Smarter Healthcare

Built on Informatics and Cybernetics

Fred A. Maymir-Ducharme

, Lee Angelelli

and Prithvi Rao

IBM US Federal CTO Office; University of Maryland University College (UMUC), Maryland, U.S.A.

Analytics Solution Center Architect, IBM US Federal CTO Office, Maryland, U.S.A.

Federal Healthcare Executive Architect, IBM US Federal CTO Office, Maryland, U.S.A.

Keywords: Healthcare, Knowledge Management, Advanced Analytics, Contextual & Cognitive Computing, Watson.

Abstract: Applying advanced analytics to big data in healthcare offers insights that can improve quality of care. This

paper focuses on the application of health informatics in care coordination, payment, wellness, and

healthcare decision management. Cognitive computing and analytics can be used to capture and extract

information from large volumes of disparate medical data. Applying natural language processing,

probabilistic computing, and dynamic learning can achieve intelligent healthcare systems that users can

interact with to drive business and medical insights across patient populations and result in greater patient

safety, care quality, wellness, and improvements in payer programs. As is the case for most organizations

with large and disparate data sets, the ability to manage information across the enterprise becomes

extremely challenging as the size and complexity of the knowledge management infrastructure grows.

Interconnecting healthcare systems and applying advanced cognitive analytics and health informatics would

provide medical organizations, clinicians, and payers with the information they need to make the best

treatment decision at the point of care. The authors address related research in the following areas: image

analysis for anomaly detection; electronic healthcare advisors for clinical trial matching and oncology

treatments options; advanced models and tools that can accelerate geo-spatial disease outbreak detection and

reporting.

1 INTRODUCTION

The healthcare industry over the past century has

evolved into a series of independent providers and

processes, focused on intervention rather than

prevention, while devaluing primary care and

population health. The problem is that many

“healthcare systems” are not run as “systems” at all.

There is little coordination of data, care, or services.

The industry is so fragmented, in fact, that in 2010

economists ranked healthcare the least efficient

industry in the world, with more than $2.5 trillion

wasted annually (IBM, 2012). Physicians have been

on information overload for decades, contributing to

the estimated 15 percent of diagnoses that are

inaccurate or incomplete (Harvard Business Review,

April, 2010). Population growth, increased life

expectancy, an increase in chronic diseases across an

aging population, health laws, and a lack of trained

clinicians exacerbate the present healthcare situation.

An over demand of patients, under supply of

healthcare workers, and financial incentives that

reward volume over value are increasing the

complexity and inefficiencies of healthcare, which

results in higher costs per patient and degraded

quality. Among healthcare executives interviewed

for the 2010 Global CEO study, 90 percent expected

a high or very high level of complexity over the next

five years, but more than 40 percent said they were

unprepared to deal with it (IBM, 2010).

This paper addresses leading-edge technology

advancements in the field of Health Informatics,

Analytics and Cybernetics. The advancements in

Health Informatics technologies to analyze Big Data,

detect patient medical risks, share reliable health and

operational information, and extract tacit knowledge

out of biomedical and healthcare data in

combination with explicit knowledge to intervene

with specific treatments across the care continuum is

a fundamental game changer for healthcare

organizations, which are challenged to increase their

quality of care, improve clinical outcomes, and at

the same time reduce costs.

Today’s Healthcare Knowledge Management

525

Maymir-Ducharme F., Angelelli L. and Rao P..

Smarter Healthcare - Built on Informatics and Cybernetics.

DOI: 10.5220/0005261705250533

In Proceedings of the International Conference on Health Informatics (HEALTHINF-2015), pages 525-533

ISBN: 978-989-758-068-0

 2015 SCITEPRESS (Science and Technology Publications, Lda.)

Systems (HCKMS) require a multi-disciplinary

approach to meet an organization’s need to innovate

and thrive. Gotz, Stavropoulos, Sun and Wang

describe the use of a predictive risk-assessment

analytics system that analyzes large volume of

heterogeneous medical data to provide various forms

of risk analytics across the patient population. By

applying text and data mining techniques, pattern

recognition, and cognitive computing techniques on

large volumes of clinical and genomic data, smarter

health informatics and analytics can drive

knowledge discovery that have a direct impact on

patients at the point of care (Gotz et al, 2012).

