Big Data: Compiling Data Across EHRs
Tim Burdick MD MBA MSc
Dartmouth-Hitchcock Health Assistant Professor Primary Care Clinic Geisel School of Medicine Heater Road Community and Family Medicine Lebanon, NH Biomedical Data Science The Dartmouth Institute Nothing to disclose. Does he expect Forget the me to document stethoscope, kid. It’s the exam in the all about EHR data! computer?! Where are we going today?
Why digitize medical records?
EHRs under the hood
Linking data in PBRNs
Projects in PBRNs
The future of truly big data
Discussion Always start with the WHY?
Why make information electronic? Environment to Vision
• This is our patient, Fred
Environment ENVIRONMENT • Measure; create structure; codify
INDEX • Fred: HA1C = 7.8;BMI = 38 Data DATA • Look for patterns; understand relationships CONTEXT/ PAST RULES • Fred has Type 2 diabetes caused by his obesity Information INFORMATION • Understand patterns; find predictability MEANING/ MODEL • Patients w/ DM have lower quality life years & cost more Knowledge KNOWLEDGE • Apply patterns; understand principles/goals
INSIGHT • DM guidelines improve Quadruple Aim outcomes FUTURE
Wisdom WISDOM • Use compassion; apply values VALUE/ PURPOSE Vision • Effective treatment of patients with DM is good for all VISION
Carpenter & Canaday (2004) The Bad and The Ugly
ANNALS OF MEDICINE NOVEMBER 12, 2018 ISSUE WHY DOCTORS HATE THEIR COMPUTERS Digitization promises to make medical care easier and more efficient. But are screens coming between doctors and patients? By Atul Gawande And the Good
Millions of patients worldwide - 83% report feeling more more in control of their health - 25% report a correction to their medical record https://www.opennotes.org/
Choosing Wisely Clinical Decision Support Adherence and Associated Inpatient Outcomes
26,500 inpatient encounter measured adherence to CDS - Cost of care 7% lower - Length of stay 6% lower
Heekin et al. Am J Manag Care. 2018;24(8):361-366 In the Beginning there was...who is this man?
Lawrence L. Weed MD UVM College of Medicine Professor of Medicine, and Community Medicine, Director, PROMIS Laboratory (Problem Oriented Medical Information System)
Touchscreen of the Megadata Terminal. c. 1977 Grand Isle Clinic, VT – PROMIS (c. 1979)
Microwave dish Health Insurance Portability and Accountability Act (1996) Title II: Administrative Simplification Provisions
● Implemented 2005 for claims submissions ● Standard electronic codes for encounters (ICD-9, ICD-10) ● Standard data transmission methods ● Security requirements ● National Provider Identification (NPI)
HIPAA, not HIPPA (a female hippo?) What drove $36 billion dollar investment in health IT?
The US Housing Market
And the US economy generally American Reinvestment & Recovery Act (ARRA) 2009
Healthcare Information Technology for Econonomic and Clinical Health (HITECH)
● Federal government reimburses hospitals and physicians for EHR adoption ● Must demonstrate “Meaningful Use” of the EHR ● Incremental requirements each year ● Clicking for dollars Standards
● ICD – International Classification of Diseases (1900) ● SNOMED CT – Systematized Nomenclature of Medicine Clinical Terms (1965) ● LOINC – Logical Observation Identifiers Names and Codes (1994) ● RxNorm (2001)
RxNorm for codeine 30mg tablet: C0123456|Codeine|Codeine Phosphate|Codeine|P|30|MG|Component
SNOMED for scalding of the left index finger Common Data Models (CDM)
Enter heart rate in EHR user interface
EHR stores in database tables
Database Extraction, Transformation, Load (ETL)
• identify what data to pull out of EHR tables
• map to Observational Medical Outcomes Project (OMOP) data model
• merge data from multiple EHR into combine database with CDM MRN VS_Date VS_time VS_measure_source VS_value
34823948-1 2020_01_24 0819 HR_monitor 92
40294820-3 2019_11_12 2354 SBP_left_arm 135
93092835-7 2018_07_15 1343 DBP_right_arm 82
83458230-7 2017_10_28 1005 Ox_Sat_finger 91
00299128-4 2019_08_01 1739 wt_kg_bed 78
MRN Name_Last Name_First DOB Zip_Home
83458230-7 Flintstone Fred 457_BC 99731
34823948-1 Flintstone Wilma 462_BC 99731
93092835-7 Burdick Tim 1981 03755
Dartmouth Coop Charts
Then
Functional Assessments - Standardization - Paper forms - Manual data extraction & entry - Subsample of patients or charts - Data sizes in the hundreds - Time-intensive Dartmouth Coop Charts
Now
LOINC codes ● Formerly Oregon Community Health Information Network ● 1 shared EHR (Epic) since 2000 ● 500+ organization partners
○ > 100 FQHCs ● 47 states ● >10,000 providers ● > 500,000 unique patients per month ● >400 staff in 35 states ● Teach and lead clinical quality improvement to member organizations ● $53 Million in external research dollars ● Hundreds of peer-review publications in top journals PCORI CDRNs
● 9 clinical research data networks ● Regional aggregation of core data from many EHR systems ● EHR, Claims, Social Determinants, Geocoded, Patient Reported Outcomes, Insurance status, pharmacy dispense, test results, immunization registries pcornet: meta-network
66 million patients ADVANCE CDRN 550 million encounters - OCHIN-affiliated clinics - Fenway Health clinics 6 billion lab results - Oregon Health & Science University clinics
1 billion prescription dispenses pcornet projects
States with PCORI Rare Disease Project leads ● Do antibiotics in early childhood increase risk of obesity?
○ 700,000 children
○ 10-year longitudinal study
● Sleeve gastrectomy is safer than Roux-en-Y gastric bypass (RYGB)
○ 33,000 patients
○ 5-year study DARTNet (Distributed Ambulatory Research in Therapeutics Network)
. Started 2007 (AAFP, Univ Colorado Family Medicine, AHRQ, private tech company) . Developed ELT tools for any EHR . Map data to OMOP CDM . 12 PBRNs contracted services ● 5 million patients ● 12 million encounter per year 3 rural: WWAMI, Appalachian, Upstate NY ● 5 billion data points
. Value-add services
Quality improvement analytics and reports
Meaningful Use /PIP registries DARTNet Academic Partners
Academic Partners •Northwestern University •Ohio State University •University of Alabama at Birmingham •University at Buffalo •University of California San Diego •University of Colorado Denver •University of Kansas •University of Minnesota •University of Texas San Antonio •University of Vermont •University of Washington
NIH Clinical Translational Science Award centers (~ Dartmouth Synergy) © Epic Systems, Verona, WI Single Alcohol Screening Question
● 77 weeks of data ● 200+ clinics ● 95,537 screenings ● 24.5 million visits ● Now you have data about alcohol use patterns across large population
Correlate w use of services, chronic disease, mental health, etc Big, Really Big Data
Genomics
Imaging studies (MR, pathology)
Social Determinants
Exposome
Proteome
Microbiome genetics
Wearables and implantables Discussion
What EHRs are you using? (Poll) eCW? Epic? athenahealth? Allscripts? Amazing Charts? Greenway? NextGen? Cerner? Other?
How are you leveraging data for quality improvement in your practice?
Research studies using your practice data from EHR?
What is keeping you from using your data more effectively?