Putting IBM Watson to Work In Healthcare

Basit Chaudhry, M.D., Ph.D. Medical Scientist IBM Research

Putting IBM Watson to Work In Healthcare

© 2012 International Business Machines Corporation

The Problem

Facts per Decision

1000

Proteomics and other effector molecules 100 Functional Genetics: Gene expression profiles 10 Structural Genetics: e.g. SNPs, haplotypes

Human Cognitive Capacity

5 Decisions by Clinical Phenotype 1990

2000

2010

2020

William Stead, IOM Meeting, 8 October 2007. Growth in facts affecting provider decisions versus human cognitive capacity.

Agenda

What is IBM Watson and why is it important?

How is IBM putting Watson to work?

What can we expect in the future?

3


Businesses are“dying of thirst in an ocean of data”

4

90%

80%

1 Trillion

of the world’s data was created in the last two years

of the world’s data today is unstructured

connected devices generate 2.5 quintillion bytes data / day

1 in 2

83%

2.2X

business leaders don’t have access to data they need

of CIOs cited BI and analytics as part of their visionary plan

more likely that top performers use business analytics © 2012 International Business Machines Corporation

Why Watson for healthcare? Diagnosis and treatment errors Shortage of MDs Demand for remote medicine

Complexity

Universal coverage

Evidence-based Medicine

Costs

Focus on Wellness and Prevention

Policy Changes

Shift from Fee-forService to ACOs

Personalized Medicine

Costs are 18% of US GDP 34% of $2.3T US spend is waste Costs can vary up to 10x

Info Overload Medical data doubles every 5 years Detailed patient biomedical markers Targeted therapies 5


Why is it sohard for computers to understandus?

Welch ran this?

Person

Organization

L. Gerstner

IBM

J. Welch

GE

W. Gates

Microsoft

“If leadership is an art then surely Jack Welch has proved himself a master painter during his tenure at GE.”

Noses that run and feet that smell? How can a house burn up as it burns down? Does CPD represent a complex comorbidity of lung cancer? What mix of zero-coupon, non-callable, A+ munis fit my risk tolerance?

6


IBM Watson combinestransformational technologies

2 Generates and 1 Understands

evaluatesevidencebased hypothesis

natural language and human communication

3 Adapts and learnsfrom user selections and responses

7

…built on a massively parallel architecture optimized for IBM POWER7


Watson enables three classes of cognitive services Ask • Leverage vast amounts of data • Ask questions for greater insights • Natural language inquiries • e.g. - Next generation Chat

Discover • Find the rationale for given answers • Prompt for inputs to yield improved responses • Inspire considerations of new ideas • e.g. - Next generation Search Discovery

Decide • Ingest and analyze domain sources, info models • Generate evidence based decisions with confidence • Learn with new outcomes and actions • e.g. - Next generation Apps Probabilistic Apps 8


Watson made incremental progress in precision and confidence IBM Watson Playing in the Winners Cloud v0.8 11/10 V0.7 04/10 v0.6 10/09

Precision

v0.5 05/09 v0.4 12/08 v0.3 08/08 v0.2 05/08 v0.1 12/07

Baseline 12/06

9


Informed decision making: search vs. Watson

Decision Maker Has Question

Search Engine

Distills to 2-3 Keywords

Finds Documents Containing Keywords

Reads Documents, Finds Answers

Delivers Documents Based on Popularity

Finds & Analyzes Evidence

Watson

Decision Maker

Understands Question

Asks NL Question

Produces Possible Answers & Evidence

Considers Answer & Evidence

Analyzes Evidence, Computes Confidence Delivers Response, Evidence & Confidence

10


Medical journal concept annotations Diseases

Medications

11

Symptoms

Modifiers


How Watson works: DeepQA Architecture Learned Models help combine and weigh the Evidence

Evidence Sources Answer Sources

Inquiry

Primary Search

Answer Scoring Candidate Answer Generation

Evidence Retrieval

1000’s of Pieces of Evidence

Balance Models & Combine

Models

Models

Models

Models

Models

Deep Evidence Scoring 100,000’s Scores from many Deep Analysis Algorithms

100’s Possible Answers

100’s sources

Inquiry Inquiry/Topic Multiple Decomposition Analysis Interpretations

Hypothesis Generation

Hypothesis and Evidence Scoring

Synthesis

Final Confidence Merging & Ranking

of a question

Hypothesis Generation

12

Hypothesis and Evidence Scoring

Responses with Confidence


Key Elements of the Clinical Diagnostic Reasoning Process Patient’s Story

Knowledge

Data Acquisition

Context

Accurate Problem Representation Generation of Hypothesis

Experience

Search for & Selection of Illness Script Diagnosis

Bowen J. N Engl J Med 2006;355:2217-2225

13

Dr. Martin S. Kohn | Clinical Decision Support: DeepQA


Watson’s Reasoning • “Shallower” reasoning over large volumes of data • Delivers weighted responses to clinicians to assist in making a informed evidence based decison ‒ Considers large amounts of data (e.g. EMR, Literature) ‒ Unbiased ‒ Learns • Hits sweet spot of human judgment (e.g. problems with bias, Big Data) • Identifies missing information • Watson’s interactive process helps clinician vector in on the appropriate decisions • Not limited by database structure

