Session 11 PD, Provider Perspectives of Values Based Payment Programs Moderator: William T. O'Brien, FSA, FCA Presenters: Donald Fry, M.D. Lillian Louise Dittrick, FSA, MAAA Colleen Audrey Norris, ASA, MAAA
UnityPoint Health Analytics SOA Spring Health Meeting Session 11 June 15, 2015 Lillian Dittrick, FSA, MAAA Assistant Director, Strategic Analytics
UnityPoint Health Overview Originally formed in 1994 2
UnityPoint Health Overview 3
Patient Stratification Overview 1. High risk patient identification and stratification 2. Regression model 3. Information Use: Care Navigators visualization tool 4. Program measurement 5. Next phase modeling to increase predictive power IBMs Watson Natural Language Processor to extract unstructured electronic medical record information Lifestyle & Socio-economic data Consumer data
1. Patient Identification and Stratification
2. Regression Model Strategy 2 Year Time Horizon Prior 12 Months Subsequent 12 Months Training Data Outcome 100+ Total Predictor Variables Demographics Specific Diagnoses Utilization of health services Emerging Low Prior High Subsequent Continuing High Prior High Subsequent
2. Regression Model Strategy Logistic Regression using Penalized Maximum Likelihood Estimation Optimum Penalty determined by maximizing Hurvich and Tsai s corrected AIC Validated using Bootstrap Resampling with no variable deletion Bias-Corrected AUC Model Type MSSP/Pioneer SIHP Continuing 0.73 0.81 Emerging 0.76 0.79
3. Information Use: Care Navigators visualization tool
3. Information Use: Care Navigators visualization tool
4. Program Measurement (sample data) 10
4. Program Measurement (sample data) 11
5. Next Phase Augment Training Data w/ Other Covariates Automatically identify information buried in unstructured electronic medial record data BMI, HDL, LDL, A1C, Pulse, Blood Pressure, Cholesterol, Tobacco, Triglycerides Pre-Diabetes, Diabetes, COPD, Hyperlipidemia (LDL >= 160, TRIG >= 500), Asthma (mild, moderate, severe persistent), Hypertension, Obesity (BMI > 30), CHF, Behavioral Health, Osteoarthritis, Rheumatoid Arthritis not on biologicals, CKD, Ischemic Heart Disease, CVA, AF, Tobacco use, Substance Abuse, Inflammatory Bowel Disease, Barrettes Esophagitis, Chronic Pain, Fibromyalgia, Endometriosis, Dementia, Parkinson, ED usage, Chronic Migraine, MS, Downs Syndrome Lifestyle and socio-economic data Consumer data 12
5. Next Phase Watson Content Analytics Annotators Annotators are used to identify valuable facts in unstructured (natural language) notes sections of the electronic medical record. Entire data sets can be analyzed for data mining. Individual records can be analyzed in real time. Annotators are configured by UPH team. DataSkill and IBM can both provide libraries of annotators. 13
5. Next Phase Watson Content Analytics Annotators Annotators are used to identify valuable facts in unstructured documents (e.g. clinician notes, consult reports, free text fields in electronic medical record) and convert to a structured form 14
5. Next Phase Watson Content Analytics Annotators Problems, Procedures Result of a series of interim annotations that identify diseases, symptoms, and disorders Normalize to standard terms and standard coding systems including SNOMED CT, ICD-9, HCC, CCS Capture timeframes of the problem determine if past or current problem Determine confidence Positive, Negative, Rule Out Negation example abdominal pain 15
5. Next Phase Watson Content Analytics Annotators Medications Result of a series of interim annotations that identify drugs, administrations, measurements Normalize to standard terms and can normalize to RxNorm
5. Next Phase Watson Content Analytics Annotators Demographic and Social Patient Age Living Arrangement Employment status Smoking status Alcohol use Compliance & Noncompliance Patient's history of medication compliance with directions such as "take all doses, even if you feel better earlier Noncompliance - Patient's history of medication noncompliance with directions. Labs results Type of lab test performed, unit of measure, result value Ejection Fraction in support of CHF use cases Coding Systems can identify these codes CPT CCS HCC NDC ( National Drug Codes) Can break out by components - example, Lortab 5 contains 5 mg of hydrocodone and 500 mg acetaminophen. This would result in 2 NdcCode annotations.
