S 4.1 STANDARDS FOR CARDIOVASCULAR DATA: A Concept Paper

Karl Hammermeister, MD, FACC

Staff Cardiologist, Denver VA Medical Center
Professor of Medicine, University of Colorado Health Sciences Center

E-mail: khammer@sembilan.uchsc.edu

Attribution: This paper represents a synthesis of concepts developed by the author in the course of his work with the Department of Veterans Affairs and the American College of Cardiology Database Committee. However, the concepts presented here do not necessarily represent official statements or policy of either the Department of Veterans Affairs or the American College of Cardiology.

Background

Despite the impressive advances in medical care in the last fifty years, medical information management has lagged seriously. Consider only that the primary method of recording the most important information about an episode of care is frequently handwritten, free-form notes. No other large-scale enterprise has such a woefully inadequate information system. A major contributor to this state of affairs has been the lack of information standards, such as agreement on the minimum information content for a specified care episode, the wording of the data elements, the wording of the definitions, and standardization of data quality assurance tools.

This state has been acceptable to the health professional so long as care was viewed as a single episode documented manually by a single care provider. However, health care is now provided by teams of care providers, and patients are demanding continuity to their care across multiple episodes by many care providers. The combination of multiple care providers with the mobility of the population requires the transfer of patient clinical information between care providers. Previously this has been done by copying portions of the medical record, creating narrative summaries, or mailing the entire record -- all expensive, inefficient, and subject to frequent failures. In the not too distant future, all clinical data will be transferred electronically. For this to occur effectively without misinterpretation of clinical concepts by the receiving care provider, there must be standardization of minimum information content, data elements and their definitions.
Furthermore, the ongoing automation of the medical record requires special attention to information standards. Much of automation efforts to date have focused on the computer storage and retrieval of narrative information, which is clearly a step forward from illegible, hand-written notes in charts that are frequently lost. However, there are at least three critical disadvantages to this approach: 1) because there is no agreement on information content, the writing of free-form clinical notes frequently results in missing data; 2) the electronic analysis of such narrative information still requires expensive, time-consuming, unreliable manual data abstraction, and 3) the automated delivery of real-time, patient-specific clinical decision aids based on the characteristics of the patient and her/his illness is not possible. The solution is to record clinical information as encoded data (information expressed as continuous or categorical variables). However, this requires the a priori identification of information content (i.e., what data elements), as well as standardization of the wording of data elements and definitions.

Current Information Standards

Information standards for health care currently exist on at least four different levels or formats: 1) electronic standards, such the DICOM imaging standards, 2) coding systems such as ICD-9-CM and CPT codes, 3) clinical lexicons, and 4) published data elements and definitions for individual clinical databases. Clearly, electronic standards are essential, but are not directly linked to clinical decision making at the interface of the care provider and the patient. Coding systems, while useful for billing and other administrative activities, have not been widely used by care providers -- perhaps because they lack the clinical detail and temporal relationships that care providers use for clinical decision making. Clinical lexicons are attempting to fill the standardization void by linking terms with similar meanings -- i.e., a type of thesaurus -- assuming that most clinical information will be recorded in a narrative, free-form format. Furthermore, most clinical lexicons lack detailed definitions of terms, and none specify minimum information content for an episode of care.

Focus Standardization Around Clinical Decision Making

The primary focus should be directed at standardization of language and minimum information content used by care providers to make the clinical decisions required for patient care. It is assumed that cost considerations, clarity of meaning, and the need for efficient mechanisms for transportation and analysis of data will drive the health care professions towards an automated medical record that uses encoded data. Furthermore, it is assumed that much of the data input into the electronic medical record will be by the care provider interacting with an input device, such as a hand-held, touch screen.

This paper discusses five topics relevant to the need and processes for achieving standardization of cardiovascular clinical information: 1) why standards are needed and what benefits may accrue, 2) what is it that must be standardized, 3) proposed steps to the standardization process, 4) the First International Cardiovascular Data Standards Conference, and 5) the need for a permanent structure for ongoing development and revision of cardiovascular data standards.

