www.swansea.ac.uk challenges in predicting patient pathways dr rajesh ransing school of engineering...

24
www.swansea.ac.uk Challenges in Predicting Patient Pathways Dr Rajesh Ransing School of Engineering Professor Mike Gravenor School of Medicine Grand Challenges in Information Driven Health Care Workshop

Upload: erik-howard

Post on 29-Dec-2015

218 views

Category:

Documents


1 download

TRANSCRIPT

Page 1: Www.swansea.ac.uk Challenges in Predicting Patient Pathways Dr Rajesh Ransing School of Engineering Professor Mike Gravenor School of Medicine Grand Challenges

www.swansea.ac.uk

Challenges in Predicting Patient

Pathways Dr Rajesh RansingSchool of Engineering

Professor Mike Gravenor School of Medicine

Grand Challenges in Information Driven Health Care Workshop

Page 2: Www.swansea.ac.uk Challenges in Predicting Patient Pathways Dr Rajesh Ransing School of Engineering Professor Mike Gravenor School of Medicine Grand Challenges

www.swansea.ac.uk

Challenges in Predicting Patient Pathways

Driving Force?

Earlier and better detection

Accurate and reliable decision making

Encouraging self-care i.e. taking patients in the decision making loop

Limited resources – Time and Money

Page 3: Www.swansea.ac.uk Challenges in Predicting Patient Pathways Dr Rajesh Ransing School of Engineering Professor Mike Gravenor School of Medicine Grand Challenges

www.swansea.ac.uk

Challenges in Predicting Patient Pathways

Data Explosion

Google world (internet, search, instant answers)

Post Genomic era

We have too much data

Goal

Self-evolving, self-learning computers to digest data and extract useful information/knowledge

Page 4: Www.swansea.ac.uk Challenges in Predicting Patient Pathways Dr Rajesh Ransing School of Engineering Professor Mike Gravenor School of Medicine Grand Challenges

www.swansea.ac.uk

Challenges in Predicting Patient Pathways

Page 5: Www.swansea.ac.uk Challenges in Predicting Patient Pathways Dr Rajesh Ransing School of Engineering Professor Mike Gravenor School of Medicine Grand Challenges

www.swansea.ac.uk

Challenges in Predicting Patient Pathways

Page 6: Www.swansea.ac.uk Challenges in Predicting Patient Pathways Dr Rajesh Ransing School of Engineering Professor Mike Gravenor School of Medicine Grand Challenges

www.swansea.ac.uk

Challenges in Predicting Patient Pathways

We can not deviate from the good old ways of Diagnosis.

Patients need professional consultation with doctors.

Early and accurate diagnosis is important

We need tools to aid their decision making process with minimum interference.

Page 7: Www.swansea.ac.uk Challenges in Predicting Patient Pathways Dr Rajesh Ransing School of Engineering Professor Mike Gravenor School of Medicine Grand Challenges

www.swansea.ac.uk

Challenges in Predicting Patient Pathways

The right treatment for the right person at the

right time

Trial and Error

Personalized MedicineCurrent Practice

One size fits all One size fits all

Page 8: Www.swansea.ac.uk Challenges in Predicting Patient Pathways Dr Rajesh Ransing School of Engineering Professor Mike Gravenor School of Medicine Grand Challenges

www.swansea.ac.uk

DiseasesDiseasesDiseases

DiseasesDiseasesDiseases

Diseases

Anatomy

Anatomy

Anatomy

Anatomy

Anatomy

Anatomy

Genes

Genes

Genes

Genes

Genes

Gen

es

Physiology Physiolog

y Physiology Physiolog

y Physiology Physiology

Diseases

PhysiologyAnatomy

Genes

Genes

GenesDiseases

Diseases

Medical Informatics

Bioinformatics

Novel relationships & Deeper insights

Page 9: Www.swansea.ac.uk Challenges in Predicting Patient Pathways Dr Rajesh Ransing School of Engineering Professor Mike Gravenor School of Medicine Grand Challenges

www.swansea.ac.uk

Challenges in Predicting Patient Pathways

Interdisciplinary Approach Health Care Providers – Hospitals – IHC

Actual patient data

Collaboration with Computer Scientists, Engineers, Clinicians, Health Informatics colleagues, Patients, Nurses