Questions
Bridging the Information Gap: Leveraging consumer data for population segmentation Presented by Colleen Norris, ASA, MAAA
Presentation Objectives 1) Applications of enhanced patient identification in a provider setting 2) Logistics of integrating these methodologies 3) Outcomes and predictive variables 4) Challenges and opportunities 2 June 15, 2015
Where are the opportunities? 40.00% 35.00% Population and Annual Claims Costs People % of People Percent % of Total of Total Cost 30.00% 25.00% 20.00% 15.00% 10.00% 5.00% 0.00% Source: 2011 and 2012 Truven MarketScan Data. 3 June 15, 2015
Where are the opportunities? 40.00% 35.00% Population and Annual Claims Costs People Percent of Total % of People % of Total Cost 30.00% 25.00% 20.00% 15.00% 10.00% 5.00% 0.00% High Cost 4 June 15, 2015
Health Status Prospective Risk Score Opportunities for Patient Identification Hidden Opportunities High Risk Critical Relatively Healthy Chronic Illness Short-term Illness Highutilizers Low Cost Mid Cost High Cost Member PMPM Cost 5 June 15, 2015
Rising Risk Low Cost Group AGE LOW COST MEDIUM COST HIGH COST <18 88.2% 11.2% 0.6% 18 24 85.1% 13.5% 1.4% 25 34 80.0% 17.6% 2.4% 35 44 79.1% 18.6% 2.3% 45 54 74.8% 22.4% 2.8% 55 64 70.5% 25.5% 4.0% Total 80.8% 17.2% 2.0% Source: 2011 and 2012 Truven MarketScan Data. Percentages indicate the likelihood of year two claim costs, given that an individual was low cost in year one, by age group. 6 June 15, 2015
Applications Identify possible highrisk patients Identify cohorts at risk for certain diseases Identify rising risks Match patients with providers who are most likely to connect. 7 June 15, 2015
What is consumer data? Geography Demographics Financials (e.g., Household Income Index) Household Automotive Data Lifestyles (e.g., Hobbies Self-Reported) Proprietary Attitudinal/Segmentation Elements (e.g., LIVING WELL, PRIZM, NICHES) 8
Where does consumer data come from? 9
Perfect world Segmentation Approach 10
Low Cost Use correlation tables to flag individuals who may be at risk. Mid Cost Develop limited predictive model High Cost Develop full blown predictive model 11 June 15, 2015
Practical approaches for using consumer data in a provider organization 1. Collect and correlate data: Purchase consumer data from a data vendor. Data on a per - individual basis is inexpensive Collect data and information from patient intake into accessible and usable EMR Create correlation matrices between patient information and risk score or other available medical information. 2. Create combined variables 3. Provider feedback 4. Develop predictive models 12 November 13, 2014
Small Steps what else may segment population risk beyond age and gender? 13 June 15, 2015
Family Composition 14 June 15, 2015
Practical approaches for using consumer data in a provider organization 1. Collect and correlate data: Purchase consumer data from a data vendor. Data on a per - individual basis is inexpensive Collect data and information from patient intake into accessible and usable EMR Create correlation matrices between patient information and risk score or other available medical information. 2. Create combined variables 3. Provider feedback 4. Develop predictive models 15 November 13, 2014
Combinations of Variables 16
Combinations of Variables 17
Top predictive variables 18
Practical approaches for using consumer data in a provider organization 1. Collect and correlate data: Purchase consumer data from a data vendor. Data on a per - individual basis is inexpensive Collect data and information from patient intake into accessible and usable EMR Create correlation matrices between patient information and risk score or other available medical information. 2. Create combined variables 3. Provider feedback 4. Develop predictive models 19 November 13, 2014
Using consumer data at the practice level 20
Practical approaches for using consumer data in a provider organization 1. Collect and correlate data: Purchase consumer data from a data vendor. Data on a per - individual basis is inexpensive Collect data and information from patient intake into accessible and usable EMR Create correlation matrices between patient information and risk score or other available medical information. 2. Create combined variables 3. Provider feedback 4. Develop predictive models 21 November 13, 2014
Using Consumer Data at a provider level: Key Takeaways 22