Mak akin ing sen ense o e of big ig da data in a in hea ealt - - PowerPoint PPT Presentation

▶
mak akin ing sen ense o e of big ig da data in a in hea
SMART_READER_LITE
LIVE PREVIEW

Mak akin ing sen ense o e of big ig da data in a in hea ealt - - PowerPoint PPT Presentation

Jul 07, 2023 188 likes •575 views

latrob obe.edu.a .edu.au Mak akin ing sen ense o e of big ig da data in a in hea ealt lthca care re wi with ef effec ectiv ive e an and d in inter erac activ ive e an anal alys ysis is Dr. Li Lianu anuhu hua (Lin

slide-1
SLIDE 1

latrob

  • be.edu.a

.edu.au

La Trobe University CRICOS Provider Code Number 00115M

Mak akin ing sen ense o e of big ig da data in a in hea ealt lthca care re wi with ef effec ectiv ive e an and d in inter erac activ ive e an anal alys ysis is

  • Dr. Li

Lianu anuhu hua (Lin ina) a) Chi

Depar artm tment ent of Comput uter er Scie ienc nce e and Informat

  • rmation

ion Techn hnolo

  • logy

gy

slide-2
SLIDE 2

latrob

  • be.edu.a

.edu.au

Slide 2 | Version 2

A b A bit it ab about

  • ut me

me

2009 2009 02/201 018

PhD Period iod at HUST (China na) ) and UTS (Aust stralia ralia) )

On-Line Analytical Processing System (OLAP)

201 2013

Data Warehouse Modeling Tool (DWDesigner)

“Best Paper Award” at PAKDD KDD20 2013

Big data fast response: real-time classification

  • f big data

stream

IBM M Resear earch ch Aust stral alia ia

Data Storming Survey and Data Visualisation

Machine ne Learni ning ng Resear search h Area

SURO: Surgical Unit Resource Optimisation (ML Section) Bendigo Health: (ML) Watson Health

Data a Mining ng Resear search Area Global ally ly Top p 200 You

  • ung

ng Resear searcher hers s Aw Awar arded ded by 4th

th

HLF Exter erna nal l Honors 2017 2017 by IBM Machine ne Learni ning ng Natura ral Langua nguage Proces essing sing You

  • ung

ng Global al Change nger by by Think nk 20 20 /G20 Romberg mberg Grant nt Aw Awar ard d by by 4th

th HLF

05/2015 015

La La Trobe University ersity AI on Digital ital Health lth

slide-3
SLIDE 3

latrob

  • be.edu.a

.edu.au

Slide 3 | Version 2

No No ind ndus ustr try cou

  • unt

nts more tha han health lthcar care

slide-4
SLIDE 4

latrob

  • be.edu.a

.edu.au

Slide 4 | Version 2

No No ind ndus ustr try cou

  • unt

nts less ss tha han he healthc lthcare are

slide-5
SLIDE 5

latrob

  • be.edu.a

.edu.au

Slide 5 | Version 2

Ever ery yea ear 18,000+ 18,000+ Aust stralia ralian de deat aths hs are are cau aused sed by by medica cal error, 12,000 12,000 of

  • f whom wer

were e dying becaus ause e of preventabl entable e events nts.

(This was published in 1995 and it is likely increasing each year) * ABS: Transport accidents: 1,402 deaths registered in 2008. See MJA: The incidence and cost of adverse events in Victorian hospitals 2003–04

A survey (publi blish shed ed in in 1999): 1999):

14,000 admissions surveyed 16.6% associated with an “adverse erse event nt” 51% of the adverse events considered preventa ntabl ble 4.9% the patient died

slide-6
SLIDE 6

latrob

  • be.edu.a

.edu.au

Slide 6 | Version 2

The total cost of adverse rse events ts (in the ye year r 2003 2003–04 04 to selecte ted d Victori rian an hospita tals ls)”:

  • $460.311 million
  • representing 15.7% of the total

expenditure on direct hospital costs

  • r an additional 18.6% of the total

inpatient hospital budget.