14 14

14 Feb. 2012

Dr. Martin S. Kohn | Clinical Decision Support: DeepQA


Where to put Watson to work Watson Capabilities Natural language understanding

Problems that require the analysis of unstructured data

Broad domain of unstructured data

Critical questions that require decision support with prioritized recommendations and evidence

Hypothesis generation and confidence scoring Iterative Question/Answering Machine learning 15

Best Fit for Watson

High value in decision support Leverage scale to maximize machine learning and improve outcomes over time © 2012 International Business Machines Corporation

Imagine if…

. . . the 1.5M people diagnosed with cancer in the US last year had a better prognosis? That’s exactly what a major health plan provider is working to accomplish. “Watson can aggregate information and give probabilities that will enable (experts) to zero in on the most likely diagnosis.” -Dr. Steven Nissen, Cleveland Clinic

16

DECIDE © 2012 International Business Machines Corporation

Imagine if… . . . new insights from medical research find their way to patient treatment programs in months instead of years? That’s exactly what a global leader in cancer care is doing today. “Watson will be an invaluable resource for our physicians and will dramatically enhance the quality and effectiveness of medical care.” -Dr Sam Nussbaum, Chief Medical Officer, WellPoint

DISCOVER 17


IBM Oncology Diagnosis and Treatment Advisor Demonstration

Shows how Watson can assist an Oncologist by: Synthesizing disparate data – patient records, clinician notes, test results, pathology reports, etc. Identifying missing pieces of data recommending tests with complete transparency Suggesting personalized, confidenceweighted, evidence-based options to improve quality of care and patient experience

18


Symptoms Family Medications Findings Patient History History

Patient

Findings

Medications History

Family History

A Her 58-year-old woman presented levothyroxine, to her medications were A 58-year-old woman complains of primary hydroxychloroquine, care physician after pravastatin, several days and anorexia, mouth, Adizziness, urine dipstick was dry positive for dizziness, anorexia, dry mouth, of alendronate. increased thirst, and frequent leukocyte esterase and nitrites. The Herurination. family history included oral and increased thirst, and frequent urination. She had also had a fever. Her history was notable for cutaneous patient was given a prescription for bladder cancer in herosteoporosis, mother, She had also hadno a fever and She reported pain herreported abdomen, lupus, hyperlipidemia, ciprofloxacin for ain urinary tract that Graves' disease in two food would “get stuck” when she left was back, and no cough, orsisters, diarrhea. frequent urinary infections,a infection. 3tract days later, patient hemochromatosis in one sister, and swallowing. She reported no pain in oophorectomy for a benign cyst, andher reported weakness and dizziness. idiopathic thrombocytopenic abdomen, back, or flankpressure and no cough, primary diagnosed Herhypothyroidism, supine blood was a purpura in one sister shortness of breath, diarrhea, ordysuria year earlier 120/80 mm Hg, and pulse was 88.

19 19

difficulty swallowing fever dry mouth thirst anorexia frequent urination dizziness noabdominal pain no back pain no cough no diarrhea Oral cancer Bladder cancer Hemochromatosis Purpura Graves’ Disease (Thyroid Autoimmune) cutaneous lupus osteoporosis hyperlipidemia frequent UTI hypothyroidism Alendronate pravastatin levothyroxine hydroxychloroquine urine dipstick: leukocyte esterase supine 120/80 mm HG heart rate: 88 bpm urine culture: E. Coli

Diagnosis Models

s ding Fin tions dica ry Me Histo y r . Pat . Histo s Fammptom Sy

Symptoms

Putting the pieces together at point of impact can be life changing

Confidence

Renal Failure UTI Diabetes Influenza Hypokalemia Esophagitis • Extract Symptoms from record Most MostConfident ConfidentDiagnosis: Diagnosis:UTI Diabetes Esophagitis Influenza • Use paraphrasings mined from text to handle ••••Extract Extract Identify ExtractMedications Patient Family negative History History Symptoms alternate phrasings and variants • •••Use Reason Use database Medical with Taxonomies mined of drug relations side-effects generalize to explain medical away Perform broad search for to possible diagnoses • • Together, symptoms conditions multiple (thirst to the is granularity diagnoses consistent may used w/ best UTI) by the explain models Score Confidence in each diagnosis based on symptoms evidence so far • Extract Findings: Confirms that UTI was present


We have only just begun to build a new era of computing powered by cognitive systems Transforming how organizations think, act, and operate Learning through interactions Delivering evidence based responses driving better outcomes

20


21
