Transforming

healthcare with
data and AI
 Varsha Buga PC03
Mansi Bafna PH15
Mrunali Pawar IP05
Luke D’Silva PC05
Gaurangi Sinkar PT11
Bhavya Bhasin PC01
Tapasya Yallatikar QA12
Himanshu QA13
Namita Raul PH12
 DATA SCIENCE AND AI
ARE TOOLS OF THE
21ST CENTURY

3
What do ‘data science’
and ‘AI’ mean?
AI can be characterized by four broad
technologies
Natural language processing (NLP)

Computer vision

Machine learning

Robotics
5
Why is now the time to
invest in data science
in health?
Three Shifts

90 percent It is now Complex

Shift 1

Shift 3
Shift 2
of the world’s possible to machine
data was store and learning can
created in process huge now interpret
the last two data sets the data that
years alone more cheaply is created

7
Consumer enthusiasm

35–40% of people in 30–60 percent of people

India and Qatar while 55–65 percent of
For people with an report that they would
45% in US and China people stating that they
electronic health record want their health data to
with internet access would be happy for
(EHR), more than 50 be shared to improve care
have access to their their clinicians to use AI
percent are active users delivery, to conduct
health records decision-support
research and to inform software in their care
health planning

8
Data availability, storage and analytics
approaches, 1990–2018

9
Advancing tools and technologies

10
Data science can help
transform healthcare
Health System Challenges

Predicting and Determining

Delivering
preventing safe and
cost effective
onset of effective
care
disease treatment

12
Use of Data Science

Contribute to Improve the

Better
efforts to quality and
engagement
optimize outcomes of
of patients
resources care

13
AI adoption in healthcare is low
 Car Automation

▪Car brake or swerve to avoid collision

and park by themselves
▪Use of compter vision and machine
learning algorithms
▪Same approach for complex
healthcare processes

15
 Clinical Research
Organization

▪Identify the optimal mix of sites in

clinical trials
▪Apply predictive risk algorithm and
machine learning to select sites to
recruit eligible participants
▪Research can be done using limited
public funds.

16
Building Blocks for Data Science and AI
Integration

Creating large, Establishing Attracting the

integrated and governance scientific talent
interoperable data structure and needed to design
sets necessary for security measures complex systems.
the development to handle sensitive
of new tools. data

17
What benefits could data science bring to
health systems?

Better
Deliver more Further
forecast
preventative personalize
population
care treatments
trends

18
What benefits could data science bring to
health systems?

Improve
Improve
productivity
user
and reduce
experience
costs

19
 A data-enabled healthcare
ecosystem

20
21
 Precision medicine in practice

22
23
Moving To A Fully Data
Enabled Learning
Health System
Data enabled learning Health System

No sector is fully data enabled

Difficult
Early question? Are people
adopters can Is personal ready to
inspire others data adopt?
confidential ?

25
Transformation to this scale

Balanced
short term
Sustained goals with
Clear Vision
Leadership long term
investment

26
Where most health systems are today?

At the start of the journey

High income Secondary Clinical

countries use care challenge?
digitalized Paper records Periodic
EHRs in + Digital experts
primary care inputs review

27
Integrated & Interoperable EHRs

Patients Estonia Patients have

health invested in ownership of
records and Personal their records
history Health &make
accessed by Records decision
physicians (PHRs) accordingly

28
Routine use of algorithm for Decision support

Innovation today is focused on algorithm to make decision

Complex Simple Transform

algorithm models built efficiency of
More accurate on large data hospital and
diagnosis than for clinical primary care
physician practice facility

29
 Personalized treatment
and Precision medicine

• Most treatment decisions follow

standardized guidelines based on
clinical trials.
• Participants in clinical trials tend to
differ from treatment populations in the
real world, which can limit the predictive
power of the published evidence.

30
Factors affecting Health outcomes
Socialcircumstances:
15–40%

Environmental and physical influences:

Exogenousfactors:
5–20% 60%

Behavior:
30–50%

Genetics: Genomicfactors:
20–30% 30%

Medicalcare: Clinicalfactors:
10–20% 10%
31
 Analyses real-world data from Uses genomic sequencing in 40 percent of
EHRs to augment evidence for leukemia and lung cancer patients to select
new therapies specific, targeted therapies.

32
 Integration with broader
system and
environmental data
• There is a great potential in the
integration of health data sets with
those that include education,
transportation, pollution and other
societal data.
• Together, these data sets will enable
more precise forecasting and more
intelligent use of scarce resources

33
 Data sources Policy and Planning

Government
Assessing
(eg: financial data)
disease burden

Data interpretation
Administrative

Data integration

Data processing
Health
system Prioritization

Clinical

Public health
surveillance Intersectoral policy
formulation
Patient-generated
data

Clinical research Resource allocation

34
 • Mobile application that allows the
consultation of the average waiting
time in the hospitals
• The user can consult, per institution,
the average time of attendance in
the emergency room.
• The average waiting time is
presented by priority: red, orange,
yellow, green and blue.