The successful establishment of cardiovascular information standards will accrue benefits in at least four areas: 1) patient care, 2) clinical quality improvement, 3) cardiovascular research, and 4) any aspect of the health care industry dependent on efficient handling of clinical information.

Patient Care Activities

Value of a common language: The primary purpose of language is to exchange information or express feelings. These purposes are best accomplished when all parties to the information exchange have a common understanding the of the information content or meaning of the words. A critical examination of the language of cardiovascular medicine reveals numerous opportunities for miscommunication, because terms may have different meanings among care providers. Consider "unstable angina": to some it means angina occurring only at rest; to others it may mean a recent exacerbation (usually not further defined) of chronic angina; to others it may mean new onset angina; and to some all three meanings are acceptable even though they identify prognostically and pathophysiologically different groups of patients. Important and commonly used terms, such as "congestive heart failure" and "left ventricular dysfunction" are similarly subject to a wide range of interpretations. A consensus on definitions for such data elements should greatly reduce miscommunication between care providers with subsequent benefit in patient outcomes.

Reduce costs of health care: The current archaic manual modes of transferring patient information are so slow and unreliable that repetition of expensive tests and other types of evaluations is common when a patient presents for care in another setting. The storage of critical patient data in a secure electronic database accessible to authorized care providers is likely to not only improve clinical decision making, but also reduce health care costs by the elimination of repetition of previous evaluations.

The costs of the delivery of health care account for about 14% of the gross national product in the United States. The administrative, non-care functions of health care are estimated to account for about 25% of health care costs in the United States. Information management is a major, essential component of virtually every aspect of the delivery and financing of health care. Our archaic, nonstandardized information systems are significant contributors to these costs.

Facilitate Clinical Quality Improvement

Much of clinical quality improvement is now based on the retrospective abstraction of medical records for clinical indicators, previously agreed upon processes of care considered appropriate to the illness, and/or adverse outcomes of care. Because there is no agreement on information content for an episode of illness, process of care information is generally poorly documented in the written medical record. For adverse outcome rates to have credibility with care providers, they must be adjusted for severity of illness. However, there is no consensus on what constitutes appropriate measures of severity of illness, and there is no agreement on the wording and definitions of data elements commonly used to describe patient risk. Standardization and electronic processing of cardiovascular information would facilitate clinical quality improvement, improve its credibility with care providers, and reduce its costs.

Facilitate Cardiovascular Clinical Research

In the United States federal and foundation grant support for clinical research has virtually dried up, except for that supporting a few large-scale clinical trials. In the future, clinical and health services research will be increasingly dependent upon databases constructed primarily for other purposes. The largest health-related databases currently available are those used for billing and other administrative purposes. Although they have been widely used for clinical and health services research, their lack of clinical detail and inability to temporally sequence processes or outcomes of care have significantly detracted from the acceptance of the results drawn from them. Virtually all other clinical databases with richer clinical detail are either regional or confined to a single institution and encompass far fewer patients, such that conclusions drawn from them may not be representative outside of their specific context. Attempts at combining information from several clinical databases have been largely frustrated by a lack of information standards. Standardization of minimum cardiovascular information content, data elements, and definitions would greatly facilitate the combining of data from local and regional databases to achieve credible, representative conclusions.

Advantages to Industry

While the information of health care encompasses a very broad range, at the core are the descriptors of the patient relevant to outcomes of care, her/his preferences for care, the nature and severity of her/his illness, the care (processes) provided, and the outcomes of that care. Standardization of these aspects of health care information should greatly reduce costs and enhance profitability for all aspects of health care -- particularly the cardiovascular informatics industry, managed care, and insurance companies.

Cardiovascular informatics industry: A large informatics industry has arisen to support cardiovascular information management -- ranging from manufacturers of large, networked, enterprise-wide computer systems for multi-hospital/clinic health care systems, to the developers of software specific to a single procedure. At present, communication between systems of different manufacturers is not possible, or possible only after prolonged, expensive custom software and hardware development. The value and demand for these products would be greatly enhanced if communication could be facilitated through information standards.

