A direct answer to the question of how major EHR data export formats compare starts with a distinction that is easy to overlook: the "export format" is rarely a single file type. It is a bundle of primary records, supporting media, and format documentation whose usability depends on the receiving environment. Athenahealth describes newline-delimited JSON or HTML 5 primary files, with XML, JPEG, PNG, TIFF, or PDF supplements included wherever relevant [1]. Epic describes a tab-separated value export native to its record structure, with content that does not fit table form such as rich text or images referenced from the tables and reviewed in a separate download [4]. Both meet the ONC requirement that export files be electronic and computable, but neither is interchangeable without mapping work. The useful comparison, therefore, centers on how machines can consume each package, whether the underlying data model is public or handed over with the files, and what ancillary artifacts must be reconciled because the receiving analyst may not count multimedia and scanned documents as finished data. The informational environment is uneven: ONC requires a public hyperlink to the export format, but does not require developers to publish proprietary data models. Technicians responsible for conversion, audit, or archive projects need to compare near-identical textual labels carefully. "JSON" at one vendor can mean heavily relational tables flattened into semantic objects, while at another it may mean a designated record set serialized directly from application tables. The following sections organize the comparison around documented containers, accessibility of format specifications, and a concrete evaluation scenario that does not require inventing procurement outcomes.

Container types and their computable interpretations

The most conspicuous difference among the vendors is the primary container type. Athenahealth states that ambulatory clinical and demographic exports are JSON, while inpatient clinical EHI files are HTML and inpatient collector files are JSON; supporting files arrive as images or PDFs [1]. Enterprise Health states that its export provides designated record set data in JSON format, including base64-encoded DICOM images within the extract [6]. Epic, by contrast, describes a computable tab-separated value file format native to Epic for the EHI Tables export, noting that some content may be referenced rather than embedded and can be reviewed in a separate download [4]. These labels matter less than their consumption logic. JSON and TSV are equally machine-readable in a narrow sense, but TSV requires a table-level schema to restore relationships, while JSON usually carries object boundaries directly.

Epic connects the TSV format to a schema index that identifies the data included in each table; the index is downloadable as a zip file, and Epic separately recommends FHIR bulk data access or C-CDA documents for standards-based exchange [4]. Oracle Health adds another category entirely by describing native database artifacts: single-tenant Millennium systems may receive a copy of the full database via an encrypted device, while Oracle Health EHR native data is provided as JSON semantic object models, with a prerequisite that exported data can be imported only into Oracle database version 23ai or later [3]. That version gate is a material decision point. A receiving team planning a staging database on another version, or on a non-Oracle relational platform, must treat the Oracle native export as source material for a separate transformation project, not as a directly loadable interoperability feed.

Format specifications: transparency versus disclosure of data models

ONC guidance imposes a specific transparency condition that shapes how these documents are published. A publicly accessible hyperlink describing the export format must be included with the exported files, and the link must allow direct access without preconditions or additional steps [2]. The regulator also states that developers do not need to make public a proprietary data model, noting that export file formats and related definitions were finalized as specific certification requirements [2]. This creates an uneven informational environment. Athenahealth's public site describes file types and the broad clinical and billing domains covered, but does not expose the relational schema behind those JSON records in the way a database vendor might [1].

Oracle Health approaches the disclosure differently. Its documentation points to a separate data format specification zip file, which includes an EHI Oracle data model containing table and column information for core Millennium data, appendices for DICOM and non-DICOM multimedia structure, and a sample manifest file [3]. Enterprise Health similarly publishes a table-style listing of designated record set content objects, such as encounters, orders, observations, and insurance plans, alongside example export files that show request strings and field values [6]. Epic's public EHI Tables index serves a comparable role for tables but Epic frames the export as a manual one-time function and directs integrators needing programmatic paths to APIs and interfaces rather than the export itself [4]. Comparing these vendors therefore requires asking not only "what format?" but "what level of schema documentation accompanies the format in this specific engagement?"

