Epic runs on Chronicles, a hierarchical, non-relational database that Connect Care describes as built on InterSystems Caché. Reporting data moves nightly into relational Clarity, then into Caboodle's dimensional warehouse alongside other sources. Analysts report from Clarity and Caboodle, and data leaves Epic through approved extracts, Reporting Workbench exports and FHIR APIs.
An analyst asked to connect a BI tool to "the Epic database" needs a more specific destination. The useful questions are how current the numbers must be, which fields the report needs and who approves access. Those answers point to Chronicles, Clarity or Caboodle, and to an export, an extract or an API.
The database name also says something about the work already done to the records. A live operational record, a detailed reporting copy and a prepared subject area each support a different kind of question. Establish which one the report needs before choosing how to extract its data.
Epic keeps three databases for three different jobs
Chronicles supports care as staff enter and use records. Clarity places extracted records into relational tables for detailed reporting, while Caboodle organizes reporting data into subject areas. The same clinical activity can therefore appear in different structures at different points in the refresh cycle.
Connect Care's glossary identifies Chronicles' hierarchical structure and operational role. UC Davis Health describes Clarity's nightly extract and Caboodle's star schema. These descriptions explain why choosing the database involves both the question's required detail and the point when that detail becomes available.
"SQL database" does not identify a single underlying product. Stanford reports different technologies for its hospitals' Clarity databases. The table names those as examples, so an engineer can distinguish the Epic reporting model from the database software that hosts it.
| Database | Built on | Structure | Refreshed | Used for |
|---|---|---|---|---|
| Chronicles | InterSystems Caché, per Connect Care | Hierarchical, non-relational | As records are entered | Live care, Reporting Workbench and Radar |
| Clarity | SQL database, such as SQL Server or Oracle | Relational | Nightly from Chronicles | Detailed reports and research extracts |
| Caboodle | SQL database | Dimensional, with a star schema | Nightly from Clarity and other systems | SlicerDicer and prepared reporting measures |
Chronicles runs live care, so analytics stay off it
Large historical analyses need a reporting source that separates those queries from clinical work. Chronicles holds the operational records staff need while caring for patients. Its hierarchical structure supports access to patients, events and encounters as that work happens.
This does not exclude Epic's own operational reporting. Connect Care describes Reporting Workbench and Radar as ways to view current Chronicles records. UC Davis's Reporting Workbench guidance says its main source is Chronicles and that reports generally suit shorter periods. A current list can therefore support an action inside Epic without becoming the source for an entire historical analysis.
For long-term reporting, Connect Care describes Clarity as a way to analyze extracted data separately from the production system. The distinction is between the jobs each source supports. An operational list needs current entries, while a historical report needs the selected records and calculations across its reporting period.
When discussing a connection, name the analytical source explicitly. A request for Clarity reporting access tells the institution which copy the proposed report will read. Calling it only an Epic connection leaves the source, refresh and intended query unresolved.
Clarity copies Chronicles into SQL tables every night
A Clarity query reads detailed records from the completed extract. Stanford Medicine STARR describes generating operational Clarity nightly from Chronicles. Its children's hospital uses Microsoft SQL Server for Clarity, while Stanford Health Care uses Oracle. A Clarity connection therefore needs the local database technology as well as the Epic model.
The relational structure supports detailed reports and custom queries, but coverage needs a local reading. Vanderbilt describes its Clarity as containing all EHR data and fully identifiable records. Stanford notes that some information visible in Hyperspace is absent from Clarity. Those descriptions should remain attached to their institutions when deciding whether a requested field is available.
Vanderbilt requires certification or proficiency in Epic's Clinical Data Model for Clarity access. That requirement illustrates the knowledge needed to interpret the records. An analyst must know how the requested population relates to the source fields before treating a returned total as the intended answer.
Queries also need to reflect the local Epic build. Epic's technical specifications explain that developers querying Clarity may need complex queries and customization for each institution. Review borrowed SQL against local fields and relationships before reuse. The database product may match another hospital's, while the report still needs different mappings.
Caboodle reshapes Clarity data into a reporting warehouse
An implemented Caboodle subject area brings related records together for analysis. Its facts and dimensions give a report writer prepared relationships to work with when the subject area covers the requested measure. That coverage determines how much of the report is already represented in the warehouse.
UC Davis describes nightly extracts from Clarity and other systems into a star schema that groups related business concepts together. For a recurring report, the question is whether those structures contain the required records, dates and measure.
Connect Care describes Caboodle as a SQL-based enterprise data warehouse that supports relationships with external sources. It places SlicerDicer primarily on Caboodle and confirms that Caboodle data can be exported to other warehouses. Our SlicerDicer guide explains how users explore populations through the available models.
Vanderbilt says its Caboodle holds most of its Clarity data and includes external links. Another institution may have implemented different subject areas. Before choosing an existing measure, inspect which records it includes and how it groups them. A familiar measure name alone cannot establish whether it matches the definition requested for your report.