Polanyi described tacit knowledge as implicit

information that is difficult to capture linguistically.

(Polanyi, 1966) This challenge is further exacerbated

when attempting to automate the capture and process

of healthcare tacit knowledge. Nonaka and Takeuchi

created a variety of models as a means of capturing

and communicating (Nonaka and Takeuchi, 1995).

Healthcare organizations are under constant

pressure to cut costs while improving patient care,

clinical outcomes, and reducing hospital re-

admission rates. IBM’s Patient Care and Insights

solution offers the ability to analyze a wealth of

structured and unstructured data from medical

records and by using similarity analytics can auto-

select a cohort of clinically similar patients for

further analysis (Gourlay, 2002). The solution

leverages healthcare annotators, natural language

processing, and analytics to track trends, patterns,

anomalies, and deviations; it produces a

personalized proactive care solution that can reduce

hospital readmission rates, predict the onset of

ailments based on a patient’s social and behavioral

patterns, ensure drug treatment effectiveness, and

help in aligning physician and care teams to patients.

With the explosion of data in terms of volume,

velocity, variety, and veracity, it is imperative and

time critical to have the ability to analyze a high

number of healthcare data points and correlate the

data in real time. Clinicians need timely and accurate

information so that they can make real time

decisions and proactively intervene to save human

lives, shorten hospital stays, reduce hospital

readmission rates, and improve the overall quality of

care. Medical devices that monitor patients’ health

conditions can generate streams of data related to

heart rate, blood oxygen saturation, and respiratory

rates. The ability to fuse real-time streaming data

from various sources and apply analytics can help

address and prevent many life threatening conditions.

Health Informatics that leverages and processes data

streams in real time with an advanced analytic

engine can provide the ability to quickly ingest,

analyze and correlate information from thousands of

real-time resources to predict and save lives in a

timely fashion across intensive care units (Dollard

2013). Big Data and analytics will play a major role

in the future of healthcare and will have a direct

impact on the medical practitioners and their care of

patients. With the onset of aging population,

managing chronic illness and conditions is becoming

a major factor in healthcare. Providers are under

pressure to deliver better and safer care at lower cost

with increased transparencies in the quality of care

and outcomes.

At the same time, payers including health

insurance companies and the US Government are

under increased pressure to monitor the accuracy of

claims and quickly detect a fraudulent claim.

Historically, a pure ‘pay and chase’ model has been

used to combat healthcare fraud. This model is

expensive to execute and thus not sustainable. A

more reliable solution would be to identify

fraudulent claims on entry by parsing the claim text,

looking for trends and anomalies that depict

fraudulent patterns, and flagging these claims early

in the payment workflow cycle for review and

analysis. Here cognitive computing in the form of

Watson Policy Advisor can play a significant role in

ingesting claim policies and guidelines. Going

through domain adaptation and training to detect

fraudulent patterns can result in significant savings

in cost, time, and resources for the payer, provider,

and the patient.

Healthcare providers are expected to be current

on the latest medical research and advancements.

Physicians are expected to be aware of the 360 view

of the patient’s medical history, allergies, drug

treatments, and clinical and lab work when they

walk into the consulting room. This puts tremendous

pressure on the practitioners’ time as they need to

not only discern their patient’s history and analyze

test results, but also maintain a current medical

knowledge base. (David Dugdale et.al., 1999).

Search is a fundamental element of a HCKMS.

With the growth of multimedia over the past decade,

new search and analytics systems have evolved that

provide the ability to search through pre-processed

metadata and automatically analyze files for the

search criteria (e.g., image recognition, speech-to-

text). These advanced search techniques enable the

capture of tacit knowledge in non-linguistic forms.

Gourlay’s knowledge management (KM) framework

stresses the value of non-verbal modes of

information, such as behaviors and processes, to

convey a variety of perspectives (Gourlay, 2002.

HEALTHINF2015-InternationalConferenceonHealthInformatics

526

There is no single Knowledge Management System

(KMS) that integrates all of the evolving KM

capabilities into a single store, and provides a single

search and analytics user interface; hence, the most

effective KMS’ today act as a federation of KM

systems.

