Instrumenting the Healthcare Enterprise for Big Data Research. Shawn Murphy MD, Ph.D. Leicester, October

Transcription

1 Instrumenting the Healthcare Enterprise for Big Data Research Shawn Murphy MD, Ph.D. Leicester, October

2 Informatics for Integrating Biology and the Bedside (i2b2) Software for explicitly integrating person-oriented data to a way that is optimized for clinical genomics research Allows integration of clinical data, trials data, and genotypic data A portable and extensible application framework Software is built in a modular pattern that allows additions without disturbing core parts Available as open source at

3 An i2b2 Environment (the Hive) is built from i2b2 Cells A B C Hive of software services provided by i2b2 cells remote local observation_fact Patient_Num Encounter_Num Concept_CD Observer_CD Start_Date Modifier_CD Instance_Num End_Date ValType_CD TVal_Char NVal_Num ValueFlag_CD Observation_Blob visit_dimension Encounter_Num Start_Date End_Date Active_Status_CD Location_CD* patient_dimension Patient_Num Birth_Date Death_Date Vital_Status_CD Age_Num* Gender_CD* Race_CD* Ethnicity_CD* concept_dimension Concept_Path Concept_CD Name_Char observer_dimension Observer_Path Observer_CD Name_Char 1 1 modifier_dimension Modifier_Path Modifier_CD Name_Char observation_fact Patient_Num Encounter_Num Concept_CD Observer_CD Start_Date Modifier_CD Instance_Num End_Date ValType_CD TVal_Char NVal_Num ValueFlag_CD Observation_Blob visit_dimension Encounter_Num Start_Date End_Date Active_Status_CD Location_CD* patient_dimension Patient_Num Birth_Date Death_Date Vital_Status_CD Age_Num* Gender_CD* Race_CD* Ethnicity_CD* concept_dimension Concept_Path Concept_CD Name_Char observer_dimension Observer_Path Observer_CD Name_Char 1 1 modifier_dimension Modifier_Path Modifier_CD Name_Char observation_fact Patient_Num Encounter_Num Concept_CD Observer_CD Start_Date Modifier_CD Instance_Num End_Date ValType_CD TVal_Char NVal_Num ValueFlag_CD Observation_Blob visit_dimension Encounter_Num Start_Date End_Date Active_Status_CD Location_CD* patient_dimension Patient_Num Birth_Date Death_Date Vital_Status_CD Age_Num* Gender_CD* Race_CD* Ethnicity_CD* concept_dimension Concept_Path Concept_CD Name_Char observer_dimension Observer_Path Observer_CD Name_Char 1 1 modifier_dimension Modifier_Path Modifier_CD Name_Char observation_fact Patient_Num Encounter_Num Concept_CD Observer_CD Start_Date Modifier_CD Instance_Num End_Date ValType_CD TVal_Char NVal_Num ValueFlag_CD Observation_Blob visit_dimension Encounter_Num Start_Date End_Date Active_Status_CD Location_CD* patient_dimension Patient_Num Birth_Date Death_Date Vital_Status_CD Age_Num* Gender_CD* Race_CD* Ethnicity_CD* concept_dimension Concept_Path Concept_CD Name_Char observer_dimension Observer_Path Observer_CD Name_Char 1 1 modifier_dimension Modifier_Path Modifier_CD Name_Char Data Model Data Repository Cell

4 An i2b2 Enterprise is built from i2b2 Hives Enterprise Research Patient Data Registry visit_dimension patient_dimension Encounter_Num 1 Patient_Num observation_fact 1 Start_Date Patient_Num Birth_Date End_Date Encounter_Num Death_Date Concept_CD Active_Status_CD Vital_Status_CD Location_CD* Observer_CD Age_Num* Start_Date Gender_CD* Modifier_CD Race_CD* observer_dimension Instance_Num Ethnicity_CD* Observer_Path End_Date ValType_CD Observer_CD TVal_Char Name_Char NVal_Num concept_dimension ValueFlag_CD Concept_Path Observation_Blob modifier_dimension Modifier_Path Concept_CD Name_Char Modifier_CD Name_Char

5 Research Patient Data Registry (RPDR) at Partners Healthcare to find patient cohorts for clinical research 1) Queries for aggregate patient numbers - Warehouse of in & outpatient clinical data million Partners Healthcare patients billion diagnoses, medications, procedures, laboratories, & physical findings coupled to demographic & visit data - Authorized use by faculty status - Clinicians can construct complex queries - Queries cannot identify individuals, internally can produce identifiers for (2) Query construction in web tool Encrypted identifiers Z731984X Z74902XX Deidentified Data Warehouse 2) Returns detailed patient data - Start with list of specific patients, usually from (1) - Authorized use by IRB Protocol - Returns contact and PCP information, demographics, providers, visits, diagnoses, medications, procedures, laboratories, microbiology, reports (discharge, LMR, operative, radiology, pathology, cardiology, pulmonary, endoscopy), and images into a Microsoft Access database and text files OR Real identifiers

