SPECIAL
ARTICLE
Framework for improved communication: The Dermatology Lexicon Project Art Papier, MD,a Robert J. G. Chalmers, MB, FRCP,b Jennifer A. Byrnes, MLS,a and Lowell A. Goldsmith, MD, MPHc Rochester, New York, Manchester, United Kingdom, and Chapel Hill, North Carolina A standardized dermatology vocabulary is central to our collective ability to gather clinical information consistently for patient care, to retrieve information for research or disease management, as well as to conduct outcomes analysis for quality improvement. The deficiencies of current classifications have been recognized by the National Institute of Arthritis and Musculoskeletal and Skin Diseases (NIAMS) in their decision to fund the development of a standard complete reference terminology for dermatology. The Dermatology Lexicon will have a major impact both on clinical care and on dermatologic research. Accuracy in medical records and telemedicine communications will be improved. Reliable terminology will provide for more complete and consistent documentation and data aggregation and reporting. In order to create a useful, sustainable lexicon, involvement of the dermatology community is essential. (J Am Acad Dermatol 2004;50:630-4.)
I
n September 2001, the National Institute of Arthritis and Musculoskeletal and Skin Diseases (NIAMS), in conjunction with the Carl J. Herzog foundation, awarded a contract to the University of Rochester Department of Dermatology to lead the development of a universally acceptable and comprehensive dermatology lexicon supporting research, medical informatics, and clinical care. Medical lexicons, or controlled vocabularies, are essential for the efficient, reliable, reproducible communication of patient data and medical knowledge. A useful computerized medical record, an effective telemedicine system, a digital image library for teaching and research all require a controlled vocabulary. Researchers and clinicians need to communicate consistently and accurately with one another. Dermatologists have a particularly difficult semantic and lexical problem. Despite the rich and complex vocabulary available to them, they frequently struggle to encapsulate visual appearances and disease processes in the words and terms at
From the Department of Dermatology, University of Rochester,a and the Dermatology Centre, University of Manchester,b and the University of North Carolina at Chapel Hill.c Funding provided by The National Institute for Arthritis and Musculoskeletal and Skin Disease, Contract #NO1-AR-1-2255, and the Carl J. Herzog Foundation. Conflicts of interest: None identified. Reprints not available from authors. 0190-9622/$30.00 © 2004 by the American Academy of Dermatology, Inc. doi:10.1016/S0190-9622(03)01571-8
630
their disposal. As a consequence, the descriptive definitions they use can translate into ambiguous meanings. The field of dermatology encompasses thousands of unique diagnostic concepts with overlapping definitions, intricate hierarchical relationships, and a variety of unique clinical presentations. The relationships between the terms used to describe visual appearances and diagnostic concepts have not been codified in any systematic fashion. Since skin disease descriptors have evolved over time and were not created systematically, there is overlap, redundancy, and vagueness built into our system of words. Computer based sources such as electronic medical records, decision support systems, and epidemiologic tools cannot function on such a foundation of different shaped and sized bricks; they necessitate a set of purpose-built tools to achieve consistent and accurate communication. In terms of efficiency, the importance of a consistent vocabulary for performing tasks can be illustrated by considering the aviation industry using regional and individual vocabularies. Would air travel be possible if communication between pilot and controller used four different terms for runway and for describing direction on a runway? Dermatology, on the other hand, is plagued by the existence of a multiplicity of synonyms within its vocabulary. Of greater importance, from a medical computing standpoint, is the situation where a single word or phrase may have more than one meaning. As a simple example in the medical field, a
J AM ACAD DERMATOL VOLUME 50, NUMBER 4
