Data Analytics for Population Health Sponsored by 1915 N. Fine Ave #104 Fresno CA 93720-1565 Phone: (559) 251-5038 Fax: (559) 251-5836 www.californiahia.org Program Handouts Monday, June 8 Track Two 2:10 pm - 3:10 pm 2016 CHIA Convention & Exhibit Speaker Susan H. Fenton, PhD, RHIA, FAHIMA California Health Information Association, AHIMA Affiliate
California Health Information Association California Health Information Association Data Analytics for Population Health June 6, 2 16 Susan H. Fenton, PhD, RHIA, FAHIMA Associate Dean for Academic Affairs UT School of Biomedical Informatics at Houston susan.h.fenton@uth.tmc.edu Copyright This material is adapted from materials developed by Johns Hopkins University, funded by the Department of Health and Human Services, Office of the National Coordinator for Health Information Technology under Award Number 90WT0005. This work is licensed under the Creative Commons Attribution- NonCommercial-ShareAlike 4.0 International License. To view a copy of this license, visit http://creativecommons.org/licenses/by-ncsa/4.0/. California Health Information Association, AHIMA Affiliate 1
Goals/Objectives or Agenda Define the terms and describe the perspectives related to population health and public health. Discuss determinants of population health. Review case studies related to population health. Explore different data sources for population health. Examine the architecture needs for population health data analytics. Population Health Informatics: An Integration of Three Disciplines 1.01 Figure. Center for Teaching and Learning, Bloomberg School of Public Health, Johns Hopkins University (2016). California Health Information Association, AHIMA Affiliate 2
Public Health is the Entire Population 1.02 Figure. Center for Teaching and Learning, Bloomberg School of Public Health, Johns Hopkins University (2016). Population Health is a Subset of the Public 1.03 Figure. Center for Teaching and Learning, Bloomberg School of Public Health, Johns Hopkins University (2016). California Health Information Association, AHIMA Affiliate 3
Working Definition Population health informatics: The systematic application of HIT, digital technologies, and information sciences to the improvement of the health and well being of a defined community or other target population. The Institute for Healthcare Improvement (IHI) Triple Aim 1.17 Figure. Stiefel & Nolan (2012). 8 California Health Information Association, AHIMA Affiliate 4
Population Health Management 1.18 Figure. Adapted by Jonathan Weiner and Center for Teaching and Learning, Bloomberg School of Public Health, Johns Hopkins University (2016) with permission from Moorhead (2010). Population Risk Pyramid: Types of Care 1 1.19 Figure. Adapted by Jonathan Weiner and Center for Teaching and Learning, Bloomberg School of Public Health, Johns Hopkins University (2016) from NHS England (2015). California Health Information Association, AHIMA Affiliate 5
Population Risk Pyramid: Types of Care 2 1.20 Figure. Adapted by Jonathan Weiner and Center for Teaching and Learning, Bloomberg School of Public Health, Johns Hopkins University (2016) from NHS England (2015). Examples of HIT Support for Population Health within an IDS Risk identification/stratification/outreach. Consumer education (at all levels of risk pyramid). Provider care process improvement: Guidelines/clinical decision support Education/benchmarking/P4P Supporting patient/consumer disease management. Monitor population outcomes. Evaluate impact of interventions. Enhance knowledge (e.g., support pragmatic and conventional clinical trials). California Health Information Association, AHIMA Affiliate 6
Determinants of Population Health Source: University of Wisconsin Population Health Institute, Robert Wood Johnson Foundation, Burness. (2016). Retrieved April 20, 2016, from http://www.countyhealthrankings.org/our approach 2.02 Graph: Center for Teaching and Learning, Bloomberg School of Public Health, Johns Hopkins University. (2016). Unified Framework of Disease Management, Population Health, and Prevention Source: Johns Hopkins University Center for Population Health Information Technology. 2.08 Figure: Center for Teaching and Learning, Bloomberg School of Public Health, Johns Hopkins University. (2016). California Health Information Association, AHIMA Affiliate 7
