320 likes | 450 Views
Document S tandards Faced and Lessons Learned. Calvin Beebe, Tom Oniki , Hongfang Liu, Kyle Marchant. Data Normalization Process. Adopt a hybrid agile process: combining both top-down and bottom-up approaches Identify gaps through normalizing real EMR data
E N D
Document Standards Faced and Lessons Learned Calvin Beebe, Tom Oniki, Hongfang Liu, Kyle Marchant
Data Normalization Process Adopt a hybrid agile process: combining both top-down and bottom-up approaches • Identify gaps through normalizing real EMR data • Modify components to close the gaps • Evaluate the CEM results • Iteratively improve
Overview • How the real EMR data looks like • by Calvin Beebe • Modeling - Lessons learned • by Tom Oniki • Process – Lessons learned • by Hongfang Liu • Database – Lessons learned • by Kyle Marchant
Overview • How the real EMR data looks like • by Calvin Beebe • Modeling - Lessons learned • by Tom Oniki • Process – Lessons learned • by Hongfang Liu • Database – Lessons learned • by Kyle Marchant
Source Inputs • HL7 V2.x - Pharmacy Orders • CDA R1 - Clinical Documents • CCD R1 - Continuity of Care Documents
HL7 2.x Encoding Rules • Message formats prescribed in the HL7 encoding rules consist of data fields that are of variable length and separated by a field separator character. • Rules describe how the various data types are encoded within a field and when an individual field may be repeated. • The data fields are combined into logical grouping called segments. • Each segment begins with a three-character literal value that identifies it within the message. • Segments may be defined as required or optional and may be permitted to repeat. • Individual data fields are found in the messages by their position within their associated segments.
HL7 2.x Pharmacy Encoded Order MSH|^~\&||116|||201105181535||RDE^O01|20110518153507018|P|2.2 PID|||87654321^^^^MPIID||PATIENT^OUR^TEST||1955-11-26.00:00:00|M||W|200 First St.^^Anytown^MN^55905||||||| PV1||E|DSCH^DSCH|2||4321|6789^Doctor^Best^D.|8888^PHYSICIAN^NOT^CHOSEN|^^^~^^^~^^^~^^^|ERS||||1|3|N|6789^Doctor^Best^D.|I|87654321|FRANKLIN BLUE CROSS||||||||||||||||3|||AV|||||201105032316|201105081600 |||||||5432^DOCTOR^NEXT^J ORC|XO|5|||CURRENT||||201105181535|EHRX-987654321|987654321 |6789^Doctor^Best^D.^|MS~$V752 RXE|^BID &0800,1730^^201105181730|070008003001^METFORMIN(GLUCOPHAGE), TABLET (1000 MG)|1000||MG|TABLET||||0||||987654321|||||||||||||^HOLD X 48 HOURS AFTER CT 3/18 RXR|ORAL ZHX||||||||||||||||||O|201105181530 ZRX|||0|||||0 Z- Segments are site-specific. Sample
Segments in Sample Reference Chapters 2, 3 & 4 in the HL7 Version 2.x Standards
Patient Context in CDA Documents Both CDA R1 & R2 support patient identification within the header of the document.
CDA Narrative Documents CDA narrative documents, support structured headers, with section narrative content. • As can be seen below the Medications Section is based on a simple HTML like syntax.
CDA Narrative Content • Both CDA R1 & CDA R2 support and actually require narrative content. • Both support Section codes, which can be used to identify sections of interest. • CDA R1 only contain narrative, while CDA R2 Documents may contain clinical statements based upon RIM Classes. • For those documents with narrative sections, content normalization requires the used of Natural Language Processing.
CCD Document – Medication Entry Template IDs specifying constraints
CCD Document – Medication Entry continued Supply reference with quantity dispensed.
Summary • Various sources were utilized to obtain Medication information. • Legacy application still leverage HL7 2.x messages to convey order information between systems. • CDA narrative and structured documents are coming on-line which convey snapshots of patient’s current state and medications.
