 9: Methoden zur medizinischen Datenmodellierung€¦ · Methoden zur medizinischen Datenmodellierung Georg Dorffner Section for Artificial Intelligence, Center for Medical Statistics,

Pharmacogenetic assays and treatment algorithms are becoming more and more numerous

Sequencing-based:PGRNseq

Microarray-based:23andMe

Affymetrix DMET chipFlorida/Stanford chip

Page 10: Methoden zur medizinischen Datenmodellierung€¦ · Methoden zur medizinischen Datenmodellierung Georg Dorffner Section for Artificial Intelligence, Center for Medical Statistics,

We are creating an ontology-based framework for representing pharmacogenomic knowledge

and providing clinical decision support

Page 11: Methoden zur medizinischen Datenmodellierung€¦ · Methoden zur medizinischen Datenmodellierung Georg Dorffner Section for Artificial Intelligence, Center for Medical Statistics,
Page 12: Methoden zur medizinischen Datenmodellierung€¦ · Methoden zur medizinischen Datenmodellierung Georg Dorffner Section for Artificial Intelligence, Center for Medical Statistics,
Page 13: Methoden zur medizinischen Datenmodellierung€¦ · Methoden zur medizinischen Datenmodellierung Georg Dorffner Section for Artificial Intelligence, Center for Medical Statistics,
Page 14: Methoden zur medizinischen Datenmodellierung€¦ · Methoden zur medizinischen Datenmodellierung Georg Dorffner Section for Artificial Intelligence, Center for Medical Statistics,
Page 15: Methoden zur medizinischen Datenmodellierung€¦ · Methoden zur medizinischen Datenmodellierung Georg Dorffner Section for Artificial Intelligence, Center for Medical Statistics,
Page 16: Methoden zur medizinischen Datenmodellierung€¦ · Methoden zur medizinischen Datenmodellierung Georg Dorffner Section for Artificial Intelligence, Center for Medical Statistics,

Alleles / haplotypes are gene variants having certainvariants (‚mutations‘) at certain characters in theirgenetic codes

Page 17: Methoden zur medizinischen Datenmodellierung€¦ · Methoden zur medizinischen Datenmodellierung Georg Dorffner Section for Artificial Intelligence, Center for Medical Statistics,

This is how it actually looks in the ontology

1 Class: 'human with CYP2C9*3'

2 EquivalentTo:

3 has some rs1057910_C

4 SubClassOf:

5 has some 'CYP2C9 *3',

6 (has some rs1057910_C) and

7 (has some rs1057911_A) and

8 (has some rs1799853_C) and

9 (has some rs2256871_A) and

10 (has some rs28371685_C) and

11 (has some rs72558188_AGAAATGGAA) and

12 (has some rs72558189_G) and

13 (has some rs9332239_C)

...

Page 18: Methoden zur medizinischen Datenmodellierung€¦ · Methoden zur medizinischen Datenmodellierung Georg Dorffner Section for Artificial Intelligence, Center for Medical Statistics,

This is how it actually looks in the ontology

1 Class: 'human with CYP2C9*18'

2 EquivalentTo:

3 (has some rs1057910_C) and

4 (has some rs1057911_T) and

5 (has some rs72558193_C)

6 SubClassOf:

7 has some 'CYP2C9 *18',

8 (has some rs1057910_C) and

9 (has some rs1057911_T) and

10 (has some rs1799853_C) and

11 (has some rs2256871_A) and

12 (has some rs28371685_C) and

...

Page 19: Methoden zur medizinischen Datenmodellierung€¦ · Methoden zur medizinischen Datenmodellierung Georg Dorffner Section for Artificial Intelligence, Center for Medical Statistics,

Dosing guideline from an FDA drug label

Page 20: Methoden zur medizinischen Datenmodellierung€¦ · Methoden zur medizinischen Datenmodellierung Georg Dorffner Section for Artificial Intelligence, Center for Medical Statistics,

Dosing guideline from an FDA drug label

1 Class: 'human triggering CDS rule 9'

2 Annotations:

3 CDS_message "0.5-2 mg warfarin per day should be considered

4 as a starting dose range for a patient with this genotype

5 according to the warfarin drug label."

