ken sikaris - melbourne pathology - previous discovery in pathology data mining

77
A/Prof Ken Sikaris 15 th October 2013 PATHOLOGY DATA MINING The discipline of pathology has electronically stored its collective experience and we now have the tools to tackle big data. A/Prof Ken Sikaris BSc(Hons), MBBS, FRCPA, FAACB, FFSc, GAICD Clinical Support Systems Director, Sonic Healthcare Chemical Pathologis, Melbourne Pathology

Upload: informa-australia

Post on 15-Jan-2015

800 views

Category:

Health & Medicine


20 download

DESCRIPTION

Associate Professor Ken Sikaris, Director of Clinical Support Systems, Sonic Healthcare and Principle Fellow in the Department of Pathology, Melbourne University presented "Previous Discovery in Pathology Data Mining" at the National Pathology Forum 2013. This annual conference provides a platform for the public and private sectors to come together and discuss all the latest issues affecting the pathology sector in Australia. For more information, please visit the conference website: http://www.informa.com.au/pathologyforum

TRANSCRIPT

Page 1: Ken Sikaris - Melbourne Pathology - Previous Discovery in Pathology Data Mining

A/Prof Ken Sikaris 15th October 2013

PATHOLOGY DATA MINING The discipline of pathology

has electronically stored its collective experience

and we now have the tools to tackle big data.

A/Prof Ken Sikaris BSc(Hons), MBBS, FRCPA, FAACB, FFSc, GAICD

Clinical Support Systems Director, Sonic Healthcare

Chemical Pathologis, Melbourne Pathology

Page 2: Ken Sikaris - Melbourne Pathology - Previous Discovery in Pathology Data Mining

A/Prof Ken Sikaris 15th October 2013

World Resources

Page 3: Ken Sikaris - Melbourne Pathology - Previous Discovery in Pathology Data Mining

A/Prof Ken Sikaris 15th October 2013

History of Sonic Healthcare

1987 Douglas Laboratories (Syd)

Colin Goldschmidt, Pathologist 1992 Clinpath (Adel)

Michael Boyd

1993 Macquarie Pathology (Syd)

1994 Clinipath (Perth)

Barry Patterson, Mining Engineer

1987 Sonic Technology Australia Ltd

1994 Sonic Healthcare

2000 Hitech (Melb)

2000 Foundation (GP)

2000 Radiology (Vic, Qld, NSW)

2002 TDL (London)

2004 Schottdorf (Germany)

2004 IPN (GP)

2005 CPL (Texas)

2006 USA (Oklahoma, Florida)

2007 USA, Germany, Switzerland

2008 USA, Germany, Switzerland

2009 USA

2010 USA, Belgium, Prime (GP)

2011 USA, Belgium, Allied (GP)

2012 Germany, Healthscope (WA)

2013 Germany

1994 Pathlab (Adelaide)

1996 Hanly Moir, Barratt & Smith (Syd)

1997 Lifescreen

1998 Silex split off

1999 Southern (Wollongong), ADL (Syd)

1999 SGS (Melb Path, SNP)

Page 4: Ken Sikaris - Melbourne Pathology - Previous Discovery in Pathology Data Mining

A/Prof Ken Sikaris 15th October 2013

Sonic Healthcare

Page 5: Ken Sikaris - Melbourne Pathology - Previous Discovery in Pathology Data Mining

A/Prof Ken Sikaris 15th October 2013

Sonic Healthcare vs. Gold Mining

Page 6: Ken Sikaris - Melbourne Pathology - Previous Discovery in Pathology Data Mining

A/Prof Ken Sikaris 15th October 2013

The Growing ‘Market’ of Pathology

Page 7: Ken Sikaris - Melbourne Pathology - Previous Discovery in Pathology Data Mining

A/Prof Ken Sikaris 15th October 2013

Page 8: Ken Sikaris - Melbourne Pathology - Previous Discovery in Pathology Data Mining

A/Prof Ken Sikaris 15th October 2013

Page 9: Ken Sikaris - Melbourne Pathology - Previous Discovery in Pathology Data Mining

A/Prof Ken Sikaris 15th October 2013

10,000 patients / day

Avg 15 tests / patients

>30 million tests / year

Melbourne Pathology

Page 10: Ken Sikaris - Melbourne Pathology - Previous Discovery in Pathology Data Mining

A/Prof Ken Sikaris 15th October 2013

Page 11: Ken Sikaris - Melbourne Pathology - Previous Discovery in Pathology Data Mining

