Weitere ähnliche Inhalte
Ähnlich wie IBM Terkko Pop-up Presentation by Pekka Leppänen (20)
Kürzlich hochgeladen (20)
IBM Terkko Pop-up Presentation by Pekka Leppänen
- 1. 1 © IBM 2018
Pekka Leppänen
IBM Healhcare
+358 40 75 88 106
leppanen@fi.ibm.com
About Watson and Healthcare
- 2. 2 © IBM 2018
1. Too much data, but not enough money
2. Data Science + Health = Watson Health
3. Examples
• Watson for Oncology, Watson for Genomics, Watson for Clinical Trials Matching
• Watson Conversation
• Watson Content Analytics
Discussion Topics
- 3. 3 © IBM 2018
We live longer – but cannot afford it
Life expectancy increases by 2 months every year (2016: Boys -2 months, Girls: 0).
Until 2012 over 1/2 of population was working, 2030 through 2040 only 1/3.
We have run out of money and the situation will get worse until 2040.
Healthcare needs to transform from labor intensive to information intensive.
Fee-for-Service model is being replaced by Value Based Care, with focus on Population Health
www.stat.fi/til/kuol/2016/01/kuol_2016_01_2017-10-27_tie_001_en.html
www.thl.fi/en/tilastot/tilastot-aiheittain/sosiaali-ja-terveydenhuollon-talous/terveydenhuollon-menot-ja-rahoitus
- 4. 4 © IBM 2018
McGovern, L., Miller, G., Hughes-Cromwick, P., Mays, G., Lantz, P., Lott, R. (2014). The relative contribution of multiple determinants to health
outcomes. Health Affairs (Robert Wood Johnson Foundation), 33, 2.
- 5. 5 © IBM 2018
It’s humanly impossible to keep up
with the knowledge and the data…
In medicine, there’s a gap between
what we know and what we do…
This rising tide of
information contains
insights critical to
your success
24 months
Frequency at
which
healthcare
data doubles2
80%
of medical data is
invisible
because it’s
unstructured1
>1PB
The amount of health-
related data a person
generates in their
lifetime3
45%
of medicine is not
evidence based4
17 years
Time it takes to translate
science to practice5
A New Reality in Healthcare
1. ASCO Releases Its First-Ever Report on the State of Cancer Care in America. Available at: http://www.ascopost.com/issues/april-15-2014/asco-releases-its-first-ever-report-on-the-state-of-cancer-care-in-america/. Accessed November 15, 2016.
2. Marconi, Katherine and Lehmann, Harold. Big Data and Health Analytics. CRC Press, 2014. Available at: http://bit.ly/1UjEtLL. Accessed June 3, 2016
3. Managed Care, January 2011 and HealthAffairs Blog, July 2011
4. Elizabeth A. McGlynn, Ph.D., Steven M. Asch, M.D., M.P.H., John Adams, Ph.D., Joan Keesey, B.A., Jennifer Hicks, M.P.H., Ph.D., Alison DeCristofaro, M.P.H., and Eve A. Kerr, M.D., M.P.H. N Engl J Med 2003; 348:2635-2645 June 26, 2003. DOI: 10.1056/NEJMsa022615. Available at:
http://www.nejm.org/doi/full/10.1056/NEJMsa022615#t=abstract
5. Slote Morris, Zoë & Wooding, Steven & Grant, Jonathan. (2011). The answer is 17 years, what is the question: Understanding time lags in translational research. Journal of the Royal Society of Medicine. 104. 510-20. 10.1258/jrsm.2011.110180. Available at:
https://www.researchgate.net/publication/51897868_The_answer_is_17_years_what_is_the_question_Understanding_time_lags_in_translational_research6.