6 EquivalentTo:

7 (has some 'CYP2C9 *1') and

8 (has some 'CYP2C9 *3') and

9 (has exactly 2 rs9923231_T)

Page 21: Methoden zur medizinischen Datenmodellierung€¦ · Methoden zur medizinischen Datenmodellierung Georg Dorffner Section for Artificial Intelligence, Center for Medical Statistics,

Describing an individual patient in OWL

1 Individual: example_patient2 Types:3 human,4 (has some rs1208_A) and (has some rs1208_G),5 (has some rs8192709_C) and (has some rs8192709_T),6 (has some rs9934438_A) and (has some rs9934438_G),7 has exactly 2 rs10264272_C,8 has exactly 2 rs9923231_T,9 has exactly 2 rs12720461_C,10 (has some ‘CYP2C9 *1’) and (has some ‘CYP2C9 *3’),11 has exactly 2 ‘CYP2C19 *1’,12 (has exactly 3 CYP2D6) and (has exactly 2 ‘CYP2D6 *1’) 13 and (has exactly 1 ‘CYP2D6 *2’)

...

heterozygous SNP variants

homozygous SNP variants

allelic variantsand copy num-ber variations

Page 22: Methoden zur medizinischen Datenmodellierung€¦ · Methoden zur medizinischen Datenmodellierung Georg Dorffner Section for Artificial Intelligence, Center for Medical Statistics,

Describing an individual patient in OWL

1 Individual: example_patient2 Types:3 human,4 (has some rs1208_A) and (has some rs1208_G),5 (has some rs8192709_C) and (has some rs8192709_T),6 (has some rs9934438_A) and (has some rs9934438_G),7 has exactly 2 rs10264272_C,8 has exactly 2 rs9923231_T,9 has exactly 2 rs12720461_C,10 (has some ‘CYP2C9 *1’) and (has some ‘CYP2C9 *3’),11 has exactly 2 ‘CYP2C19 *1’,12 (has exactly 3 CYP2D6) and (has exactly 2 ‘CYP2D6 *1’) 13 and (has exactly 1 ‘CYP2D6 *2’)

...

heterozygous SNP variants

homozygous SNP variants

allelic variantsand copy num-ber variations

"0.5 - 2 mg warfarin per day should be considered as a starting dose range for a patient with this genotype according to the warfarin drug label."

OWL Reasoner

Page 23: Methoden zur medizinischen Datenmodellierung€¦ · Methoden zur medizinischen Datenmodellierung Georg Dorffner Section for Artificial Intelligence, Center for Medical Statistics,

BUT HOW CAN PERSONALIZED MEDICINE BE PUT INTO PRACTICE?

The Medicine Safety Code  initiative

Page 24: Methoden zur medizinischen Datenmodellierung€¦ · Methoden zur medizinischen Datenmodellierung Georg Dorffner Section for Artificial Intelligence, Center for Medical Statistics,

We are creating a barrier‐free system for storing and interpreting personal pharmacogenetic information (based on 2D barcodes and web‐based decision support)

Page 25: Methoden zur medizinischen Datenmodellierung€¦ · Methoden zur medizinischen Datenmodellierung Georg Dorffner Section for Artificial Intelligence, Center for Medical Statistics,

We are creating a barrier‐free system for storing and interpreting personal pharmacogenetic information (based on 2D barcodes and web‐based decision support)

Page 26: Methoden zur medizinischen Datenmodellierung€¦ · Methoden zur medizinischen Datenmodellierung Georg Dorffner Section for Artificial Intelligence, Center for Medical Statistics,

Alternatively, data can also be entered manually

Page 27: Methoden zur medizinischen Datenmodellierung€¦ · Methoden zur medizinischen Datenmodellierung Georg Dorffner Section for Artificial Intelligence, Center for Medical Statistics,

Matching treatment recommendations are displayed

Page 28: Methoden zur medizinischen Datenmodellierung€¦ · Methoden zur medizinischen Datenmodellierung Georg Dorffner Section for Artificial Intelligence, Center for Medical Statistics,

Take‐home messages

• Genomic CDS ontology + Medicine Safety Code system provide a comprehensive solution for clinical pharmacogenetics

• 2D barcodes can be used to quickly provide genetic data (or other patient data) at the point‐of‐care

• RDF/OWL 2 is capabale of providing both decision support functionality as well as flexible knowledge bases for personalized medicine

Page 29: Methoden zur medizinischen Datenmodellierung€¦ · Methoden zur medizinischen Datenmodellierung Georg Dorffner Section for Artificial Intelligence, Center for Medical Statistics,

Thanks!

Local team:

Jose Antonio Miñarro Gimenez

W3C partners:

Richard Boyce (University of Pittsburgh)

Robert R. Freimuth (Mayo Clinic)

Michel Dumontier (Carleton University)

Simon Lin (Marshfield Clinic)

Robert L. Powers (Predictive Medicine, Inc.)

