Data-driven medicine: Actionable insights from patient data Session #2, February 20, 2017 Turner Billingsley, MD, CMO, InterSystems Randy Pallotta, Manager, InterSystems Charlie Harp, CEO, Clinical Architecture 1
Speaker Introduction M. Turner Billingsley, MD, FACEP Chief Medical Officer, InterSystems Charlie Harp CEO, Clinical Architecture Randy Pallotta Manager End User Healthcare Sales Engineering, InterSystems 2
Conflict of Interest M. Turner Billingsley, MD, FACEP Charlie Harp Randy Pallotta Have no real or apparent conflicts of interest to report. 3
Agenda Longitudinal Patient Health and Care Record Ontologies Innovative Point of Care Provider Tools, Actionable Insight Data Normalization Inferencing and Logical Reasoning Pilot Overview Q & A 4
Learning Objectives Describe how presenting medically relevant information in an innovative CliniGraphic is of high value to providers Discuss how state-of-the art inferencing technology can synthesize complex & disparate patient information State how a large HCO identified 4800+ previously unrecognized high risk patient conditions in 6 months Discuss the ways connected health records can enhance care delivery, improve patient outcomes, and manage population health and risk more effectively Identify the role both structured and unstructured data can take in providing clues for completing and correcting patient information 5
STEPS Value Category HCO with an advanced integrated medical record used clinical inferencing technology to reason over medical records to: Advance clinical awareness Identify 95 patients with undocumented high risk conditions on day one 6
Data Driven Medicine Today well beyond tipping point of EHR installation Challenge: risk of data overload How do we Get to what matters? Extract and deliver value from the electronic health records and systems? Keep promise to clinicians - it will be worth it 7
Senior center Prison Social service agency Hospital Pharmacy Government Rehab Child protective services Family Laboratory Payer Home care agency Physician Researcher Nursing home 8 Ambulance Medical school Pharma/device company
Data Driven Medicine: Data, Data and More Data Disparate data sources Structured and unstructured data Information overload vs. What am I missing? Expanding access to patient records Clinicians must consider increasing volumes of data from clinical research Important information may be unstructured The volume of unstructured data present in most clinicbased systems is estimated at 80 percent and growing. Source: FY16 HIE inpractice Task Force (2016). Blending Structured and Unstructured Data to Develop Healthcare Insights. 9
Where can data be leveraged to make a difference? Providers and healthcare organizations need Right information At the right time In the right format Provide relevant knowledge at the point of care Improve patient care delivery, increase efficiency Meet organizational goals and regulatory requirements Support population health initiatives 10
How do we enable providers to achieve these goals? Make it part of their normal workflow Within a comprehensive care record Provide relevant, actionable insight and value Tell me something I didn t know / need to know 11
Partnership = Remarkable Results Large Health System 3 Hospitals & 1 Million+ Patients InterSystems HealthShare Information Exchange + + Clinical Architecture Symedical & Advanced Clinical Awareness Suite Six Months 12
Smarter Systems using Ontologies 13
Introducing Ontologies What is an ontology? An ontology is a collection of relationships specific to a domain For instance, we could have the following ontologies defined as subsets of the Type II Diabetes ontology: Type 2 Diabetes Medications Type 2 Diabetes Comorbidities Type II Diabetes Related Lab Results 14
Leveraging Ontologies Consolidated views of clinical data Building out clinical alerts (for gaps in care, missed procedures, vaccinations, labs, missing diagnoses, etc.) Send alert if patient is on a diabetes medication, has a high glucose OR a high A1C, and has at least one diabetes comorbidity As opposed to: Send alert if patient.medication contains ('12345', '54331', '4455'...) AND patient.labs contains ('556677','554433', '332211'...) and lab. result > 6 OR patient.labs contains ('83838','02020','20020',...) and patient.diagnoses contains ('83838', '92929', '01010',...) 15
Legacy Standards Connected Health Solutions Normalization Legacy Standards Emerging Standards Emerging Standards Proprietary formats Proprietary formats Unstructured Unstructured HealthShare with Embedded Extensible Data Model 16
Integration at Point-Of-Care 17
Clinical Inference: CHF 18
Clinical Inference Workflow: CHF 19
Consolidated Views of Patient Data Why is this important? With longitudinal patient record, we solve the missing data problem; how do we make it efficient for providers? Ontologies allow aggregated data - from multiple clinical/financial/claims sources - to be displayed in a way that is meaningful to clinicians - Normalized - Consolidated - De-duplicated With one-click, real-time in the provider workflow 20
CliniGraphic Available 21
CliniGraphic Presentation: CHF 22
Pilot CliniGraphics Four CliniGraphics currently deployed at a customer site: Hypertension COPD CHF Type II Diabetes More to follow: High Risk Pregnancy, Renal Failure, etc. 23
The Pilot 24