Search and analytic capabilities have evolved

considerably over the past few decades. Search was

the initial focus, since one must logically “find” the

data before one can analyze it. The early work

focused on a variety of “relevancy ranking

techniques” (e.g., key word proximity, cardinality of

key words in documents, thesaurus, stemming, etc.)

and federated search, which provides the ability to

search across a variety of DBMS, XML-stores, file

stores, and web pages from a single search interface.

In parallel, but at a slower pace, analytics

capabilities have followed suit in their evolution by

extending traditional search engine reverse-indices

to store and exploit search criteria and other

metadata captured during previous searches or data

crawls.

A variety of analytics have been developed that

extend traditional business intelligence and data

warehousing techniques to provide predictive

analysis, and even stochastic analysis. Predictive

models depend on a large corpus of data and can be

calibrated by modifying the model attributes and the

parameter ranges. Stochastic modeling applies a

variety of techniques to discover new aggregations

of data and other data affinities that can later be used

to augment the organization’s predictive models.

These techniques rely on capabilities such as object-

based computing, contextual computing, and

cognitive computing.

The remainder of this paper elaborates on the

concepts relevant for HCKMS, including Predictive

Analytics, Stochastic Analytics, and technology like

Cognitive Computing. In addition the paper

documents the use of these technologies in

exploitation of tacit knowledge in HCKMS. IBM’s

WATSON (Ferrucci, 2012) is built on these

technologies, and the company’s research continues

to exploit these capabilities for the healthcare

industry.

2 HEALTH CARE

INFORMATICS

There are numerous analytic concepts and

technologies one can apply to HCKMS. Traditional

Data Warehousing and Business Intelligence (BI)

technologies, such as Master Data Management and

crawling or indexing data sources, provide the

ability to integrate multiple data sources into a single

repository and lexicon to better understand the data

an organization has in their KM repositories and

thus better understand and manage the associated

assets, products, and services.

The next level of maturity is embodied by

Predictive Analytics, which can leverage data

models, such as entity and relationship maps, to

predict current or future events, meet objectives, and

identify risks such as fraud and market changes.

Predictive analytics models are typically represented

as aggregations of data in models that generally

represent “observable insights.” When sufficient

data is available and mapped to a model, leading a

level of assurance, an organization is able to make

predictions. Organizations in mature industries with

a large corpus of empirical data can effectively

calibrate their predictive models by changing the

models’ attributes and ranges – and then assessing

the accuracy and fidelity of the model (e.g., “false-

positives” and “false negatives”) against the corpus

of data.

Stochastic modeling goes beyond the

deterministic aspects of predictive modeling by

introducing non-determinism to create or discover

new models. Stochastic approaches generally

include affinity link analysis (e.g., petri-net or

semantic modeling) that extends predictive models

with new data clusters (e.g., new associations that

co-occur, are temporally or geodetically associated,

or are statistically related by an additional data

element or parametric such as data attributes and

ranges).

2.1 Cognitive Analytics

Advanced analytics has been a primary focus of the

authors for the last decade. Having previously

worked on traditional DBMS On-Line Analytical

Processing and a variety of Data Warehousing

projects prior to 2000, the new millennium challenge

has been unstructured data and multimedia.

(Maymir-Ducharme and Angelelli, 2014).

This section describes cognitive analytics, which

the authors define to include contextual analytics.

Contextual analytics is used to assess information

within a confined set of data sources, e.g., a set of

200 documents resulting from a federated search

query and ingest. Contextual analytics applies

techniques such as relevancy ranking, entity

extraction, entity-relationship modeling, parts of

speech tagging, etc., to analyze the data from a

SmarterHealthcare-BuiltonInformaticsandCybernetics

527

federated search. Cognitive analytics extends the

scope of the analyses to include “implicit knowledge”

and perspectives that may be represented by lexicons,

taxonomies, models, or rules-based computations

tailored to the areas of interest represented by the

data in the KMS. These frames of reference are then

used to build understanding and insights about the

explicit knowledge in the KMS. Cognitive

Analytics builds on four key capabilities: collection

of data, contextual computing, cognitive computing,

and exploitation of results (cybernetics).

Figure 1: Cognitive Analytics Architecture.

