FHIR Bundle Processing in AWS HealthLake: Understanding the Difference Between Batch and Transaction

Details

A FHIR Bundle is a container that holds multiple FHIR resources. AWS HealthLake offers different processing methods based on the nature of the data. The 'Batch' mode treats each resource independently; this is suitable when partial success is acceptable, meaning remaining data can still succeed even if one resource fails (e.g., bulk downloading patient records from various facilities). Conversely, the 'Transaction' mode handles all included operations as a single atomic unit. This guarantees strict data integrity—either all operations succeed, or none do—and is essential when referential integrity is required (e.g., creating a patient record along with related observations and conditions simultaneously). Key differences include execution order and failure handling. In 'Transaction' mode, the operations are processed in the specified order, and local reference integrity is applied. In contrast, 'Batch' mode processes each operation individually, allowing for partial success. The choice between these modes depends on whether the application requires guaranteed consistency across related resources (transaction) or if independent processing with tolerance for individual failures is acceptable (batch).