The Pilot InterSystems role Aggregate data across multiple clinical sources and messaging formats Uniquely identify each patient Provide a normalized composite health record for each patient at the point of care Apply patient consent policies Allow for secure clinical messaging Clinical Architecture s role Semantic Operating System Provide standard terminologies and ontologies Support interoperability and normalization Support unstructured text processing Support complex ontological reasoning (CliniGraphic) Support clinical inferencing 25
Advanced Clinical Awareness Leverage Encapsulated Knowledge to Improve Provider Awareness Summarize patient information relative to a particular condition Alert providers of potential issues Proactively look for gaps in patient information Improve outcomes with proactive quality interventions Identify patient cohorts for disease management Identify patient cohorts for clinical trial recruitment The potential is limitless 26
Chaos = Uncertainty Good advice requires good information Build the most complete picture of the patient as possible Aggregate information from all available sources Normalize structured data to reduce dissonance Make the most of unstructured data where necessary Summarize, remove noise and fills gaps in data where possible 27
Aggregation and Normalization PRACTICES First, build a solid information foundation Normalize The only people who see the whole picture are the ones who step outside the frame. Problems HOSPITALS Aggregate Meds Labs CLAIMS Patient LABS Subscribe Manage Interoperate Normalize 28 Observations
Unstructured Information Then scour all sources for critical insights Observation PRACTICES Code System: SNOMEDCT Code : 250908004 Term: Left Ventricle Ejection Fraction Result Value: 55 Result Unit: % 29
Ontological Reasoning Ontologies allow software to summarize data and understand how the different pieces relate to one another Congestive heart failure Acute congestive heart failure Problems Measured by rolls up to Congestive heart failure Losartan Meds Labs Ejection Fraction Treated with Patient Has comorbidity Observations Atrial Flutter 30
Ontological Reasoning 31
Logical Reasoning Problems Inferences leverage patient information, ontologies and logical reasoning to look for patterns of interest Has rollup IF Type 2 Diabetes Mellitus is not present Has rollup AND Meds Labs Hemoglobin A1c is greater than 7 % Patient Observations OR Sulfonylurea is present THEN! Patient may be an undocumented diabetic Member of Class 32
Logical Reasoning I agree clinically with the above concern. I DO NOT agree clinically with the above concern. I am aware of the concern and am monitoring the patient. I have not seen this patient before. This patient is no longer under my care. 33
Pilot: Staffing & Build Approach Team of three RN clinical informaticists - Research best practices for standards of care - Identification of rules, exceptions, logic flow First builds were completed by Clinical Architecture - Reviewed, tested, and validated by informaticists and QA team - Discovered a small number of false positives/negatives - Tuned the rules and algorithms Training for self service - 2 days training of clinical informaticists - Built 2 Conditions with CA supervision - Built the remaining independently 34
What We Learned in the Process Ontologies and logical reasoning must be localizable and portable Encapsulated reasoning must incorporate all relevant information, including unstructured text Encapsulated reasoning should support complex time and longitudinal reasoning Encapsulated reasoning must collect relevant evidence and dynamically build a narrative that support the assertion 35
Pilot Use Case - Results Three hospitals, ~100 clinics, 1 clinical lab, 14 diagnostic imaging groups Over a period of six months, involving 1 million+ patients Identified Patients Undocumented Diagnosis 5 Moderate COPD 6 Severe COPD 36 Hypertensive Disorder ~1300 Congestive Heart Failure ~3500 Diabetes Mellitus Type II 36
Summary / Wrap Up 37
Data-driven medicine: Actionable insight from patient data Uncover undiagnosed patient conditions / undocumented diagnoses Broaden the circle of knowledge Improve the information available to other care providers Expand the information available for population health efforts Quality improvement, gaps in care, etc. Disease registries Care coordination Avoid unintended consequences 27.8% US diabetics undiagnosed 1 $ ~cost of $2864/pp/yr. 2 1. National diabetes statistics report: estimates of diabetes and its burden in the United States, 2014 Atlanta: US Centers for Disease Control and Prevention; 2014. Available: https://www.cdc.gov/diabetes/pdfs/data/2014-report-estimates-of-diabetes-and-its-burden-in-theunited-states.pdf (accessed 2017 Jan. 25). 38 2. Zhang Y1, Dall TM, Mann SE, Chen Y, Martin J, Moore V, Baldwin A, Reidel VA, Quick WW. The economic costs of undiagnosed diabetes. Population Health Management. 2009 Apr;12(2):95-101.
Wrap Up Power in collaboration / partnership Longitudinal, extensible source - neutral community-wide health and care record Clinigraphic, Clinical Inference tools available real-time, in provider workflow Added value Providers - actionable, relevant information at point of care Organization - manage risk, PH strategies, quality initiatives, etc. 39
A Summary of How Benefits Were Realized for the Value of Health IT Clinical Inferencing and the CliniGraphic address all 5 STEPS 40
Contact Info / Questions Turner Billingsley, MD turner.billingsley@intersystems.com @InterSystems https://www.linkedin.com/company/intersystems Charlie Harp charlie_harp@clinicalarchitecture.com @ClinicalArch Randy Pallotta randy.pallotta@intersystems.com Please complete the online session evaluation Booth #1561 Booth #3171 41