Figure 1 illustrates the functional architecture of

IBM’s Content Analytics (ICA), a Cognitive

Analytics solution. The architecture above does not

reflect networks, physical or logical views, or

storage, which is traditionally a combination of

DBMS for the structured data and a content

management system for the unstructured text and

multimedia. The Crawler Framework provides the

ability to identify and ingest data from a variety of

sources and in various formats. ICA uses an IBM

search engine that extends the open source Lucene

indexer for efficient storage of the information

content, metadata, and other analytical results.

The first phase of cognitive computing involves

ingesting and indexing a data set. General indexing

may include metadata such as URL, URI, document

name, document type, date stamp, and date indexed.

The original data and documents may or may not

become persistent in the KMS depending on

organizational policy, storage costs, and legal data

rights.

The “Document Processor components provide

the

contextual analytics capabilities. The second phase

of cognitive computing involves applying various

analytics to individual documents, and ultimately

across all documents. This includes the ability to

implicitly or explicitly extract features and

relationships from diverse data and data sources,

creating metadata to continuously build and update

context around the data elements. The figure above

illustrates the “stand-alone” ability to apply a unique

set of analytics on a select document or group of

documents, which is at times is a valuable capability

outside of the broader analytical framework. The

Search and Index API (SIAPI) is used to analyze one

or more documents, resulting in the “annotations”

illustrated as output from the SIAPI.

Figure 2: UIMA Framework.

The UIMA framework illustrated in Figure 2

provides the core contextual analytics through a

variety of annotators. Annotators support an array of

analytical processing capabilities, such as language

identification, entity extraction, entity type

extraction, parts of speech tagging, tokenization,

machine translation, speaker identification, and

tagging. The annotations are used for both

contextual analytics and cognitive analytics.

The UIMA framework can be used for structured

and unstructured data – and for multimedia. IBM has

created numerous translingual annotators that

enhance search, e.g., supporting transliterations,

foreign character sets in UTF-16, multi-lingual

search, and relevancy ranking algorithms such as

stemming and polymorphic analysis.

Figure 3 illustrates some of the results of text

analytics annotators. The keywords are labeled and

color coded to facilitate user’s finding the terms of

interest within a document. The “Relations in this

text” also identified relationships between the

entities extracted. Behind the scenes tokenizers

recognize multiple spellings of the same name

(refining associative analytics), and parts of speech

tagging recognizes pronouns and is therefore able to

include name-relationships associated with an

individual. Applying these analytics resulted

mapping “she” and “her” to Robin Cook.

IBM has extended these text analytics to support

multiple foreign languages, including English,

Russian, Chinese, and Arabic. The company’s

translingual technology includes two types of

machine translation (MT): Rules Based MT, and

Statistical MT. The search engine supports searching

in English, in a foreign language, or in multiple

languages, and it can search foreign language

HEALTHINF2015-InternationalConferenceonHealthInformatics

528

transliterations in the native language vernacular and

foreign character sets.

To apply text analytics to multimedia, one uses

multimedia annotators such as:

 Language Identifier to identify one of 128

languages within the first three phonemes;

 Speech to text translations to convert speech

into text and then apply the appropriate text

analytics;

 Speaker Identifier to identify and distinguish

multiple speakers in an audio or video clip

and tag conversations appropriately;

 Speaker Authentication to authenticate a

speaker if their voice has been enrolled into

the system as an identity’s voice.

An important observation is that this system now

enables users to search multimedia natively – as

opposed to the traditional limitations of static

tagging techniques – and apply the same analytics,

e.g., implicit knowledge and discovery, to any

combination of structured text, unstructured text,

images, audio, and video – concomitantly.

The Analytics Server in Figure 1 illustrates the

ability to extend the text and multimedia analytics

previously discussed, to support lexicons,

taxonomies, and other analytical models. The “text

miner” in the analytics server uses a multi-element

mapping structure to create analytic models. An

element can be an entity, an entity type, a

relationship, or an attribute, e.g., tagged or derived

metadata. Various analytic tools provide an entity-

relationship (E-R) model that one can then visualize

a number of individuals and their direct and

transitive relationships graphically. One can then

create a variety of models (e.g., entity social network

models, hierarchical, organizational, risk/threat, etc.)

to represent an organization’s predictive models or

categorization schemes. Valuable attributes such as

time, geo-location, and demographics can be used to

provide unique analytic models and visualizations.

Experience shows that the quality of the lexicon

and taxonomy created for the KMS directly impacts

the soundness of the resulting models and analyses.