Joanne S. Luciano (Rensselaer Polytechnic Institute)

Eric Prud’hommeaux (W3C)

M. Scott Marshall (MAASTRO Clinic)

Funding:

Austrian Science Fund (FWF): [PP 25608‐N15]

Links:

http://www.genomic‐cds.org/

http://safety‐code.org/

http://samwald.info/

Page 30: Methoden zur medizinischen Datenmodellierung€¦ · Methoden zur medizinischen Datenmodellierung Georg Dorffner Section for Artificial Intelligence, Center for Medical Statistics,

Beispiel 2: Kontinuierliche Schlafmodellierung

Roman Rosipal, Achim Lewandowski, GD

Page 31: Methoden zur medizinischen Datenmodellierung€¦ · Methoden zur medizinischen Datenmodellierung Georg Dorffner Section for Artificial Intelligence, Center for Medical Statistics,

Inst

itut f

ür A

rtific

ial I

ntel

ligen

ceZe

ntru

m fü

r Med

izin

isch

e S

tatis

tik,

Info

rmat

ik u

nd In

telli

gent

e S

yste

me

Continuous sleep profile

Automatic sleep analysis

Page 32: Methoden zur medizinischen Datenmodellierung€¦ · Methoden zur medizinischen Datenmodellierung Georg Dorffner Section for Artificial Intelligence, Center for Medical Statistics,

Inst

itut f

ür A

rtific

ial I

ntel

ligen

ceZe

ntru

m fü

r Med

izin

isch

e S

tatis

tik,

Info

rmat

ik u

nd In

telli

gent

e S

yste

me

Continuous model

Continuous probability vectors

Preprocessing and feature extraction

Supervised learning of class-conditional GMMs

Unsupervised reshuffling of GMMs

AR(10) coefficients

GMMs

Calculate posteriors

GMMs

Polysomnographic recordings

Single channel EEG data

R&K labels

Spindle

Process

Artifacts

Detection

Page 33: Methoden zur medizinischen Datenmodellierung€¦ · Methoden zur medizinischen Datenmodellierung Georg Dorffner Section for Artificial Intelligence, Center for Medical Statistics,

Inst

itut f

ür A

rtific

ial I

ntel

ligen

ceZe

ntru

m fü

r Med

izin

isch

e S

tatis

tik,

Info

rmat

ik u

nd In

telli

gent

e S

yste

me

b003302

0 50 100 150 200 250 300 350 400 4500

1

tim e [m in]

deep

0

1

s2

0

1

s1

0

1

wak

e

C4

C3

otherswake

s1s2s3s4

REM

B003302: Female, 76 years

Page 34: Methoden zur medizinischen Datenmodellierung€¦ · Methoden zur medizinischen Datenmodellierung Georg Dorffner Section for Artificial Intelligence, Center for Medical Statistics,

Inst

itut f

ür A

rtific

ial I

ntel

ligen

ceZe

ntru

m fü

r Med

izin

isch

e S

tatis

tik,

Info

rmat

ik u

nd In

telli

gent

e S

yste

me

“Subjective sleep quality” versus “Objective sleep quality”R&K

SSA-1 Number of stage shifts (/hr TST)

-10.00 0.00 10.00

s_qua_21

-30.00

-20.00

-10.00

0.00

10.00

20.00

fsts

t00c

fstst00c = -0.42 + 0.53 * s_qua_21R-Square = 0.19

20-39 40-59 >=60

-10.00 0.00 10.00

s_qua_21

fstst00c = -1.30 + 0.21 * s_qua_21R-Square = 0.03

-10.00 0.00 10.00

s_qua_21

fstst00c = -1.53 + 0.01 * s_qua_21R-Square = 0.00

Page 35: Methoden zur medizinischen Datenmodellierung€¦ · Methoden zur medizinischen Datenmodellierung Georg Dorffner Section for Artificial Intelligence, Center for Medical Statistics,

Inst

itut f

ür A

rtific

ial I

ntel

ligen

ceZe

ntru

m fü

r Med

izin

isch

e S

tatis

tik,

Info

rmat

ik u

nd In

telli

gent

e S

yste

me

“Subjective sleep quality” versus “Objective sleep quality”hGMM

SSA-1 Number of stage shifts “deep – S2”

-10.00 0.00 10.00

s_qua_21

-0.0200

0.0000

0.0200

0.0400

sc_d

_s2c

sc_d_s2c = 0.00 + -0.00 * s_qua_21R-Square = 0.15

20-39 40-59 >=60

-10.00 0.00 10.00

s_qua_21

sc_d_s2c = 0.00 + -0.00 * s_qua_21R-Square = 0.14

-10.00 0.00 10.00

s_qua_21

sc_d_s2c = 0.00 + -0.00 * s_qua_21R-Square = 0.13

Page 36: Methoden zur medizinischen Datenmodellierung€¦ · Methoden zur medizinischen Datenmodellierung Georg Dorffner Section for Artificial Intelligence, Center for Medical Statistics,

Inst

itut f

ür A

rtific

ial I

ntel

ligen

ceZe

ntru

m fü

r Med

izin

isch

e S

tatis

tik,

Info

rmat

ik u

nd In

telli

gent

e S

yste

me

Result• Measures for sleep continuity and

architecture based on R&K showed significant correlations with subjective sleep quality only in young subjects.

