Large Scale Predictive Analytics for Chronic Illness using FuzzyLogix s In-Database Analytics Technology on Netezza

Similar documents
Commercial Health Insurance Claims Data. for Studying HIV/AIDS Care. Senior Scientist, Innovus Epidemiology. David D.

Predictive Models: Current and Future. November 9, Steve Wickstrom Vice President, Research and Methods OptumInsight

Collective Impact Report

Leveraging Data for Targeted Patient Population Health Improvements

Final Report DIABETES VALUE METRIC. valueworks. Executive summary. The need for a measure of value

Presenter. Rebecca Susic Director Account Management MEDai

PREVENTATIVE COMMUNITY PHARMACY DIABETES MANAGEMENT PROGRAMS BROOKE HUDSPETH, PHARMD, CDE, MLDE KROGER DIABETES CARE

Risk Classification Modeling to Combat Opioid Abuse

SmartLife - Addressing the Needs of People with Diabetes (PwD) Using Machine Learning

SUPPORTING EMPLOYEES WITH CANCER: THE CANCER CARE HUDDLE. March 26, 2018

Provider Service Model. Collaborating for Success Jodi Stockslager, Sr. Provider Advocate, Provider Relations

May 16, Division of Dockets Management (HFA-305) Food and Drug Administration 5630 Fishers Lane, Room 1061 Rockville, MD 20852

Nebraska Diabetes Prevention Action Plan

Do OurHealth primary care clinics improve health & reduce healthcare costs? OurHealth Patient Engagement Analysis June 2018

Social Determinants of Health

SPOTLIGHT ON SENIOR HEALTH

Submitted to the House Energy and Commerce Committee. Federal Efforts to Combat the Opioid Crisis

California HealthCare Foundation Health Care Leadership Program 2010 Evaluation Focus: Social Network Analysis (Part 1)

Challenges for U.S. Attorneys Offices (USAO) in Opioid Cases

Boosting the Value of Lab Testing: How HEDIS Uses Lab

The Architecture of Performance Measurement

MULTIPLE LINEAR REGRESSION 24.1 INTRODUCTION AND OBJECTIVES OBJECTIVES

Innovator Case Studies: Oncology Networks

The National Vaccine Advisory Committee: Reducing Patient and Provider Barriers to Maternal Immunizations

Factors associated with unsuppressed viral load in HIV-1 infected patients on 1 st line antiretroviral therapy in South Africa

Advancing the management of Chronic Kidney Disease. Employee Benefits Planning Association- December s Program 12/6/2017 1

PHARMACY BENEFITS MANAGER SELECTION FAQ FOR PRODUCERS

Making Diabetes Prevention a Reality: The National Diabetes Prevention Program

PROVIDER CONTRACT ISSUES

Complex just became comfortable.

3/20/2013. "ICD-10 Update Understanding and Analyzing GEMs" March 10, 2013

Policy Writing and Advice 2 Manager Training 2 Employee Awareness 3 Rehabilitation 3

Agenda. Immunization Registry Reporting in Community Health Centers. Presented by: Ben Pierson Program Manager Health Information Exchange

Applied Medical. Statistics Using SAS. Geoff Der. Brian S. Everitt. CRC Press. Taylor Si Francis Croup. Taylor & Francis Croup, an informa business

Corporate Presentation Fourth Quarter 2017

Webinar Series: Diabetes Epidemic & Action Report (DEAR) for Washington State - How We Are Doing and How We Can Improve.

Low Back Pain Report October 2013: Cost and Utilization of Health Care in Oregon

Evidence-based Health Program Overview. yourjuniper.org. Today

VISION CARE INVESTMENT PAYS BIG BENEFITS.

caspa Comparison and Analysis of Special Pupil Attainment

Facilitating Cross-System Data Sharing for Psychotropic Medication Oversight and Monitoring

Applying Six Sigma Principles to Drive Healthcare Behavior Change:

The Increasing Number of Opioid Overdose Deaths in the United States. A Brief Overview

Note: This is an authorized excerpt from 2016 Healthcare Benchmarks: Population Health Management. To download the entire report, go to

Performance Analysis:

Overview of the NC Diabetes Prevention and Management Guide. Ronny Bell, Ph.D., MS, Chair Jan Nicollerat, MSN, RN, ACNS-BC, CDE, Vice Chair

Creating Better Health

Data for Healthy Insights

FAMILY & CHILDREN S SERVICES STRATEGIC PLAN

1.4 - Linear Regression and MS Excel

ONLINE CHRONIC DISEASE SELF-MANAGEMENT PROGRAM BETTER CHOICES BETTER HEALTH INTRODUCTION FOR ACL OPPORTUNITY

Call for Proposals: Demonstration Projects and Champion Development for Providers to address Type 2 Diabetes Prevention

Jefferies Healthcare Conference. June 2016

Making a difference through health How PwC is helping to change lives

Real World Patients: The Intersection of Real World Evidence and Episode of Care Analytics