Adverse events are associated with significant costs. Administrative datasets are a cost-effective source of information that can be used for a range of clinical governance activities to prevent adverse events.

slide-7
SLIDE 7

latrob

  • be.edu.a

.edu.au

Slide 7 | Version 2

The he Return turn On On In Invest estment ment (R (ROI OI) on

  • n big

big data data approa

  • aches

ches wi will be be even hi high gher er for

  • r he

healthcare thcare tha han ot

  • ther

her ind ndus ustrie tries*.

* Source from Harvard Medical School.

slide-8
SLIDE 8

latrob

  • be.edu.a

.edu.au

Slide 8 | Version 2

https://www.aihw.gov.au/reports- statistics/health-welfare-overview/health- welfare-expenditure/overview

slide-9
SLIDE 9

latrob

  • be.edu.a

.edu.au

Slide 9 | Version 2

Necessar essary con

  • nditions

ditions for

  • r data to

to ta take on

  • n new li

life in in Health althcare: are:

Incenti entives es Ac Access ess to to Da Data ta

slide-10
SLIDE 10

latrob

  • be.edu.a

.edu.au

Slide 10 | Version 2

Digi gital tal Health th Data in in Aus ustra trali lia

slide-11
SLIDE 11

latrob

  • be.edu.a

.edu.au

Slide 11 | Version 2

Digi gital tal Health th in in Aust stra rali lia

My Health Record Health Identifier Service TeleHealth

slide-12
SLIDE 12

latrob

  • be.edu.a

.edu.au

Slide 12 | Version 2

My My Health th Record

  • rd

Population ulation registr gistration ation: As of 23 March 2016, there was a total of 2,642,278 active digital records, approximately 11% of the current Australian population.

From: Evolution of eHealth in Australia, Achievements, lessons, and opportunities

slide-13
SLIDE 13

latrob

  • be.edu.a

.edu.au

Slide 13 | Version 2

My My Health th Record

  • rd

Organi anisat atio ion regist gistrat ration ion: As of 23 March 2016, a total of 8,139 organisations were registered in the My Health Record system. This number continues to increase steadily

slide-14
SLIDE 14

latrob

  • be.edu.a

.edu.au

Slide 14 | Version 2

Health th Id Ident ntifier ifier (H (HI) I) Servic vice

Organisations registered in the HI Service by state / territory and type – 29 Feb 2016

slide-15
SLIDE 15

latrob

  • be.edu.a

.edu.au

Slide 15 | Version 2

Teleheal ehealth th

For the 2014-15 financial year, there were over 84,000 MBS claims for telehealth consultations1. A CSIRO telehealth and monitoring study found a 37% reduction in mortality with the use of the Telemedcare Clinical Monitoring Unit in the management of chronic disease in the home.

1 Source: Medicare Statistics, Medicare Item Reports July 2014–July 2015. 2 Results of the CSIRO multi‐site national trial of telehealth for the management of chronic disease in the home, in Health Informatics Conference, Brisbane , 2015.

slide-16
SLIDE 16

latrob

  • be.edu.a

.edu.au

Slide 16 | Version 2

In Incenti entives es in in Aus ustra trali lia

slide-17
SLIDE 17

latrob

  • be.edu.a

.edu.au

Slide 17 | Version 2

The Aus ustralian tralian heal althcare thcare sy syst stem em is s pot

  • tentia

ntially lly de deal aling ng with h two

  • ma

main n prob

  • blems1

lems1

1 A review of the Australian healthcare system: A policy perspective, 2018

? Resource source Alloca location tion ? Perf rformanc

  • rmance

e and and Patien Patient t Out utcomes comes Impro provements ements

slide-18
SLIDE 18

latrob

  • be.edu.a

.edu.au

Slide 18 | Version 2

Data Ana nalytic ytics

slide-19
SLIDE 19

latrob

  • be.edu.a

.edu.au

Slide 19 | Version 2

Wh Which diabetic etics are are likely ely to to benefi nefit fr from

  • m

int nter erventi ention?

