Integration Challenges of Clinical Information Systems Developed Without a Shared Data Dictionary

Transcription

1 MEDINFO 2004 M. Fieschi et al. (Eds) Amsterdam: IOS Press 2004 IMIA. All rights reserved Integration Challenges of Clinical Information Systems Developed Without a Shared Data Dictionary Val N. Hicken a,b, Sidney N. Thornton a,b, Roberto A. Rocha a,b a Department of Medical Informatics, Intermountain Health Care, Salt Lake City, Utah b Department of Medical Informatics, University of Utah, Salt Lake City, Utah Val N. Hicken, Sidney N. Thornton, Roberto A. Rocha Abstract Legacy systems have proven to be long-term integration challenges for Intermountain Health Care (IHC) despite commitment and attention to share clinical information across settings and among clinicians. This study measures the extent of the disparity of data elements across three independent data systems in current use. A sample of relevant data elements was selected across systems covering prenatal, labor and delivery, and newborn intensive care units (NICU). The findings revealed only 17% of these sample data elements had compatible structure across all three systems. The implications from differences in granularity, missing data, and duplicate data entry, include diminished data quality, greater risk for medical error, increased costs of integration and inefficient use of clinician time. Retrospective guidelines for managing conceptual context and granularity are given to assist in designing an integrated longitudinal patient electronic medical record. Keywords: Systems integration, longitudinal computerized medical records systems, information management, terminologies, data quality, common data elements, data model reconciliation, EHR systems standardization, ontology development. Introduction Intermountain Health Care (IHC), an integrated delivery network of 22 hospitals and 72 clinics in Utah and southern Idaho, is designing and implementing an enhanced information systems infrastructure on an enterprise-wide scale. 1 Over the past 19 years IHC has developed women and newborn specific clinical information systems for Prenatal, Labor and Delivery (L&D), and newborn intensive care unit (NICU). Original functional requirements for the L&D system were narrowly focused on nurse-user documentation needs and did not consider interoperability. 2 Design constraints led to the use of an independent system. Consistent data models were not used and concept meanings were embedded within the L&D information system itself. In 1995 IHC began the focused development of a common Health Data Dictionary (HDD) to identify and maintain core clinical data concepts, in collaboration with 3M HIS. 3 Since that point, new applications use concepts defined in the HDD and new concepts are added as needed. The L&D system at IHC was developed before the advent of the HDD and evolved autonomously, having its own definitions for its coded data elements. Having collected information on over 300,000 births to date, a rich body of clinical knowledge has been obtained. Because of the independent nature of the L&D system development and design, integrating information into and from this system has been very difficult. Presently, an integrated longitudinal patient view of prenatal and maternal information does not exist. Clinicians must access the separate specialty systems and compare data from each in order to get the longitudinal patient view. IHC is reviewing the requirements of building an infant-centric view of a longitudinal Electronic Health Record (EHR) to cover the care continuum through prenatal, delivery, and discharge, and also links mother and infant data. IHC has adopted an open architecture approach and is actively working to interface multiple best-of-breed individual systems [3]. The costs, however, are substantial to create interfaces with older systems that were developed without using a common data model or data dictionary. Presently much of the information captured in the prenatal record is re-entered manually into other data systems. The prevalence of data variation as represented by mismatched meanings as well as missing or conflicting data across systems has not been measured. Failure to share patient information across data systems may lead to inefficiency and reduce quality of care [4]. Studies show that redundant records lead to errors and extra effort, misdirected data, over-reliance on the spoken word, inaccuracies and information loss, limited standardization, miscommunications, decision changes, and limited outcomes evaluations [5]. The 1991 Institute of Medicine (IOM) report, published by The Committee on the Quality of Health Care in America, brought attention to the need for the implementation of computer-based patient records [6]. A subsequent IOM 2001 study emphasized that what is needed for safer, higher-quality care is the appropriate use of information technology to support clinical and administrative processes. Recommendations have been made for clinicians to cooperate with each other to ensure an appropriate exchange of information and coordination of care [7]. Information system infrastructures must facilitate access to all relevant components of the patient medical record and to provide clinicians and multidisciplinary teams with the information necessary to deliver evidence-based care consistently and safely. 1053