- 6. 6 © IBM 2018
Value-based care presents a problem of scale
FULL RISK
Brings new
responsibility for
managing overall health
Optimized
health
outcomes,
maximized
revenue
OPTIMAL CLINICAL
DELIVERY
Requires transformative
care management as the
population expands
Population under
management
Value-based
reimbursement
Scaling up to value-based care
Max transformation
Max risk
- 7. 7 © IBM 2018
1. Too much data, but not enough money
2. Data Science + Health = Watson Health
3. Examples
• Watson for Oncology, Watson for Genomics, Watson for Clinical Trials Matching
• Watson Conversation
• Watson Content Analytics
Discussion Topics
- 8. 8 © IBM 2018
Watson Health is forging a partnership
between humans and machines
Together, we can…
Generate Remarkable Outcomes
Help Accelerate Discovery
Create Essential Connections
Enable Heightened Confidence
People excel at:
Common
sense
Dilemmas Morals Compassion Imagination Dreaming Abstraction Generalization
Cognitive systems excel at:
Natural
Language
Pattern
Identification
Locating
Knowledge
Machine
Learning
Eliminate
Bias
Endless
Capacity
- 9. 9 © IBM 2018
Cognitive systems are generally defined by the ability to
understand, reason, learn, and interact
UNDERSTAND
Cognitive systems can
understand unstructured
information the same
way humans do
REASON
They can reason, grasp
underlying concepts,
form hypotheses, and
infer to extract ideas
LEARN
Each data point,
interaction and outcome
helps to continuously
sharpen expertise
INTERACT
With abilities to see,
talk and hear, cognitive
systems interact with
humans in a natural way
- 10. 10 © IBM 2018
What if a solution could help you derive the
insights you need, when you need them?
Turn information
into insights
Keep up with growing
volumes of data
Acquire information
from many disparate
sources
- 11. 11 © IBM 2018
In light of this disruption, digital technologies have become a
source for creating new value in healthcare
Artificial intelligence
Simulation of human intelligence processes
Robotics
Conception, design,
manufacture, and
operation of robots
Machine learning
systems
Ability to learn and
improve without explicit
instructions
Natural language
processing
Ability to understand human
speech as it is spoken
Deep
learning
Machine learning with artificial
neural network algorithms
Predictive
analytics
Predicting outcomes
using statistical
algorithms and
machine learning
Recommendation
engines
Analyze data and
suggest something as
per user’s interest
Cognitive Technologies
Create new ways of interacting with
customers
Reveal more powerful on-demand business
insights through real-time access to data
Enable business model – and ecosystem –
transformation
Business Value
- 12. 12 © IBM 2018
Today, cognitive computing provides both strategic and
financial value while improving quality care to a patient
Re-invent
client
engagement
Digitize and
streamline
processes
Deploy
disruptive
business
models
§ Customer satisfaction
§ Revenue growth
(shorter sales cycle)
§ Cost reduction
(headcount)
§ Customer satisfaction
§ Customer retention
§ Cost reduction
(operational)
§ Revenue growth (new
product development)
§ Revenue acceleration
(speed to market
entry)
Scaling
human
expertise
cost-
effectively
Natural
language
interactions
Streamlining,
standardizing,
and improving
decision
processes
Business
Challenge
Example of how Cognitive helps today
Primary Strategic
Value
Primary Financial Value Client Impact
Time-
consuming
search
Evidence-
based
responses
Yielding
insights
unattainable
by humans
ENGAGEMENT
§ Complex patient engagement
scenarios with different past histories
§ Provide patients comprehensive
evidence based answers to complex
questions
DISCOVER
§ Help patients with insights far above
human levels
§ Finds insights and connections,
understands the vast amounts of
information available
DECIDE
§ Offer evidence-based
recommendations
§ Evolve continually towards more
accuracy based on new information,
outcomes, and actions
- 13. 13 © IBM 2018
Five pillars enabled
through a platform
that has data,
knowledge, analytics
and industry specific
solutions supported
on a secure cloud.