computerized record may not identify the correct meaning of the word cold, as this word can ambiguously mean common cold or refer to cold temperature. In our specialty, dermatitis not only has different meanings to clinical dermatologists and dermatopathologists but also may be used to describe a range of unrelated conditions which neither would think of as dermatitis (eg, dermatitis herpetiformis, dermatitis artefacta, perioral dermatitis). To be useful in computer terms, words and phrases cannot have ambiguous meanings; each must be defined in terms of the concept which it is intended to convey. Defined concepts rather than the words used to identify them can then be assigned unique codes by which information can be reliably stored and retrieved.1 This means that the diagnostic word or words that are shared across different disease concepts, for example diagnoses sharing the word dermatitis (dermatitis herpetiformis, dermatitis artefacta, perioral dermatitis), can be made unambiguous by assigning a code to each unique diagnostic concept, not to the individual words. Unless a computerized system correctly codes for concept (and can therefore correctly identify context), true intended meaning cannot be captured. Imagine a telemedicine system that relies on the varied experiences of physicians to enter presumptive diagnoses. Regional and individual preference will result in great variety of diagnostic terms used.2 Is a congenital nevus with change stored under congenital dysplastic nevus, congenital nevus with atypia, atypical mole, or perhaps changing congenital mole? How do we represent concepts and order them in a framework which is machine readable? Tragic error can occur when everyone uses their own favorite words and abbreviations, and as a result, chart notes, laboratory values, or medication instructions are misinterpreted. Ambiguity, misidentification, and misrepresentation of intended concepts are inevitable when individuals have different meanings for the same word or phrase and the medical record cannot thus accurately represent the intended concept. This problem will not be resolved by the mere digitization of words, but on the contrary, will tend to get worse as the volume of electronic medical communication increases. Reliable and consistent clinical research in dermatology hinges not only on an ability to describe and codify diagnostic information accurately, but also on the ability to handle complex objective data, which is often visual in nature. Visual patterns and images are at the core of dermatology practice, education, and research. The new imaging and computing technologies are also facilitating remote diagnosis through telemedicine. Advances in
Papier et al 631
electronic pattern recognition software open up new possibilities in both clinical diagnosis and patient management.3 Digital technology is also creating new opportunities for enhancing medical education and patient care through digital image databases and computerized medical records and knowledge sources. The discussion of definitions is not new to the field of dermatology. A. Bernard Ackerman has frequently pointed out its lack of consistent, clear terminology.4 He has argued for greater precision and clarity in diagnostic definitions, for “a complete dictionary of dermatology.” Respondents concur, urging the abandonment of unhelpful and outmoded diagnostic terms.5-7 While the recent debate has supported the vision of a comprehensive dermatology terminology, it has not moved forward to consider the importance of a properly designed computerbased information architecture to provide the framework for a universal, high quality modern “dictionary” of dermatology. What distinguishes a computer-based lexicon from a standard text-based dictionary are formal definitions, the highly structured associations which define the position of each concept within a semantic framework.1 Thus, in medicine, diagnostic concepts can be linked by a series of associations such as “is a type of,” “complicated by,” and “predisposes to,” to produce a terminology which can be manipulated symbolically by a computer. For example, Gardner’s Syndrome “is a type of” genetic disease that is “complicated by” keratinous cysts and “predisposes to” colon cancer. The construction of a lexicon using formal definitions thus calls for a consistent, reproducible methodology which is not seen in traditional dictionaries.
IMPACT ON PATIENT CARE The shortcomings of existing dermatologic coding systems, especially International Classification of Disease (ICD), are readily apparent to the clinical dermatologist. ICD diagnostic codes were created before the capabilities of computers were available. The lack of a concept-oriented relational database structure means ICD cannot utilize the computer to relate synonyms and create multiple hierarchies. ICD, which encompasses 145 diagnostic entries in the abridged dermatology version,8 also lacks the breadth and depth needed in dermatology. It does not, for instance, distinguish between basal and squamous cell carcinoma of the skin. The availability of a comprehensive, logically ordered, standardized terminology for dermatology will improve patient care by enhancing the quality and accuracy of clinical documentation. This was
632 Papier et al
recognized by Bernstein et al9 who noted the need for a robust set of procedure definitions and terminologies to support patient records, training, and patient communication in dermatologic surgical practice. By standardizing aspects of the patient care chart, they found that therapies and procedures could be analyzed and assessed more effectively as a means to improving self-assessment. Similar arguments could be applied to all branches of clinical dermatology.