2 Figure 1 - Overview of all data elements involved. The majority of prenatal elements are pertinent to both L&D, and NICU according to recommendations by ACOG and IHC clinician review. Approximately one third of the elements in Labor and Delivery are also relevant in NICU. Research by Shortell et al. showed how coordination and communication among clinicians across settings resulted in greater efficiency and better clinical outcomes [8]. Variation in data across systems occurs in part from the lack of standard datasets for specific domains. Although there are efforts by many including IOM and Health Level Seven (HL7) to standardize EHR functions and clinical message format, the specific content for perinatal data varies even within each institution [4]. Professional societies such as ACOG (American College of Obstetricians and Gynecologists); AAP (American Academy of Pediatrics), AWHONN (Association of Women s Health, Obstetric and Neonatal Nurses) have published their own versions of forms and lists of guidelines [9,10]. These are often used to develop the element content of the computerized record data sets for specific areas of obstetrical and perinatal care. Discussions with representatives at ACOG and the Department of Maternal and Child Health Bureau (MCHB), confirmed that there is no single national standard for content for a perinatal longitudinal patient record [11,12]. Sharing information about prenatal risk assessment allows for anticipatory planning, individualized education, and appropriate referral. Information used for outcomes of risk assessment, facilitate the refinement of guidelines by which the effectiveness of the care can be evaluated. Maloni, et al. point out that comprehensive care involving multidisciplinary caregivers can lead to an improvement in maternal and infant care outcomes. 13 The care needs to be coordinated from the prenatal period through the first year of infant life when the needs of the mother, infant, and family adjustment are the greatest. This care must be combined with information systems that allow the effective sharing of patient needs and treatment across the perinatal care continuum, further justifying an integrated longitudinal record. Methods This paper reports our descriptive study to detect the extent of the disparity between common data elements in three separate data systems: Prenatal, L&D, and NICU. To begin, all data elements used in three data systems were identified. A subset of thirty representative elements was selected. Each element was evaluated for relevancy to the longitudinal EHR using knowledge engineering methods to devise a model for data relevancy, compare to national standard references, and achieve consensus of clinical domain experts. Elements for each system were obtained from prenatal record forms, the L&D system chartable item dictionary, and SQL database queries for appropriate NICU EMR data elements. The overlap of these elements is illustrated in the Ven diagram shown in Figure 1. The Prenatal record contains approximately 400 unique data elements, over 2800 elements are found in the L&D system and approximately 2000 elements are in the NICU system. The high number of data elements in the L&D system reflects the inconsistent data models used by that system, which blurs the distinction between data element versus data value. Sample data element election criteria included: 1) commonly used data elements from th current prenatal record, that are relevant across the prenatal, delivery, discharge care continuum, and, 2) data elements, which if missing or contradicting other sources of clinical documentation, would have a high potential to lead to a medical error. External standard references included data sets from Perinatal Nursing from AWHONN and the Guidelines for Perinatal Care from AAP and ACOG. The panel of clinical experts at IHC that refined and validated the sample data set, was comprised of two physicians (medical directors of the IHC Women s and Newborn Clinical Program) and three nurses (directors and managers of Women s and Newborn Nursing, Data Management, and Health Data Dictionary teams). The elements were evaluated to see if they were present in each clinical system and if they matched in structure and meaning across the three systems. When comparing the Prenatal to L&D, the Prenatal would be considered the source system and the L&D would be the destination system. The term More Granular would be used when the options to define the element are either more numerous or contain more information about the data value it represents. For example, Chronic Hypertension is more granular than Hypertension. Dates that describe when a laboratory was ordered would make that test element more granular. The listing of several specific drugs used would be more granular than the term Drug Use. The results of the comparisons were validated by clinical experts. Summary descriptive statistics demonstrate the proportions of matches, mismatches, and missing data elements ( total of 90 comparisons). If the systems were perfect-ly integrated and shared common concepts, the result would be perfect matches in all 90 comparisons. A perinatal nurse administrator and a neonatal nurse practitioner, very familiar with the use of these clinical systems, validated the element match results. 1054