Managed care: Managed care -- more than any other health care delivery structure -- is dependent upon valid, transportable clinical information. Managed care leaders are proposing to select and assess care providers by profiling the costs and quality of their care. This requires the aggregation and comparison data from multiple practice settings. With out information standards, this process will be severely restricted in scope and lacking in credibility.

Insurance companies: A major portion of health care insurance companies' expenditures outside of payments for care is spent on information management. Included are the documentation of the need for care through preapproval processes that collect information on the clinical characteristics of the patient and her/his illness, and the documentation of the care provided. These activities are both a source of great frustration to care providers and inconvenience (or worse) to patients, because of the time required and errors made due to archaic information management. Information standardization would initially facilitate this process and ultimately do away with it, as the databases of the automated medical record could be accessed electronically for this information.

Minimum Information Content

Perhaps the most important standardization issue, and simultaneously the most difficult to solve, is the problem of what should constitute minimum information content. Information content is critically important because it defines adequate documentation of and/or communication regarding an episode of care. It will be difficult to achieve consensus, because of the well-documented, marked variation in the processes of care for patients with apparently similar illnesses between groups of care providers. Clinical practice guidelines, the development of which has been largely driven by these process variations, provide a structure for defining information content. Clinical practice guidelines are broadly disseminated statements regarding the processes of care that should be provided for specific clinical situations; they are often based on scientifically-valid studies showing improved patient outcomes with the specified process of care. Therefore, a clinical practice guideline describes both the characteristics of the patient and her/his illness, as well as the processes of care to be provided -- i.e., the minimum information content for a single clinical decision. What remains to be accomplished is to link a series of clinical decisions into a logical sequence that represents the usual pathway of care for a group of patients presenting with a common illness or symptoms.

FIGURE1.GIF illustrates a pathway of care consisting of linked clinical decisions leading to a recommendation for coronary arteriography for patients presenting with chest pain or other symptoms suspected to be due to myocardial ischemia; this is taken from the VETERANS HEALTH ADMINISTRATION CAREGUIDE (Care Aided and REfined with GUIDElines) FOR ISCHEMIC HEART DISEASE, a set of clinical practice guidelines collated by a multidisciplinary working group. The oval at the top describes the population of patients under consideration, all of whom come from other clinical decision nodes in other pathways denoted by the numbers; the hexagons represent the dichotomous clinical decision nodes; the rectangles represent recommended processes of care; the small circles containing numbers are direction finders indicating that the pathway goes to the hexagon or rectangle denoted by the number. Each hexagon denoted by a number with an asterisk has an associated table of clinical decision criteria as illustrated in TABLE1.GIF for node 3.04, "Priority Candidate for Coronary Arteriography". Seven additional pathways to encompass the full range of care for patients with ischemic heart disease from initial presentation with chest pain to cardiac rehabilitation have been completed or are in process: 1) emergency department evaluation of chest pain, 2) early treatment of acute myocardial infarction, 3) revascularization, 4) the initial ambulatory care evaluation of chest pain, 5) the noninvasive evaluation in the ambulatory care setting, 6) the routine outpatient follow-up, and 7) cardiac rehabilitation. The criteria for the more than 40 clinical decision nodes in these eight pathways should serve as the basis for the determination of the minimum information content for the full range of care of ischemic heart disease patients.

The Data Object Concept

With the advent of comprehensive electronic data management, including direct entry into an a automated medical record by care providers, the concept of an item of information can be expanded well beyond just the data element and its definition without additional data input effort. These additional data characteristics include the time and location of data input and the identity of the entering individual (an essential part of data security), whether the data source is primary with the entering individual or is a default from another source, the distinction between an unknown and missing value, data completion requirements, and built-in data quality checks. The data object concept, in which each item of information includes some or all of the above characteristics (FIGURE2.GIF), is analagous to the principle of object oriented programming (program once; use many times). Once specified, the data object could be used repetitively as the need arises. For example, the application of the data object concept would allow information on angina severity at multiple episodes of care to be recorded from a single data input field for each episode, as information on date, time, location, and care provider are automatically added by the computer. Thus, the standardization process should include all of the components of the data object.