Life Sciences Oncology/
genomics
Imaging Value
based care
Government
Analytics/Insights Platform Image Analytics Cognitive Knowledge Platform
IBM Watson Health Cloud
200M+
lives
100M+
patient
records
Images
15M+
pages of
medical
literature
40+ M
research
documents
- 14. 14 © IBM 2018
1. Too much data, but not enough money
2. Data Science + Health = Watson Health
3. Examples
• Watson for Oncology, Watson for Genomics, Watson for Clinical Trials Matching
• Watson Conversation
• Watson Content Analytics
Discussion Topics
- 15. 15 © IBM 2018
“In 30% of the cases, Watson had found something new”
“These were things that by our own definition, we would’ve considered actionable had we known about it”
— Dr. Ned Sharpless, former Director of the Lineberger Cancer Center, on 60 Minutes
Data & Evidence – Concordance & Additional Treatment Recommendations
A case study with UNC Lineberger Comprehensive Cancer Center compared the human tumor board and
Watson for Genomics’ analysis of tumor sequencing data:
1: Cancer Diagnostics and Molecular Pathology: Enhancing Next-Generation Sequencing-Guided Cancer Care Through Cognitive Computing. The Oncologist first published on November 20,
2017; doi:10.1634/theoncologist.2017-0170. Accessed at: http://theoncologist.alphamedpress.org/content/early/2017/11/20/theoncologist.2017-0170.full.pdf+html?sid=0703cdfd-
db36-45fb-b561-a81544688384
1,022
patients
analyzed
Watson was
>99% accurate
in identifying tumor
board findings
Watson identified
additional options in
335 patients
(33%) of the patients
42 patients
with highly actionable
mutations1
- 16. 16 © IBM 2018
Data & Evidence – Efficiency
16
In a recent comparison study by the New York Genome Center, researchers using a beta
version of Watson to help scale the interpretation of whole genome sequencing found that:
10 minutes
Watson provided a report of
potential clinically actionable
genomic insights.
Whole genome sequencing
identified more clinically
actionable mutations than the
current standard of examining a
limited subset of genes, known
as a targeted panel.1
9,600 minutes
Human analysis and curation
arrived at similar conclusions
for this patient.
1: Comparing sequencing assays and human-machine analyses in actionable genomics for glioblastoma. Neurology Genetics Aug 2017, 3 (4) e164; DOI:
10.1212/NXG.0000000000000164. Accessed at: http://ng.neurology.org/content/3/4/e165
- 17. 17 © IBM 2018
Data & Evidence — Operational
Efficiency
Cognitive technology addressing optimal
cancer clinical trials: matching and
protocol feasibility in a community
cancer practice
During a 16-week trial period, data from 2,620 visits by
lung and breast care patients were processed in the
Clinical Trial Matching (CTM) system.
Watson for Clinical Trial Matching successfully
demonstrated the ability to expedite patient screening
for clinical trial eligibility, reducing processing time from
1 hour and 50 minutes to 24 minutes.
Increased efficiency
(Compared to manual work by a clinical trial coordinator at Highlands Oncology Group)
78%
Reduced pre-screening
wait time by 78%
94%
Omitted 94% of non-matching
patients automatically – reducing
screening workload dramatically
2017 ASCO Annual Meeting. Cognitive technology addressing optimal cancer clinical trial
matching and protocol feasibility in a community cancer practice. DOI:
10.1200/JCO.2017.35.15_suppl.6501 Journal of Clinical Oncology 35, no. 15_suppl (May
2017) 6501-6501. Accessed at:
http://ascopubs.org/doi/abs/10.1200/JCO.2017.35.15_suppl.6501#affiliationsContainer
- 18. 18 © IBM 2018
Natural conversation is key to effective patient engagement
Maija has been
pondering how her new
hobby – ultra running –
impacts her nutritional
needs.
She decides to seek
professional advise.
Via her app, she
accesses the chat
function and initiates a
conversation.
MAIJA’S
JOURNEY
1
2 I live in Ruoholahti Helsinki
We have dieticians available
at these locations. Which
location do you prefer?
Where can I get advise on
nutrition?
In that case I suggest
Porkkalankatu. Make a
booking here.
Hi! How can I help you?
Customer approaches with a
specific intent. Conversational
system identifies intent and
relevant entities.
1
Through intents and entities conversational
system is able to follow conversational
logic, ask follow-up questions, and direct to
correct pages.
2
UNDER
THE HOOD
- 19. 19 © IBM 2018
Creating the intent model starts
with identifying probable
utterances expressed by chatbot
users. Utterances can be
collected from historical customer
service data and new data from
real end users. The solution is
taught to detect correct intents
and entities from the utterance,
and trigger a matching action.
Intents are extracted
from the utterances.
Intent definition and
categorization is a
critical part of intent
model creation. Intent
definitions should be
aligned with the
purpose of the
chatbot.
In addition to intents,
conversational system
recognizes entities,
which are classes of
objects or data types
relevant to a user's
purpose. Entities help
the system to select
the most appropriate
action to be triggered.