3 Table 1: Conceptual comparison of data elements. A categorization scheme explaining the nature of the discrepancies when elements were compared between systems. Results Substantial conceptual differences in the elements were found. Only five of 30 elements (17%) had the same meaning and structure across all three data systems. Out of 90 comparisons tested, 20 (22%) comparisons resulted in a complete match between two systems. Of the 30 data elements used, there were 10 (33%) unique data elements where there was at least one match between systems. The results of the conceptual comparisons are found in Table 1. Figure 2 shows the proportional distributions of matched data elements between three perinatal systems. The low match proportions illustrate a problem larger than just data-to-data mapping. Currently many instances remain in the NICU application that require the users to review paper printouts and re-enter prenatal patient information into a free text box. When the NICU system is involved, more than one third of the elements tested are manually entered into non-coded text boxes. Taking coded data from one system and transforming it into text in another system often results in loss of meaning, and decreases the clinician s face time with the patient, while introducing potential for errors because of human interpretation and manual data entry. This study shows the extent of the variation in concepts when different clinical data systems are developed without using a common shared data dictionary. Data gathered in the prenatal stage such as maternal and family medical history, laboratory test results and risk factors should be accessible to appropriate clinicians across the perinatal care continuum. Because this study focuses on the process by which a larger more complete data set for a longitudinal patient EHR can be built, a small test data set needed to be defined and used for the comparison. This data set is a representative list to test the effectiveness of the existing systems in passing key and relevant information. Discussion IHC clinical application development teams have spent eight years refining and building consensus on standardized content and tying concepts to a common Health Data Dictionary. Much progress has been made, yet interfacing legacy clinical information systems has proven to be a larger challenge that was earlier anticipated. It is very time-consuming to deal with concepts across different contexts and granularities. Difficult decisions about correct interpretations of meaning must be frequently made so that these assumptions can be minimized. Although efforts are on-going, the electronic medical record still does not contain all the information that is available and envisioned. Currently huge opportunities exist to reduce the charting error if a common data dictionary is used. Our research provides a pilot study and process for building consensus on core concepts. This brings to the forefront content management issues that affect data quality, patient safety, reduced costs of integration and efficient use of clinician time. Hindsight from the development during different clinical systems will benefit IHC and others who are engaged in the implementation of a longitudinal patient record. From our analysis, we have formulated seven ontological guidelines that may assist in the effective definition and maintenance of core concepts in a longitudinal patient EHR: 1. Agree on necessary clinical core data elements 2. Determine appropriate personnel to define the element values (domain of concepts for each data element) 3. List appropriate personnel/instruments to enter the element values (instances of concepts reflecting the conditions observed in the patient) 4. Identify optimal time to capture the element values (stage of the clinical process where the individual concepts are first observed or measured by a clinician) 5. Distinguish potential users of the data element by role 6. Associate potential uses of the element both at point of care and later for outcomes research 7. Discover effective utilization of the element with other elements to derive new elements and/or concepts We concur with the formal ontology spoken of by Stead and others who describe methods used by system developers to make it easier for the users to interact with the computer [14]. Ontologies can facilitate consensus in understanding a domain between different partners, along with preventing the meaning of concepts from drifting over time [15]. The problem of various meanings in disparate systems has important implications for clinical monolithic systems that have been created from acquisitions of separately developed modules. Unless a disciplined approach using a common data dictionary was used, concerns are warranted that those systems may face similar integration and data quality problems. The use of archetypes is another proposal to close the semantic gap between generic data models such as 1055

