There is a pressing need for timely, reliable, and generalizable information to guide infection control efforts directed against methicillin-resistant Staphylococcus aureus
(MRSA) within hospitals. This microorganism frequently causes abscesses, bloodstream infections, post-surgical infections, and sometimes deaths; estimates from existing research and census data suggest that 17,000 attributable deaths occurred in 2008 [1
]. With the objective of reducing MRSA transmission in hospitals, the Department of Veterans Affairs (VA) implemented the National MRSA Prevention Initiative in October 2007 [3
]. The program included VA-wide MRSA testing upon admission to, discharge from, and transfers between acute care wards; rules for contact precautions; hand hygiene; a change in culture to one of shared responsibility; and new reporting systems [4
]. The VA Inpatient Evaluation Center (IPEC) gathered data to evaluate this program by employing coordinators at each facility to review MRSA results. The current mode of data collection could be augmented and made more efficient with detailed electronic microbiology data. These electronic data could also be used for algorithmic surveillance, which has the advantage of reliability over time and place [5
Microbiology data are increasingly collected electronically throughout the United States and could eventually provide a powerful means of infectious disease surveillance, but the synthesis and utilization of databases across large networks remains a daunting endeavour both in and out of the VA. Barriers include differing data models [6
], messaging strategies, and security issues [7
]. The VA medical centers have had an electronic medical record system for over 20
years. This includes, but is not limited to, microbiology data maintained at 152 hospitals currently active worldwide. These data are siloed at each hospital, complicating the process of compiling and integrating enterprise-wide data. Re-engineering the system to capture standardized, structured data will eventually be performed, but was a prohibitively large undertaking at the time. Hence, our objective was to evaluate methods that permitted rapid extraction and validation of these microbiology data.
The VA stores patient-level microbiology and most other types of data in a hierarchical health information system called the Veterans Health Information Systems and Technology Architecture (VistA). VistA uses a programming language and database called MUMPS (Massachusetts General Hospital Utility Multi-Programming System). Although all VA medical centers use the same software programs, they may have distinct naming conventions and some variation in data structure [8
]. This allows flexibility but also permits redundancy and idiosyncrasies to creep into the data. There has been some consolidation of VistA instances among medical centers, but most continue to maintain their own VistA system. Because a core system integrating microbiology data across VistA systems was not otherwise available, we utilized an available system developed by VA Patient Care Services (PCS).
The PCS system used Medical Domain Objects, an approach similar to the process that retrieves records during the course of clinical care (see Figure ). Healthcare providers normally access data using the CPRS (Computerized Patient Record System) graphical user interface. The CPRS interacts with the core MUMPS databases through a number of established remote procedure calls (RPCs) that execute patient data objects (objects that assemble data to form reports or components of reports for visual display). The process for using a RPC is identical at all medical centers and is highly reliable. VistaWeb, developed a number of years ago to access VistA data off-site, also uses RPCs. VistaWeb is in daily use nationwide and enables off-site access to medical records [9
Figure 1 Data Network Diagram. RPC - remote procedure call. CPRS - Computerized Patient Record System. RDW- regional data warehouse. CDW-corporate data warehouse. PCS – VA Patient Care Services. NLP- natural language processing. Laboratory data are generated (more ...)
The PCS system uses the VistaWeb interface to execute RPCs at each medical center and then uploads the data to a Structured Query Language (SQL) relational database in a secure VA data center. The VA login and network security processes are based upon the same approaches required for providers system-wide. Because the PCS system uses existing, reliable data processes, it requires no independent maintenance and can be run as a background process.
The separation of RPCs and patient data objects can have a valuable role in maintaining data validity. Data extracted through RPCs represent a coupling of both VistA data and the patient data object, which means that information extracted this way retains context from the patient data object that would not otherwise be present. RPCs invoke universal commands to local patient data objects that incorporate the meta-data necessary to make VistA data intelligible to providers. If data structures or reporting formats change within an implementation of VistA then local programmers must also update patient data objects so that RPCs continue to retrieve appropriate data to display for healthcare workers. Thus, the data that healthcare workers see, interpret, and report errors about are the same data extracted through the PCS process. When raw VistA data are pulled into a central database, often by different teams than those that built the local patient data objects, the data must be carefully evaluated for changes in structure and semantics. This is critical because microbiology records often contain multiple tests and a hierarchical structure of cultures, microorganisms, and susceptibilities.
Direct data retrieval can retain native data structure, but is unable to retrieve ‘misplaced’ data. This is because 1) not all of the complex aspects of microbiology reports were anticipated when the data model was developed; 2) microbiology reports from client laboratories may have formats incompatible with the data model; and 3) MUMPS has only one data type, so there are no data type checks. As a result, data may be systematically or sporadically entered in the wrong place in VistA, but still be correctly represented to the provider from RPCs used by CPRS.
The methodology that we investigated retrieved VA-wide, patient-level data in the same format used by health care providers. These records are in a semi-structured, free-text form that is as “human-readable” as an official microbiology report. Even though standardized, individual fields were lost by using this format, the record can be inspected visually to interpret its meaning. But as the data were already assembled and could be updated daily, they represented a valuable resource in need of formal validation.