• In contrast, measures for sleep continuity and architecture based on hGMM showed significant correlations in all age-groups

© Alle Rechte liegen bei den Autoren.

Page 37: Methoden zur medizinischen Datenmodellierung€¦ · Methoden zur medizinischen Datenmodellierung Georg Dorffner Section for Artificial Intelligence, Center for Medical Statistics,

Inst

itut f

ür A

rtific

ial I

ntel

ligen

ceZe

ntru

m fü

r Med

izin

isch

e S

tatis

tik,

Info

rmat

ik u

nd In

telli

gent

e S

yste

me

Continuous probability model II

• z: state• x: AR(10) vector• c: RK class• s: spindle class

Model assumption: given the state z, x,c, and s are independent

Page 38: Methoden zur medizinischen Datenmodellierung€¦ · Methoden zur medizinischen Datenmodellierung Georg Dorffner Section for Artificial Intelligence, Center for Medical Statistics,

Inst

itut f

ür A

rtific

ial I

ntel

ligen

ceZe

ntru

m fü

r Med

izin

isch

e S

tatis

tik,

Info

rmat

ik u

nd In

telli

gent

e S

yste

me

Applying the modelExpress the current sleep as R&K posterior for given x and s

Page 39: Methoden zur medizinischen Datenmodellierung€¦ · Methoden zur medizinischen Datenmodellierung Georg Dorffner Section for Artificial Intelligence, Center for Medical Statistics,

Inst

itut f

ür A

rtific

ial I

ntel

ligen

ceZe

ntru

m fü

r Med

izin

isch

e S

tatis

tik,

Info

rmat

ik u

nd In

telli

gent

e S

yste

me

Applying the modelExpress the current sleep as ‚raw‘ state posterior for given x and s

Page 40: Methoden zur medizinischen Datenmodellierung€¦ · Methoden zur medizinischen Datenmodellierung Georg Dorffner Section for Artificial Intelligence, Center for Medical Statistics,

Inst

itut f

ür A

rtific

ial I

ntel

ligen

ceZe

ntru

m fü

r Med

izin

isch

e S

tatis

tik,

Info

rmat

ik u

nd In

telli

gent

e S

yste

me

Relative time spent in a state

N

i ii NaxzpzRTS1

/),|()(

• subject with observations (x1,a1),...,(XN,aN)

• Looking for a measure judging the time spent in a state z, weighted by ‚intensity‘ of a visit:

e.g. calculate sum of posteriors for each state z, relative to length of recording

Page 41: Methoden zur medizinischen Datenmodellierung€¦ · Methoden zur medizinischen Datenmodellierung Georg Dorffner Section for Artificial Intelligence, Center for Medical Statistics,

Inst

itut f

ür A

rtific

ial I

ntel

ligen

ceZe

ntru

m fü

r Med

izin

isch

e S

tatis

tik,

Info

rmat

ik u

nd In

telli

gent

e S

yste

me

Correlation to outside criteria• can compare given sleep quality

variable (e.g. result of concentration test) with RTS(z) for a list of subjects

=> use Spearman-Rank correlation to detect monotonic relationships

Page 42: Methoden zur medizinischen Datenmodellierung€¦ · Methoden zur medizinischen Datenmodellierung Georg Dorffner Section for Artificial Intelligence, Center for Medical Statistics,

Inst

itut f

ür A

rtific

ial I

ntel

ligen

ceZe

ntru

m fü

r Med

izin

isch

e S

tatis

tik,

Info

rmat

ik u

nd In

telli

gent

e S

yste

me

Results: sleep stages

Page 43: Methoden zur medizinischen Datenmodellierung€¦ · Methoden zur medizinischen Datenmodellierung Georg Dorffner Section for Artificial Intelligence, Center for Medical Statistics,

Inst

itut f

ür A

rtific

ial I

ntel

ligen

ceZe

ntru

m fü

r Med

izin

isch

e S

tatis

tik,

Info

rmat

ik u

nd In

telli

gent

e S

yste

me

Results: probabilistic sleep model

Rosipal et al., Biol Psychol, 2013

Page 44: Methoden zur medizinischen Datenmodellierung€¦ · Methoden zur medizinischen Datenmodellierung Georg Dorffner Section for Artificial Intelligence, Center for Medical Statistics,