Data element: The data element is the question being asked. Using an analogy from particle physics, the data element is the fundamental particle of information (quark), which is unique and cannot be easily divided. The wording of the data element, perhaps the most critical component of the data object, is that which will appear on the data input screen (or data form for manual data entry). The concept conveyed by these words is critical to information reliability across observers and to consistency by a single observer at multiple care episodes. As familiarity with the data input format grows, these words will usually be the sole trigger of the response from the data recorder.

Data element responses: The allowed responses to the question (data element) are almost as critical to high quality data as the question itself -- particularly for categorical variables other than "yes" and "no" responses. It is usually preferable to order the responses to categorical variables (e.g., the Canadian Cardiovascular Society classification for severity of angina) if possible. If the information recorder is to pick from a list of categorical responses, it is preferable to design the responses to be mutually exclusive so that only one can be selected, as opposed to allowing the selection of one or more responses. The latter allows for ambiguities; if no response is chosen, it is not possible to determine whether none of the responses applied to the question or that the data recorder simply skipped the question.

Continuous variables should have the number of significant figures and/or number of decimal places specified in the software, or on the data form by providing a rectangle for each digit to be recorded with the decimal point shown. Units must also be specified for continuous variables. When software is used for data input, it is preferable to calculate some variables from their primary data (e.g., age from birth date), as opposed to recording each separately.

Data definitions: The writing of comprehensive, unambiguous definitions for data elements is difficult and time consuming. Revisions are often required following initial trials of data entry and analysis, making pilot testing an essential feature. We propose that each data element have a short definition that automatically appears at the bottom of the screen when the cursor is placed over the data element and its responses, and a long definition that can be brought up by a click or double click over the data element and its responses.

Audit trail: An audit trail is the automatic recording of the identity of the data recorder or data reader and the date, time and location that the data file was opened. The audit trail entry serves two critical functions: 1) it automatically stores in the database the identity of the data recorder and the time of the observation, analogous to the dating and signing of notes in the current medical record (undated and unsigned notes become virtually extinct), and 2) it enhances data security. It has been shown that if users of automated medical records know that their identity can be traced if they record or read data, confidentiality and security breaks are much less likely to occur.

Data source: A major advantage of the automated medical record built around standardized, encoded data elements, is that it can markedly reduce redundancy in data recording -- an activity that currently adds very substantially to the cost of health care. A response to a data element entered once can be the default value each time the same data element is requested again. The data recorder has the option of keeping the default value or recording a new response if it has changed since the initial recording. However, the interpretation of the data element may be dependent on whether its source was default from a previous episode of care or primary with the present episode of care. Similarly, interpretation may vary if the source was the patient interacting with a computer program or was the care provider taking a history in the usual fashion. Therefore, these data source characteristics need to be included as components of the data object.

Internal data quality checks: One of the major strengths of direct data entry into a computer is that real-time data quality checks can be built into the software that prompts the data recorder to reconsider her/his entry if it fails to meet a quality test. The most commonly used are warning and refusal ranges for continuous variables, and logical cross-checks between variables. For a patient undergoing coronary artery angioplasty, the recording of a birth date of 11/15/75 (20 years of age) might result in the following warning to the data recorder: "It would be unusual for a patient of this age to undergo PTCA; please check your entry of birth date"). Similarly, the recording of a birth date of 11/15/90 for a patient undergoing PTCA in a non-pediatric setting would result in a refusal to accept the birth date and the following message: "This is not an allowable birth date; please record another birth date". An example of a logical cross-check might be the display of a warning if an attempted angioplasty was recorded for a coronary artery segment for which the degree of stenosis was recorded as 20%.