Advances in Population Health Analytics Ways to integrate disparate numerators and denominators to define true populations. Ways to identify those at risk, both at the community and patient panel levels. Advanced tools for extracting unstructured data from many sources using text mining (NLP). Real time signals and dynamic modeling. Models and tools to help medical care systems move away from FFS toward population value perspectives. Data Type Specs 1 Variables: list of common variables for this data type. Background: general trend in the population and relationship to health care utilization (if available and applicable). Analytic: serving as dependent or independent variable (or both). Derived variables: different methods (e.g., direct query, algorithmic) used to extract derived variables from this data type (e.g., extract prescription pattern and adherence from medications). Coding standards: common coding conventions for this data type (e.g., RxNorm, National Drug Code or NDC, Anatomical Therapeutic Chemical Classification System or ATC, Systematized Nomenclature of Medicine or SNOMED). California Health Information Association, AHIMA Affiliate 8
Data Type Specs 2 Data sources: common data sources containing this data type (e.g., claims, EHR, HIE). Data quality: typical data quality issues with this data type (if any). Data interoperability: specific interoperability issues that may affect the collection, integration, and sharing of this particular data type. Legal considerations: issues with privacy and security that may impact accessing or utilizing this data type (e.g., Health Insurance Portability and Accountability Act or HIPAA). Health Care Determinants 5.01 Figure. Adapted from What are Population Health Determinants or Factors? Improving Population Health California Health Information Association, AHIMA Affiliate 9
Common Data Types Demographics (e.g., age, sex, gender). Diagnoses (e.g., diagnosis, severity). Medications (e.g., prescription, dispense). Procedures (e.g., inpatient, outpatient). Surveys (e.g., HRA, Patient Health Questionnaire version 9 or PHQ 9). Utilization (e.g., cost, hospitalization). Grouper variables (e.g., advanced clinical groups or ACGs, diagnosis related groups or DRGs). Demographics: Sample Data 5.02 Table. Courtesy Hadi Kharrazi 2016 Sample Age and Sex Data (synthetic data) California Health Information Association, AHIMA Affiliate 10
Demographics: Population Age Trend 1 Population by Age and Sex: 2000 and 2010 5.03 Figure. U.S. Census Bureau. Census 2000 summary file 1 and 2010 Census summary file 1 (as cited in Howden & Meyer, 2011). Demographics: Population Age Trend 2 Age Distribution and Median Age: 1960 to 2010 5.04 Table. Adapted from U.S. Census Bureau (as cited in Howden & Meyer, 2011). California Health Information Association, AHIMA Affiliate 11
Demographics: Health Care Cost Commercial [Payer] Aging Curve (age and sex) 5.05 Chart. Adapted by Hadi Kharrazi and the Center for Teaching and Learning, Bloomberg School of Public Health, Johns Hopkins University (2016). Copyright 2013 Society of Actuaries, Schaumburg, Illinois. Reproduced with permission. Demographics: Medicare Cost Medicare Aging Curve (age and sex) 5.06 Chart. Adapted by Hadi Kharrazi and the Center for Teaching and Learning, Bloomberg School of Public Health, Johns Hopkins University (2016). Copyright 2013 Society of Actuaries, Schaumburg, Illinois. Reproduced with permission. California Health Information Association, AHIMA Affiliate 12
Demographics: Medical Expenditure Panel Survey (MEPS) Cost 1 35% 30% 29% Total Cost of Top 5% Spenders 25% 20% 15% 10% 5% 5% 9% 10% 15% 18% 14% 0% <18 19 34 35 44 45 54 55 64 65 79 80+ Age Groups Percentage of healthcare expenses incurred by the top 5% within age groups (based on Medical Expenditure Panel Survey 2002 data) 5.07 Chart. Adapted from Stanton (2006). Demographics: MEPS Cost 2 Age Distribution of Low versus High Spending Groups (based on Medical Expenditure Panel Survey 2009 data) 5.08 Chart. Adapted from Schoenman & Chockley (2012). California Health Information Association, AHIMA Affiliate 13
Demographics: Population Distribution of Cost 5.09 Table. Adapted from Duncan (2011). Relative cost per member per year (PMPY) by age and sex (numbers are generated based on a large commercial claims database) Demographics: Distribution of Risk Factors 5.10 Table. Adapted from Duncan (2011). Relative risk per member per year (PMPY) by age and sex (numbers are generated based on a large commercial claims database; number of patients in each age range varies) California Health Information Association, AHIMA Affiliate 14
