Unpacking AI for Healthcare

Post on 13-Apr-2017

96 views 1 download

transcript

#healthpredicted

Unpacking AI for Healthcare@ashdamle

Image from http://bryanchristiedesign.com/

We have very little control over health and care. From doctors to insurers to patients – we are all struggling with making sense of health.

our health is complex37+ Trillion Cells

Image from http://bryanchristiedesign.com/

We have have no control, and very little visibility into how health evolves

As a result, care management and coordination is broken & imprecise, leading to:

higher and higher costs of care with little improvement in health outcomes.

We have an opportunity.

High quality data and analytics can drive precision into healthcare, reducing costs of medical care while improving health outcomes.

The challenge: Healthcare has one of the most complex data sets in existence.

High volume. High dimensionality . Heterogeneous. Varied formats. Multi-faceted relationships. Noisy.

And yet, we are still using 19th century solutions for a 21st

century problem!

Why not healthcare?

voice recognition, image recognition, natural language processing, deep learning & machine learning

AI has helped many other industries achieve unprecedented levels of efficiency in overcoming data complexity

$6B $2B

The AI market in healthcare will hit $6 billion by 2020 (Frost and Sullivan)

$2 billion can be saved annually with a tech-enabled processes (Accenture)

AI is best positioned to solve the health data challenge

AI surfaces the signal from the noise in health dataallowing us to understand what to do, for whom, when, and why

+

giving everyone more control and precision over health and care

Automated information processing

45% of routine, manual tasks that can cost up to $90 million can be automated by adapting current AI technologies (McKinsey).

1

Precise disease management

Machine learning could increase patient outcomes at by 50% at about half the cost (Indiana University).

2

Efficient provider-patient

encounters

Virtual health apps can save physicians 5 mins per patient encounter (Accenture)

3

Social robots for patient

engagement

Robots like PARO have been found to reduce patient stress and interaction with caregivers(World Economic Forum)

4

What if we could use AI to predict future health with precision, timeliness and speed?Could we significantly reduce costs of care while creating more improving outcomes: less complex, real-time feedback loops, more personalized?

How do we get there?

We need real-time machine-based systems that leverage data to predict health with precision, timeliness and confidence, so we can deliver high-value personalized care at scale.

It requires…

1.Deep domain expertise in medicine to build robust, clinically-relevant models

Data science expertise to handle complexity of health data and apply advanced machine learning techniques

Access to large data sets for supervised and unsupervised training of models

Infrastructure that can prepare terabytes of data for analysis with speed

Industry collaboration to build solutions that can be seamlessly applied into clinical workflows

Introducing Lumiata:an example of Medical AI

that handles the complexity of health data

We want to radically transform the way health data is put to work.1. Power data-driven precision in predicting health to

reduce costs and improve health outcomes2. Bring clarity, control and confidence to all health actors

Lumiata leverages Medical AI to precisely predict and manage risk at the individual level. We drive the personalization and automation needed to make health predictable.

Data Scientists

Utilize the latest in AI & deep learning to evolve Lumiata’s

Medical Graph

Design & deploy new models for targeted use cases

Clinical Scientists

Adjudicate ongoing clinical inputs into Lumiata’s Medical

Graph

Ensure clinical relevance of predictive analytics & rationale

DS CS

To build Lumiata, we combine deep domain expertise

330M+ data points describing the relationships between…

• Hundreds of protocols & guidelines• 40K+ Symptoms & Signs• 4K Diagnoses• 3K Labs, Imaging, Tests• 3K Therapeutic Procedures• 7K Medications

across age, gender, durations, lifestyle

Our AI is powered by a learning probabilistic Medical Graph & Deep Learning

3TB+unstructured  

data

175M+patient   record  

years

39K+physician  curation  hours

that predicts individual health risks, and helps embed personalization and automation in risk

management operations.

Input(Data)

Analyses(FHIR+AI)

Output(Insights)

Delivery(API)

ImpactAction

Risk Matrix + Clinical RationaleRISK MATRIX& CLINICAL RATIONALE

MEDICAL GRAPH

It augments our ability to identify and capture value in data

by bringing clinical precision, giving everyone

the confidence to act with precise health

predictions

by automating labor-intensive risk

management operations to reduce costs

(data gathering + data synthesis + analysis + planning + messaging +

decision + fulfill)

&

symptoms diagnoses labs Images

therapyprocedures

meds

environ. factors,

seasonality

lifestyle + demo. profile

geography

past medical history

genetics

family history

vitalscomplaints

∫(age, gender, duration, ethnicity, …)

∫(age, gender, sensitivity, specificity, …)

Generating per patient models of health, making healthcare delivery predictable and personalized.

Our Medical Graph maps multi-dimensional relationships to handle the complexities of health data

and by mapping out the relationships of health data, the Medical Graph address many of the data complexities

in systematic, scalable way

Demographics

Lumiata Medical Graph

Procedures

Physical Exam & Tests

Medical & Social Hx

Sensors & Wearables

Genomics

High volumeHigh dimensionality

HeterogeneousVaried formats

Multi-faceted relationshipsNoisy

Multiple Coding SystemsGraphs not Trees/DAGs

PUBMED  ReferencesPUBMED  References

Lumiata   Risk  MatrixCondition 1 2 3 4 5 6 7 8 …

0-­‐1  Year Y N N Y Y N N N …

1-­‐2  Years Y N N Y Y Y N N …

2+  Years Y N N Y Y Y N Y …

Clinical  Rationale

Clinical  Rationale

Past  Med  History

Diagnoses

Abnormal  Labs

Procedures

Medications

where each prediction is supported with medical evidence, bringing confidence, control and clarity to health operations

36,000+Physician

Curation Hours

Clinical Integration Engine Clinical Analytics Engine API & Web Platform

Real-Time Data Clinical

FinancialSocial

Environmental

DescriptiveIntrospective

PredictivePrescriptive

Discovery

Operationalize Data

Data Unification

Insight & Action Generation

Data & Action Distribution

and transforms data to insight to action

Fast-tracking healthcare toward value-based care

Automated risk stratification to drive population health management

Precise & personalized care management interventions

Clinical alignment and agreement between payers and providers

Reduced costs by removing labor-intensive, redundant tasks

+

True Clinical State & Risk EvolutionDifferential Diagnosis and Triage

Missing DiagnosisData Driven GuidelinesClinically Right Coding (ICD, HCC)

Risk AdjustmentQuality MaximizationPredict High Cost Claimants Utilization PredictionCare Coordination

with clear practical use cases available via an API or web app

Through AI, we are giving everyone the confidence to act on data in a way that improves care, automates processes and reduces costs.Health plans become more cost-effective and collaborative.Caregivers deliver more precise and timely care. Patients get personalized treatment plans.

Image from http://bryanchristiedesign.com/

powering clear, predictable health outcomes

#healthpredicted

Unpacking AI for Healthcare@ashdamle