THE POWER OF. Savings through the largest dentist network. Hometown expertise. Measurably superior service

Provider Bulletin 2016 Fourth Quarter

National Drug and Alcohol Treatment Waiting Times Report

ICD-10 Contingency Planning Thinking through Step Up Step Down Translation for Contingency Planning Ryan McDermitt, VP compliance Products, Edifecs

Injecting Equipment Provision in Scotland Survey 2011/12

Updating immunization schedules to reflect GSK vaccines

Data-Driven Study Feasibility Assessment and Impact on Successful Execution of Clinical Trial Protocols

ASO core offerings. Self-funded groups, sized 100+

Supplementary Online Content

Harold Rogers Update Melissa McPheeters, PhD, MPH

International Journal of Research in Science and Technology. (IJRST) 2018, Vol. No. 8, Issue No. IV, Oct-Dec e-issn: , p-issn: X

Camden Citywide Diabetes Collaborative

Physician Engagement and Prediabetes

Predicting New Customer Retention for Online Dieting & Fitness Programs

Jefferson Healthcare Rural Health Dental Clinic

Pediatric Restorative Benefits: Potential for Fraud & Abuse

Leveraging Electronic Health Data in a Multinational Clinical Trial: Early Learnings from the HARMONY- OUTCOMES EHR Ancillary Study

The NCQA Population Health Management Resource Guide. Natalie Mueller, MPH Manager, Product Development NCQA

Combination therapy compared to monotherapy for moderate to severe Alzheimer's Disease. Summary

Team-Based Decision Support in Diabetes Outcomes and Costs

Uses of the NIH Collaboratory Distributed Research Network

Thank you for joining today, please wait while others sign in.

Using IBM Unified Data Model for Healthcare to Maximize the Value of Unstructured Data in a Population Healthcare Management Program

Workshop Overview. The Problem. National Milestones. The Registry Solution. Benefits of Registry/Managed Care Collaboration

Clicking on the New Patient button allows the user to enter or edit patient and subscriber information to be stored for future use.

Analysis and Interpretation of Data Part 1

Mental Toronto Hydro

Archimedes, Medicare, and ARCHeS

At the Intersection of Public Health and Health Care: CDC s National Asthma Control Program

About The Report. imarc. Key Questions Answered in this Report:

Linking Public Interests to Ensure Sustainable Statewide Quitlines

WA PMP Access by Public Payers. PDMP North Regional Meeting St. Louis, MO April 23-24

2017 Drug Trends Series

Dementia-Capable North Carolina A Strategic Plan for Addressing Alzheimer s Disease and Related Dementias

2 Surgeries, a Spinal Cord Stimulator and the $500,000 Difference

INSIGHTS INTO ADHD CARE IN GERMANY BASED ON SHI CLAIMS DATA 03. March Results of the CoCA Study (Horizon 2020)

Appendix C CHANGING THE TRAJECTORY:

ARKANSAS MEDICARE CHRONIC CONDITIONS REPORT. July 1, T. Mac Bird, PhD, APCD Analytic Lead Kenley Money, APCD Director. Version

Table of Contents. EHS EXERCISE 1: Risk Assessment: A Case Study of an Investigation of a Tuberculosis (TB) Outbreak in a Health Care Setting

INSIDE BROKER THE SAFE CHOICE IS OFTEN THE

What Can We Do to Help Physicians Get into the Business of Immunization? A Preliminary Environmental Scan and Brainstorming Session

Reaching Out Model Programs Fact Sheet

Cleveland Clinic Home Monitoring Pilot

Insurance Providers Reduce Diabetes Risk Through CDC Program

Multi Parametric Approach Using Fuzzification On Heart Disease Analysis Upasana Juneja #1, Deepti #2 *

Transcription:

Large Scale Predictive Analytics for Chronic Illness using FuzzyLogix s In-Database Analytics Technology on Netezza Christopher Hane, Vijay Nori, Partha Sen, Bill Zanine Fuzzy Logix Contact: Cyrus Golkar, EVP, Business Development, Fuzzy Logix, Mobile: 408-858-7979, cyrus.golkar@fuzzyl.com, http://www.fuzzyl.com/

Netezza TwinFin & Fuzzy Logix The Simple Appliance Built for Serious Analytics Ingenix, Inc. 2

About Ingenix Part of UnitedHealth group of companies, #21 in Fortune 500 Industry leader in health care information and technology Enable secure delivery of health claims and clinical information for more than 1 in 7 Americans, and touching over 100,000 health care professionals Largest private health care database with 90+ million patient lives over 17 years Work with over 6,000 hospitals, 250,000 physicians, and 350 state and Federal agencies We are passionate advocates for the use of health information to save lives, improve care, and solve fundamental problems in health care. We Serve Hospitals and Delivery Systems Physicians Commercial Payers Government Payers Government Regulators Life Sciences Employers Ingenix, Inc. 3

