Any scientific activity is based on rational thinking and reasoning. We are observing the real world; we are interpreting the observed data deriving information that will justify our future actions. Then, mostly by deductive reasoning, we may reached knowledge. This new acquired knowledge will be added to the existing one and we will used it to interpret other data.

In healthcare, the main source of data is the patient. The physician or the nurse observes the patient; they collect data. Data are interpreted and information is derived. The physician or the nurse uses this information, jointly with the information obtained from other similar groups of patients, to achieve new knowledge. Again, this new knowledge will be used to interpret other data.

In order to be able to interpret observed data they have to be structured, grouped in a finite number of categories or classes, in accordance to objective criteria. How many classes we use, how much we refined the classification, this is depending upon the goal of the classification. We have different classification systems when we want to make statistics about the observed diseases and when we want to plan care in a hospital ward.

Moreover, the syndrome of the Information Society has a direct impact on health communication and indirectly this is a catalyst for developing better terminology systems. Health information have to be structured, grouped, in such a way that they can be easily captured in the electronic health record and then communicate. Only if these requirements are satisfied, the quality of care can be improved trough deep analysis of patient data. Health on the net, the new paradigm, is asking for sharing and integrating health data and the tools offered by the new technologies are enabling this and, if wisely applied, the results can be rewarding. This why national and international bodies are exploring through research projects the extent to which existing vocabularies and classifications serve as accurate sources of knowledge for health information systems and their clinical applications.

A Classification is the systematic assignment of objects in different categories in function of precise criteria.

A terminology is a collection of words and/or phrases used to describe a concept or phenomenon.

A vocabulary is the stock, the glossary, the repository or repertoire of words from which to name or describe phenomena within a language or knowledge base (Lang et al, 1995).

Language is a mean of communicating, the particular style of verbal or written expression characteristic of a person, group or profession. In the context of this paper, a more appropriate definition of a language is as a set of characters, conventions, and rules used to convey ideas and information (Lang, et al, 1995).

A common language starts with building a common terminology. The standard terms and concepts of a domain (medicine, nursing etc) are gathered in a thesaurus or nomenclature system. In such a system, there is a code for every term or concept. Terms and concepts can be combined to form complex terms or concepts. A nomenclature system is different from a classification, for the user may build any code combination. This is a burden making retrieval difficult.

A terminology data base is a collection of interrelated files with records containing terms and definitions of the terms, organised and stored together in a computer system (Lang, et al, 1995).

The word taxonomy has its origins in Carl von Linnaeus work. Linneaus was an 18^th century Swedish biologist that left us a classification of plants in twenty-four classes based on the number and pattern of stamens and also a classification of the animal reign. Often the word 'taxonomy' is used as a synonym of 'classification'. Taxonomy specifies on what rules, procedures and basic principles or criteria objects of the universe of discourse are assigned to different groups or classes. Classes are not overlapping. Objects of a class have common characteristics. A class may be divided in sub-classes and so on. There are no empty classes. For example, in the ICD-10 classification, diseases are grouped by organs or aetiology. The main disease or etiological categories are subdivided into groups, each group into 3-digit classes and so on.

Formally, the universe of discourse U (i.e. all the objects of interest) is partitioned in a finite number of classes: U = 4 C_i with i=1 to n, where n is the number of classes. An object x is in a class C (x c C ) if and only if P(x) (x has the propriety P) and for any two classes C_i and C_j, with i ! j , C₁₃ C₂=Â . Also, for every class C from the universe U , C! Â.