Demographics: Apply Age/Sex Risk Factors to a Population Predicting relative cost by using age and sex risk factors (note that relative costs for different age/sex categories can be expressed as relative risk factors, enabling us to assess the average risk of an individual, or the overall (relative) risk of a population) 5.11 Table. Adapted from Duncan (2011). Demographics: Predicting Future Health Costs Emp. Num. of Lives Age/Sex Risk Fact. Y0 (base) Age/Sex Risk Fact. Y1 (future) Risk Fact. Ratio % Predict Cost Y1 Actual Cost Y1 Diff $ Predict/ Actual Diff % Predict/ Actual 1 73 1.37 1.42 138% $4,853 $23,902 ($19,049) 392.5% 2 478 0.74 0.76 74% $2,590 $2,693 ($102) 3.9% 3 37 0.86 0.87 84% $2,965 $1,339 $1,626 54.8% 4 371 0.95 0.97 95% $3,331 $3,325 $6 0.2% 5 186 1.00 1.03 100% $3,516 $3,345 $170 4.8% 6 19 1.80 1.85 180% $6,328 $10,711 ($4,383) 69.3% 7 359 0.95 0.97 94% $3,315 $3,401 ($87) 2.6% 8 543 0.94 0.96 93% $3,269 $3,667 ($398) 12.2% 9 26 1.60 1.64 159% $5,595 $5,181 $414 7.4% Avg. 1.00 1.03 1.00 $ $ $ % Sum of absolute differences $26,235 Using demographic factors to predict future healthcare cost (Traditional (Age/Sex) risk prediction is somewhat accurate at the population level. Larger group costs are more predictable than smaller groups. Predicted Cost = Baseline Cost x Trend x Subsequent Year/Baseline Year Relative Age/Sex Factor) 5.12 Table. Adapted from Duncan (2011). California Health Information Association, AHIMA Affiliate 15
Diagnoses: Chronic Conditions and Cost Information about a patient s condition, particularly chronic condition(s) is potentially useful for predicting risk compared to age/sex alone 5.17 Table. Adapted from Duncan (2011). 31 Diagnoses: Condition-based Relative Risk Improvement in prediction of cost when condition based relative risk is applied. (Commercial risk groupers are available that predict relative risk based on diagnoses which is particularly helpful for small groups.) 5.18 Table. Adapted from Duncan (2011). California Health Information Association, AHIMA Affiliate 16
Diagnoses: Disease Severity and Cost Effect of disease severity on cost not all diabetics represent the same level of risk. Different diagnosis codes help identify levels of severity. (numbers are generated based on a large commercial claims database) 5.19 Table. Adapted from Duncan (2011). Traditional Data Sources of Population Health Insurance data: Hospital claims. Physician/professional claims. Pharmacy claims (prescription benefit management). Clinical data: Electronic health record (EHR) data. Registry data. California Health Information Association, AHIMA Affiliate 17
Nontraditional Data Sources of Population Health Patient provided/ generated data sources. Public health and vital records data sources. Social services data sources. Environmental and geographical data sources. Resource availability data sources. Consumer and nonmedical data sources. Health information exchange (HIE) data sources. Other medication data sources: Surescripts. PDMP. Other potential data sources. Factors Affecting Population Health Data Sources Increased end user adoption of health IT solutions (e.g., health care providers, consumers). Expanding data variety generated by emerging population health data sources (e.g., electronic health records [EHRs], personal health records [PHRs], mobile health [mhealth]). Enhanced continuity of data flow among health IT solutions across various health related domains (e.g., interoperability standards). Increased coverage of health IT solutions (e.g., larger population wide data repositories). California Health Information Association, AHIMA Affiliate 18
Variety of Data Sources 5.51 Figure. Adapted from Weiner (2012). Continuum of Data Sources 5.52 Figure. Courtesy Hadi Kharrazi 2016 California Health Information Association, AHIMA Affiliate 19
Data Linkage and Integration Challenges 1 Interoperability: the ability of a system to exchange electronic health information with and use electronic health information from other systems without special effort on the part of the user. The Roadmap [Online image]. Office of the National Coordinator for Health Information Technology. (2015). Data Linkage and Integration Challenges 2 MPI is needed to match patients when population health data are derived from multiple sources (e.g., matching EHR records with insurance claims). Creating and utilizing MPI is a complex process and may introduce error and bias in population health databases. The following data attributes are often used for MPI creation (note that most of them are protected health information): Name (first, middle, last, previous, suffix). Date of birth. Address (current and previous street address, city, state, ZIP code). Phone number (current, previous). Gender. California Health Information Association, AHIMA Affiliate 20