6 2014 s usage of RPDR > 5100 registered users over 12 years, 655 new in teams/year gathering data for research studies 2634 detailed patient data sets returned to these teams in 2014, containing data of 24.7 million patient records. From a survey of 153 teams Importance of the data received from the RPDR was evaluated in relation to the study it was supporting. $ million total research support critically dependent on RPDR from patient data received throughout life of funding. Usefulness of Detailed Data 106 Total Responses Not Useful 15% Useful 42% Critical 43% ~300 data marts were created to support hospital operations, representing about 80 million patient records

7 Theory of Kimball translated to Healthcare Data Concept DIMENSION concept_key concept_text search_hierarchy Encounter DIMENSION encounter_key encounter_date hospital_of_service Star schema Patient-Concept FACTS patient_key concept_key start_date end_date practitioner_key encounter_key value_type numeric_value textual_value abnormal_flag Patient DIMENSION patient_key patient_id (encrypted) sex age birth_date race deceased ZIP Pract. DIMENSION practitioner_key name service Binary Tree start search 2200 million

8 Adapting to Big Data Fact Table Works Well for Heath Data Integration! BUT People need to agree to give you their data When database gets very large Can be difficult to update and not fragment index Expensive to pay for database and high performance server Innovation tends to suffer when need to integrate into a large mission-critical system

9 An i2b2 Enterprise is built from i2b2 Hives Enterprise of different i2b2 Hives observation_fact Patient_Num Encounter_Num Concept_CD Observer_CD Start_Date Modifier_CD Instance_Num End_Date ValType_CD TVal_Char NVal_Num ValueFlag_CD Observation_Blob visit_dimension Encounter_Num Start_Date End_Date Active_Status_CD Location_CD* patient_dimension Patient_Num Birth_Date Death_Date Vital_Status_CD Age_Num* Gender_CD* Race_CD* Ethnicity_CD* concept_dimension Concept_Path Concept_CD Name_Char observer_dimension Observer_Path Observer_CD Name_Char 1 1 modifier_dimension Modifier_Path Modifier_CD Name_Char observation_fact Patient_Num Encounter_Num Concept_CD Observer_CD Start_Date Modifier_CD Instance_Num End_Date ValType_CD TVal_Char NVal_Num ValueFlag_CD Observation_Blob visit_dimension Encounter_Num Start_Date End_Date Active_Status_CD Location_CD* patient_dimension Patient_Num Birth_Date Death_Date Vital_Status_CD Age_Num* Gender_CD* Race_CD* Ethnicity_CD* concept_dimension Concept_Path Concept_CD Name_Char observer_dimension Observer_Path Observer_CD Name_Char 1 1 modifier_dimension Modifier_Path Modifier_CD Name_Char A observation_fact Patient_Num Encounter_Num Concept_CD Observer_CD Start_Date Modifier_CD Instance_Num End_Date ValType_CD TVal_Char NVal_Num ValueFlag_CD Observation_Blob visit_dimension Encounter_Num Start_Date End_Date Active_Status_CD Location_CD* patient_dimension Patient_Num Birth_Date Death_Date Vital_Status_CD Age_Num* Gender_CD* Race_CD* Ethnicity_CD* concept_dimension Concept_Path Concept_CD Name_Char observer_dimension Observer_Path Observer_CD Name_Char 1 1 modifier_dimension Modifier_Path Modifier_CD Name_Char observation_fact Patient_Num Encounter_Num Concept_CD Observer_CD Start_Date Modifier_CD Instance_Num End_Date ValType_CD TVal_Char NVal_Num ValueFlag_CD Observation_Blob visit_dimension Encounter_Num Start_Date End_Date Active_Status_CD Location_CD* patient_dimension Patient_Num Birth_Date Death_Date Vital_Status_CD Age_Num* Gender_CD* Race_CD* Ethnicity_CD* concept_dimension Concept_Path Concept_CD Name_Char observer_dimension Observer_Path Observer_CD Name_Char 1 1 modifier_dimension Modifier_Path Modifier_CD Name_Char

10 Inside a Parent PICI Ontology Patient Resolution P1 H1 H2 H3 P2 H1 H2? Pa Pb QUERY P3 H1? H3 Pc SUBQUERIES Query Distribution Engine

