Healthcare analytics
Healthcare delivery systems, payors, and service providers use analytics for a range of purposes, such as revenue cycle management, quality management, and process improvement. These entities generate, analyze, and exchange large volumes of data. The health data processed spans a diverse set of domains, including clinical, finance, supply chain, human resources, research and more. The volume and variety of data processed by analytics is steadily increasing, making extensible, elastic, cloud-based architectures increasingly attractive.
Healthcare analytics architectures have the following characteristics:
-
Source data is ingested from upstream sources such as EHRs using bulk and streaming protocols. The raw data is oftentimes persisted for data lineage and reprocessing. For example, raw HL7 v2 messages, EDI transactions for claims, genomic variant data, medical images, faxed or scanned documents, and so on are ingested and stored for downstream processing, oftentimes for multiple, disparate use-cases (such as input data for AI/ML inferencing, dashboards, forecasting, and sharing).
-
Data transformations create derived datasets by applying structural and semantic transforms and linking data from different tables, systems, or domains. During this process, the data can be mapped to standard clinical terminologies (such as SNOMED, ICD, CPT, and NDC) to provide consistency for downstream consumers of the data. Transformation jobs may include machine learning steps such as OCR of scanned or faxed documents, transcription of voice recording, or natural language processing of clinical text. Additionally, these transformations may include steps to reduce the fidelity of the data for specific use cases, such as generating de-identified or limited datasets for research use cases.
-
Business logic runs on the derived data to generate actionable insights. Late binding approaches for the data schema enable a more agile approach and avoids large, upfront investments in heavy ETL. Results are sent to one or more purpose-built data stores.
-
Various stakeholders consume the data to glean insights. Data consumers are often diverse, ranging from non-technical reviewers of business intelligence dashboards, to researchers running deep learning algorithms.
-
Data remains encrypted at-rest and in-transit throughout the entire process. Identity and access controls are enforced so that access to sensitive health data is limited appropriately. All data assets can be recorded in a centralized Data Catalog to promote data discovery and reuse.
