Learning Objectives
This tutorial page is intended for i2b2 Administrators who are new to i2b2 and clinical informatics. Maybe you've been asked to deploy i2b2 for your institution. Maybe your institution already has an i2b2 instance, and you've just inherited it. You are curious about how things work and how to set up or modify the ontology database.
This page will help you achieve these goals:
- Understanding what an ontology is for i2b2
Choosing an appropriate ontology tree for i2b2
Deploying an ontology tree for i2b2
- Understanding how the ontology enables queries in i2b2
Part 1: Ontology Basics
What is an ontology?
"Ontology" represents the branch of Philosophy that discusses "existence." But in Information Science, an "ontology" is a way of listing the attributes of a subject area and showing how these attributes are interrelated. An ontology for a subject area or "domain" consists of a set of terms that represent the entities in (or attributes of) that domain, along with a set of expressions that define how the entities in that domain are related to each other.
In i2b2's domain of "Clinical Research Informatics," an ontology is a collection of healthcare-related "concepts" or terms. These terms represent the various categories of information in clinical and translational science. In an i2b2 ontology, categories such as Patient Demographics, Diagnoses, Procedures, Laboratory Tests, and Medications are described using hierarchical lists of "concepts."
Unlike some other ontologies in Information Science, an i2b2 ontology does not typically define relationship expressions among the concepts or entities. Therefore, strictly speaking, an i2b2 ontology is generally more akin to a "taxonomy," which is a hierarchical group of terms or concepts, without any expressions defining the relationships among them.
For instance, if you recall (from your high-school biology class) Linnaeus's "binomial nomenclature" model of how to name living organisms, then you already know what a taxonomy is. As another example, a Camry is a Toyota, which is an automobile, which is a type of motorized, wheeled conveyance.
Why does i2b2 have an ontology?
An i2b2 instance requires an ontology so that researchers can locate concepts of interest and locate patients whose medical records contain these concepts.
The healthcare concepts contained in the ontology are the building blocks of an i2b2 query. The i2b2 ontology is represented in the upper-left corner of the i2b2 webclient (user interface). See the screenshot below.
These concepts are rendered as a collection of nested folders and generally represent child-parent relationships. Researchers locate these concepts by "drilling down" through a hierarchical tree, or by using the "search" tool. When you use the demo i2b2 site, you will see the i2b2 ontology tree on the upper-left side of the user interface.
In this screenshot, we would say that i2b2 is displaying an ontology tree with 13 top-level ontology trees or categories, namely: Clinical Trials, Custom Metadata, Demographics, Diagnoses, etc.
Here are some examples of child-parent (or is-a) relationships found in i2b2's ontologies of healthcare concepts:
- diabetes mellitus is an endocrine disorder, which is a type of diagnosis; it will be found in both the "Diagnoses" tree and in the "Diagnoses (ICD10)" tree;
- aspirin is a non-steroidal anti-inflammatory drug, which is an analgesic, which is a type of medication; it will be found in the "Medications" tree;
- tonsillectomy is a type of surgery, which is a procedure; it will be found in the "Procedures" tree;
- hemoglobin A1c is a blood test, which is a type of laboratory test; it will be found in the "Laboratory Tests" tree; and
- race is a patient-demographic descriptor; it will be found in the "Demographics" tree.
Here is another screenshot, where you can see the expansion for "Race" in the "Demographics" tree:
This diagram shows the high level workflow:
Where do i2b2 ontology "concepts" come from?
Typically, the concepts that make up the various categories are based on medical terminology standards published by various institutions. Concepts can also be created ad hoc for a specific i2b2 instance by the i2b2 administrator for that site. (For more information on custom metadata, please see Ontologies 201.)
The commonly used standards representing basic structured clinical patient data are represented by a number of different "coding schemes" — ICD-9, ICD-10, CPT-4, HCPCS, SNOMED CT, LOINC, RxNorm, UMLS, and VA Classes. We mention this here because these codes appear inside the ontology tables, and they are also referenced in the user interface. See Appendix B for further information.