IMPACT ON RESEARCH AND EPIDEMIOLOGY Dermatologic diagnoses are most frequently based upon gross morphologic clinical findings rather than diagnostic tests or pathophysiologic mechanisms. There are, however, striking inconsistencies in the descriptive terminology used in the different major textbooks within the specialty.10 A well-defined vocabulary for morphological parameters such as configuration, distribution, texture, and color should improve the precision of morphological description and thus the quality of education and information exchange between health care professionals and clinical researchers. As diagnosis is often dependent upon clinical features which have no clear definitions, sound epidemiologic or multicenter clinical research is difficult to conduct in many dermatologic diseases. Consider clinical trials requiring the precise measurement of cutaneous response to a medication. A clinical trial of a new medication for the treatment of herpes zoster would be likely to require scoring of skin lesions for vesicle size, degree of crusting, scarring, pain, etc. A study of a new treatment for leg ulcers would have the same need for recording details such as morphology, color, size, depth, undermining, and width of inflammatory borders of lesions. The lack of precisely defined morphologic and diagnostic terminology makes it impossible to create data sets which can be interpreted between researchers. Please refer to Appendix 1: Vague terminology and assessing the risks of smallpox vaccination (the appendixes may be found online only at www.eblue.org).
IMPACT ON MEDICAL LITERATURE INDEXING AND RETRIEVAL The inadequacies of controlled vocabularies responsible for the indexing and retrieval of the biomedical literature are well known to end-users. Free text searching by disease name is not a suitable answer given the large numbers of synonyms in use for many skin diseases. As an example, consider a MEDLINE literature search using Medical Subject Headings (MeSH) for transient acantholytic dermatosis and the com-
J AM ACAD DERMATOL APRIL 2004
Table I. MEDLINE: 1966-2002; Date of Search: March 22, 2002 Term
Hits
Acantholysis (MeSH) Grover’s disease (keyword) Transient acantholytic dermatosis (keyword) Grover’s disease or Transient acantholytic dermatosis (Grover’s or Transient acantholytic dermatosis) or Acantholysis Missed citations
261 49 81 110 292 31
MeSH, Medical Subject Headings.
monly used eponym Grover’s disease (Table I). Neither of these two synonymous diagnostic terms are Medical Subject Headings in the MEDLINE database. This means that the MEDLINE user is referred to the nearest index term, which happens to be acantholysis. A search of acantholysis yields relevant citations, but also many acantholytic disorders other than Grover’s disease/transient acantholytic dermatosis. Such a search today of the MeSH heading acantholysis results in 261 citations. In contrast, a keyword search of Grover’s disease provides 49 citations, while a search of transient acantholytic dermatosis provides 81. After removing the duplicate citations from the independent searches of Grover’s disease and transient acantholytic dermatosis (20), there are 110 unique citations related to this diagnostic concept. The search of acantholysis which yielded 261 citations failed to detect 31 of these 110 unique citations. Therefore, in order to retrieve relevant information on this single concept, one would have to search on at least three terms. This example is by no means unique but rather the norm for dermatological diseases. A comprehensive set of hierarchically ordered diagnostic terms will facilitate literature searching by allowing the user to rapidly view related terms and synonyms and to construct precise yet comprehensive keyword, title, and abstract searches. Using the example above, a hierarchy would allow a search for all of the diagnoses that have acantholysis as a feature, or a more precise search of Grover’s disease specifically. A dermatologist will be able to enter any diagnostic term into the search window and automatically generate the appropriate search terms to retrieve the 110 unique citations encompassing the concept of Grover’s disease or transient acantholytic dermatosis. Current users of MEDLINE are unlikely to be aware of the imprecision of their searches. With the implementation of a diagnostic concept search tool, dermatologists would utilize all the necessary search terms to retrieve the complete body of
Papier et al 633
J AM ACAD DERMATOL VOLUME 50, NUMBER 4
relevant reference citations. Please refer to Appendix 2: DLP prototype, diagnoses organized by pathophysiology (at www.eblue.org).
