Access and Use of EHR Data for CTR Research - PowerPoint Presentation

Access and Use of EHR Data for CTR Research Scott Campbell, PhD, MBA Associate Professor, Pathology and Microbiology Sr. Director – Research Information Technologies October 23, 2019

Acknowledgements James McClay, MD James Campbell, MD Bryant England, MD Jay Pedersen, MS Yeshwanth Narayana, MS Dr. W. Scott Campbell, Dr. James R. Campbell, Dr. James C. McClay partially supported by NIH Award: 1U01HG009455-01; Patient Centered Outcomes Research Institute (PCORI) Award CDRN-1306-04631); Funding from UNMC Departments of Pathology and Microbiology and Internal Medicine

Questions that need answers Find all patients diagnosed for lower GI adenocarcinoma Find all patients with Chronic Kidney Disease subsequently diagnosed with cancer. What is incidence rate at UNMC compared to state and national rates of cancer by type Find all cases of lower GI adenocarcinoma with MMR IHC results, MSI testing and any *RAS mutation by pathogenicity. Did these patients receive any gene targeted therapy AND did it followed FDA guidelinesFind all positive opioid screens in urine or blood performed in ED and “repeat” patient encountersFind all viral infections in post-transplanted patients, identify frequency by organism, associated treatments and subsequent outcomes Answers contained in EHR

Primary Data – Bridge the Gap Translational Junction Bedside Bench

EHR Data Lake? Use of clinical terminologies (standards) in a standard way: useful EHR data lakes C haracterized (annotated) data is cleaner data Dirty data needs to be cleaned Not all data is clean or “good”

Where is the EHR data? EPIC Clarity EPIC Caboodle EPIC Slicer Dicer CRANE (Clinical Research Analysis Environment) EHR Access Core

EPIC Clarity Epic is a transactional database (hierarchical database) Fast at point of care Slow for data analysis EPIC Chronicles EPIC Clarity Clarity is built relational database (MS SQL) Good for data analysis, usually Data dump of all Epic information No classification of information

EPIC Caboodle Clarity extract, relational database Categorized Data, not in clinical terms External data sources Enterprise Data Warehouse (EDW) Accessible by EDW teamEPIC Chronicles EPIC Clarity EPIC Caboodle

Slicer Dicer Epic ad hoc query tool Identified and de-identified Requires Epic Access A provider’s patients only for identified search Business orientationUnderlying data model not clinically oriented

EHR Data Access Core EHR Access Core Interacts with Clarity Must wrangle data from Clarity Requires knowledge of Epic, Epic workflow and clinical context More data than CRANENo data annotation/characterization (Clarity issue NOT core’s)CRANE is NOT the EHR Access Core but can work hand-in-glove Starting point for any future CRANE data augmentation

CRANE Clarity extract, relational database Categorized Data using national and international medical data standards External data sources (NAACCR; SSDI) Identified and de-identified versions Many Epic data elements but not all. Based on user-driven use casesConnected to PCORI Common Data Model and GPC ConsortiumSelf-serve and supported service EPIC Chronicles EPIC Clarity CRANE

CRANE architecture

I2B2 Query & Analysis Tool

CRANE Contents Main Categories – Organized by standards-based metadata

The focus on standards EHR data created and stored based on the workflow of the care provider Stored as a “value” in a data cell or location in the database. The data cell has an identifier but nothing more Workflow provides critical context about the stored “value” and how to interpret the value Standards are used to provide the context of the “value” Standards enable consistent, broad-based understanding of the “value” and how to interpret it. Standards allow grouping of similar “values” for different patients regardless of EHR database design – All EHR’s data locations are different…even when the vendor is the same!

Nebraska Medicine i2b2 Data Architecture I2b2 Information Class Standard Metadata Ontology ADT historyEpic FacilityCancer registryICD-O Clinical measurements Social history LOINC; SNOMED CT Demographics LOINC; SNOMED CT Diagnoses (Encounter dx; Problems; Past Med History) (ICD-9-CM); ICD-10-CM; SNOMED CT Encounters Epic Facility; Encounter classes Laboratory results LOINC, SNOMED CT (pathology) Medications (Orders and Rx; Dispense records) RXNORM; NDC Procedures (Professional services; Hospital procedures; Procedure history ) CPT; (ICD-9-CM); ICD-10-PCS; HCPCS; SNOMED CT

Example Value = 80 Epic Cell# = 1234 Blood pressure Diastolic Sitting Outpatient Each data point in Epic must be contextualized and labeled for consistent and correct interpretation and discovery

Understanding the data What is the data model? Star-schema How is the data “characterized” Standards What am I really looking for?The phenotype

I2b2 Star schema: One fact per record (= One question + answer) i2b2 Observation fact Observation Fact Instance_num Modifier_CD Concept_CD VALTYPE_CD UNITS_CD TVAL_CHAR NVAL_NUM OBSERVATION_BLOB Patient Encounter Provider Start_date End_date

What is the patient hemoglobin? “13.2 mg/dl” i2b2 Observation fact Observation Fact Modifier_CD Concept_CD LOINC:2951-2 Patient Encounter Provider 11/1/2016 7:30AM End_date 13.2 Mg/dl

What is the patient problem? “Breast cancer” i2b2 Observation fact Observation Fact Modifier_CD DX |PROB\ACTIVE Concept_CD SNOMEDCT:254837009 (Malignant tumor of breast) Patient Encounter Provider Start_date End_date