There are many terms related to i2b2 ontologies that are new to me. Is there a Glossary?
Yes! Here is a Glossary that provides explainers for many of the key terms related to i2b2 ontologies and the i2b2 architecture.
Part 2: Choosing Your Ontology
Do all i2b2 instances share the same i2b2 ontology?
No. Several i2b2 ontologies have been developed and are openly available for use. Any organization may also modify an existing ontology for its own use, or develop a new ontology. (See Ontologies 201 – Custom Metadata – Additions and Modifications.)
The i2b2 software includes i2b2 ontologies that you can use right away. There are additional ontology trees that can substitute for, or can be added to, the standard i2b2 ontology trees. (See Appendix F – Advanced Ontologies.)
Which ontologies can I use right away ('out of the box')?
The i2b2 database-loading modules come with at least 3 sets of "metadata" or ontology trees. These are the Demo Ontology, the ACT Ontology, and the ACT-on-OMOP Ontology. Inside the i2b2 software release folder you will find the three ontology tree folders (demo
, act
, and act-omop
) at this location (linux syntax): i2b2/edu.harvard.i2b2.data/Release_m-n/NewInstall/Metadata
, where m
is the version number and n
is the release identifier.
You may review/download the database-loading modules on Github: https://github.com/i2b2/i2b2-data. You may also download from i2b2's Software Download page: https://www.i2b2.org/software/.
Each ontology is designed to be used with some version of the i2b2 Common Data Model.
Name | Folder | Description | When To Use | Target Data Model |
---|---|---|---|---|
i2b2 Demo Ontology | demo | default metadata from i2b2 authors; legacy categories | demonstration | i2b2 Common Data Model (star-schema) |
ACT Ontology | act | modern categories and concepts, supplied by the ENACT project | production | i2b2 Common Data Model (star-schema) |
ACT-on-OMOP Ontology | act-omop | modern categories and concepts, supplied by the ENACT project | production, at sites using OMOP CDM | multi-fact-table-enhanced i2b2 Common Data Model (star-schema) |
How should I choose which ontology to use?
This is a key question/concept.
The ontology you choose depends on your institution's research goals and patient data.
Research Goals
If your goal is simply to understand and show how i2b2 works, then the i2b2 demo ontology is sufficient. If your goal is to set up i2b2 for research use, one of the ACT ontologies will be far more useful. The ACT ontologies are more modern, more robust, and will satisfy the needs of more researchers.
Furthermore, any institution that is planning to join the ENACT Network will need to use an ACT ontology to ensure compatibility with the other institutions in the network.
Patient Data
First, a little more terminology. The i2b2 "patient data" are stored in the i2b2 Clinical Research Chart database, or "CRC database." The ontology categories and concepts that describe those patient data are known as the "metadata." The metadata are stored in the i2b2 Metadata database, or "ONT database."
When setting up the patient data for i2b2, your institution needs to decide how it will conduct the ETL mapping from the EHR to the i2b2 CRC database. If your patient data are going to be set up only for i2b2 and SHRINE use, then many institutions set up the patient data in the default i2b2/tranSMART Common Data Model (CDM). In this case, they typically use the ACT Ontology with its robust collection of coding standards; the patient data from the EHR system need to be mapped to the multiple coding schemes in the ACT Ontology as they are loaded into the CRC database.
If your patient data are going to be set up in a database for queries by other systems besides i2b2, then many institutions use the OMOP Common Data Model (CDM) for their patient data in the CRC database. In this case, they typically use the ACT-on-OMOP Ontology, which relies chiefly on the OMOP Vocabulary standard; the patient data from the EHR system need to be mapped to the appropriate OMOP Vocabulary coding scheme as they are loaded into the OMOP CDM in the CRC database.
It is highly recommended for newcomers to begin their i2b2 journey by first setting up the Demo Ontology with the Demo CRC patient data.
Setting up i2b2 can be a complex undertaking. You can learn a lot about how it works, and get the system up and running most quickly, by first setting up your i2b2 instance with the i2b2 Demo Ontology and demo patient data. Those databases will comprise a "demo" project in your i2b2 instance.