35
 Fully data-enabled
learning health systems

• Technologies – including machine learning,

wearable's, robotics, networked smart
sensors, bioengineering and molecular
biology will transform the way care is
planned, delivered and experienced.

36
 ROBOTIC ASSISTED SURGERY

ROBOT PORTERS

ROBOT DISPENSERS

37
 ARTIFICIAL PANCREAS

This device then can sense a rise in sugar levels triggering

38 automatic delivery of insulin to regulate the patient's system
Five building blocks of
transformation
 Organization
Data governance
wide data
and security
repositories

Interoperability Data science

of data capabilities

Use and
repeated reuse
of data

40
1. Organization wide data repositories
Data sets of skin
melanoma images can be
Gather and store data used to develop a tool that
digitally can diagnose malignant
tumors more accurately
than human specialists
Estonia in 2008
implemented an eRecord
system that digitally
records and stores every
interaction a patient has
with the health system.
2. Data governance and security
Clearly, data repositories that hold sensitive and potentially identifiable
patient information are a security concern

Patients understandably worry about the inappropriate sharing and use

of their data

Healthcare is particularly vulnerable to cyberattack due to its limited

resources, fragmented governance structures and cultural
behaviors. Compared with other critical sectors, healthcare has
chronically underinvested in IT
2. Data governance and security

▪ 1. Who can access which data and under what circumstances?

▪ 2. How do we prevent and detect unauthorized access to data?
▪ 3. How do we ensure the accuracy and consistency of data?
▪ 4. How do we ensure traceability and accountability for each
interaction made with the data?
▪ 5. Where does liability rest for breaches of data security or
erroneous recommendations arising from automated algorithms?
44
 Principles of
data system
Only collect, store and give access to essential data