When the objects are concepts, a classification will separate in disjunctive classes and subclasses the domain of concepts (for example - in medicine ‘all diseases’ or ‘all surgical procedures’; in nursing 'all phenomena' or 'all actions'). In fact, classification means ordering the observed data based on what we know about the real world (prior knowledge) and in accordance with an objective. When we are partitioning the universe of discourse, we are approaching it from a specific point of view. If we are interested in the population of humans that can vote, we will use only two classes of ages:

non-adult age < 18

adult age ³ 18

If we are instead interested in the distribution of measles on age groups, the classifications will be different:

baby age 0 - 1

child age > 1 and age < 5

group 3 age 5 - 14

group 4 age 15 - 24

group 5 age 25 - 34

group 6 age 35 - 44

group 7 age ³ 45

Classification axes

In the former example (with the age groups), the classification was simple, being done in accordance with only one criteria or parameter. If the classification is made for more than one criteria or parameter, it is said that there are several axes of classification - the classification is multi-axial. It is the case of most of the classification in healthcare, when we need support for expressing the complexity of the concepts. A disease for example can be approached from the aetiology, morphology or functionality point of view. Observed data will be distributed into classes in accordance with these criteria but also depending on the intended usage. For example, ICPC is a bi-axial classification, with the two axes: tractus (oriented to the body system - the tract) and disease. ICNP is multi-axial classification for nursing. Phenomenon axes are: focus of nursing practice, judgment, degree, frequency, chronicity, topography/laterality, body site and likelihood. Terms of these axes are combined in order to express a nursing diagnosis. In the following, the steps for defining risk for chronic, extremely disturbed sleep are described:

Diagnosis : disturbed sleep

extremely disturbed sleep

chronic disturbed sleep

risk for disturbed sleep

The nursing phenomenon is sleep.

Axes Selected term

Focus on nursing practice --> sleep

Judgment --> disturbed

Degree --> extremely

Chronicity --> Chronic

Likelihood --> Risk for

Action axes are : action type, objects, target or recipient, methods and instrument. Terms from different axes are combined to describe nursing interventions. For example:

Interventions : Alleviating an individual's pain by applying a cold pack

Teaching the family about nutrition using instructional

materials

Reducing anxiety using a guided imagery technique

Axes Selected term

Action type ----> alleviating reducing teaching

Object ---> pain anxiety nutrition

Target ---> individual individual family

Methods ---> cold pack GIT instructional material

In figure 3 there is a diagram showing the place of a health classification within a person-based information strategy.

The International Classification of Diseases (ICD) is maybe the most widely known and used medical classification. This is a statistical classification, based on the importance and prevalence of diseases entities. ICD is using the notion of compaction in order to aggregate into one class similar diseases having statistical value for the control of the disease process. Related to this is the ICIDH classification (International Classification for Incapacity, Disablement and handicap).

ICD-10, which is developed, maintained and published by the World Health Organisation, has been available for implementation since 1993. It represents the broadest scope of any previous ICD revision to date. ICD-10 is more comprehensive than current standards and extends well beyond the traditional causes of death and hospital admission. The expansion of content and specificity to conditions and situations that are not diseases are particularly relevant for use of the classification system outside the hospital setting. Table 6 gives examples of some of the subcategories provided in ICD-10 for the capture of risk factors to health, such as lifestyle, life management, psychosocial circumstances, and the occupational or physical environment. Such an expanded scope may attract new users to ICD-10 and may increase the number of databases in which the codes appear. This is important given the evolution of integrated health information systems.

In the next table there are listed some examples of subcategories provided in the Tenth Revision of the International Statistical Classification of Diseases and Related Health Problems for capture of risk factors to health:

In Australia, a comparison of the last two versions of ICD has been carried out. This has been accompanied by a critical study of the adequacy of the classification to current problems. The results of a mapping showed that, of a total of 13,600 ICD-10 codes, 50.8% were more specific than the ICD-9-CM codes, 31.5% were as specific, and only 11.5% either were less specific or could not be compared. This increased specificity contributes to more relevant data for epidemiological research and decision-support purposes. Gains in the level of specificity also increase the sensitivity of the classification when making refinements in applications, such as grouping methods.