  • n?
slide-20
SLIDE 20

latrob

  • be.edu.a

.edu.au

Slide 20 | Version 2

Repor

  • rting

ting

Which diabetics are likely to benefit from intervention? Select count(PatientID) from PatientTable join PatientID from Claims where ICD9 is (‘250*’) or NDC is (‘25545565’, ‘34982738’) and VisitDate between (“1/1/2010”, ’31/12/2010’) = null

slide-21
SLIDE 21

latrob

  • be.edu.a

.edu.au

Slide 21 | Version 2

Risk sk Scor

  • res

es

Whi hich diabeti betics cs are likely ely to to bene nefit fit from int nterventi ntion? n?

Claim aim data data Logist istic ic Regre ress ssion ion Code or

  • r

Rules es

  • Leukemi

emia

  • Diabetes

abetes

  • Heart failure
  • Dementia
  • COPD
  • etc
slide-22
SLIDE 22

latrob

  • be.edu.a

.edu.au

Slide 22 | Version 2

  • He

Heavy avy de dependen endence ce on

  • n di

dise sease ase co code des

  • Risk

sk sc scor

  • res

es of

  • ften on
  • ne si

size ze fits al all – al all di disease sease, pop

  • pulati

ulation,

  • n, etc
slide-23
SLIDE 23

latrob

  • be.edu.a

.edu.au

Slide 23 | Version 2

How to to answer er th these se question estions fr from

  • m Do

Docto tors? rs?

  • I know who cost more last year, I need to know which of my diabetics is

most likely to end up in the emergency department soonest.

  • I already know these five people I am going to see them in any way.

Tell me who’s pre-diabetics that’s heading toward diabetes that I can do something about that’s very different.

slide-24
SLIDE 24

latrob

  • be.edu.a

.edu.au

Slide 24 | Version 2

How Su Super pervised vised Machine chine Learning arning Wor

  • rks

ks

Whi hich h of my y diabetics betics is most t likely ly to to end nd up i in th n the emergen rgency y department? tment?

Diabetic abetics with ED ED Diabetic abetics without hout ED ED Mach chine ine Learning ning Classif assific icatio ation Model el

slide-25
SLIDE 25

latrob

  • be.edu.a

.edu.au

Slide 25 | Version 2

How Su Super pervised vised Machine chine Learning arning Wor

  • rks

ks

Clas assifi ificat cation ion Model

95% 95% Diabetic abetics 90% 90% 89% 89%

slide-26
SLIDE 26

latrob

  • be.edu.a

.edu.au

Slide 26 | Version 2

Machin chine Learning arning con

  • nsiders

siders mo more re data ta

Risk of

  • f

re re-admission admission Diabeti betic likely ely to to be be admi mitted tted

  • Length of Stay
  • Admitted via ED
  • # ED visits
  • Length of Stay
  • Admitted via ED
  • # ED visits
  • All co-morbidities
  • Substance Issues
  • Specialty Provider Types
  • Behavioral Health
  • Nurses’ impressions
  • Etc……
slide-27
SLIDE 27

latrob

  • be.edu.a

.edu.au

Slide 27 | Version 2

  • More tolerant

erant to to missing, g, inaccurat curate data

  • Models

ls learn with new data

Machine chine Learning arning

slide-28
SLIDE 28

latrob

  • be.edu.a

.edu.au

Slide 28 | Version 2

Resource source Allocatio location – Ho Hospit spital al Waiting ting Time mes

Problem: blem:

How many patients should we operate on every day for the next month in order to increase the hospital revenue, in respect to the limited number of beds and operating theatres?

slide-29
SLIDE 29

latrob

  • be.edu.a

.edu.au

Slide 29 | Version 2

https://www.aihw.gov.au/reports/aust ralias-health/australias-health 2018/contents/indicators-of- australias-health/waiting-times-for- elective-surgery