Use unique, consistent pseudonyms

Consider using local or cloud-based repositories

Use encryption and enterprise-grade security system

45
3. Interoperability Of Data Within And
Across Health Systems
▪ AI, and in particular machine-learning techniques, is most powerful when it uses
different data types from a range of sources.
▪ Flu-forecasting tools can combine geolocation data, past flu trends and Twitter
feeds to forecast influenza outbreaks with up to six weeks’ lead time.
▪ Further studies have shown that disease outbreaks can be observed through
Google search terms
▪ For these purposes, data can be integrated at two scales: within an individual
tool or device, and at a larger system scale
Single tool or device:
▪ The integration of multiple different data sources, even if collected locally, can give
powerful results.
Larger system scale
▪ Data can be integrated across an entire national health system and then combined
with other public records.
Qatar’s Estonia’s
focus on data-enabled
e-health System
healthcare
Since the
Qatar haslate
implemented
1990s, Estonia
one Electronic
has been moving
Health Record
its 1.3 million
(HER) system
citizens across
to a digital
the
country, making
government model. the
Underentire
the leadership
population’s
of a national
health Chief
records
Information
interoperable.
Officer,
Furthermore,
citizens can now government
vote, manage
funding
theiris tax,
beingmanage
used their
to improve
education andandprioritize
public
patients’ and
services, interactions
control their
with health
the carerecords
system.online.
For example,
The benefits
the to
Qatar
the Computing
citizen are
Research
clear: transparency,
Institute (QCRI)
speedareand
developing
ease ofAIuse.toolsFor
thatthe
canstate,
translate
the into
benefits
different
are
languages for
enormous: Estonia
international
estimatespatients,
that itwith
saves
around
2 percent
90 percent
of its accuracy.
GDP eachThey yearare
in
also developing
salaries and expenses
mobilethrough
apps that
digitization
can estimate body mass index (BMI) from facial
images, and monitor children’s lifestyle, food habits and sleeping patterns using
47 wearable sensors.
▪ Data sets are stored locally, they can be accessed remotely and linked using
unique identifiers for each citizen. Integration like this requires significant mutual
co-operation and policy oversight. While this may be challenging in healthcare, it
is not unachievable.
▪ Policymakers have an important role to play in promoting common standards
and open source tools, to support the development of publicly accessible
resources.
▪ Successful examples from healthcare include:
1. Digital Imaging and Communications in Medicine (DICOM) standards used in
medical imaging,
2. open-source clinical management systems, such as Open Source Clinical
Application Resource (OSCAR)54 used in parts of Canada.
4. Data Science Capabilities
▪ Most health The UK’s Beth
Commitment
systems Israel
do notDeaconess
To Developing
have Healthcare
advanced Health System
IT Capabilities
capabilities in data science.
The
▪ In UK
Bethwas
the Israel
theterm,
short Deaconess
first country Healthcare
skill gapsin can
the world System
tothrough
be filled develop
includes
a national
2,600 health
partnerships physicians
IT strategy.
and talent across
fouracquisition.
However,
hospitals.
this strategy
It is one largely
of the failed
leadingtocentres
realize its
of data
potential
science
to improve
innovationcareinand
US
healthcare.
health
▪ In the outcomes.
Its success
medium- Part of this
hasfailure
to longer-term, beenhasdriven,
beenwill
countries attributed
in
needpart,
toto by
focusing
train amandating
broad toopool
muchoftight
on
interoperability
implementing
health dataEHRs of EHRs,
without
scientists. and investing
by moving in its
howdata
thehosting
digital data
to the
would
cloud.orHowever,
could be
crucially,
used to drive
the hospital
transformational
network change.
has alsoThe included
UK Government
skilled datahas engineers
learned infromits
1. Attracting
organization
this and retaining
and otherstructure.
international talented
Amazon staff
experiences,
and Google
and has employees
reprioritizedwere
digital
brought
maturity
in and
2. Partnership
given
clinicalaccess
and data
to 7 petabytes
science capabilities
of data. Theinstrength
the NHS. of this
It hascollaboration
created four hastraining
greatly
increased
organizations:
3. Training innovation.
the NHS Within
Digital
twoAcademy
years, operational
to train chieftools
clinical
wereinformation
developed to: officers
and
1.
chief information
Automate repetitive officers
tasks in data science skills and leadership, the Farr
Institute, Health Data Research UK (HDR UK), and the Alan Turing Institute. These
2. Improve efficiency in operating theatres by 30 percent
organizations are simultaneously developing academic health data science
Reduce on
capabilities
3. human errorscale.
a national in the detection of breast cancer from lymph node
biopsies by 85 percent using a deep-learning image analysis tool.
5. Use and reuse of data to improve decision-
making and care
▪ The fifth component that health systems must get right is the regulatory
environment.
▪ To ensure that systems can continue to learn and improve based on the data
collected, they must be able to use and repeatedly reuse the data for multiple
purposes.
▪ Regulatory issues in data science are related to two areas: consent and device
approval mechanisms
Consent
▪ Traditionally, patients have been asked to give informed approval for their
data to be used for a specific reason, for example to develop a new cancer-
screening tool.
▪ However, this model prevents the patient’s data from being reused for a
different research purpose, limiting the potential of large, integrated data sets.
▪ A new approach, termed broad consent, asks patients to agree to their data
being held for a much longer period, and being used for a wider range of
applications
 Image analysis tools for clinical decision support
AI Device
is now approval mechanisms
able to process images with a high degree of accuracy, even outperforming humans
in •some cases. This
Introducing capability
new softwarecan
forbe usedclinical
aiding to diagnose patient
decisions images from anywhere in the
and operational
world.practices
There aremay several examples:
require regulatory approval. This can involve clinical trials to
• prove
Israeli that these
start-up Zebratools are Vision
Medical at leastoffers
as effective as existing
image analysis toolstechniques.
to help diagnose osteoporosis,
emphysema and brain haemorrhages. Since algorithms are cheap to run, this clinical support can cost
as little as $1 per scan.
• Ping An Healthcare and Technology developed world-leading AI-based lung nodule detection systems
for CT scans, which are being rolled out across hospitals in China.
• A Stanford University platform uses mobile phone images to detect skin cancers as accurately as a
dermatologist. This provides a route for quick, accessible screening for melanoma worldwide.
• QuantX is a computer-aided breast cancer diagnosis platform for magnetic resonance imaging (MRI)
scans. QuantX received fast-tracked regulatory approval because the US FDA prioritizes treatments
that can offer substantial increases in care quality.
• ET Medical Brain, Alibaba’s cloud-based solution, combines data hosting and image diagnostics. It is a
leading AI-enabled health solution in China, working in partnership with hospitals to access data and
develop tools. One tool detects thyroid cancer from ultrasound images with an 85 percent success
rate, better than the human rate of 60–70 percent.
Policymakers can act
now to start the journey
 ▪ 1. Provide national leadership for data science
and AI in healthcare
▪ 2. Identify and gain ‘quick win’ opportunities
(first 12 months)
▪ 3. Set strategic priorities for the medium term
(one to three years)
▪ 4. Pursue a longer-term transformation plan
(three to 10 years).

54
1. Provide national leadership for data science
and AI in healthcare
▪ Communication: Tell the public about the potential risks of the program,
▪ such as security breaches, and how these are being mitigated, as well
▪ as the benefits, including greater convenience, more transparency and
better patient care
▪ Engagement: Consult and partner with critical stakeholders, especially
▪ within the private sector, which is likely to be a major source of
investment
▪ and innovation
Strategic direction-setting: Develop and execute a strategy for data
▪ science and AI
An advisory board: Set up a national policy group for data and
55 informatics to provide guidance and advice.
2. Identify and gain ‘quick win’ opportunities
(first 12 months)
 Launching apps that give patients on-demand video
consultations with primary care providers, or launching
chatbots to guide patients to appropriate primary care
resources
 Adding AI-supported diagnostics to radiology, especially
where these reduce waiting times for the scan and the
results
 Using triage algorithms in emergency departments to
improve flow, reduce waiting times and keep patients
informed.
56
3. Set strategic priorities for the medium
term (one to three years)

1. Digitization and integration of data.

2. Data governance and stewardship.

3. Capabilities and skills.

57
4. Pursue a longer-term transformation
plan (three to 10 years)

1. Continually improving data integration.

2. Refining regulatory frameworks.

3. Building capabilities.

58
 That’s it from our side. Thank you for
the patient listening!

JAI JIO!

59