Inst

itut f

ür A

rtific

ial I

ntel

ligen

ceZe

ntru

m fü

r Med

izin

isch

e S

tatis

tik,

Info

rmat

ik u

nd In

telli

gent

e S

yste

me

HMM trajectories

Same subject(1st/2nd night)

Page 45: Methoden zur medizinischen Datenmodellierung€¦ · Methoden zur medizinischen Datenmodellierung Georg Dorffner Section for Artificial Intelligence, Center for Medical Statistics,

Inst

itut f

ür A

rtific

ial I

ntel

ligen

ceZe

ntru

m fü

r Med

izin

isch

e S

tatis

tik,

Info

rmat

ik u

nd In

telli

gent

e S

yste

me

Individual differences

• Q: Is it really justified to ask for more correlations?

Page 46: Methoden zur medizinischen Datenmodellierung€¦ · Methoden zur medizinischen Datenmodellierung Georg Dorffner Section for Artificial Intelligence, Center for Medical Statistics,

Inst

itut f

ür A

rtific

ial I

ntel

ligen

ceZe

ntru

m fü

r Med

izin

isch

e S

tatis

tik,

Info

rmat

ik u

nd In

telli

gent

e S

yste

me

Conclusions• PSG/EEG objectively describes

physiology of sleep• Visual approaches lead to „fuzzy“

ground truth, automation leads to reliability

• Data-based approaches can extract more information

• But relationship to outside criteria about sleep quality due to other effects (context, individual characteristics

Page 47: Methoden zur medizinischen Datenmodellierung€¦ · Methoden zur medizinischen Datenmodellierung Georg Dorffner Section for Artificial Intelligence, Center for Medical Statistics,

Beispiel 3: Vorhersage der Mortalität nach Herzstillstand

Fritz Sterz, Stefan Aschauer, GD

Page 48: Methoden zur medizinischen Datenmodellierung€¦ · Methoden zur medizinischen Datenmodellierung Georg Dorffner Section for Artificial Intelligence, Center for Medical Statistics,

Inst

itut f

ür A

rtific

ial I

ntel

ligen

ceZe

ntru

m fü

r Med

izin

isch

e S

tatis

tik,

Info

rmat

ik u

nd In

telli

gent

e S

yste

me out of hospital cardiac arrest

(OOHCA)

• major health problem

• 500.000 patients in United States and Europe /year

• overall mortality: 8% - 11%

background

Page 49: Methoden zur medizinischen Datenmodellierung€¦ · Methoden zur medizinischen Datenmodellierung Georg Dorffner Section for Artificial Intelligence, Center for Medical Statistics,

Inst

itut f

ür A

rtific

ial I

ntel

ligen

ceZe

ntru

m fü

r Med

izin

isch

e S

tatis

tik,

Info

rmat

ik u

nd In

telli

gent

e S

yste

me

background• OOHCA has a very uncertain

outcome

• no valid outcome scoring system• problem in giving reliable outcome

estimation • delicate decisions

based on experience and gut feeling

Page 50: Methoden zur medizinischen Datenmodellierung€¦ · Methoden zur medizinischen Datenmodellierung Georg Dorffner Section for Artificial Intelligence, Center for Medical Statistics,

Inst

itut f

ür A

rtific

ial I

ntel

ligen

ceZe

ntru

m fü

r Med

izin

isch

e S

tatis

tik,

Info

rmat

ik u

nd In

telli

gent

e S

yste

me

aim

• to assess the predictability of outcome after OOHCA, based on a number of observational variables

• to identify variables with high predictive power

• to assess whether a multivariateapproach is superior to a univariateone

• to derive a OOHCA outcomeprediction score tool

Page 51: Methoden zur medizinischen Datenmodellierung€¦ · Methoden zur medizinischen Datenmodellierung Georg Dorffner Section for Artificial Intelligence, Center for Medical Statistics,

Inst

itut f

ür A

rtific

ial I

ntel

ligen

ceZe

ntru

m fü

r Med

izin

isch

e S

tatis

tik,

Info

rmat

ik u

nd In

telli

gent

e S

yste

me

benefit• improvement of the predictability of

patient’s survival would be of major medical and socioeconomic interest.

