Table 3.

Comparison of data flow and transformation from EHR to aggregated analysis. Grey cells indicate when data first leaves control of local site.

CER Project Name	Raw data	Natural Language Processing	Data Normalization	Data aggregation	Data Analysis Tools
SPAN	Data from local EHR, billing, and other sources are accessed in their native form, or extracted and stored in research data warehouse	No	Each site transforms local data to the common data model (HMORN VDW) using native coding systems (NDC, ICD9, HCPCS, LOINC); transformations checked centrally for consistency	Centrally-developed application (analytic programs, queries) runs at local site, results sent to Data Coordinating Center or study lead site	Menu-driven query tools for encounter-level VDW datasets are under development; tools available querying of aggregate data
WICER	Extracted from local EHRs, sent to central data warehouse where it is combined, by patient, with data from multiple organizations (HIE warehouse – model)	Yes, using general purpose MedLee²⁸	Central DB uses Columbia's Medical Entities Dictionary (MED)²⁹ to map to LOINC, ICD, SnoMed	Central site	Tool under development to allow end-users to identify patients with characteristics of interest, specify query constraints, and combine data elements
CER-HUB	Data extracted from Local EHR, stored in local data warehouse	Yes, using project-specific, knowledge-based MediClass applications¹¹	Local site uses centrally-developed data processor – MediClass application–to populate Clinical Research Documents using Unified Medical Language System as standard knowledge base.	Centrally-developed processors are downloaded to create study-specific clinical events that are transmitted to central site for aggregation and analysis	Tools available for researchers to create new and review existing NLP query modules and test them against new data sets. Tools in development to enable investigators to analyze and review final aggregated results
RPDR	Data extracted from local systems (some daily, some monthly), aggregated centrally in data warehouse.	Yes, using project-specific pattern recognition models^30,31	Data mapped to ICD-9-CM and COSTAR (for diseases), NDC (for medications), LOINC (for labs), CPT and HCPCS (for procedures) data stored centrally in local coding systems and dynamically mapped in query tool.	All aggregate data is derived from queries generated through query tool accessing central data repository, detailed data on patients are gathered from central repository and through enterprise web services.	Drag-and-drop web Query Tool allows users to construct ad hoc, Boolean queries for hypothesis generation from structured data, to get aggregate totals and to graph age, race, gender and vitals
INPC COMET-AD	Extracted from local EHR (or payer), sent to central data warehouse, stored distinctly but can be combined at patient-level across multiple organizations; HIE-model	Yes, using project-specific pattern recognition models³².	Data are normalized and coded using standard vocabularies (LOINC, ICD9, CPT, NDC, etc.) when/where appropriate	Specially developed data aggregation routines developed and run centrally	All SAS and other statistical data analysis done at the central site; new data analysis tools are under development
SCOAP-CERTN	Data are extracted from source systems and transmitted via VPN channels to the central data warehouse servers; message stream is queued then parsed;	Text mining tools operate on the data within the central data warehouse; use hybrid text mining (rules base and statistical) to extract and tag the text and derive data elements needed for QA/QI and CER	Parsed data stream is either directly mapped onto the HL-7 data model or into a staging table from which it is mapped to the HL-7 derived and augmented data model.	Data extracted from local systems and sent to secure remote server cluster running Amalga	Sites and SCOAP-CERTN personnel will use secure web sites and interfaces with appropriate authentication, authorization and logging to query the Amalga data warehouse