These multi-element mappings are processed by the

text miner and result in an XML tabularized list that

can be used for analysis or as input to a variety of

visualizations.

There are numerous types of visualizations one

can derive from the multi-element mapping results

(models). ICA includes Entity - Relationship Model

Views, Entity-type – Relationship Model Views,

Hybrid Entity-type – Relationship Model Views,

Categorization Model Views, and traditional

Measurement and Metrics Views, e.g., Pie charts.

Another view is Facets, which represent different

views a user can define to visualize either search

results (e.g., views based on a taxonomy) or

analytics results (e.g., multi-element mappings

viewed as models. ICA Facets present the results of

a search query that is mapped to the user’s

categorization scheme. It provides a quick glance at

the taxonomy nodes where the majority of the search

results fall, and other related keywords from the

lexicon that were not explicitly in the search query.

Facets provide the ability to model various clusters

of data elements and help intuitively guide the

prioritization, evaluation, and focus areas of search

results and analytics.

Recognizing the speed with which analytics

technology is emerging and maturing, a KMS

Analytics solution needs to be extensible and

provide interfaces for new inputs – as well as output

to other systems. ICA has the ability to input and

crawl a variety of data sources. The system also

provides the ability to export data and results to

other analytic components, either in an XML

document or directly into a relational DBMS. There

are three different export opportunities in the ICA

architecture:

1. Crawled data can be exported before

indexing or performing any other text

analytics;

2. Indexed data can be exported before applying

multi-element mappings and other analytics;

and

3. Analyzed data can be exported to be used by

different visualization technologies, or

additional analytic engines.

Section 1 discusses stochastic analysis, which is

not supported by this Cognitive Analytics solution.

One approach to stochastic analytics could be to

export the predictive models from ICA into another

modeling engine such as SPSS Modeler, which

includes petri-net and semantic models that could be

used to discover new aggregations, thereby

extending the exported multi-element mappings and

relevant metadata.

2.2 Stream Processing

There are many other data attributes to consider, and

one more in particular is worth noting. A new

paradigm has evolved over the last decade. Stream

computing differs from traditional systems in that it

provides the ability to analyze data in motion (on

the network), rather than data at rest (in storage).

The parallel design of stream processing and the

“pipe and filter” architecture allows a variety of

SmarterHealthcare-BuiltonInformaticsandCybernetics

529

analytics to be applied to data in motion (Maymir-

Ducharme, 2013). Stream processing dynamically

supports analytics on data in motion. In traditional

computing, you access relatively static information

to answer evolving and dynamic analytic questions.

With stream processing, one can deploy an

application that continuously applies analysis to an

ever-changing stream of data before it ever lands on

disk – providing real-time analytics capabilities not

possible before.

Stream computing is meant to augment current data

at rest analytic systems. The best stream

processing systems have been built with a data

centric model that works with traditional

structured data and unstructured data, including

video, image, and digital signal processing.

Stream processing is especially suitable for

applications that exhibit three characteristics:

compute intensity (high ratio of operations to

I/O), data parallelism allowing for parallel

processing, and ability to apply data pipelining

where data is continuously fed from producers to

downstream consumers. As the number of

intelligent devices gathering and generating data

has grown rapidly alongside numerous social

platforms in the last decade, the volume and

velocity of data that organizations can exploit

have mushroomed. By leveraging Stream

Processing practitioners can make decisions in

real-time based on a complete analysis of

information as it arrives from monitors and

equipment (measurements and events) as well as

text, voice transmissions, and video feeds.

2.3 IBM Watson System

Some organizations are using KMS based on

Cognitive Analytics to capture, structure, manage,

and disseminate knowledge throughout their

organization and thus enabling employees to work

faster, reuse best practices, and reduce costly rework

from project to project. Society has now entered an

era where the complexity of our world and the risks

thereof demand a capacity for reasoning and

learning that is far beyond individual human

capability.

Watson takes the Cognitive Analytics technology

described in Section 2.1 to a new level of innovation,

adding advanced natural language processing,

automated reasoning, and machine learning to the

Cognitive Analytics components, e.g., information

retrieval, knowledge representation, and analytics.

Watson used databases, taxonomies, and ontologies

to structure its knowledge, enabling the processing

of 200 million pages of unstructured and structured

text (stored on four terabytes of disk storage.)