Representation of unknown and missing values: The definitions of unknown and missing values have been the source of considerable confusion, and their representation in databases has been nonuniform. We propose the following definitions: "unknown" means that the observation or test has not been performed; while "missing" means that the data recorder either gave no attention to the data element (i.e., skipped it) and/or did not determine whether the observation or test had been performed. For example, the information content is quite different if a patient has never had an assessment of left ventricular function as opposed to the data recorder skipping over the data element. This problem is exacerbated by the tendency of care providers to not record normal values or pertinent negatives, which has frequently resulted in the substitution of normal values/responses for missing values. Careful attention to minimum data content as described above, and the distinction between missing and unknown values should ameliorate these current database problems.

Data completion requirements: The incomplete recording of data is a universal problem and creates significant analytic challenges beyond the missing information content. Many databases are now specifying minimum data completion rates that vary with the perceived importance of the data content. Three levels of data completion proposed: 1) 100% completion for a few mandatory data elements before the record can be submitted to the database (e.g., birth date and gender), 2) > 95% completion of the core data set defined as the minimum information content required for the clinical decision at hand (vide supra), and 3) an extended data set which would require > 80% completion in both the field and record or consideration of deletion of the data element from the data set.

Data object identifier: This would be a code (perhaps limited to eight characters initially until more flexible database software comes into wide-spread usage) that uniquely identifies the data element. The data object identifier would be combined with the date and time of recording and the identity of the recorder to uniquely identify a fundamental piece of information (quark). This identifying information could be used to automatically store the information quark in a predefined locus in the database. This would greatly reduce the effort in both setting up database management software and in the combining of information from two or more databases.

Data Security and Protection of Confidentiality

The advent of the electronic medical record has raised considerable concern among both care providers and the public at large over possible breaks in confidentiality of sensitive clinical information (e.g., HIV status). A few widely publicized incidents have fueled the fires of these concerns. What is not widely realized is that despite the greater ease of accessibility to and transfer of electronic information, the potential for protection of confidentiality is greater with electronic records than with paper records. While data security is a whole area of endeavor that cannot be detailed here, several principles seem to be emerging: 1) the requirement for dual passwords, 2) the knowledge by database users that their viewing or recording of information can be tracked, 3) the stripping of patient and care provider identifiers from databases used for noncare functions, 4) the encryption of data -- particularly during electronic transmission, and most importantly, 5) the careful education of users of the database in regard to the importance of and methods for data security.

A six-step standardization process is proposed (FIGURE3.GIF): 1) agreement on pathways of care consisting of linked clinical decision nodes (FIGURE1.GIF), 2) identification of minimum information content, 2) creation of a database of currently used data element wordings and definitions, 3) consensus process to standardize all components of the data object, 5) publication of results, and 6) continuous up-dating and revision.

Creation of Pathways of Linked Clinical Decision Nodes

This is a complex process that requires both a knowledge of relevant clinical practice guidelines and extensive clinical experience in the topic area. The guidelines can be used to identify the major clinical decisions that must be made for an episode of care. Clinical experience is needed to link the decision nodes together in temporal sequence that reflects the way most clinicians provide care.

Identification of Minimum Information Content

The second step in this process is to specify the minimum information content. The major clinical decision nodes making up pathways describing the care of the patient (FIGURE1.GIF) provide a logical basis for defining minimum information content. Clinical practice guidelines describe the characteristics of patients who should receive a specified process of care. These patient characteristics are the criteria for the clinical decision nodes (TABLE1.GIF). Conversion of these decision criteria to the data elements, responses, and definitions of the minimum information set is relative straight-forward given adequate knowledge of the scientific studies from which the practice guidelines were drawn.