Data Analysis and Machine Learning software tools

MetaCause – Machine Learning

GeneCIS – Clinical Data Capturing System

Autonomy – Meaning based symbolic processing

Page 10: Www.swansea.ac.uk Challenges in Predicting Patient Pathways Dr Rajesh Ransing School of Engineering Professor Mike Gravenor School of Medicine Grand Challenges

www.swansea.ac.uk

MetaCause: Swansea University Spin Out

ObjectiveObjective: Develop Self-learning Process Optimisation and Diagnosis Software.

Financial Supporters:Financial Supporters: (~£1M, 10 Person Years)

Engineering and Physical Sciences Research Council (EPSRC)

KEF Collaborative Industrial Research Project (Welsh Assembly

Government)

Industrial Partners:Industrial Partners:

Consortium of 7 foundries and Cast Metal Federation

Rolls Royce Plc, Tritech Precision Components Ltd

Blaysons Olefins Ltd, Wall Colmonoy Ltd, MB Fine Arts Ltd

Kaye Presteigne Ltd, MA Edwards Ltd

Page 11: Www.swansea.ac.uk Challenges in Predicting Patient Pathways Dr Rajesh Ransing School of Engineering Professor Mike Gravenor School of Medicine Grand Challenges

www.swansea.ac.uk

DiseasesDiseasesDiseases

DiseasesDiseasesDiseases

Diseases

Anatomy

Anatomy

Anatomy

Anatomy

Anatomy

Anatomy

Genes

Genes

Genes

Genes

Genes

Gen

es

Physiology Physiolog

y Physiology Physiolog

y Physiology Physiology

Diseases

PhysiologyAnatomy

Genes GenesDiseases

Diseases

MetaCause is proven for Aerospace Applications

Novel relationships & Deeper insights

Page 12: Www.swansea.ac.uk Challenges in Predicting Patient Pathways Dr Rajesh Ransing School of Engineering Professor Mike Gravenor School of Medicine Grand Challenges

www.swansea.ac.uk

Mission Statement

Earlier and better detection

Identify high risk patient groups and monitor them

Recognise patterns in genetic/clinical data and medical history

Identify main effects/interactions to predict risk factors

Develop a self-evolving software

Accurate and reliable decision making

Combine risks together and aid decision making

Reduce overall cost for NHS

Page 13: Www.swansea.ac.uk Challenges in Predicting Patient Pathways Dr Rajesh Ransing School of Engineering Professor Mike Gravenor School of Medicine Grand Challenges

www.swansea.ac.uk

1). Validation Studies

Data: Fitness and metabolic measures in childrenOn-going population studies, SAIL linked

Risk Outcomes: precursors of diabetic and cardiac conditions

Fairly well defined and understood system

Page 14: Www.swansea.ac.uk Challenges in Predicting Patient Pathways Dr Rajesh Ransing School of Engineering Professor Mike Gravenor School of Medicine Grand Challenges

www.swansea.ac.uk

1). Validation Studies: Metabolic Syndrome

Risk factors for metabolic syndrome

Factor Effect Size Confirmed with logistic regression

Waist moderate ✓

Hips moderate ✓

Skinfolds moderate ✓

HDL-C moderate ✓

Interactions

Waist * Hips Strong ✓

Page 15: Www.swansea.ac.uk Challenges in Predicting Patient Pathways Dr Rajesh Ransing School of Engineering Professor Mike Gravenor School of Medicine Grand Challenges

www.swansea.ac.uk

1). Validation Studies: correlates of fitness

Factors associated with fitness test scores (top 10)

1. BMI 6. HOMA

2. Hips 7. Waist

3. Skinfolds 8. LDL-C

4. DBP 9. Total Cholesterol

5. Triglycerides 10. SBP

Possible advantages1. Detection of interactions (automatic, very large number of interactions) expected and detected)2. Non-linear trends in quantitative variables(good at detecting threshold effects when linear model doesn’t fit very well)

