2019 HSR&D/QUERI National Conference
4049 — Distributed Research Networks: Implementation of a Virtual Data Warehouse for Pragmatic Pain Trial Development within the VHA
Lead/Presenter: Sesh Mudumbai, COIN - Palo Alto
All Authors: Mudumbai SC (VA Palo Alto HCS; Center for Innovation to Implementation;Stanford University, Department of Anesthesiology), Fan Wu (VA Palo Alto HCS; Cooperative Studies Program), Justin Chambers (VA Palo Alto HCS;Center for Innovation to Implementation) Aditi Kapoor (VA Palo Alto HCS;Center for Innovation to Implementation) Jennifer S. Lee VA Palo Alto HCS; Cooperative Studies Program) David Clark (VA Palo Alto HCS; Stanford University, Department of Anesthesiology) Edward R Mariano (VA Palo Alto HCS; Stanford University, Department of Anesthesiology) Randall Stafford (VA Palo Alto HCS; Stanford University, Department of Medicine)
The National Institutes of Health suggests that distributed research networks (DRN) are essential for extracting and pooling large volumes of data for multi-center pragmatic trials. A key component of a DRN is the virtual data warehouse (VDW) that fits local data elements to a common data model. As part of a multi-healthcare system study on chronic pain, our objectives were to describe 1) implementation of the Health Care Systems Research Network (HCSRN; i.e., Kaiser Permanente) VDW; and 2) evaluation metrics.
We assembled a VDW within VINCI and local informatics networks, including all patients from one facility from 2015 to present. We initiated extract-transform-load processes: Extraction from the VA Corporate Data Warehouse; Transforming into HCSRN data model; and Loading into SAS datasets. Key design considerations included distributed queries; executing queries against local datasets; returning aggregated, de-identified results; and supporting simple and complex queries. Using SQL server "shell" structures, the VDW normalized VA data elements for tables including demographics and pharmacy. Primary keys involved medical record and real social security numbers. Our process involved target-source mapping; data queries; and documentation for meta-data dictionaries. Two types of queries were run to validate data characteristics: data structure and primary keys; then formatting, completeness of data and accuracy. Evaluation of VDW involved 1) an iterative quality assurance process 2) whether data could be updated weekly; 3) characterization of base population; and 4) ability to identify new cohorts.
After passing QA checks for tables, our VDW demonstrated dynamic weekly updating, ensuring near-real time analytics. Our initial VDW population (n = 72,399) was primarily older (mean age = 62.5), male (92.3%), with race categories including white (69.6%), black (9.5%) and unknown (12.2%). Our base cohort identification algorithm (patients newly started on long-acting opioids) was functional with consistent, valid results across study period. Additional cohorts identified included patients prescribed > 3 months either buprenorphine for opioid use disorder (n = 186); or co-prescribed opioids and benzodiapenes (n = 2,708).
We report the first implementation of the HCSRN VDW and common data model within a VHA facility.
The development of a distributed research network data warehouse has created opportunities for multi-center, pragmatic clinical pain trials.