HealthImpact - Summary Build mathematical models for identifying individuals at risk of developing a certain disease or condition > similarity of historical claims with others whose history and outcome is known Early intervention for people at risk so as to treat, delay or even prevent the onset of the disease Employer or health plan asks us to score their members > Data is de-identified before Ingenix sees it. > We return scores to client s wellness vendor and some reports to client > Wellness vendor does outreach to enroll at-risk patients > Ingenix may get feedback on test results Ingenix, Inc. 4

Research Data Warehouse Medical, Pharmacy and Enrollment from 1993 to today Over 70M people In 2009 13.3 million individuals with both medical and pharmacy coverage + 8.7 million with just medical coverage Ingenix research data mart Ingenix, Inc. 5

Why is this important? In 2008, CDC estimated that 23.6 million Americans, (7.8%), had diabetes and another 57 million adults had prediabetes. > As many as 1 in 3 U.S. adults could have diabetes by 2050 if current trends continue About 27 percent of those with diabetes do not know they have the disease > Only half of the adults classified as being at high risk for serious vision loss, visited an eye doctor in the past 12 months Diabetes costs $174 billion annually, including $116 billion in direct medical expenses. > $9,677 for each person diagnosed > $700 for each American Ingenix, Inc. 6

Overview Classification model used to predict the class of new data Input for each individual in model > diagnosis, procedure, drug claim codes over three-year period > demographics (age, gender, zip code, ) Output > probability of that individual developing diabetes Collect Medical Facts for at most 3 years Patient Time Line Ignore 90 days First Detected Diabetes Event For controls, a day >= 90 days before end of enrollment Ingenix, Inc. 7

Some reasons why we use IBM Netezza TwinFin P12 Physician claims: ~1 TB, 3.6 Billion rows Facility Claims: ~325 Gb, 1.4 Billion rows Pharmacy Claims: ~375 Gb, 1.3 Billion rows Enrollment ~100 Gb, 0.6 Billion rows Physician and Facility claims have up to 9 diagnoses on each row. > We want to pivot this table into Individual, diagnosis, service date > Self-union the tables to each other 9 times > Result is table 55 Gb & 4.1 Billion rows created in 4m34s Enrollment data has multiple enrollments that needs to be sewn together to determine longest continuous spans > Complex lead and lag window logic > Created in ~14m I write queries without worrying about optimization. Ingenix, Inc. 8

Identifying Population with Diabetes & without Healthcare Effectiveness Data & Information Set (HEDIS) > measures developed by National Committe for Quality Assurance (NCQA) An individual has diabetes if he/she has claim(s) with specific > diagnosis and procedure codes on same day, OR > drug code(s), insulin Initial Data Counts: > 3.5M people with diabetes (with at least 1 HEDIS criteria match) > 68M without diabetes In the model > 2.2M rows, use 20% for out of sample testing Ingenix, Inc. 9

Statistical Model Compute mathematical relationship (logistic regression) diabetic(1,0) ~ demographics (gender, age, minority, ) + codes (diagnosis, procedure, drugs) Codes considered for model must meet filter criteria (density and odds ratio) and clinical criteria > We also merge codes into more informative sets Individuals considered for model (cohort matching) > for every person with diabetes, choose n without diabetes with similar geographic (census) division years of claims Ingenix, Inc. 10

Running the Model Model Matrix > #columns (codes) 1,540 > #rows (individuals) 1,774,759 > #non-zero entries 38,397,099 > very sparse, density ~1% If in-memory statistical software was used > matrix one-tenth the size takes several hours to run > cannot compute key statistics (gini coefficient) even for that size > need to sample and combine results > computing measures of matrix stability such as variance inflation factor would not be meaningful Ingenix, Inc. 11

Using FuzzyLogix In-Database Analytics with Netezza TwinFin P12 Developed SQL code which runs end-to-end > population identification and cohort matching > claims extraction, merging, filtering > model matrix building and solving > model diagnostics (correlations) > test data accuracy statistics (data not used in training) RUNS IN < 30minutes Advantages > can run multiple scenarios to see impact of assumptions should there be a 3 claims requirement or can that be relaxed? should we have multiple models for each age group or just one? > custom models for different customers If pharmacy data is not available all individuals being scored are in the southeast only Ingenix, Inc. 12

Plot shows median, inter-quartile range (box covers 50% of population) and whiskers out to 1.5*IQR > Thicker line is actually dots for outliers beyond 1.5*IQR Gini coefficient is 0.87, cstatistic is 0.92 on out of sample data Ingenix, Inc. 13

Conclusions Ingenix could not do this work without IBM Netezza or FuzzyLogix in-database analytics. We have already scored over 2 million people for a partner who uses this service to enroll members in diabetes specific wellness programs. Fuzzy Logix Contact: Cyrus Golkar, EVP, Business Development, Fuzzy Logix, Mobile: 408-858-7979, cyrus.golkar@fuzzyl.com, http://www.fuzzyl.com/ Ingenix, Inc. 14