Finally, intent model
creation includes defining
correct actions for the
chatbot to retrieve based
on the detected intent and
entities. Functionalities like
response variation
randomization and slot
filling help creating a more
human-like experience.
Response quality
can be scored by
users (e.g.
thumbs up/down
or 1-3 stars).
Collected
feedback can be
used to improve
accuracy through
re-training.
Design and build of the intent model is a fundamental element
of creating natural conversation with Watson Conversatio
Where can I get
advise on nutrition?
#find_location @nutrition
We have dieticians
available at these
locations. Which
location do you prefer?
UTTERANCE → INTENT → ENTITY → ACTION → FEEDBACK
- 20. 20 © IBM 2018
Natural conversation expands and integrates
to cover both pre and post care processes
As Maija is running,
suddenly she feels pain
in her knee. Frustrated,
she is forced to stop.
Maija reaches for her
phone, opens the app,
logs in with touch id,
and opens the chat
function.
As the customer logs in, the
conversational system is aware
of her profile. For instance
whether she is an occupational
or private customer.
1
2
Dr. Mallikas has an appointment
time on Tuesday 3.10. 9:00 am
at Porkkalankatu. Would you like
to reserve it?
The intent and entities in
the message are identified
by the conversational
system.
2
Integration to the patient
information system and
business logic enable
managing of appointments
without human operators.
3
Maija is able to make a
booking by answering the
questions in the
conversation. The night
before her appointment,
she gets a reminder
message.
Despite the unfortunate
circumstances Maija is
happy to see Dr.
Mallikas.
1
I’m sorry to hear that Maija.
Would you like to see the same
orthopedist as previously?
That sounds like a good idea.
My knee hurts. I need to see a
doctor.
Is she available around noon?
She is free at 12:15 pm. Should I
book it?
3
UNDER
THE HOOD
- 21. 21 © IBM 2018
Kela – Specific official support in a demanding domain in Finnish language
FINNISH LANGUAGE
IMPLEMENTATION1 OFFICIAL SUPPORT IN A
CHALLENGING DOMAIN2HIGHLIGHTS
Key considerations:
• Solid Finnish language implementation in a
juridically challenging domain
• Very simple UI utilizing pre-build functionalities
fitted to KELAs official public image.
• Provides “Official and Legally checked” answers
to questions related to students benefits.
• Links easily to additional information and
functionalities like calculators etc.
LINKS CUSTOMERS TO
ADDITIONAL SOURCES OF
INFORMATION
3
- 22. 22 © IBM 2018
Single log in
Healthcare service
providers and other
stake holders
Patient Information
Systems
Kanta
Data lake
Patient-360°
crawl, index and integrate to a chosen viewpoint
- 23. 23 © IBM 2018
Medical report
TYÖTERVEYSHUOLLON KÄYNTI KIRJAUS : ”... 58-vuotias nainen saapui vastaanotolleni
8.2.2017 sen jälkeen kun hän oli kärsinyt useamman päivän huimauksesta,
ruokahaluttomuudesta, anoreksiasta, kuivasta kurkusta, lisääntyneestä janon tunteesta sekä
toistuvasta virtsaamisen tarpeesta. Hänellä oli ollut aiemmin kuumetta ja kertoi
nielemisvaikeuksista ja ruuan jäämisestä helposti kiinni kurkkuun. Hän ilmoitti myös kivusta
vatsassaan, selässsä sekä kyljessä. Ei yskää, hengenahdistusta, ripulia tai dysuriaa.
Hänellä on suvussa ilmennyt suu sekä virtsarakkosyöpää äidin puolelta, gravesin tautia
kahdella sisaruksella, hemochromatosis yhdellä sisaruksella ja idiopaattinen thrombocytopenic
purpura yhdellä sisaruksella.
Hänen potilashistoriassaan on mainintoja huomattavasta cutaneous lupuksesta,
hyperlipidemiasta, osteoporoosista, usein toistuvuvista virstatien tulehduksista, kolmesta
keisarin sektiosta jossa ei ilmennyt komplikaatiota, vasemman ophorectomyn hyvän laatuisesta
kystasta, primääri kilpirauhasen vajaatoiminnasta joka oli diagnisoitu vuosi kystan löytämisen
jälkeen.