• valid outcome estimation could facilitate decision-making for persons in authority and could save medical resources

Page 52: Methoden zur medizinischen Datenmodellierung€¦ · Methoden zur medizinischen Datenmodellierung Georg Dorffner Section for Artificial Intelligence, Center for Medical Statistics,

Inst

itut f

ür A

rtific

ial I

ntel

ligen

ceZe

ntru

m fü

r Med

izin

isch

e S

tatis

tik,

Info

rmat

ik u

nd In

telli

gent

e S

yste

me

methods

• based on a cardiac arrest-registry with > 4000 patients which were resuscitated from OOHCA and which were admitted to the Department of Emergency Medicine at a large University Hospital

• multivariate logistic regression was applied on 20 variables before ROSC deemed to have high predictive power

Page 53: Methoden zur medizinischen Datenmodellierung€¦ · Methoden zur medizinischen Datenmodellierung Georg Dorffner Section for Artificial Intelligence, Center for Medical Statistics,

Inst

itut f

ür A

rtific

ial I

ntel

ligen

ceZe

ntru

m fü

r Med

izin

isch

e S

tatis

tik,

Info

rmat

ik u

nd In

telli

gent

e S

yste

me

methods

• the framework of machine learning was chosen

• a 10-fold cross-validation was done for reliable estimates and confidence intervals

• main performance parameter was the area under the ROC curve (AUC)

Page 54: Methoden zur medizinischen Datenmodellierung€¦ · Methoden zur medizinischen Datenmodellierung Georg Dorffner Section for Artificial Intelligence, Center for Medical Statistics,

Inst

itut f

ür A

rtific

ial I

ntel

ligen

ceZe

ntru

m fü

r Med

izin

isch

e S

tatis

tik,

Info

rmat

ik u

nd In

telli

gent

e S

yste

me

variablesVariable name Description Value Scalesex Sex of the patient Male=0, Female=1 binaryage Age of the patient In years, at the time of cardiac arrest metricbmi Body Mass Index Weight (kg) / Size (m) Squared metricdiabetes Previous diagnosis of diabetes Diabetes = 1, no diabetes = 0 binarysmoker Patient is a smoker Smoker=1, nonsmoker=0 binarymyocinfarct Patient previously had a myocardial infarction Infarction=1, no infarction=0 binarykhk Previous diagnosis of Coronary Artery Disease CAD=1, no CAD=0 binaryhypertension Previous diagnosis of hypertension Hypertension=1, no hypertension=0 binaryheartfail Previous diagnosis of heart failure Heart failure=1, no heart failure=0 binarycvi Previous diagnosis of chronic venous insufficiency CVI=1, no CVI=0 binary

copd Previous diagnosis of chronic obstructive pulmonary disease COPD=1, no COPD=0 binaryopcpre OPC score prior to cardiac arrest Score 1 to 5 ordinal, treated as metricnyh5pre NYH5 score prior to cardiac arrest Score 1 to 5 ordinal, treated as metricnoflow Minutes between cardiac arrest and first aid (length of "no

flow" time) in minutes metricmin2srosc Minutes between cardiac arrest and SROSC in minutes metriccause Main cause of cardiac arrest Cardiac=1, non-cardicac=0 binary

firstaidFirst aid performed by physician, family member, paramedic or layman Physician=1, non-Physician=0 binary

nodefi Number of defibrillation shots Count of shots metricadrenaline Amount of adrenaline applied Total amount (in …) metric

shockable Shockability of rhythm in first defibrillation Shockable=1, non-shockable=0 binarydefireaction Reaction tp the first defibrillation Not shockable=0, shockable and VT/VF (as reaction

to first defi)=1, shockable and PEA=2, shockable+Asystole=3, shockable+SR/RHY/SVES/VES/AVES+ no pulse=4, shockable+pulse=5

ordinal, treated as metric

Page 55: Methoden zur medizinischen Datenmodellierung€¦ · Methoden zur medizinischen Datenmodellierung Georg Dorffner Section for Artificial Intelligence, Center for Medical Statistics,

Inst

itut f

ür A

rtific

ial I

ntel

ligen

ceZe

ntru

m fü

r Med

izin

isch

e S

tatis

tik,

Info

rmat

ik u

nd In

telli

gent

e S

yste

me

Initial variables, histograms

0 0.5 10

500

1000

1500

0 0.5 10

500

1000

1500

0 0.5 10

500

1000

1500

0 0.5 10

1000

2000

-5 0 50

200

400

0 0.5 10

500

1000

1500

-5 0 50

500

1000

1500

0 0.5 10

1000

2000

-10 0 100

500

1000

0 0.5 10

500

1000

1500

-5 0 50

500

1000

-10 0 100

500

1000

1500

0 0.5 10

500

1000

1500

0 0.5 10

1000

2000

0 0.5 10

500

1000

1500

-10 0 100

500

1000

1500

0 0.5 10

500

1000

1500

0 0.5 10

1000

2000

0 0.5 10

1000

2000

0 0.5 10

500

1000

1500

0 0.5 10

500

1000

1500

0 0.5 10

500

1000

1500

0 0.5 10

1000

2000

0 0.5 10

500

1000

Page 56: Methoden zur medizinischen Datenmodellierung€¦ · Methoden zur medizinischen Datenmodellierung Georg Dorffner Section for Artificial Intelligence, Center for Medical Statistics,