A/Prof Ken Sikaris 15th October 2013

Medical Learning 1st Phase: Masters

Hippocrates

400BC

Galen

150AD

Avicenna

980AD

Paracelsus

1520AD

Osler

1880AD

Page 12: Ken Sikaris - Melbourne Pathology - Previous Discovery in Pathology Data Mining

A/Prof Ken Sikaris 15th October 2013

Medical Learning 2nd Phase: Journals

NEJM

1812

BMJ

1840

JAMA

1883

Lancet

1823

Arch Int Med

1908

Page 13: Ken Sikaris - Melbourne Pathology - Previous Discovery in Pathology Data Mining

A/Prof Ken Sikaris 15th October 2013

Medical Learning: 3rd Phase Databases

Framingham

Heart

Study

1948

n=4,486

National

Health

And

Nutrition

Examination

Survey

1971

n=17,227

AusDiab

1999

n=79

Canadian

Community

Health

Survey

2000

n=450

Busselton

Health

Study

1966

n=177

ABS

2013

n=….

Journal Articles

Page 14: Ken Sikaris - Melbourne Pathology - Previous Discovery in Pathology Data Mining

A/Prof Ken Sikaris 15th October 2013

“Database” articles on Medline

Page 15: Ken Sikaris - Melbourne Pathology - Previous Discovery in Pathology Data Mining

A/Prof Ken Sikaris 15th October 2013

Computer Foundations of Knowledge Acquisition

• Newell, A.; Simon, HA. Computer science as empirical

inquiry: symbols and search. In: Haugeland, J., editor. Mind

Design. MIT Press/Bradfor Books; Cambridge: 1981. p. 35-

66.

• Compton P, Jansen R. “A philosophical basis for

knowledge acquisition.” Knowledge Acquisition

1990;2(3):241–257.

• Symbols & Relationships

Page 16: Ken Sikaris - Melbourne Pathology - Previous Discovery in Pathology Data Mining

A/Prof Ken Sikaris 15th October 2013

Pathology: The study and diagnosis of

disease.

Masters

Evidence

Based

Literature

Database

Patterns

Pathology

Conclusions

Stored

Knowledge Experience Knowledge

Experience

Page 17: Ken Sikaris - Melbourne Pathology - Previous Discovery in Pathology Data Mining

A/Prof Ken Sikaris 15th October 2013

Pathology: The study and diagnosis of

disease.

Masters

Evidence

Based

Literature

Database

Patterns

Pathology

Conclusions

Stored

Knowledge Experience Knowledge

Experience

Page 18: Ken Sikaris - Melbourne Pathology - Previous Discovery in Pathology Data Mining

A/Prof Ken Sikaris 15th October 2013

Pathology: The study and diagnosis of

disease.

Masters

Evidence

Based

Literature

Database

Patterns

Pathology

Conclusions

Stored

Knowledge Experience Knowledge

Experience

Page 19: Ken Sikaris - Melbourne Pathology - Previous Discovery in Pathology Data Mining

A/Prof Ken Sikaris 15th October 2013

Knowledge discovery from databases

Fayyad U, Piatetsky-Shapiro G, Smyth P,

From Data Mining to Knowledge Discovery in Databases

Artificial Intelligence Magazine 1996;Fall:37-53

Page 20: Ken Sikaris - Melbourne Pathology - Previous Discovery in Pathology Data Mining

A/Prof Ken Sikaris 15th October 2013

Page 21: Ken Sikaris - Melbourne Pathology - Previous Discovery in Pathology Data Mining

A/Prof Ken Sikaris 15th October 2013

• 1. Heterogeneity of Data

– Volume & Complexity – MRI, CBC

– Physicians Interpretation – English, Synonyms

– Sensitivity & Specificity – All diagnoses imprecise

– Mathematical Models – Gaussian(?), Qualitative

– Canonical Forms – Different expressions (units)

• Liver secondaries, metastatic liver disease

• SNOMED, Synoptic reporting

Page 22: Ken Sikaris - Melbourne Pathology - Previous Discovery in Pathology Data Mining

A/Prof Ken Sikaris 15th October 2013

Page 23: Ken Sikaris - Melbourne Pathology - Previous Discovery in Pathology Data Mining

A/Prof Ken Sikaris 15th October 2013

• 2. Ethical / Legal / Social

– Data Ownership – Who can sell? (Not for sale.)

– Fear of Lawsuits – Unnecessary tests?

– Privacy and Security – Concealed identifiers

– Expected benefits – How big? Rare diseases?

– Administrative – Contractual agreements, Audits

Page 24: Ken Sikaris - Melbourne Pathology - Previous Discovery in Pathology Data Mining

A/Prof Ken Sikaris 15th October 2013

• 3. Statistical Philosophy

– Ambush – expected not found – but new one is!