Population Health System Architecture and Design Centralized architecture: data is accumulated and managed in a single and centralized repository. Advantages: Simplicity/efficiency. Data are consistent, with easier patient linkage. Disadvantages: Doesn t scale up well. Single point of control must trust the custodian. Requires exceptional leadership. Everyone has to accept the same identifier. Needs robust communication infrastructures. Population Health System Architecture and Design 2 Centralized Population Health Data Warehouse Architecture 5.58 Figure. Courtesy Hadi Kharrazi 2016 California Health Information Association, AHIMA Affiliate 21
Population Health System Architecture and Design 3 Federated architecture: network permits users access only when needed. An MPI matching process is needed. Advantages: Data ownership can be managed by defining policies. Individual organizations able to control their own data. Benefits of scale. Builds on existing infrastructure. More opportunities for creativity. Disadvantages: Requires more coordination. May be slower than a monolithic database. The patient identifier problem has to be solved. Population Health System Architecture and Design 4 Federated Population Health Data Warehouse Architecture 5.59 Figure. Courtesy Hadi Kharrazi 2016 California Health Information Association, AHIMA Affiliate 22
Population Health at the Community Level: Case Study 1 Case Study 1: Understanding Community Level Factors Related to Obesity among Veterans. Builds on the largest EHR database in the U.S., with more than 30 million patients. Potentially the largest historical database in the U.S., with historical BMI information and clinical risk factors. Based on residence of veteran project links in public health, socioeconomic, and community level data relevant to the obesity epidemic at the population level. Population-Based Analytic/ Predictive Model for Obesity Electronic BMIs derived from EHRs of national care system (obesity levels based on 29,000 BMIs captured in one day in EHRs). 2.11 Figure: Johns Hopkins Center for Population Health IT (CPHIT). Map showing 2014 poverty rates by county. 2.12 Figure: U.S. Department of Commerce Economic and Statistics Administration, U.S. Census Bureau. California Health Information Association, AHIMA Affiliate 23
Geographic-Level Population Decision Support Obesity within the VA Health Care System 2.13 Figure: Image compilation from U.S. Department of Veterans Affairs (VA), Veterans Health Administration (VHA), Office of Analytics and Business Intelligence (OABI). Population Health at the Community Level: Case Study 2 Case Study 2: Using Health Exchange and Geo Data to Develop Community Level Health Metrics in Baltimore, Maryland. The CRISP HIE in Maryland is one of few in the U.S. with 100% participation of all hospitals. Maryland now has all payer global budget that rewards hospitals for shifting focus to the community. Collaborative team is linking medical, geographic, and social data to address various health issues at the community level. Projects include hot spotting outlier hospitalizations, falls in the elderly, and opioid addiction. California Health Information Association, AHIMA Affiliate 24
Conceptual Model for the Maryland Population Health Information Network (M-PHIN) 2.14 Chart: Jonathan Weiner and Center for Teaching and Learning, Bloomberg School of Public Health, Johns Hopkins University. (2016). Hot-Spotting Baltimore Hospitalizations Using HIE Data 2.15 Figure: Maryland Department of Health and Mental Hygiene for U.S. Department of Health and Human Services, Centers for Medicare and Medicaid Services State Innovation Models (SIM) Initiative. California Health Information Association, AHIMA Affiliate 25
Population-Based Predictive Analytic/Intervention of Elders Falls Linking EHRs and Social Data Rates of falls by neighborhood, derived from real time health information exchange data in Baltimore City (draft results). 2.16 Figure: Adapted by Kharrazi et al. (2016). Johns Hopkins Center for Population Health IT (CPHIT). GIS map of housing types in Baltimore by census tract. 2.17 Figure: City of Baltimore Zoning Districts. Data Analytics for Population Health Summary Population and public health are not the same. Data for population health analytics comes from a wide variety of sources. Explored challenges utilizing the data for population health analytics California Health Information Association, AHIMA Affiliate 26
Question/Answer Thank you for your attention! California Health Information Association, AHIMA Affiliate 27