Page 16: Www.swansea.ac.uk Challenges in Predicting Patient Pathways Dr Rajesh Ransing School of Engineering Professor Mike Gravenor School of Medicine Grand Challenges

www.swansea.ac.uk

Page 17: Www.swansea.ac.uk Challenges in Predicting Patient Pathways Dr Rajesh Ransing School of Engineering Professor Mike Gravenor School of Medicine Grand Challenges

www.swansea.ac.uk

Page 18: Www.swansea.ac.uk Challenges in Predicting Patient Pathways Dr Rajesh Ransing School of Engineering Professor Mike Gravenor School of Medicine Grand Challenges

www.swansea.ac.uk

2). Whole Genome Studies

Data: 1434 Single Nucleotide Polymorphisms in DNA samples

Risk Outcomes: Diabetes (type I), case (n=895) control (n=817)

SNP effects not previously well known

Aim is to create short list of most important SNPs

Page 19: Www.swansea.ac.uk Challenges in Predicting Patient Pathways Dr Rajesh Ransing School of Engineering Professor Mike Gravenor School of Medicine Grand Challenges

www.swansea.ac.uk

2). Whole Genome Studies: statistical approaches

Standard methods : n separate individual 2 testsrank by p-valueDetermine cut-off for significance after correcting for multiple testing

MetaCause: Consider all SNPs together (and interactions)

As expected both Methods identify strongest signal (1 SNP, odds ratio = 3.0, large sample size (few missing values))

What is the effect of method choice on ‘short list’ of candidate genes?

Page 20: Www.swansea.ac.uk Challenges in Predicting Patient Pathways Dr Rajesh Ransing School of Engineering Professor Mike Gravenor School of Medicine Grand Challenges

www.swansea.ac.uk

2). Whole Genome Studies: comparison of methods

‘Significant’ with Statistical Tests

‘Significant’ with MetaCause

Yes No

Yes 5 6

No 15 1408

Where do they differ and Why?

Page 21: Www.swansea.ac.uk Challenges in Predicting Patient Pathways Dr Rajesh Ransing School of Engineering Professor Mike Gravenor School of Medicine Grand Challenges

www.swansea.ac.uk

2). Main Categories of Misclassification (So far!)

1. p-value vs odds ratio (clinical vs statistical significance)Closer correlation between MetaCause and SNPs ranked by odds ratio than p-valueThose SNPs short listed by MetaCause but not statistically significant were found to

have large odds ratios 2. Consideration of interactions (automatically searched for in MetaCause)

interactions involving ‘non-significant’ SNPs. 3. Consideration of population size. Risky rare genotypes have less “impact“

at the population level.

Challenge: Need to clearly define study questions (and hence functions of risk to optimised):

Individual SNP effects or interactions?Individual or population risk?

Page 22: Www.swansea.ac.uk Challenges in Predicting Patient Pathways Dr Rajesh Ransing School of Engineering Professor Mike Gravenor School of Medicine Grand Challenges

www.swansea.ac.uk

PubMed

Medical Informatics

Patient Records

Patient Records

Disease Database

Disease Database→Name→Synonyms→Related/Similar Diseases→Subtypes→Etiology →Predisposing Causes→Pathogenesis→Molecular Basis→Population Genetics→Clinical findings→System(s) involved→Lesions →Diagnosis→Prognosis→Treatment→Clinical Trials……

Clinical Trials

Clinical Trials

Bioinformatics

Genome

Transcriptome

Proteome

Interactome

Metabolome

Physiome

Regulome Variome

Pathome

Ph

arm

acog

enom

e

Disease

World

OMIM

►Personalized Medicine►Decision Support System►Patient Pathways►Diagnostic Test Selector►Clinical Trials Design►Hypothesis Generator…..

Data Mining

the Ultimate Goal…….

Page 23: Www.swansea.ac.uk Challenges in Predicting Patient Pathways Dr Rajesh Ransing School of Engineering Professor Mike Gravenor School of Medicine Grand Challenges

www.swansea.ac.uk

Page 24: Www.swansea.ac.uk Challenges in Predicting Patient Pathways Dr Rajesh Ransing School of Engineering Professor Mike Gravenor School of Medicine Grand Challenges

www.swansea.ac.uk