INTEGRATION WITH EXISTING AND FUTURE STANDARDS A dermatology reference terminology must integrate with existing medical lexical standards. Computerized medical records, telemedicine applications, and decision support systems contain vocabulary generic to all of medicine, not specifically dermatology. Therefore a dermatology lexicon must share common language (and codes) so that the dermatology terms relate to medicine as a whole. The Unified Medical Language System (UMLS) resource can be used to facilitate such integration of a specialty specific nomenclature to the broader medical vocabulary.11 The UMLS assists software and information system developers in “mapping” words to MeSH, ICD, World Health Organization (WHO), and The Systematized Nomenclature of Human and Veterinary Medicine (SNOMED)12 terminologies. Consider the UMLS database as a warehouse for the words of medicine. The warehouse has a mix of private and public lexical products which a researcher can gain access to by registering with the National Library of Medicine.13 UMLS is not a lexicon itself; it is a central repository for individual vocabularies. The Dermatology Lexicon Project terminology will become part of this “warehouse,” meaning that its terms will be cross-indexed to the terms of MeSH, ICD, WHO and other UMLS participants.
CURRENT WORK IN DERMATOLOGIC TERMINOLOGY The modern history of nomenclature in dermatology dates to 1977 when Crawford Brown, MD, and a task force formed by the American Academy of Dermatology created SNODERM,14 an abstraction of the original SNOMED, a term set created by the College of American Pathologists. One must recognize that affordable, high-speed computing was only just beginning to be realized at this time. The subset of terms derived from SNOMED was incomplete and there was no opportunity for dermatologists to supplement the lexicon. Between then and the early 1990s, there was little effort to organize a modern lexicon in the United States. In the United Kingdom, however, the British Association of Dermatologists (BAD) has been closely involved with the development of diagnostic codes for dermatology for over 20 years. In the early 1990s, the Association formally recognized the need for a new look at dermatological disease coding and secured a grant from the United Kingdom’s National
Health Service to develop a comprehensive, hierarchically ordered dermatological disease index. A Working Group devised a new hierarchical structure that was designed to be much more logical than that imposed by ICD-9 or ICD-10, although terms were cross-mapped to ICD-10 which is required for National Health Service documents. Additionally, terms were submitted to its Clinical Terms Project, now subject to amalgamation with SNOMED as SNOMED Clinical Terms. Please refer to Appendix 3: BAD Diagnostic Index (at www.eblue.org). In Germany, the Deutsche Dermatologische Gesellschaft set up a Working Group in 1991 to create a catalogue of all dermatological diagnoses including synonyms. The Group consists of about 20 dermatologists including both independent practitioners and representatives of university dermatological departments. The Group published its first Dermatologischer Diagnosenkatalog in 1994. Since 1994, three further editions have been published, the latest of which is based on the ICD-10 classification.15
A LEXICON CREATED BY DERMATOLOGISTS A lexicon holds no meaning or value if it does not accurately reflect the discipline of dermatology. As such, the needs of the end-user are paramount to ensure that the lexicon is designed to satisfy real needs. Feedback from all facets of the dermatology community is needed to ascertain user needs and how the lexicon would be integrated into their research or clinical practice. Dermatologists throughout the United States, both those in academia and private practice, are taking an active role in the creation of the Lexicon. Consultants and editors are placed into one of three working groups: morphology, diagnoses, or therapy/procedures. The Morphology Working Group is developing succinct morphology definitions both textually and with illustrations. These are created by a professional medical illustrator by manipulating suitable digital images and combining them with computer rendered drawings. The definitions will include, as appropriate: size, color, depth/elevation, contents of internal matter, consistency, border, shape, texture, and palpability. Please refer to Appendix 4: draft definition and illustrations for vesicle (at www. eblue.org). The members of the Diagnosis Working Group act as editors for the diagnoses belonging to each of the “branches” of the pathophysiology tree, according to their particular areas of expertise. The diagnoses are organized by the following categories: neoplastic, infectious, genetic, neuropsychiatric, nu-