Example: Phenotype of Metastatic Colorectal Cancer Possible sources of truth: NAACCR registry; Problem List; Encounter Dx ;CEA levels > 25   True Positives False Negatives NAACCR Stage IV 45 N/A SNOMED CT Problem List 50 25 ICD-10-CM Encounter 27 38 CEA level > 25 49 26 Chart review cases excluded and computational data supporting inclusion     Stage SNOMED CT ICD-10-CM CEA > 25 Chart Review Correction           Case 1     X     Case 2     X X   Case 3       X   Case 4 X X     Nodal Involvements           Case 5   X X     Case 6     X     Case 7   X   X   Case 8     X   Anal Cancer           Case 9       X Myxoma           Case 10 X     X Cases with Single Computational Item suporting inclusion by reason Number Stage IV 2 SNOMED CT Problem 3 ICD-10-CM Encounter 7 CEA level > 25 5 Understanding of data to support the use case. Not as simple as “what is the ICD10 code”

Use Case – Cohort Identification Delivery of a retrospective data feed for 1/1/2017 – 05/31/2019 of raw, de-identified EMR data Distinct LBCL/DLBCL cancer patients, 18 years of age and over, will be identified using the following IDC 10 Codes: LBCL/DLBCL ICD 10 Codes: C82.00-C82.69C82.80-C82.99C83.30-C83.39C83.50-C83.59C85.20-C85.29Patients diagnosed with LBCL/DLBCL that have received Lenalidomide (Revlimid) treatment in three lines of therapy or less will be broken out by year.Where possible, at least 1 of the 3 treatment lines should be an anti-CD20 containing therapy (i.e. Rituximab, Rituxan).

Patient Count with Diagnosis

Adult Patients with Dx

First Treatment Criteria

Second Treatment Criteria

Data Beyond Patient Counts Patient count associated with patient list Patient list uses obfuscated patient identifier Data elements (facts) extracted for patients in list Series of .csv files for import into analytics software R, Tableu, SAS PATIENT_NUM CONCEPT_CD PARAMETER RESULT UNITS START_DATE END_DATE 2054 LOINC:1975-2 Serum Bilirubin 0.6 mg/dL [2019/03/11:03:51:00 AM] [2019/03/11:06:47:00 AM] 2054 LOINC:1975-2 Serum Bilirubin 1.2 mg/dL [2019/03/12:09:40:00 AM] [2019/03/12:10:21:00 AM] 2054 LOINC:2160-0 Serum Creatinine 0.83 mg/dL [2019/03/08:03:41:00 AM] [2019/03/08:05:08:00 AM] 2054 LOINC:2160-0 Serum Creatinine 1.07 mg/dL [2019/03/09:06:31:00 AM] [2019/03/09:07:13:00 AM] 2054 LOINC:2160-0 Serum Creatinine 1.09 mg/dL [2019/03/09:09:53:00 AM] [2019/03/09:10:18:00 AM] 2054 LOINC:2160-0 Serum Creatinine 1.08 mg/dL [2019/03/09:02:02:00 PM] [2019/03/09:02:34:00 PM] 2054 LOINC:2160-0 Serum Creatinine 1.22 mg/dL [2019/03/10:04:36:00 AM] [2019/03/10:05:10:00 AM] 2054 LOINC:2160-0 Serum Creatinine 1.3 mg/dL [2019/03/10:09:48:00 PM] [2019/03/10:10:17:00 PM] 2054 LOINC:2160-0 Serum Creatinine 1.18 mg/dL [2019/03/11:03:51:00 AM] [2019/03/11:04:32:00 AM] 2054 LOINC:2160-0 Serum Creatinine 0.95 mg/dL [2019/03/11:10:27:00 AM] [2019/03/11:11:19:00 AM] 2054 LOINC:2160-0 Serum Creatinine 0.77 mg/dL [2019/03/11:03:11:00 PM] [2019/03/11:04:45:00 PM] 2054 LOINC:2160-0 Serum Creatinine 0.71 mg/dL [2019/03/11:06:36:00 PM] [2019/03/11:07:23:00 PM] 2054 LOINC:2160-0 Serum Creatinine 0.67 mg/dL [2019/03/12:12:11:00 AM] [2019/03/12:12:46:00 AM] 2054 LOINC:2160-0 Serum Creatinine 0.58 mg/dL [2019/03/12:05:51:00 AM] [2019/03/12:06:35:00 AM] 2054 LOINC:2160-0 Serum Creatinine 0.57 mg/dL [2019/03/12:09:40:00 AM] [2019/03/12:10:21:00 AM]

How to access Access to query tool requires UNMC data use agreement (standard for UNMC employees) Username/Password – Dr. Jim McClay ( jmmclay@unmc.edu ) De-identified, patient counts only De-identified data elements extractable by super-users and CRANE analystsQuery number used by analysts for patient list Data can be re-identified IF IRB approval granted and verified

Learning, Support and Coming Attractions Super user training Ability to access and extract data from CRANE Consent from Chair to support users in department Use case project for experiential learning Formal course for credit (in development)Documentation (in development)User guide Data dictionary Intake coordinator/contacts Jarrod Anzalone ( alfred.anzalone@unmc.edu) Yuning Zhang ( yuning.zhang2@unmc.edu) On-going development to self-serve tooling

Questions