Inst

itut f

ür A

rtific

ial I

ntel

ligen

ceZe

ntru

m fü

r Med

izin

isch

e S

tatis

tik,

Info

rmat

ik u

nd In

telli

gent

e S

yste

me

Data sets

Page 57: Methoden zur medizinischen Datenmodellierung€¦ · Methoden zur medizinischen Datenmodellierung Georg Dorffner Section for Artificial Intelligence, Center for Medical Statistics,

Inst

itut f

ür A

rtific

ial I

ntel

ligen

ceZe

ntru

m fü

r Med

izin

isch

e S

tatis

tik,

Info

rmat

ik u

nd In

telli

gent

e S

yste

me

witnessed Trainingset median 25% percentile

75% percentile Percent 1 Percent 0

sex 27.53% 72.47%age 59 49 69 0 0bmi 26.12 23.88 29.22 0 0diabetes 16.20% 83.80%smoker 30.90% 69.10%myocinfarct 12.92% 87.08%cad 21.91% 78.09%hypertension 32.21% 67.79%heartfail 11.05% 88.95%cvi 5.99% 94.01%copd 9.74% 90.26%opcpre 1 1 1 nyh5pre 1 1 2 noflow 1 0 6.5 min2srosc 20 10 30 cause 69.76% 30.24%firstaid 34.18% 65.82%nodefi 2 0 4 adrenaline 2 0 4 defireaction 1 0 2 shockable 59.83% 40.17%cpc30d 3 1 5 mortality 39.89% 60.11%

Testset median 25% percentile

75% percentile Percent 1 Percent 0

sex 27.84% 72.17%age 61 50 71 bmi 26.23 24.11 29.41 diabetes 20.62% 79.38%smoker 31.62% 68.39%myocinfarct 14.09% 85.91%cad 24.74% 75.26%hypertension 41.92% 58.08%heartfail 14.78% 85.22%cvi 4.81% 95.19%copd 6.53% 93.47%opcpre 1 1 2 nyh5pre 1 1 2 noflow 1 0 5 min2srosc 19 12 32 cause 62.54% 37.46%firstaid 49.49% 50.52%nodefi 1 0 3 adrenaline 1 0 3 defireaction 1 0 2 shockable 54.30% 45.70%cpc30d 3 1 5 mortality 42.27% 57.73% 

Page 58: Methoden zur medizinischen Datenmodellierung€¦ · Methoden zur medizinischen Datenmodellierung Georg Dorffner Section for Artificial Intelligence, Center for Medical Statistics,

Inst

itut f

ür A

rtific

ial I

ntel

ligen

ceZe

ntru

m fü

r Med

izin

isch

e S

tatis

tik,

Info

rmat

ik u

nd In

telli

gent

e S

yste

me

Non-witnessedTrainingset median 25% percentile 75% percentile Percent 1 Percent 0 sex 30.46% 69.54%age 56 41 68 bmi 25.94 22.86 28.72 diabetes 14.37% 85.63%smoker 27.01% 72.99%myocinfarct 9.77% 90.23%khk 16.09% 83.91%hypertension 26.44% 73.56%heartfail 10.92% 89.08%cvi 4.02% 95.98%copd 8.62% 91.38%opcpre 1 1 1 nyh5pre 1 1 1 cause 43.68% 56.32%nodefi 1 0 4 adrenaline 4 2 6.5 defireaction 0 0 1 shockable 36.78% 63.22%cpc30d 5 5 5 mortality 76.44% 23.56%Testset median 25% percentile 75% percentile Percent 1 Percent 0 sex 20.00% 80.00%age 54 44.5 64.25 bmi 26.23 23.32 28.18 diabetes 0.00% 100.00%smoker 32.00% 68.00%myocinfarct 4.00% 96.00%khk 12.00% 88.00%hypertension 44.00% 56.00%heartfail 4.00% 96.00%cvi 4.00% 96.00%copd 16.00% 84.00%opcpre 1 1 1 nyh5pre 1 1 1 cause 60.00% 40.00%nodefi 1 0 6.25 adrenaline 4 2 8 defireaction 1 0 1 shockable 60.00% 40.00%cpc30d 5 1.75 5 mortality 68.00% 32.00% 

Page 59: Methoden zur medizinischen Datenmodellierung€¦ · Methoden zur medizinischen Datenmodellierung Georg Dorffner Section for Artificial Intelligence, Center for Medical Statistics,