Database of Currently Used Data Element Wordings and Definitions

To facilitate the review of existing wordings of data elements and their definitions, we propose the creation of a database of currently used cardiovascular data elements and definitions. The APPENDIX illustrates one record (for congestive heart failure) from a database that we are compiling from several major, actively used cardiovascular databases: the Veterans Health Administration (VHA) Continuous Improvement in Cardiac Surgery Program and the VHA Processes, Structures, and Outcomes in Cardiac Surgery Study (VHA CICSP/PSOCS); the New York Cardiac Surgery and Angioplasty Reporting System (NY CSRS); the Society for Thoracic Surgery database (STS) ; the Health Care Financing Administration's Cooperative Cardiovascular Project (HCFA CCP); the American College of Cardiology PTCA database (ACC PTCA), and the Society for Cardiac Angiography and Intervention (SCAI). Also included are records from MEDISGROUPS (a large generic database), a standard medical dictionary (Dorland's Definition), and the VHA clinical lexicon. Several major cardiovascular databases (including those of the Northern New England Cardiovascular Study Group and Duke University Medical Center) remain to be included.

Consensus Process

The third step is a consensus process to agree upon wordings of the data elements, possible responses to the data elements, their definitions, associated quality checks, and other components of the data object. A number of working groups of five to seven cardiovascular clinicians with interests in databases and informatics could be formed -- each focusing on a specific limited topic. Their first task would be to specify information content by working from care pathways or clinical algorithms, such as illustrated in FIGURE1.GIF. A technical support group would then prepare the database of all possible wordings of data elements and definitions for each data concept. This database would be used by the working groups to arrive at a consensus on the wording of each data element, definition, and quality control characteristics.

Publication

The end-product of this standardization process, a dictionary of consensus-derived standardized data elements and definitions together with other data object characteristics lie in the public domain, as language should belong to all -- particularly to the patient. Therefore, this data dictionary should be published in major professional journals and made available in electronic format to all at cost.

Continuous Up-Dating and Revision

Both medicine and its language are continuously evolving. Any attempt at information standardization will be out of date within a few years or less. Furthermore, there is likely to be a steep learning curve in the standardization process -- meaning that pilot testing leading to early revisions will be essential. Therefore, a permanent structure for the continuous revision and up-dating of the information standards will be necessary.

Because data are so closely linked to processes of care, data standardization will be a complex process requiring special attention to achieving consensus where possible. The first step towards data standardization is agreement on what standards are necessary, the processes by which standardization is undertaken, and the structure to accomplish these goals. These will be three of the primary goals of the First International Cardiovascular Data Standards Conference to be held June 27 - 28, 1996, at Heart House, the American College of Cardiology Headquarters in Bethesda, Maryland.

Participants

Representatives from all of the major cardiovascular professional societies including, the American College of Cardiology, the American Society of Echocardiography, the Canadian Cardiovascular Society, the European Society of Cardiology, the North American Society for Pacing and Electrophysiology, the Society for Cardiac Angiography and Intervention, and the Society for Thoracic Surgery, will be invited. Representatives from the major cardiovascular databases, including: The American College of Cardiology databases, Duke University Medical Center, Emory University, the Health Care Financing Administration's Cooperative Cardiovascular Project, Mayo Clinic, New York Cardiac Surgery Reporting System, the Northern New England Cardiovascular Study Group, the Society for Thoracic Surgery cardiac surgery database, and the Veterans Health Administration's Continuous Improvement in Cardiac Surgery Program and the CAREGUIDE for Ischemic Heart Disease Project, will also be invited. Developers of clinical lexicons, such as SNOMED and the National Library of Medicine, should participate, as there is much to learn from these efforts. Existing standardization bodies, such as ANSI, should also be represented. Finally, there needs to be a broad representation from industry, which will be a major user of the standardization products and, hopefully, provide the financial support for the standardization process. These should include manufacturers of cardiovascular hardware and software, vendors of cardiovascular databases, managed care companies, and insurance companies.

Pilot Project: Standardization of Variables Predictive of Short-Term Outcomes from CABG and PTCA

A fourth major goal of the First International Cardiovascular Data Standards Conference will be a pilot project: the standardization limited data sets for the prediction of short-term outcomes from PTCA and CABG. These topics were chosen because of ongoing cardiovascular data standardization efforts led Robert Jones, MD, and David Pryor, MD (CABG) and Peter Block, MD (PTCA).