A concrete evaluation exercise: same patient, four export packages

A hypothetical exercise can force the comparison into a testable sequence without inventing performance results. Suppose a health system requests an export for one complex patient who has structured labs, two scanned consents, a DICOM image series, and dictated notes. The evaluation team should first inventory what each documented package would contain and how the relational glue is represented. For athenaOne, the team would expect JSON primary data plus referenced image or PDF files, and would check whether the supporting XML files are themselves structured or simply documents [1]. For Epic, the team would inspect the EHI Tables index for which table carries the lab results and which columns reference the scanned consents or images, then verify the separate download contains those objects [4]. For Oracle Health Millennium data, the team would locate the corresponding database tables in the provided EHI Oracle data model and, when native EHR data is present, confirm the Oracle 23ai-or-later staging prerequisite before attempting import [3]. For Enterprise Health, the team would look for DICOM base64 content and dictation_documents linking audio to transcriptions inside the JSON extract [6].

This exercise is a suggestion, not a documented test result, and the outcome depends on site configuration. Epic's own documentation warns that content varies by applications, version, practices, and configuration decisions at the generating health system [4]. A team should therefore run the same inventory against a sample production export, not only the vendor's published specification, and record any contents that are referenced but absent, any binaries whose metadata disconnects from the primary index, and any fields that require a separate schema request. The comparison document should list four dimensions per vendor: primary container, schema or index availability, multimedia handling, and any import prerequisite. Where a dimension is not published, it should be marked unknown and requested explicitly rather than inferred from another vendor's behavior.

Multimedia and document artifacts change the comparison

A comparison that ignores supporting files will understate conversion effort. Athenahealth says human-readable files in XML, JPEG, PNG, TIFF, or PDF may contain encounter summaries, imaging results, scanned clinical documents, or insurance card images, and that these are included wherever relevant [1]. Enterprise Health's object list includes dictation voice files, dictation_documents linking transcriptions to original audio, DICOM images provided in base64, and separate DICOM study and series tables [6]. Epic states that some electronic health information, such as rich text documents or images, might not be available in table format and might be referenced from the EHI Tables with the actual files reviewed in a separate download [4]. The operational difference is subtle but significant.

In a JSON-first environment like Enterprise Health, some imaging data is embedded as base64 within the primary extract, which simplifies single-file handling but increases payload size and requires decoding before clinical review. In an Epic TSV export, the table may carry a pointer while the binary object sits elsewhere in the package, shifting integrity checking to the relationship between index and artifact. Athenahealth's phrase "wherever relevant" leaves the incidence of PDF encounter summaries versus structured JSON records unspecified in its public documentation; a technician should not infer a fixed ratio from the wording [1]. Oracle Health goes furthest by separating core EHR database data, multimedia storage DICOM and non-DICOM data, document imaging content, and native EHR data into distinct specification appendices, with a manifest mapping multimedia files to database entries through supplied queries [3]. These documented structures, not abstract file extensions, drive the comparison.

What the documentation does not establish

Several consequential facts remain unknown from the public documents and should be treated as gaps rather than advantages for any vendor. Athenahealth's public documentation does not state how relationships among newline-delimited JSON records are expressed, whether through foreign keys, embedded IDs, or manifest files [1]. Epic states that exported content can vary based on software applications in use, version, documentation practices, configuration decisions, and whether the health system included materials not sourced from Epic [4]. That statement means the same Epic-defined EHI Tables spec may produce materially different content at two customer sites, and no published document reviewed here establishes uniform field coverage across installations.

Oracle Health's native export imposes an Oracle 23ai prerequisite for import [3], but the source does not describe performance characteristics, extract runtimes, or downstream mapping effort. Enterprise Health documents object names generously, yet its publication does not state how referential integrity is conveyed to a receiving non-Enterprise Health system beyond the JSON structure shown [6]. AdvancedMD, representing a different segment, lists multiple output types including C-CDA XML, CSV transaction reports, PDF chart prints, Microsoft Access for self-service practice management data, and scanned documents in native format, but it also states that bulk exports of notes in PDF or Word format cannot be provided [7]. That specific limitation, documented by AdvancedMD, is not documented for the other vendors and should not be assumed to apply to them. The honest comparison works at the level of documented container, availability, and prerequisite, while flagging the rest for verification during a vendor engagement.