Concerning adequacy, the Australian research team concluded that ICD-10 introduces both new terminology and new clinical concepts, giving it a higher level of clinical credibility and acceptance. Unlike previous revisions, ICD-10 allows for enhancements to accommodate newly discovered diseases, such as AIDS. WHO has established an ongoing maintenance and updating process that ensures input from member states as well as from interested professional bodies. In addition, there are plans to share updates internationally by means of electronic technology. This will enhance the life-expectancy of the classification system.

Another well-known classification is the International Classification of Procedures in Medicine (ICPM). ICPM, developed by WHO, was intended to be a classification for all medical procedures, services and activities. The mono-hierarchical structure of the ICPM considers anatomy and topography but not medical specialties, patient ages or diagnoses (which are coded by other classifications). The class of surgical procedure (such as incision, excision etc) is mentioned on the lower levels of ICPM. If the surgery ia a very complex operation, as in the case of a polytrauma or microsurgery, there are necessary multiple codes, but in normal cases, one operation is coded by only one to five digit notation. Remarks and descriptions about terms are used to explain the correct application of the ICPM.

ICD and ICPM are mono-hierarchical, conceptual classifications and include disjunctive and extensional vocabularies with about 7,000 canonical and 20,000 lexical variant terms (ICD-9).

Another classification is the International Classification for Primary Care (ICPC). ICPC was designed for epidemiological use.

The most important axis is the "root operation" which specifies the underlying objective of the procedure: 0=Bypass, 1=Change, 2=Creation, 3=Destruction, 4=Dilatation, 5=Division, ..., 9=Extraction, B=Fragmentation, C=Insertion, D=Inspection, ..., R=Revision, S=Transfer, T=Transplantation. There are 30 different root operations. Composite terms like appendectomy are never considered root operations. PCS does not involve information about the diagnostic or the pathology and morphology. The "not otherwise specified" (NOS) and "not elsewhere classified" classes are not allowed and eponyms (Billroth’s OP) are not included. PCS avoids multiple meanings for the same term by a standardised terminology. Different procedures have a unique term.

1 Circulatory system disease G 390-459

2 ischaemic heart disease G3 410-414

3 acute myocardial infection G30 410

4 anterior myocardial infarct G301 410.1

The ATC (Anatomic Terapeutical Chemical) code is a hierarchical classification for drugs. ATC does not cover combined products or dermatological preparations.

The DRGs (Diagnosis Related Groups ) classification and the CPT codes are intended to provide cost evaluations of the care process. CPT is used only in USA. DRGs is based on the ICD classification and it groups ICD entities about 400 classes) on factors that affects the cost of treatment and length of stay in hospital (see more_links).

SNOMED (Systematised Nomenclature of Medicine )- http://snomed.org is a multi-axial coded nomenclature, not a classification. It allows the recording of all diseases entities (causal agent, specific site, diagnostic criteria, therapeutic procedures etc), regardless of prevalence, as well as observations related to individual cases. SNOMED axes are: topography, morphology, function, aetiology, disease. The architecture of SNOMED allows the automated construction of simple sentences using the nouns of each axe. The architecture of SNOMED permits also the elaboration of procedures and algorithms incorporating diagnostic criteria of disease.

For example :

lung + granuloma + fever + Mycobacterium tuberculosis = tuberculosis

(site) (morphologic change) (function) (etiologic agent) (disease or syndrome)

T M F

And this can be read as:

" If in any site T, with some M and F changes, the Mycobacterium tuberculosis is found, then a disease is present and it is tuberculosis".

The third version of SNOMED is called ‘SNOMED International’, the Systematized Nomenclature of Human and Veterinary Medicine (Coté et al. 1993) (A previous version has been translated to German and was published by Wingert 1984). SNOMED was originally developed by pathologists and does not belong directly to the WHO family but strong links are recognisable, for example in the above mentioned morphology axis of SNOMED is used also in the ICD-O of WHO. SNOMED refers a medical expression or term on orthogonal axes such as M=Morphology, T=Topography/Localization, P=Procedure, E=Etiology/Agent, F=Function, D=Disease. The idea behind the nomenclature is that a procedure for a disease or a morphologic alteration caused by an agent is combined with a functional disturbance. According to this structure, a medical expression is transformed into the SNOMED statement D=T+M+E+F+P, with each axis abbreviation followed by the hierarchical numerical code.