11 Flow of Distributed Query Parent Child Child

12 Creating Enterprise Wide Query and Analysis System for Big Data RPDR Redcap Survey Data Public Health (CMS) i2b2 Navigator Imaging DICOM Partners Biobank Genomics Sets Notes / Text Repository Big Data Commons Integrates disparate islands of patient data (clinical and research data) onto a Common Big Data Platform 12

13 RPDR Search for Diabetes, A1C, and SGLT2

14 RPDR Enhance w/ LMR Note Search

15 RPDR Additional Patients Returned

16 I2b2/RPDR Text-based Queries Transparent to users - add text based searching to query tool Researchers could search for phrases of interest in a variety of clinical notes housed in a separate notes repository Cardiology, Discharge Summaries, Endoscopy, LMR, Operation, Pathology, Pulmonary, and Radiology Notes are updated nightly in a separate notes repository Several steps taken to conceal PHI through query Researchers will not be able to query on numbers of any kind Names are excluded from text index making them unsearchable Search terms are also matched against a PHI database Separate database is securely maintained with latest patient PHI, including names, addresses, numbers, nicknames, etc. Search terms are queried in Notes Repository and PHI database Patients that are returned from both databases are excluded from results Final aggregate counts are obfuscated to hide true totals and limit ability to identify one individual

17 Creating Enterprise Wide Query and Analysis System for Big Data RPDR Redcap Survey Data Public Health (CMS) i2b2 Navigator Imaging DICOM Partners Biobank Genomics Sets Notes / Text Repository Big Data Commons Integrates disparate islands of patient data (clinical and research data)onto a Common Big Data Platform 17

18 The Partners Biobank Samples Consent The Partners Biobank provides samples (plasma, serum, and DNA) collected from consented patients. Data Research Discoveries 25,000 patients have consented to date! Samples are available for distribution to Partners investigators* to help identify novel Personalized Medicine opportunities that reduce cost and provide better care *with required approval from the Partners Institutional Review Board (IRB). Improved Clinical Care for All Patients

19 Biobank Integrative Genomics Strategy Partners BioBank Samples (Whole Blood Extracted DNA/RNA) Genotyping Transcriptome Epigenome Profiling Illumina MEGArray: Multi-Ethnic GWAS/Exome SNP Array Whole Transcriptome Analysis: RNA-seq Methylation Analysis: HumanM450K Array Array Cost: $59/ sample Array Cost: $40-50/sample Array Cost: $150/sample Genome/Transcriptome Analysis: ~$100/sample Genome/Transcriptome/Epigenome Analysis: ~$260/sample

20 Data Integration Phenotype Discovery Center Electronic Medical Record (EMR) Data Coded Data Demographics Diagnoses Lab Results Additional Data Other Research Data Survey Data Samples RPDR Medications Procedures Etc. Biobank Data Text Data (Notes/Reports) Physician Notes Imaging Reports Pathology Reports Etc. Genetic Data (coming in 2015) GWAS Consent Biobank Portal Utilities Validated Phenotypes Type II Diabetes Coronary Artery Disease Congestive Heart Failure Rheumatoid Arthritis Calculated Controls (Charlson Index) Data Visualization Annotation IBD Multiple Sclerosis Bipolar Disorder Data Queries Extract Data Natural Language Processing Data and Sample Requests Links all the data: Sample, consent, EMR, GWAS, survey, phenotypes, controls DNA Serum Plasma Recontact Consent Status Research 2

21 Curating a Disease Algorithm 1. Create a gold standard training set. 3. Develop the classification algorithm. Using the data analysis file and the training set from step 1, assess the frequency of each variable. Remove variables with low prevalence. Apply adaptive LASSO penalized logistic regression to identify highly predictive variables for the algorithm 2. Create a comprehensive list of features (concepts/variables) that describe the phenotype of interest 4. Apply the algorithm to all subjects in the superset and assign each subject a probability of having the phenotype 95% Specificity 21

22 Use Phenotyping Algorithms to define cohorts of treatmentresistant and treatmentresponsive depression Initially: AUC = 0.54 Finally: AUC = 0.87 Depressed at Encounter Well at Encounter Initially: AUC = 0.55 Finally: AUC = 0.86 Clinical Status Model Specificity Sensitivity Precision AUC Depressed Billing Codes (0.03) 0.57 (0.14) 0.54 (0.02) Depressed NLP (0.05) 0.78 (0.02) 0.88 (0.02) Depressed NLP + Billing Codes (0.06) 0.78 (0.02) 0.87 (0.02) Well Billing Codes (0.02) 0.26 (0.27) 0.55 (0.03) Well NLP (0.06) 0.86 (0.02) 0.85 (0.02) Well NLP + Billing Codes (0.07) 0.85 (0.02) 0.86 (0.02)