Watson used the UIMA framework as well as the

Apache Hadoop framework to support the required

parallelism of the distributed system, which

consisted of ninety IBM Power 750 processors and

sixteen terabytes of RAM.

Figure 3 illustrates the various components of the

Watson architecture. More than 100 different

techniques and technology were used to provide

natural language analytics, source identification,

hypothesis generation and discovery, evidence

discovery and scoring, and hypotheses merging and

ranking.

Figure 3: IBM Watson Architecture.

IBM is leveraging its Watson technology to create

the concept of Cogs computing, which is designed to

follow and interact with people and other cogs and

services inside and across cognitive environments. A

“cog” represents a specific frame of reference and

the associated data. IBM is using Cogs computing

to create “Industry of Knowledge” expert advisors.

3 HEALTH CARE

CYBERNETICS

Watson can be used to analyze structure and

unstructured Healthcare data, such as medical

journals, patient medical records, physician notes,

lab results, and produce confidence weighted and

evidence based responses to queries. In the process,

Watson will become progressively smarter and help

the healthcare industry reduce cost and deliver

quality healthcare.

3.1 Watson Oncology Case Study

Memorial Sloan-Kettering Cancer Center (MSKCC)

is using Watson to help Oncologists battle cancer.

HEALTHINF2015-InternationalConferenceonHealthInformatics

530

Traditionally, Oncologists diagnose cancer using a

patient’s chart, x-rays, laboratory data, a few

medical books; and they might then recommend

either the general radiation therapy or three types of

chemotherapy. Today, Oncologists face a

perpetually growing sea of data in their efforts to

effectively deal with every aspect of their patients’

care. The associated medical information doubles

every five years. An Oncologist must maintain

awareness and understanding of the ever expanding

medical books and articles; electronic patient and

family medical records; more than 800 different

cancer therapies sequencing 340 cancer tumors

(each with multiple mutations), 20,000 genes,

correspondence with more than 1,000 physicians.

Traditional processes for cancer prognosis and the

recommendation of therapies are no longer able to

effectively harness all of the available data. Hence

physicians are turning to Watson to develop

precision based medicine in cancer.

MSKCC and IBM are training Watson to

compare a patient’s medical information against a

vast array of treatment guidelines, published

research, and other insights to provide individualized,

condensed, scored recommendations to physicians.

Watson’s Natural Language Processing capabilities

enable the system to leverage this sea of

unstructured data, including journal articles, multiple

physicians’ notes, as well as the guidelines and best

practices from the National Comprehensive Cancer

Network.

The evolving IBM Watson Oncology Diagnosis

and Treatment Advisor includes supporting evidence,

and with every suggestion, it provides transparency

and assists physicians with decision-making and

patient discussions. Watson interactively points out

when more information is needed and updates its

suggestions as new data is added (IBM, February

2013).

For example, MSKCC had a cancer case

involving a 37 year-old Japanese non-smoking

patient who was diagnosed with lung

Adenocarcinoma cancer. The physician asked

Watson for a recommend therapy. Watson’s initial

case analysis recommended Chemo-Erlotinib

treatment at a 28% confidence interval. Watson

needed more information and recommended the

physician perform a molecular pathology test to

detect if there were any EGFR mutations because

Watson’s knowledge base indicated that 57% of all

EGFR mutation in women with Adenocarcinoma

cancer are missed. The Lab results identified the

presence of an EGFR exon 20 mutation. Watson

referenced a medical paper citing an exception,

where the EGFR exon 20 mutation did not respond

to Erlotinib treatment. Analyzing the new lab

information along with the medical article, Watson

then recommended Cisplatin / Pemetrexed treatment

with 90% confidence. “There are only about two or

three physicians in the world who would know this

information” said Dr. Jose Baselga (Physician-in-

Chief at MSKCC) at the IBM Watson Group Launch

in New York Event (9 January, 2014).

3.2 Watson Medical Insurance Case

Study

WellPoint Inc., is the largest for profit managed

health care company in the Blue Cross and Blue

Shield Association. WellPoint identified that a

significant amount of time, and monetary resources

were not being used optimally to deliver quality care

to their constituents. For example, WellPoint needed

to examine and pre-approve a high number of

medical claim forms before a patient could be

treated. The pre-approvals required WellPoint claim

processors to process a high volume of claims

containing both structured and unstructured data,

assimilate and coordinate across many external

systems to track and validate claims across clinical

and patient data, which was a labor intensive and

time consuming process. Long wait times frustrated

both the provider and patients, especially when time

sensitive information and approvals were requested.