When you have proven i2b2's deployment with the demo project, you can add a separate, new project for research. In this case you could use actual patient data in a second CRC database and an ACT Ontology in a second ONT database. Those new databases will comprise a "research" project in your i2b2 instance.
Actually, I inherited an i2b2 instance. How can I tell which ontology it's using?
You may compare the tables in your ONT database with the variations shown in Appendix D – i2b2 Ontology Tables.
Part 3: Deploying Your Ontology
How do I deploy my chosen ontology tree?
See the installation instructions here: 3.7 Metadata Tables
What should my metadata database look like when I am done?
See the list of metadata tables here: Appendix D – i2b2 Ontology Tables
What else is pertinent to setting up my first ontology?
Two very important topics for understanding the relationship between patient data (CRC) and metadata (ONT) are i2b2 Projects and CRC-Based Metadata. And it is critical to understand that because these two databases work in tandem, the full deployment of the ontology is not complete until the CRC database is deployed as well.
i2b2 Projects
The pairing of a metadata database with a patient database (ONT plus CRC) defines a "project" in i2b2. For instance, the pairing of the i2b2 demo ontology with the i2b2 demo patient data would define a "demo" project in i2b2. In that same i2b2 instance, an institution could create a second ONT database for the ACT ontology, and a second CRC database for the institution's actual patient data; the pairing of those two new databases would define a genuine research project in i2b2.
It's important to note that each ONT database is designed to be associated with only one CRC database; that is, each ONT database is designed to belong to only one i2b2 project. This is because the ONT database actually includes patient counts from a CRC database; and for that reason can be tied to only a single CRC database.
The converse is also true. Each CRC database is designed to be paired with only one ONT database. This is because the CRC database actually includes some very specific metadata that must match the metadata in the ONT database.
The instructions for creating multiple projects are outside the scope of this Ontologies tutorial page.
CRC-Based Metadata
There is a special metadata table that resides in the CRC database. This metadata table is called the CONCEPT_DIMENSION table. It resides in the CRC patient database, and it is part of the i2b2/tranSMART patient CDM, but it is still metadata. This table is crucial for the success of the i2b2 query mechanism. You may consider this the "secret sauce" that joins the metadata with the patient data. There are settings in this table that must match certain values in the ONT database. For a complete explanation of how the CRC database's metadata table is used, please see Ontologies 103 – Ontology Table Structure and Query Mechanism.
A second metadata table also resides in the CRC database. This is called the QT_BREAKDOWN_PATH table. This defines how certain queries are performed. Again, there are settings in this table that must match certain values in the ONT database.
Part 4: How Ontologies and Patient Data Are Tied Together
How does it work? How does the i2b2 ontology make the patient data queryable?
This is another key question/concept.
The patient data in the CRC database work in tandem with the metadata in the ONT database to allow i2b2 to conduct queries for the researcher. The metadata spell out all the various healthcare concepts or terms that a researcher may wish to query for in the user interface, and the patient data must be coded in such a way that they reflect the codes defined in the metadata.
Only when a patient's codes match a query's ontology codes will the patient be counted as part of a query result. Each patient "observation" in the CRC database must have a code associated with it, and that code must match a healthcare concept — diagnosis, procedure, medication, lab test, demographic descriptor, etc. — that is defined in the ONT database. So the CRC patient data will be queryable in i2b2 only if the patient facts in the CRC database are recorded utilizing standard codes that are defined in the i2b2 metadata (ONT database).
For instance, if the patient's electronic health record indicates that the patient had the diagnosis "Acute tonsillitis
," then that fact needs to be recorded in the CRC database using the standard code for that particular diagnosis as defined in the ONT database. In the case of using the default i2b2 Diagnoses metadata tree, that code would be "ICD9:463
". When a researcher makes a query in the user interface for "Acute tonsillitis
," then i2b2 will query the CRC database for all patients who have the code ICD9:463
in their data records.