Median ian Wait iting ing Times mes for Selec lected ted Elec ectiv tive Surger gery, 2016 2016-20 2017 17

slide-30
SLIDE 30

latrob

  • be.edu.a

.edu.au

Slide 30 | Version 2

Pati tient ent Flo low in in a Ho Hospital spital

slide-31
SLIDE 31

latrob

  • be.edu.a

.edu.au

Slide 31 | Version 2

  • Time Horizon
  • Operating rooms
  • Ward beds
  • ICU beds
  • Departments
  • Master Surgery Schedule
  • Number of changes to

MSS

In Input ut Paramet rameters ers

  • Surgeons and specialists
  • Surgery type
  • Department
  • Specialist type
  • Urgency category
  • Duration of Surgery
  • Length of stay
  • Chance of using ICU

Hosp spital tal Resources

  • urces

Pat Patient ient Wai aitin ting List st

slide-32
SLIDE 32

latrob

  • be.edu.a

.edu.au

Slide 32 | Version 2

How Unsuper supervised vised Machine chine Learning arning Wor

  • rks

ks

Surger gery Clusteri ering Pati tient nt Clas assifi ificat cation ion to to surger gery clusters ters Adjust st Waiti ting ng List

slide-33
SLIDE 33

latrob

  • be.edu.a

.edu.au

Slide 33 | Version 2

Effect ctiv ive Da Data ta An Analytics alytics

  • Needs to be usable - not just to scientists
  • You cannot settle for a one-size-fits-all approach to

understanding data

  • Can not ignore the free text. Results must be interpretable by

clinicians (interactive)

  • Embed as software into existing clinical/operational

workflows and systems

slide-34
SLIDE 34

latrob

  • be.edu.a

.edu.au

Slide 34 | Version 2

An Anatom tomy y of

  • f a He

Healthcare althcare In Inte tell lligent igent Sy Syste tem m – Watson tson for

  • r Be

Benefits efits

►Problem:

blem: Documents describing health insurance coverage are long and complex and exclusion criteria are not easily accessible

►Solution

ution: : Automate processing of documents to extract and summarise information (ML and NLP)

slide-35
SLIDE 35

C o m p a r i s o n V i s u a l i z a t i o n H e a l t h c a r e S y s t e m

slide-36
SLIDE 36

latrob

  • be.edu.a

.edu.au

Slide 36 | Version 2

Time

Comparison mparison Vi Visua suali lization ation

Tempor

  • ral

al stati tistics stics comparison rison of health th systems ems

Overview Statistics Visualisation

  • Chronic/Non-Chronic
  • Stratified by age, gender, geo
  • Temporal trend

Disea ease se- Mortalit ality y Rate

  • Patient paid
  • Government paid
  • Gap paid

Disease ease- Expen enses ses

  • Prescribed drug pattern - Disease

Disea ease se-Drugs Drugs

  • Geo Distribution
  • Patient visit frequency

Service vice Provide vider- Patient ent Visits its

slide-37
SLIDE 37

latrob

  • be.edu.a

.edu.au

Slide 37 | Version 2

Ou Our Digital tal Hea ealth th Tea eam:

Diabetes betes predic ictio tion: Morta tali lity ty rate te predict iction ion and nd co cont ntrol: l:

Prof. Phoebe Chen

Int nteractiv ractive Visuali lisa satio tion Ment ntal l Health th Research arch (Auti tism sm) ) : Drink nking ing outc tcome predicti iction:

  • n:
  • Dr. Dennis

Wollersheim

  • Dr. Oliver

Stanesby Prof. Emmanuel Kuntsche Prof. Henry Duh

Ou Our Par Partn tners: ers:

slide-38
SLIDE 38

latrob

  • be.edu.a

.edu.au

Slide 38 | Version 2

My My Co Contact tact De Deta tail ils:

Dr Lianhua ua Chi Lecturer turer in In Informati ation

  • n Techn

hnolo

  • logy

gy School

  • l of Engineering

eering an and M d Mat athem hematical atical Sciences ences La Trobe be Un University ersity | Victoria

  • ria

E: E: L.chi@l i@latrobe atrobe.edu.au .edu.au T: T: 03 9 9479 2454 W: W: https ps://lia //lianhua1 nhua1221. 221.gi github thub.io/ .io/