634 Papier et al
tritional/metabolic/endocrinologic, environmental/ external agents, and immunologic/inflammatory. The tree will allow users to view a diagnosis in a hierarchical manner and infer etiological relationships. Similarly, the Procedures and Therapy Working Group will organize treatments, including medications, surgical procedures, phototherapy, and complementary and alternative medicine, in a hierarchical structure. A lexicon is not of value if it cannot be sustained. The Dermatology Lexicon will be a dynamic entity and must be able to evolve to accommodate new knowledge. It will take advantage of modern computing techniques to enable single diagnostic concepts to be displayed in more than one axis (ie, on more than one “branch” of the hierarchical tree; eg, by morphology, by causative agent, and by pathophysiologic mechanism) and for the diagnostic concepts to be moved from one “branch” of the tree to another as our understanding of the etiology of disease changes with time. The database will be created to facilitate constant upgrading, and partnerships will be developed to support long-term maintenance. This is a key point. The success of the project is not defined by creating a static dictionary, but rather through the construction of sustainable information management tools. Connectivity to international efforts such as the British Association of Dermatologists’ Diagnostic Index and the German Dermatologischer Diagnosenkatalog will be maintained. The Internet and E-mail will support a distributed editorial team and promote rapid review of terminology subsets. In addition to a core Advisory Board composed of experts in medical informatics and consultants in dermatology, numerous dermatologists act as expert editors of the content. The process of validating the lexicon will take place online, allowing for more widespread comment. Ultimately, the lexicon will be available
J AM ACAD DERMATOL APRIL 2004
on the Internet at no cost to ensure access to all dermatologists. Please contact
[email protected]. edu with questions or comments. We thank The National Institute for Arthritis and Musculoskeletal and Skin Diseases and the Carl J. Herzog Foundation for their support. REFERENCES 1. Cimino JJ. Desiderata for controlled medical vocabularies in the twenty-first century. Methods Inf Med 1998;37:394-403. 2. Stewart WD. Geographic dermatology. Int J Dermatol 1990;29: 477-8. 3. Binder M, Puespoeck-Schwarz M, Steiner A, Kittler H, Muellner M, Wolff K, et al. Epiluminescence microscopy of small pigmented skin lesions: short-term formal training improves the diagnostic performance of dermatologists. J Am Acad Dermatol 1997;36:197-202. 4. Ackerman AB. Need for a complete dictionary of dermatology early in the 21st century. Arch Dermatol 2000;136:23. 5. Malak JA, Kibbi AG. Revised terminology in dermatology: a call for the new millennium. Arch Dermatol 2001;137:93-4. 6. Hurt MA. On the importance of definition in dermatology and all fields of human endeavor. Arch Dermatol 2001;137:664-5. 7. Gniadecki R. Rigid definitions restrict the evolution of understanding. Arch Dermatol 2000;136:1271. 8. McCaffree DL. Abridged ICD-9-CM codes for dermatologists. Department of Health and Human Services; 1996. 9. Bernstein RM, Rassman WR, Seager D, Shapiro R, Cooley JE, Norwood OT, et al. Standardizing the classification and description of follicular unit transplantation and mini-micrografting techniques. Dermatol Surg 1998;24:957-63. 10. Jackson R. Definitions in dermatology. A dissertation on some of the terms used to describe the living gross pathology of the human skin. Clin Exp Dermatol 1978;3:241-7. 11. Lindberg DA, Humphreys BL, McCray AT. The Unified Medical Language System. Methods Inf Med 1993;32:281-91. 12. Lussier YA, Rothwell DJ, Cote RA. The SNOMED model: a knowledge source for the controlled terminology of the computerized patient record. Methods Inf Med 1998;37:161-4. 13. Humphreys BL, Lindberg DA, Schoolman HM, Barnett GO. The Unified Medical Language System: an informatics research collaboration. J Am Med Inform Assoc 1998;5:1-11. 14. Brown CS. Systematized nomenclature of dermatology. Baltimore: College of American Pathologists; 1978. 15. Graubner B, Brenner G. German adaptations of ICD-10. Stud Health Tech Inform 1999;68:912-7.