• Training set and testing set

– Superset statistics

• Qualitative, changes with time, missing data

Page 25: Ken Sikaris - Melbourne Pathology - Previous Discovery in Pathology Data Mining

A/Prof Ken Sikaris 15th October 2013

8.7% of clinical

data may be

Unusable

Page 26: Ken Sikaris - Melbourne Pathology - Previous Discovery in Pathology Data Mining

A/Prof Ken Sikaris 15th October 2013

• 3. Statistical Philosophy

– Ambush – expected not found – but new one is!

• Training set and testing set

– Superset statistics

• Qualitative, changes with time, missing data

– Established Procedures

• Scientific Method

Page 27: Ken Sikaris - Melbourne Pathology - Previous Discovery in Pathology Data Mining

A/Prof Ken Sikaris 15th October 2013

Scientific method

• Verification Driven Data Search

• Graphs, tables, descriptions

Hypothesis

Observation

Experiment

• Discovery Driven Data Mining

• Linkage / Cluster analysis

• Finding similar segments

• Finding deviations in a segment

Page 28: Ken Sikaris - Melbourne Pathology - Previous Discovery in Pathology Data Mining

A/Prof Ken Sikaris 15th October 2013

Data Searching vs Data Mining

• Data Searching

– You know what you are looking for

• and where it is:

–SQL searches:

» How often does hyponatraemia cause death?

• Data Mining

– You know you are looking for (mortality)

• but don’t assume you know where it is.

Page 29: Ken Sikaris - Melbourne Pathology - Previous Discovery in Pathology Data Mining

A/Prof Ken Sikaris 15th October 2013

Page 30: Ken Sikaris - Melbourne Pathology - Previous Discovery in Pathology Data Mining

A/Prof Ken Sikaris 15th October 2013

Page 31: Ken Sikaris - Melbourne Pathology - Previous Discovery in Pathology Data Mining

A/Prof Ken Sikaris 15th October 2013

Page 32: Ken Sikaris - Melbourne Pathology - Previous Discovery in Pathology Data Mining

A/Prof Ken Sikaris 15th October 2013

110 120 130 140 150 160

Sodium

0

10

20

30

40

50

60

70

80

90

100

Mo

rta

lity

(%

)

Lowest Highest

Sodium & Inpatient Mortality

Page 33: Ken Sikaris - Melbourne Pathology - Previous Discovery in Pathology Data Mining

A/Prof Ken Sikaris 15th October 2013

1 2 3 4 5 6 7

Potassium

0

20

40

60

80

100

Mo

rtali

ty (

%)

Lowest Highest

Potassium & Inpatient Mortality

Page 34: Ken Sikaris - Melbourne Pathology - Previous Discovery in Pathology Data Mining

A/Prof Ken Sikaris 15th October 2013

Cholesterol and Inpatient Mortality

0 2 4 6 8 10 12

CHOLESTEROL

0

10

20

30

40

50

60

INP

AT

IEN

T M

OR

TA

LIT

Y %

Page 35: Ken Sikaris - Melbourne Pathology - Previous Discovery in Pathology Data Mining

A/Prof Ken Sikaris 15th October 2013

• 4. Special Status of Medicine

– Life and death

• Not luxury, pleasure or convenience product

– Long apprenticeship

– Medical Research

• Community responsibility

• Scientific truths can be used for ‘good’ or ‘evil’

Page 36: Ken Sikaris - Melbourne Pathology - Previous Discovery in Pathology Data Mining

A/Prof Ken Sikaris 15th October 2013

• ‘Markets’ for Laboratory Information

– Individual Test Results – Reference Intervals

– Profile of Test Results – Diagnostic algorithms

– Rare Tests – Accumulated experience

– Critical Results – For emergency planning

– Follow Up tests – To guide usefulness

– Trended results – For treatment guidance

Page 37: Ken Sikaris - Melbourne Pathology - Previous Discovery in Pathology Data Mining

A/Prof Ken Sikaris 15th October 2013

Data Elements in Pathology Data Mining

Patient Demographics

Gender

DOB

Patient Status

Physiological

Childhood,Pregnancy

Pathological

Analyte Results

Numerical

Numbers / Ordinal

Qualitative

Groupings / Text

Temporal Data

Date of Test, Date of repeat

Time of day, Season,

Gestation, Admission

Page 38: Ken Sikaris - Melbourne Pathology - Previous Discovery in Pathology Data Mining

A/Prof Ken Sikaris 15th October 2013

Folate deficiency trends

Mets & Sikaris, et al. MJA 2002; 176

(Since 1995)

Brown et al. MJA 2011; 194 (2): 65-67.