(IBM,

June 2014). WellPoint explored rigid code

based procedures and claim processing guidelines

and discovered that they too had limitations and

inefficiencies.

WellPoint turned to Watson to leverage its power

to easily handle natural language processing to

streamline healthcare insurance pre-approvals for

patients and providers. WellPoint trained Watson

with 25,000 historical cases to help their claim

processors with tools to streamline their pre-

approval claims processing and assist them with an

automated decision making process. Watson on its

part ingests the plethora of treatment guidelines,

policies, medical notes, journals and documents and

using its natural language processing capability

processes the unstructured data such as text based

treatment requests, leverage its hypothesis and

evaluation processes to provide confidence-scored

and references backed evidences to help WellPoint

make better and more efficient decisions about the

treatment authorization requests. This iterative

process enables Watson to improve iteratively as

payers and providers use it in their claims processing

and help in the proper decision making outcome to

SmarterHealthcare-BuiltonInformaticsandCybernetics

531

either approve or deny a pre-authorization claim for

medical procedures. The new system provides

responses to all requests in seconds, as opposed to

72 hours for urgent pre-authorization and three to

five days for elective procedures with the previous

UM process.

Elizabeth Bingham, Vice President of

Health IT Strategy at WellPoint Inc., stated, “We

know Watson makes us more efficient and is helping

us turn around requests faster: It also ensures we are

consistent in our application of medical policies and

guidelines.” (IBM,

June 2014)

3.3 Watson Neonatal Care Case Study

IBM is applying Streaming computing and advanced

analytics such as machine learning to develop state-

of-the- art multimodal medical informatics that

combine analyses of multiple information sources,

e.g., text, image, molecular to help intensive care

units (ICU) improve prediction, prevention and

treatment of diseases. IBM Stream computing and

advance analytics are making great advancements in

the development of an instrumented, interconnected,

and intelligent ICU platform that integrates human

knowledge with medical device data and enables

real-time automated holistic analysis of patient’s

vital signs to detect early subtle patient’s warning

signs before symptoms become critical and life

threatening. This is another example of leveraging

advanced analytics and cybernetics to achieve

smarter healthcare for humans.

ICU clinicians review outputs from multiple

medical devices to continuously monitor patient

vital signs such as blood pressure, heart rate, and

temperature. The medical devices are designed to

issue a visual and/or audio alert when any vital sign

goes out of the normal range, which notifies

clinicians to take immediate action. The medical

devices produce massive amounts of data: Up to

1,000 readings per second are summarized into one

reading every 30 to 60 minutes. Some data is stored

for up to 72 hours before being discarded (

Gotz et al,

2012)

. Even the most skilled nurses and physicians

have difficulty reading all the medical devices

monitoring a patient’s vital signs to detect subtle

warning signs that something is wrong before the

situation becomes serious. “The challenge we face

is that there’s too much data,” says Dr. Andrew

James, staff neonatologist at The Hospital for Sick

Children (Sick Kids) in Toronto. “In the hectic

environment of the neonatal intensive care unit, the

ability to absorb and reflect upon everything presen-

ted is beyond human capacity, so the significance of

trends is often lost.” (Gotz et al, 2012)

It is difficult to detect early signs of hospital-

acquired infections. IBM T.J. Watson Research

Center’s Industry Solutions Lab (ISL) is working

with hospitals and medical research labs including

the University of Ontario Institute in Project Artemis

to develop a first-of-its-kind, stream-computing

platform that ingests a variety of unstructured

monitoring-device data (instrumented); integrates

clinical/medical knowledge from multiple sources

(interconnected) for automated analysis using

advance machine learning algorithms to detect

subtle patient vital signs (intelligent); and alert

hospital staff to potential health problems before

patients manifest clinical signs of infection or other

issues. The Project Artemis brings human

knowledge and expertise together with medical

device-generated data captured in machine learning

algorithms to produce a holistic early detection

patient analysis. The Project Artemis stream

computing platform capabilities, which enable

nurses and physicians to detect medical significant

events even before a patient exhibit symptoms, will

enable proactive treatment that can increase the

success rate and potentially saving lives. Physicians

at the University of Virginia can detect nosocomial

infections 12 to 24 hours in advance before

“preemies” exhibit symptoms. (IBM, December

2010)