Inst

itut f

ür A

rtific

ial I

ntel

ligen

ceZe

ntru

m fü

r Med

izin

isch

e S

tatis

tik,

Info

rmat

ik u

nd In

telli

gent

e S

yste

me

Page 60: Methoden zur medizinischen Datenmodellierung€¦ · Methoden zur medizinischen Datenmodellierung Georg Dorffner Section for Artificial Intelligence, Center for Medical Statistics,

Inst

itut f

ür A

rtific

ial I

ntel

ligen

ceZe

ntru

m fü

r Med

izin

isch

e S

tatis

tik,

Info

rmat

ik u

nd In

telli

gent

e S

yste

me

Page 61: Methoden zur medizinischen Datenmodellierung€¦ · Methoden zur medizinischen Datenmodellierung Georg Dorffner Section for Artificial Intelligence, Center for Medical Statistics,

Inst

itut f

ür A

rtific

ial I

ntel

ligen

ceZe

ntru

m fü

r Med

izin

isch

e S

tatis

tik,

Info

rmat

ik u

nd In

telli

gent

e S

yste

me

Score

Page 62: Methoden zur medizinischen Datenmodellierung€¦ · Methoden zur medizinischen Datenmodellierung Georg Dorffner Section for Artificial Intelligence, Center for Medical Statistics,

Inst

itut f

ür A

rtific

ial I

ntel

ligen

ceZe

ntru

m fü

r Med

izin

isch

e S

tatis

tik,

Info

rmat

ik u

nd In

telli

gent

e S

yste

me

Simplifed scorePredictor Points Predictor Points

1. Age group 3. Minutes until SROSC>80 32 >100min 35>70 27 >50min 21>50 23 >40min 13>60 20 >30min 10>40 16 >20min 7≤40 11 >10min 4

>0min 12. Adrenalin administered 0min 0>10mg 24>5mg 12 4. Shockable rhythm?>4mg 7 Yes ‐15>3mg 5 No 0>2mg 4>1mg 2>0mg 10mg 0 Total score

Total score Probability for mortality

<13 10%13‐22 30%23‐30 50%31‐40 70%>40 90%

Page 63: Methoden zur medizinischen Datenmodellierung€¦ · Methoden zur medizinischen Datenmodellierung Georg Dorffner Section for Artificial Intelligence, Center for Medical Statistics,

Inst

itut f

ür A

rtific

ial I

ntel

ligen

ceZe

ntru

m fü

r Med

izin

isch

e S

tatis

tik,

Info

rmat

ik u

nd In

telli

gent

e S

yste

me

predicted mortality

Page 64: Methoden zur medizinischen Datenmodellierung€¦ · Methoden zur medizinischen Datenmodellierung Georg Dorffner Section for Artificial Intelligence, Center for Medical Statistics,

Beispiel 4: Simulation molekularer Dynamiken

Bernhard Knapp, Wolfgang Schreiner, GD

Page 65: Methoden zur medizinischen Datenmodellierung€¦ · Methoden zur medizinischen Datenmodellierung Georg Dorffner Section for Artificial Intelligence, Center for Medical Statistics,

Inst

itut f

ür A

rtific

ial I

ntel

ligen

ceZe

ntru

m fü

r Med

izin

isch

e S

tatis

tik,

Info

rmat

ik u

nd In

telli

gent

e S

yste

me

Molecular dynamics simulation –T-cells

Page 66: Methoden zur medizinischen Datenmodellierung€¦ · Methoden zur medizinischen Datenmodellierung Georg Dorffner Section for Artificial Intelligence, Center for Medical Statistics,

Inst

itut f

ür A

rtific

ial I

ntel

ligen

ceZe

ntru

m fü

r Med

izin

isch

e S

tatis

tik,

Info

rmat

ik u

nd In

telli

gent

e S

yste

me

Patterns within 50ns

Knapp et al., Plos One, 2013

Page 67: Methoden zur medizinischen Datenmodellierung€¦ · Methoden zur medizinischen Datenmodellierung Georg Dorffner Section for Artificial Intelligence, Center for Medical Statistics,

Inst

itut f

ür A

rtific

ial I

ntel

ligen

ceZe

ntru

m fü

r Med

izin

isch

e S

tatis

tik,

Info

rmat

ik u

nd In

telli

gent

e S

yste

me

Statistical significance

Page 68: Methoden zur medizinischen Datenmodellierung€¦ · Methoden zur medizinischen Datenmodellierung Georg Dorffner Section for Artificial Intelligence, Center for Medical Statistics,

Inst

itut f

ür A

rtific

ial I

ntel

ligen

ceZe

ntru

m fü

r Med

izin

isch

e S

tatis

tik,

Info

rmat

ik u

nd In

telli

gent

e S

yste

me

Affected regions