(Since Sept 2009)

Page 39: Ken Sikaris - Melbourne Pathology - Previous Discovery in Pathology Data Mining

A/Prof Ken Sikaris 15th October 2013

LUNG COLON

PROSTATE BREAST

Page 40: Ken Sikaris - Melbourne Pathology - Previous Discovery in Pathology Data Mining

A/Prof Ken Sikaris 15th October 2013

10 100 100020 30 40040

Ferritin

80

85

90

95

Media

n M

CV

YW

OW

M

ALL MEN

OLD WOMEN

YOUNG WOMEN

•Sikaris K.A., “Combining Clinical Biochemistry and Haematology Databases to define Predictive Values for Ferritin.” Clin Biochem Rev 1997;18:81.

Page 41: Ken Sikaris - Melbourne Pathology - Previous Discovery in Pathology Data Mining

A/Prof Ken Sikaris 15th October 2013

0 10 20 30 40 50 60 70 80 90 1000

Vit D

2.28

2.29

2.3

2.31

2.32

2.33

2.34

2.35

2.36

2.37

Post

menopausa

l C

orr

ecte

d C

alc

ium

2.23

2.24

2.25

2.26

2.27

2.28

2.29

2.3

2.31

2.32

Pre

menopausal C

orr

ecte

d C

alc

ium

Lu ZX, Dahanayaka K, Lambrianou J, Ratniake S, Sikaris KA, “How much Vitamin

D is sufficient? An evidence based approach.” Clin Biochem Rev 2007; 28:S29

Page 42: Ken Sikaris - Melbourne Pathology - Previous Discovery in Pathology Data Mining

A/Prof Ken Sikaris 15th October 2013

442

622

1002

1293

1584 15

83

1811

1952

1976

2079

2036

1950

1824

1709

1516

1371

1119

870

770

597

454

407310246

179

151

133

98

426

930

1196

1391

1691

1828

1649

1391

90561

5

384

196

136

950

1589

2268

2754

2907

3463

3605

3589

3940

4213

4058

4204

3915

3476

3220

2896

2307 19

3215

171153

949

678

590

464319

265

201

0 50 10065

Vitamin D

50

60

70

80

90

100P

rem

enopausal

Geom

etr

ic M

ean A

LP

3

4

5

6

7

8

All

Geom

etr

ic M

ean P

TH

Lu ZX, Dahanayaka K, Lambrianou J, Ratniake S, Sikaris KA, “How much Vitamin

D is sufficient? An evidence based approach.” Clin Biochem Rev 2007; 28:S29

Page 43: Ken Sikaris - Melbourne Pathology - Previous Discovery in Pathology Data Mining

A/Prof Ken Sikaris 15th October 2013

Sonic Reference Intervals 2009

Page 44: Ken Sikaris - Melbourne Pathology - Previous Discovery in Pathology Data Mining

A/Prof Ken Sikaris 15th October 2013

Page 45: Ken Sikaris - Melbourne Pathology - Previous Discovery in Pathology Data Mining

A/Prof Ken Sikaris 15th October 2013

Alkaline Phosphatase in Childhood

Page 46: Ken Sikaris - Melbourne Pathology - Previous Discovery in Pathology Data Mining

A/Prof Ken Sikaris 15th October 2013

Triglycerides in Childhood

Page 47: Ken Sikaris - Melbourne Pathology - Previous Discovery in Pathology Data Mining

A/Prof Ken Sikaris 15th October 2013

Cholesterol in Pregnancy

Page 48: Ken Sikaris - Melbourne Pathology - Previous Discovery in Pathology Data Mining

A/Prof Ken Sikaris 15th October 2013

Triglycerides in Pregnancy

Page 49: Ken Sikaris - Melbourne Pathology - Previous Discovery in Pathology Data Mining

A/Prof Ken Sikaris 15th October 2013

Hb in Pregnancy

Page 50: Ken Sikaris - Melbourne Pathology - Previous Discovery in Pathology Data Mining

A/Prof Ken Sikaris 15th October 2013

Lymphocytes in Pregnancy

Page 51: Ken Sikaris - Melbourne Pathology - Previous Discovery in Pathology Data Mining

A/Prof Ken Sikaris 15th October 2013

Neutrophils in Pregnancy

Page 52: Ken Sikaris - Melbourne Pathology - Previous Discovery in Pathology Data Mining