Let's view a high-level illustration of the mechanism behind a typical query. Here's that concept's definition from the diagnoses table in the ONT (metadata) database:
Excerpt from Diagnoses Table in Demo Ontology (I2B2 table in ONT metadata database) | ||
---|---|---|
Concept Name, displayed to user | Concept Path, a unique identifier | other fields |
Acute tonsillitis |
| ... |
When the user selects that Concept Name in the user interface, the i2b2 Query Tool associates that with the Concept Path. When running the query, i2b2 looks for that path in the concepts table in the CRC database, which looks like this...
Excerpt from Concepts Table in Demo Ontology (CONCEPT_DIMENSION table in CRC database) | ||
---|---|---|
Concept Path, a unique identifier | Concept Code, used in facts table | other fields |
|
| ... |
...and i2b2 associates that Concept Path with the Concept Code for that concept. i2b2 then locates those patients in the facts table in the CRC database that share that Concept Code...
Excerpt from Facts Table in Demo Patient Data (OBSERVATION_FACTS table in CRC database) | |||
---|---|---|---|
Patient ID | Visit ID | Concept Code | other fields |
|
|
| ... |
|
|
| ... |
|
|
| ... |
|
|
| ... |
|
|
| ... |
|
|
| ... |
...and reports a count of those patients back to the user at the conclusion of the query. In this illustration, there were 2 unique patients (over 6 encounters) that had this Concept Code in the facts table, so the user's query for patients with a diagnosis of "Acute tonsillitis
" will return a count of 2 patients.
The query mechanism illustrated above applies to most queries, but not all. For more details, see Ontologies 103 – Ontology Table Structure and Query Mechanism.
It's important to understand that each institution will have its own protocols for coding diagnoses, procedures, medications, etc., in the patient electronic health records (aka EHR, as in Cerner or Epic databases), and that these institutional protocols may use standard, non-standard, or proprietary codes. So, when loading patient data into the CRC database for i2b2, it's necessary for the data curators who perform the ETL process (extract-transform-load) to map the codes existing in the patient EHR into the codes that are defined in the i2b2 metadata (ONT database).
For instance, let's say a patient's EHR record includes an NDC code for a medication. And let's say that your institution's i2b2 ontology tree only has an RxNorm code for that particular medication. Then the medication record from the EHR would need to be mapped into a patient record in the i2b2 CRC database in such a way that the ontology's RxNorm code (not the EHR's NDC code) appears in the patient record. If the patient record in the CRC database has the NDC code from the EHR, then it would not be matched in a query, since the i2b2 query would be using the RxNorm code.
What if my institution records new concepts into the EHR that are not already included in the ontology?
It's not unusual for institutions to include coding in their EHR to reflect novel diagnoses, new medications, etc. And some institutions prefer to use proprietary codes for some clinical information. If your institution is using concepts or codes that are not reflected in your i2b2 ontology tree, then your researchers will not be able to query for those concepts in i2b2.
The remedy for this is to customize your ontology trees to include the new concepts that are missing from the i2b2 ontologies. See Ontologies 201 – Custom Metadata – Additions and Modifications.
If your new concepts are not strictly proprietary, then please consider contacting the appropriate coding authority, asking them to include your new concepts into the next release of their terminology.
What if my institution's EHR is missing concepts that are found in one of the ontology trees?
Never fear! No institution will be using all of the concepts in every ontology tree. It is normal to have concepts in the ontology tree that match none of your patient records.
To prevent your institution's researchers from inadvertently choosing an "unused" concept in their queries, we recommend adding patient counts to each concept in the ontology tree. You can learn about how to do this in Ontologies 102 – Patient Counts ("totalnum
").
Suggested Next Steps
- Visit Appendix D – i2b2 Ontology Tables to learn more about the structure of the metadata.
- Visit Appendix E – Test Queries to learn how to run some "sanity check" queries on your new ONT and CRC databases.
- Visit Ontologies 102 – Patient Counts ("
totalnum
") to learn how to add patient counts to your ontology concepts in the user interface.
Add Comment