23 Biobank Portal Curated Diseases Validated Phenotype Count* Predictive Positive Value Bipolar Disease 71 89% Congestive Heart Failure % Coronary Artery Disease 2,420 97% Crohn s Disease % Multiple Sclerosis 94 90% Rheumatoid Arthritis % Type 2 Diabetes Mellitus 1,887 97% Ulcerative Colitis % Healthy Controls based on Charlson Index * Based on 15,880 patients ** Based on 21,300 patients Count** 0 10-year survival probability is >98.3% 2, year survival probability is >95.87% 4, year survival probability is >90.15% 6,545

24 Partners Biobank Portal Login Screen

25 Partners Biobank Portal Splash Screen

26 Biobank Portal Run a Query Step 1: Run a query in the Biobank Portal Two criteria: a. The patient is a smoker (from the Health Information Survey) b. The patient agrees to be re-contacted

27 Biobank Portal Request Samples Based on Prior Query Step 2: Request Samples in the Biobank Portal based on the prior query

28 Partners Biobank Portal Defining a Query

29 Partners Biobank Portal Viewing Query

30 Partners Biobank Portal Running a Query

31 Partners Biobank Portal Download De-Identified Data

32

33 Partners Biobank Portal Specify Data for Download

34 Download table and select patients

35 Partners Biobank Portal Request Genetic Data

36 Navigating NGS Variant Data with Sequence Ontology Combination of concepts and modifiers to identify: An SNV/SNP located on a 3 UTR An SNV/SNP associated with a certain gene An SNV/SNP of specified zygosity

37 DEMO 37

38 38 Log on to Parent

39 39 Parent show all available ontologies

40 40 Formulate Query as usual in Parent

41 Parent combines child patient sets Child #1 has Diagnoses Child #2 has Lab tests Answers is Intersection 41

42 42

43 43 Child # 2 has Lab tests (and meds/demog)

44 44 Query to Child #2 returned 2 patients

45 45

46 Child #1 has Diagnoses (and proc/demog) Query to Child #1 returned 89 patients 46

47 I2b2 Community Software distributed as open source

48 Patient Data Registries Innovative Tools & IT solutions Research Partners Biobank Quality Patient Care

49 Innovation Platform Secure computational platform for Big Data Providing analytic tools to gather and compute upon big data within a secure virtual machine Working with EMC to provide storage, computation, and virtualization Enabling growth using secure cloud services Connecting to clinical care For effective recruitment of patients to clinical trials For providing innovative insights into clinical care using big data resources at Partners

50 Innovation Platform Linking to EMR with I2b2 Apps Published 2011

51 Substitutable Medical Apps and Reusable Technologies SMART REST SMART Connect

52

53 SMART Apps link i2b2 to the EMR Substitutable Medical Application and Reusable Technology Commissioned form the Office of the National Coordinator Allows Big Data from i2b2 to integrate Clinical Apps into the Epic EMR. Paradigm is similar to Mobile Apps with a proposed standard interface using FHIR sponsored by the National Argonaut EMR Project.

54 SMART-i2b2 Creates the Clinical Innovation Platform Connecting Partners Big Data Commons with the practice of Precision Medicine Navigates complex issues with Intellectual Property and integration into Epic Standard, highly regulated Application Program Interface in HL7 FHIR (Fast Healthcare Interoperability Resources) to be provided for clinical applications to dip into Partners Big Data Commons with potential to scale to National level Pre-built plug-in architecture for Apps that are built to SMART specifications will allow providers in EMR to receive Decision Support from Big Data Commons

55 Treating Cancer Patient with Precision Medicine Partners Big Data Commons

56 Enabling Innovation to reach into EMR SMART Apps run in EMR Got Statins? BP Centiles Cancer Risk SMART-Enabled DW - I2b2/RPDR Partners Big Data Commons SMART-Enabled Document/Pivotal SMART-Enabled EMR - Epic Research Applications run directly in Data Lake

57 Tribute to I2b2 Core Team Isaac Kohane Susanne Churchill Shawn Murphy Griffin Weber Paul Avillach Michael Mendis Lori Phillips Janice Donahoe Nich Wattanasin David Wang Christopher Herrick Bill Wang Vivian Gainer Andrew Cagan SMART i2b2 Team Nich Wattanasin Kenneth Mandl Joshua Mandel i2b2 SHRINE Team William Simons Douglas MacFadden I2b2/tranSMART Team Paul Avillach Michael McDuffy Jeremy Easton-Marks mi2b2 Team Randy Gollub Christopher Herrick Bill Wang

58 I2b2, SHRINE, and SMART Information and Software on the Web i2b2 Homepage ( i2b2 Software ( i2b2 Community Site ( SHRINE at Harvard ( SHRINE Software: ( SMART Platforms Homepage (