A/Prof Ken Sikaris 15th October 2013

Platelets in Pregnancy

Page 53: Ken Sikaris - Melbourne Pathology - Previous Discovery in Pathology Data Mining

A/Prof Ken Sikaris 15th October 2013

Page 54: Ken Sikaris - Melbourne Pathology - Previous Discovery in Pathology Data Mining

A/Prof Ken Sikaris 15th October 2013

Page 55: Ken Sikaris - Melbourne Pathology - Previous Discovery in Pathology Data Mining

A/Prof Ken Sikaris 15th October 2013

Page 56: Ken Sikaris - Melbourne Pathology - Previous Discovery in Pathology Data Mining

A/Prof Ken Sikaris 15th October 2013

Page 57: Ken Sikaris - Melbourne Pathology - Previous Discovery in Pathology Data Mining

A/Prof Ken Sikaris 15th October 2013

Page 58: Ken Sikaris - Melbourne Pathology - Previous Discovery in Pathology Data Mining

A/Prof Ken Sikaris 15th October 2013

Page 59: Ken Sikaris - Melbourne Pathology - Previous Discovery in Pathology Data Mining

A/Prof Ken Sikaris 15th October 2013

Page 60: Ken Sikaris - Melbourne Pathology - Previous Discovery in Pathology Data Mining

A/Prof Ken Sikaris 15th October 2013

Page 61: Ken Sikaris - Melbourne Pathology - Previous Discovery in Pathology Data Mining

A/Prof Ken Sikaris 15th October 2013

Brain

Mental Heart Dermatitis

Arthritis

Diabetes

Urinary

Drugs/Liver/HIV Respiratory

Page 62: Ken Sikaris - Melbourne Pathology - Previous Discovery in Pathology Data Mining

A/Prof Ken Sikaris 15th October 2013

Page 63: Ken Sikaris - Melbourne Pathology - Previous Discovery in Pathology Data Mining

A/Prof Ken Sikaris 15th October 2013

Page 64: Ken Sikaris - Melbourne Pathology - Previous Discovery in Pathology Data Mining

A/Prof Ken Sikaris 15th October 2013

Page 65: Ken Sikaris - Melbourne Pathology - Previous Discovery in Pathology Data Mining

A/Prof Ken Sikaris 15th October 2013

Page 66: Ken Sikaris - Melbourne Pathology - Previous Discovery in Pathology Data Mining

A/Prof Ken Sikaris 15th October 2013

Page 67: Ken Sikaris - Melbourne Pathology - Previous Discovery in Pathology Data Mining

A/Prof Ken Sikaris 15th October 2013

Page 68: Ken Sikaris - Melbourne Pathology - Previous Discovery in Pathology Data Mining

A/Prof Ken Sikaris 15th October 2013

Page 69: Ken Sikaris - Melbourne Pathology - Previous Discovery in Pathology Data Mining

A/Prof Ken Sikaris 15th October 2013

Page 70: Ken Sikaris - Melbourne Pathology - Previous Discovery in Pathology Data Mining

A/Prof Ken Sikaris 15th October 2013

Page 71: Ken Sikaris - Melbourne Pathology - Previous Discovery in Pathology Data Mining

A/Prof Ken Sikaris 15th October 2013

Public-Private Partnerships

Funding

Demonstrations for opinion leaders

Coding scheme standards

Multidisciplinary teams

Benchmark other ‘markets’

Page 72: Ken Sikaris - Melbourne Pathology - Previous Discovery in Pathology Data Mining

A/Prof Ken Sikaris 15th October 2013

Page 73: Ken Sikaris - Melbourne Pathology - Previous Discovery in Pathology Data Mining

A/Prof Ken Sikaris 15th October 2013

Page 74: Ken Sikaris - Melbourne Pathology - Previous Discovery in Pathology Data Mining

A/Prof Ken Sikaris 15th October 2013

Page 75: Ken Sikaris - Melbourne Pathology - Previous Discovery in Pathology Data Mining

A/Prof Ken Sikaris 15th October 2013

Page 76: Ken Sikaris - Melbourne Pathology - Previous Discovery in Pathology Data Mining

A/Prof Ken Sikaris 15th October 2013

Page 77: Ken Sikaris - Melbourne Pathology - Previous Discovery in Pathology Data Mining

A/Prof Ken Sikaris 15th October 2013

SUMMARY

• Pathology/Medical Databases

– Unique technical issues

– Complex legal, ethical, confidentiality ownership issues

– New paradigms in scientific/statistical analysis

– Special status: Life/death, medicine, community values

• Data mining and pathology databases

– A repository of established knowledge

– A source for new knowledge

– A framework for clinical decision making