Hänellä on määrätty lääkitys levotyroksiiniin, hydroksiklorokiiniin, pravastatiiniin ja
alendronaattiin. Virtsanäyteestä ilmeni positiivisa leukosyyttiesteraasia ja nitriittejä. Potilaalle oli
määrätty resepti fo ciprofloxacinia virtsatieinfektiota varten. Kolme päivää myöhemmin potilas
kertoi heikkoutta ja huimausta. Hänen verenpaineensa oli 120/80 mmHg, ja pulssi oli 88...”
SAIRAALAN KEUHO-OSASTON YÖ VUORO KIRJAUS :
HAPETUS : Tavoitteena normoventilaatio
hengityskoneessa, limaa nousee jnkv. sekä trakeasta että
nielusta HEMODYNAMIIKKA: Tavoitteen RR-syst 140-120
mmHg -> Noradrenaliinaa titrattu. 0 sen mukaisesti CVP-
tavoite 6-10 mmHG. Eks:ssä *null* 0 siisti SR. DIUREESI :
OK
Tajunta : Sedaatiotauolla nopeasti pintaan vetää.
Mielekkäästi kädet kohti intubaatioputkea. Avaa silmät
puheelle. OMAISET : Mies keskustellut tilanteesta EA-
pkl:llä lääkärin kanssa. TAJUNTA : Potilaan tajunta ennen
koilausta tauon aikana hyvä ; liikutteli kaikkia raajojaan
pupillat keskikokoiset ja valolle reagoivat. Kädet menee
mielekkäästi kohi putkea. Pyyntöjä ei noudata / kielimuuri?
Hoitamaton lienee vuotava aneurysma. Nimotop ja
Caprilon jatkuvatKOTIHOITO KONTROLLI KIRJAUS : Vas jalkapöydän kipua ja huimausta. RR 169/92, p. 80,
SpO2 96.5%, lämpö 36,1.Lab+ Tytär käynyt illalla. Kontrolloitu natiivirtg eikä mielestäni kipua
selittävää kuvassa. crp 38, Hb alhainen, 90, mikä ei uutta, mutta etiologia ilmeisesti avoinna
- 24. 24 © IBM 2018
Text Mining
Watson Content Analytics
UIMA
CRAWL & IMPORT PARSE & INDEX
SEARCH &
ANALYTICS
Data is ingested from more than 40 datasource
R represents a C1-12 alkyl group or alkoxyl group, C2-12 alkenyl
group or alkoxylalkyl group ... Z represents a fluorine atom,
chlorine atom, bromine atom, cyano group, --OCF3, --OCF2 H, --
CF3, -- ... alkoxyl group, C2-12 alkenyl group or alkoxylalkyl
group or C3-12 alkenyloxy group, which comprises ... the
following general formula (IV): [Figure] wherein Z represents a
fluorine atom, chlorine atom, ... H, --CF3, --OCH2 CF3, C1-12
alkyl group or alkoxyl group, C2-12 alkenyl group or alkoxylalkyl
group ... group which may be substituted by fluorine atom, trans-
1,4-cyclohexylene group, pyrimidine-2,5-diyl ... -1,3-dioxane-2,5-
diyl group; and m represents an integer of 0 or 1 and a
compound represented by the ... R represents a C1-12 alkyl
group or alkoxyl group
Natural Language
Processing
• Tokenization
• Morphological Analysis
• Part-of-Speech Detection
• Entity Extraction
• Semantic Analysis
• Sentiment Analysis
• Clustering
• etc.
Key information is extracted using NLP
Toimii myös suomenkielellä
Business user obtains insight using intuitive
and unique mining application
What products
are increasing in
recent
problems?
What are the
requests and
claims that stand
out for specific
products?
What is the
correlation
between
products and
defects?
Other data source via API
Export to RDB
Deep Inspection
Alerts
Business User
text index
Linguistics – WCA
Studio
Linguistic rules based
Watson Knowledge
Studio
Machine Learning based
DEVELOP – TEACH MACHINE
Analyzed Content & data
Analyzing Natural Language with UIMA
- 25. 25 © IBM 2018
Finnish Language Support