Overview • How the real EMR data looks like • by Calvin Beebe • Modeling - Lessons learned • by Tom Oniki • Process – Lessons learned • by Hongfang Liu • Database – Lessons learned • by Kyle Marchant
Lessons Learned - Modeling • Open tools would be a great contribution to the interoperability. Examples: • mapping terminology, e.g., local codes to LOINC/HL7/SNOMED • mapping models, e.g., HL7 messages/CDA documents to CEMs, CEMs to ADL, etc. • generating sample instances • communicating information • browsers • generating documentation
Lessons Learned - Modeling • Documentation is essential – we didn’t produce enough of it. But . . .
Lessons Learned - Modeling • Documentation is essential – we didn’t produce enough of it. But . . . • It’s hard to communicate (verbally or in written word) the intricacies and complexities needed to make an effort like this work (much less keep information up-to-date)
Lessons Learned - Modeling • “One model fits all” won’t work • Clinical Trials (e.g., CDISC CSHARE) vs Secondary Use (e.g., SHARPn) • Proprietary EMR (e.g., GE Qualibria) and Open Secondary Use (e.g., SHARPn) • value set differences
Lessons Learned - Modeling • The root of all modeling questions: Precoordination vs. postcoordination and what to store in the model instance vs. leave in the terminology • Clinical drug vs. drug name/form/strength/ route • LOINC code vs lab test/method • Display names • Drug classes
Overview • How the real EMR data looks like • by Calvin Beebe • Modeling - Lessons learned • by Tom Oniki • Process – Lessons learned • by Hongfang Liu • Database – Lessons learned • by Kyle Marchant
Lessons Learned - Process • The design of the pipeline needs to be flexible enough to accommodate all kinds of changes – (agile) • UIMA is a nice architecture • Configurable • Model-driven • e.g., taking an XSD specification of CEM and translating into UIMA types • Seamless integration with NLP pipeline
Lessons Learned - Process • Diverse input formats • Structured - semantics may be different from different institutions • Needs to understand the data • Unstructured - there is a gap between the semantics of free text and the semantics of standards • Semantics in free text may be at coarse granularity level or can be paraphrased into different expressions
Lessons Learned - Process • Different requirements for different use cases - All normalization tasks but the necessary fields can be different for different use cases • Medication Rec (ClinicalDrug - critical) • Phenotyping (Gender, Race, AdministrativeDiagnosis, Lab, …)
Lessons Learned - Process • Too many standards to choose when implementing HL7 standards • Mapping from local codes to standard value sets – non-trivial • Versioning of standards is crucial • Do not assume the mapping will be trivial if the EMR data has already adopted the same standard as SHARPN value sets
Lessons Learned - Process • Different granularities between CEMs and original structures • Dose Strength “50-mg” • NotedDrug CEM: Unit=MG Value=50 • Inference • TakenDoseUpperLimit needs to be inferred from TakenDoseLowerLimit
Overview • How the real EMR data looks like • by Calvin Beebe • Modeling - Lessons learned • by Tom Oniki • Process – Lessons learned • by Hongfang Liu • Database – Lessons learned • by Kyle Marchant
Database – CEM to DB mapping • Relational structure for the Demographics data worked well • This provided a nice view into the patients data without having to have a lot of knowledge of the Patient CEM structure. • Allowed for both adds and updates • Was not intended to be a full MPI but does meet the minimal need of linking a given patients records for a given institution.
Database – CEM to DB mapping • XML Sample data for the Clinical CEMs • XML samples proved very valuable for validation against the XSD's and for providing an initial set of test messages to Channels. • The more complete these records were the more useful. • Assumed all mappings to code sets were done prior to receiving on the CEM to DB channel.
Database – CEM to DB mapping • Code re-use across the Clinical CEM Channels proved very useful • Standardized CEM DB Structure - IndexData, SourceData, and PatientData • Common patient “matching” code used • Currently only supports Add but Updates now possible due to SourceSystemID that was added recently
Database – CEM to DB mapping • Issues and Challenges • Date formatting was one example of needing to understand how the data was being received and used. • Field level storage VS storing of full XML - Tradeoff - Decided to always store full XML – May need to look at additional relational fields on clinical CEMs for better searching support
Database – CEM to DB mapping • A few Miscellaneous Items • Mirth support for XML, HL7, etc proved very useful for traversal of structures in code and for field validations. • Supporting Add and Updated for Patient (base) record was useful. • Always create the Patient base record first regardless if Patient or Clinical CEM was received as first CEM to the DB.