4 Figure 2 - Proportional distributions of matched data elements between perinatal data systems: Prenatal to L&D, Prenatal to NICU, L&D to NICU. the HL7 Reference Information Model and the medical terminology [16,17]. Archetypes allow modeling of domain concepts external to the system data model and preserve meaning that would be lost if the semantic representation were embedded into the information system like what was done with the L&D system. Effectively implementing a longitudinal EHR requires the use of a common data dictionary. To the degree that a data dictionary is populated with clinical concepts drawn from standard medical terminologies, data variation will be decreased and interoperability within the institution and across institutions will be increased. Conclusion Our study showed a substantial degree of data variation across clinical data systems in the Prenatal, L&D, NICU care continuum. Only 17% of our sample data elements had the same meaning and structure across all three data systems. Only 22% of the data element comparisons resulted in a complete match between two systems. The use of centralized concept-based data dictionaries has a substantial effect on data sharing across clinical data systems. When variation of data structures and concept meaning exists across systems, decision support tools cannot be consistently triggered, and outcomes research is compromised. The need to use consistent data models and define relevant concepts in a shared data dictionary should be addressed early in the development process. Acknowledgments We acknowledge the expertise and insight of the IHC data managers, analysts and clinicians in the Women and Newborn Clinical Program. References [1] Clayton PD, Narus SP, Huff SM, Pryor TA, Haug PJ, Larkin T, et al. Building a comprehensive clinical information system from components. The approach at Intermountain Health Care. Methods Inf Med 2003;42(1):1-7. [2] Twede M, Gardner R, Hebertson R. A PC-based system for intrapartum monitoring. Contemporary OB/GYN 1984(Special Issue -- Technology 1985):24: [3] 3M HIS. Making Sense of the Data: Using a Medical Data Dictionary to Integrate, Share, and Understand Clinical Data. White Paper. Salt Lake City, Utah: 3M Health Information Systems; [4] Institute of Medicine, Committee on Data Standards for Patient Safety, and B.o.H.C. Services, Key capabilities of an electronic health record system: letter report. 2003, Washington, D.C.: National Academies Press. [5] Rosenal TW, Forsythe DE, Musen MA, Seiver A. Support for information management in critical care: a new approach to identify needs. Proc Annu Symp Comput Appl Med Care 1995:2-6. [6] Institute of Medicine (U.S.). Committee on Improving the Patient Record., Dick RS, Steen EB. The computer-based patient record : an essential technology for health care. Washington, D.C.: National Academy Press; [7] Institute of Medicine. Crossing the Quality Chasm: A New Health System for the 21st Century. Washington D.C.: National Academy Press; [8] Shortell SM, Zimmerman JE, Rousseau DM, Gillies RR, Wagner DP, Draper EA, Knaus WA, Duffy J. The performance of intensive care units: does good management make a difference? Med Care 1994;32(5): [9] American Academy of Pediatrics (AAP), American College of Obstetricians and Gynecologists (ACOG). Guidelines for Perinatal Care. 5th ed. Elk Grove Village, IL , Washington, DC: American Academy of Pediatrics (AAP), American College of Obstetricians and Gynecologists (ACOG),; [10]Simpson KR, Creehan PA, (AWHONN) AoWsHOaNN. Perinatal Nursing. 2nd ed. Philadelphia: Lippincott Williams and Wilkins; [11]Williams S. Vice President Education, American College of Obstetricians and Gynecologists (ACOG),. Interview by author, 2 July 2003, Washington D.C. [12]Kogan MD. Director of Office of Data and Information Management, Maternal and Child Health Bureau (MCHB), Health and Human Services (HHS). Interview by author, 11 July 2003, Rockville MD 1056

5 [13]Maloni JA, Cheng CY, Liebl CP, Maier JS. Transforming prenatal care: reflections on the past and present with implications for the future. J Obstet Gynecol Neonatal Nurs 1996;25(1): [14]Stead WW, Miller RA, Musen MA, Hersh WR. Integration and beyond: panel discussion. J Am Med Inform Assoc 2000;7(2): [15]De Keizer NF, Abu-Hanna A, Zwetsloot-Schonk JH. Understanding terminological systems. I: Terminology and typology. Methods Inf Med 2000;39(1): [16]Heard S, Beale T, Freriks G, Mori AR, Pishev O. Templates and Archetypes: how do we know what we are talking about?: HL7; /12/2003. [17]Health Level Seven I. HL7 Data Model Development. See Address for correspondence Val N. Hicken 4646 W Lake Park Blvd Salt Lake City, UT 84120, USA lpvhicke@ihc.com 1057