Conference Organization

Steering Committee: At a preliminary organizational meeting January 19, 1996, in Denver, a Conference Steering Committee was formed: David Ataide, Hewlett-Packard Corporation; Peter Block, MD, Portland, Oregon; Ross Davies, MD, Ottawa, Canada; Karl Hammermeister, MD, Co-Chair, Denver, Colorado; Mark Hlatky, MD, Co-Chair, Stanford, California; Robert Jones, MD, Durham, North Carolina; and William Weintraub, MD, Atlanta, Georgia.

Working Groups: Five working groups have been formed: 1) Domains of Standardization: the Data Object Concept (Chair: Karl Hammermeister, MD), 2) Processes to Achieve Standardization (Chair: William Weintraub, MD), 3) Variables Predictive of Short-Term Outcome Following CABG (Chair: Robert Jones, MD), 4) Variables Predictive of Short-Term Outcome Following PTCA (Chair: Peter Block, MD), and 5) Planning a Permanent Structure for Ongoing Standards Development and Revision (Chair: Robert Califf, MD). The Working Group Chairs will select five to seven subject matter experts to assist them in the preparation draft documents to be presented at the Conference.

Conference agenda: The one and one-half day conference will begin with presentations on current status and overall direction for standardization by several internationally-recognized, subject-matter experts. This will be followed by initial presentations from the four Working Groups. The afternoon of the first day will be spent in small group deliberations of each of the working groups; all conference attendees will have an opportunity for individual input at these sessions. There will be more in-depth reports from each of the Working Groups on the morning of the second day, followed by responses from the leaders of the major professional societies.

Products of conference: The principle product of the conference will be a document delineating areas of agreement, as well as areas where consensus could not be achieved. The major topics to be covered in this document include the mission and goals for standardization of cardiovascular data, which standards are important and achievable, the processes of standardization, proposed standards for variables predictive if short-term outcomes from CABG and PTCA, and the establishment of a more permanent structure for standardization in new topic areas and ongoing revision and updating. It is proposed that this document will be published simultaneously and in identical form in the journals representing the major professional societies participating in the Conference.

It should be clear from the foregoing that the data standardization process is complex and laborious; a permanent standards center will be necessary to support future efforts at standardization of cardiovascular data elements in other topic areas and ongoing revision of data elements previously standardized. Therefore, we propose the formation of an international cardiovascular data standards organization, which would be a collaborative organization between cardiovascular professional societies, industry, existing standardization groups, and other interested parties.

Mission

The mission of this organization would be to facilitate the development of international cardiovascular clinical data standards with a particular focus on variables used by care providers and patients in making clinical decisions.

Specific Responsibilities

The responsibilities of this organization would include: 1) creation and revision of pathways of care (FIGURE1.GIF) and criteria lists (TABLE1.GIF) from published clincal practice guidelines, 2) collation of existing data elements and definitions into a database of databases, 3) support of the activities of individual working groups as described above, 4) facilitation of the development of consensus through supporting meetings of working group chairs and soliciting additional input from interested organizations, and 5) assistance in writing and publication of consensus reports.

TABLE1.GIF
Criteria for Determining Patients for Whom Coronary Arteriography Should Be Recommended without Further Noninvasive Testing.
To view this picture full-size or to download it just click on the thumbnail.

FIGURE1.GIF
An Example of a Pathway of Care with Major Clinical Decision Nodes from the Veterans Health Administration CARE GUIDE (Care Aided and REfined with Guidelines) for Ischemic Heart Disease.
To view this picture full-size or to download it just click on the thumbnail.

FIGURE2.GIF
Components of a Data Object.
To view this picture full-size or to download it just click on the thumbnail.

FIGURE3.GIF
Proposed Steps for the Process of Standardization of Cardiovascular Data.
To view this picture full-size or to download it just click on the thumbnail.

EXAMPLE OF A RECORD (FOR CONGESTIVE HEART FAILURE) FROM A DATABASE OF CURRENTLY USED CARDIOVASCULAR DATA ELEMENTS AND DEFINITIONS

DEFINIT1.GIF DEFINIT2.GIF