Epic data leaves through four main routes
Choose the export route by the records recipients need and how often they need them. A tabular list can serve an operational review. A recurring finance or research report may require detailed fields and calculations that a reporting team maintains in SQL.
Epic's export documentation describes several electronic formats and SQL exports from relational reporting databases. It also links to FHIR Bulk Data for population-level exports. Epic on FHIR lists standard resources, including Encounter, and developer documentation for API access.
The API route needs its own coverage review. Match the required fields to the resources the API exposes before choosing it for the report. An available resource establishes an exchange route, while the institution's report definition determines whether that route supplies enough information.
For a repeated request, establish how the source will refresh and how the recipient will retain the definition. A file that answers today's question may need scheduled extracts and shared calculations when the same review becomes routine.
| Route | What you get | Fits |
|---|---|---|
| Reporting Workbench export | A tabular list from current entries | Short operational lists and CSV input to a raw data layer |
| Clarity extract | Detailed reporting fields from the nightly copy | Warehouses, research cohorts and finance reports |
| Caboodle export | Prepared subject-area records | Reports covered by the institution's implemented model |
| FHIR APIs | Standard resources over REST, including bulk exports | Applications whose required fields are exposed by the API |
Warehouse projects start by extracting Clarity each night
A recurring warehouse report needs a retained source load, a record of its schema and definitions that can be reused. Keeping the original extract helps analysts trace a result back to the records available for that reporting period. Keeping its schema helps identify mappings affected by an Epic upgrade.
Databasin's team starts with Clarity and uses Caboodle subject areas where they help. Our Epic connector was created at WashU Medicine. FHIR resources can supplement the reporting sources, and Reporting Workbench CSV exports can enter the raw layer. We never query Chronicles for analytics because those queries would compete with clinical work.
We keep each load as extracted, with its timestamp and schema version, then apply local mappings and define shared measures. Those definitions give subsequent reports a consistent basis for selecting records and calculating results. The raw history remains available for examining a source change or tracing a calculation.
The result can land in your Snowflake, Databricks Unity Catalog or an Iceberg lakehouse in your Azure tenant, under access your institution approves. Read about moving Epic Clarity into Snowflake or Databricks for that route. Our healthcare data warehouse page covers the broader work of joining clinical records with the other sources a report needs.
Example: getting last month's encounters out of Epic
This illustrative request asks for last month's inpatient encounters by unit in a spreadsheet or BI tool, with each route supplying a different form of the records.
| Route | What you get | Watch for |
|---|---|---|
| Reporting Workbench export | A list drawn from current entries | Generally suited to shorter periods |
| Clarity query | Detailed encounter fields from the completed nightly extract | Needs knowledge of the local tables and definitions |
| Caboodle subject area | Prepared encounter facts and related dimensions | Requires an implemented subject area covering the request |
| FHIR Encounter resources | Standard encounter fields through an API | Check coverage of the required local fields |
Choose the route after establishing the fields needed and how often the request repeats. A single list and a monthly report may need different extract and maintenance arrangements, even when both concern the same encounters.
Five checks before you connect to Epic data
- Name the database you needState whether the report needs current operational records, detailed Clarity fields or a prepared Caboodle subject area.
- Confirm who approves accessIdentify the institution's access owner and the records the report needs, including the approved use of identifiable information.
- Check which database runs ClarityConfirm the local database technology and completed refresh schedule before selecting the connection that will read the reporting copy.
- List your Caboodle subject areasMatch the requested fields and measures to the subject areas already implemented before deciding which records need another route.
- Plan for Epic upgradesRetain raw loads and their schema versions so analysts can trace changed fields and adapt the mappings behind recurring reports.
Sources
Alberta Health Services Connect Care, Chronicles, 2026 · ehealth.connect-care.ca/epic-systems
Alberta Health Services Connect Care, Clarity, 2026 · ehealth.connect-care.ca/epic-systems
Alberta Health Services Connect Care, Caboodle, 2026 · ehealth.connect-care.ca/epic-systems
Stanford Medicine STARR, Electronic Health Record, 2026 · starr.stanford.edu/data-types
Vanderbilt Clinical Informatics Center, Epic Data Resources, 2026 · vumc.org/vclic/epic-data-resources
UC Davis Health, Clinical Data Sources, 2026 · health.ucdavis.edu/data/sources.html
UC Davis Health, Epic Reporting Workbench, 2026 · health.ucdavis.edu/data
Epic open.epic, Technical Specifications, 2026 · open.epic.com/TechnicalSpecifications
Epic, Epic on FHIR, 2026 · fhir.epic.com
Discuss getting your data out of Epic
We can discuss which Epic records your teams need, where those records should land and the reports they should support. In the demo, we show a working example that answers your questions and discuss what delivering it would involve.