4 CONCLUSIONS

This paper provides an overview of advanced

analytics technology such as cognitive computing

that can be applied in support of Healthcare

Informatics. The interactive Watson system

demonstrates the use of cognitive analytics and

illustrates various Healthcare Cybernetic automa-

tions as medical and insurance decision support

systems

Although the current Watson emphasis is on

"validated" data & sources from mature industries,

there is further research focused on broadening the

types of data and applying cognitive computing in

other mature industries as well as in less stable

industries such as entertainment.

ACKNOWLEDGEMENTS

The authors wish to thank Joe Angelelli, Ph.D.,

Director, Health Services Administration at Robert

Morris University and Susan Carlson, MBA, PMP

for providing technical consulting.

HEALTHINF2015-InternationalConferenceonHealthInformatics

532

REFERENCES

“Redefining Value and Success in Healthcare” IBM

Healthcare and Life Sciences. February 2012.

http://www.ibm.com/

Capitalizing on Complexity: Insights from the 2010 IBM

Global CEO Study” IBM Institute for Business Value.

May 2010. http://www.ibm.com/ceostudy.

David Gotz et al “ICDA: A Platform for Intelligent Care

Delivery Analytics”, American Medical Informatics

Association Annual Symposium (AMIA), Nov 3, 2012.

Wes Rishel et. al ‘IBM’s Patient Care and Insights: A

Promising Approach to Healthcare Big Data’.

Gartner,Research, 27 November 2012.

David C. Dugdale, Ronald Epstein, Steven Z. Pantilat

‘Time and the Patient-Physician Relationship’, Journal

General Internal Medicine, Volume 14, January 1999.

Polanyi, M. (1966) The Tacit Dimension, London:

Routledge & Kegan Paul.

Nonaka, I. and Takeuchi, H. (1995) The Knowledge-

Creating Company, Oxford: Oxforbd University Press.

Gourlay, Stephen (2002) “Tacit Knowledge, Tacit

Knowing or Behaving,” 3rd European Organizational

Knowledge, Learning and Capabilities Conference; 5-

6 April 2002, Athens, Greece.

Gal, Y. et al. “A Model of Tacit Knowledge and Action,”

IEEE Computational Science and Engineering, 2009

International Conference p. 463-468.

Maymir-Ducharme, Fred and Angelelli, Lee “Cognitive

Analytics: A Step Towards Tacit Knowledge?”

proceedings of the 8

International Multi-Conference

on Society, Cybernetics and Informatics, July 2014.

Maymir-Ducharme, F. and Ernst, R. “Optimizing

Distributed and Parallel TCPED Systems,” US

Geospatial Intelligence Foundation (USGIF) Technical

Workshop, Denver CO, July 17-19, 2013.

Ferrucci, D. “Introduction to ‘This is Watson’,” IBM

Journal of Research and Development, Vol 56

May/July 2012.

Memorial Sloan-Kettering Cancer Center IBM Watson

Helps Fight Cancer With Evidence-Based Diagnosis

and Treatment Suggestions, Feb. 8, 20132/8/1313,

http://www.ibm.com/services/multimedia/MSK_Case

_Study_IMC14794.pdf.

IBM Watson – WellPoint Inc, IBM Watson Platform. June

2014 http://www.ibm.com/smarterplanet/us/en/ibm

watson/assets/pdfs/WellPoint_Case_Study_IMC14792

.pdf.

Vince Dollard, Emory University Hospital Explores

‘Intensive Care Unit of the Future’, Emory University

Health Sciences, Woodruff Health Science Center,

Nov 11, 2013.

Shirley S. Savage, ‘How Watson and Cognitive

Computing are changing healthcare and customer

service’, IBM Systems Magazine, October, 2013.

IBM Corporation - University of Ontario Institute of

Technology. December 2010 http://www.ibmbigdata

hub.com/sites/default/files/document/ODC03157USE

N.PDF.

SmarterHealthcare-BuiltonInformaticsandCybernetics

533