For example :

D 01810 (Streptococcal pharyngitis) = T 61100 (Pharynx) + M 4000 (Inflammation) + E 25000 (Streptococcus).

SNOMED international has now 11 axes and syntactical links such as associated-with, followed-by, generated-in etc. Special relationships between medical terms can be explained in order to build diagnosis. Information qualifiers have also been introduced, for example - patient history, family history, necessity of treatment, main diagnosis, further diagnoses and similar factors.

A SNOMED coding example (not all axes)

The development of nursing in the last century had lead to a strong trend toward the development of a professional language for nursing documentation and record keeping and of course for better communication.

NANDA, North American Nursing Diagnoses Association [Carroll-Johnson (1989)] developed a one-dimensional hierarchical classification (and terminology) of nursing diagnoses. The top level of hierarchy is built from nine human response patterns and up to 4 levels can be added. For example, the nursing diagnosis "activity intolerance" is coded by 6.1.1.2. with the first digit ‘6’ referring to the response pattern category "movement".

The OMAHA System is a three-dimensional classification of 44 nursing diagnoses representing the first dimension, mostly for visiting nurses in a community (at first in Omaha, USA). The problems are divided at the top level in "environmental", "psycho-social", "physiological" and "health-related problems". The two other dimensions are nursing activities with an additional four categories, and nursing results with 3 categories in a 5 level scale.

Another classification for nursing diagnoses is the HHCC (Home Health Care Classification). This is a three-dimensional system with 145 categories for nursing diagnoses, 160 nursing intervention and three modifiers for the outcome (improved, constant, aggravated). Four action types such as "assess/evaluate", "direct care/perform", "teach/educate" and "manage" are used to characterised nursing interventions. The HHCC includes not only a framework (20 care components) which can be used to link the three steps/phases of the nursing process, but also a methodology to document the care process. The HHCC also includes definitions and coding structure that can be used to compare care across health care settings. The codes are similar to the ICD-10 digit structure.

The NIC (Nursing Intervention Classification) [McCloskey, Bulechek (1992)] is a one-dimensional coding system for activities. It includes not only names and codes for interventions but also definitions, explanations and literature. The 433 interventions are divided into 6 domains.

The ICNP International Classification of Nursing Practice (Nursing Intervention Classification) is under construction by the International Council of Nurses [ICN (1993)]. The ICNP promises to provide a structure for the classification of nursing diagnoses, interventions and outcomes, and it also takes into account all important available nursing classifications. Today the Beta version has been released (ICNP and Telematic Applications for Nurses in Europe, IOS Press, Health Technology and Informatics, vol. 61, 1999). A short history of ICNP and usefull links can be found on : http://atlas.ici.ro/ehto/telenurse/ . The volume ICNP and Telematic Applications for Nurses in Europe, IOS Press, Health Technology and Informatics no. 61, edited by Randi A. Mortensen (ram@diss.dk), is providing interested points of view throughout a collection of valuable papers.

Walter Sermeus was a promoter of the NMDS ( Nursing Minimum Data Set) in Belgium as a standard for measuring care and the cost of care. Sermeus defines nursing minimum data sets as follows :

"systematic registration of the smallest number possible of unequivocally coded data items, with respect to or for the purpose of nursing practice; making information available to the largest possible group of users according to a broad range of information requirements".

In Germany, Ingenerf (1996) is maintaining an Internet-accessible collection of annotated pointers (English and German) to Internet resources providing a general view in the field of medical terminology, classifications, nomenclatures, coding tools and the like. Good quality training materials are provided by the IT-EDUCTRA project. Also additional information can be found on the next sites:

http://atlas.ici.ro/ehto/telenurse

UMLS (Unified Medical Language System) is a project started in 1986 and conducted by NLM (National Library of Medicine -USA). The objective is to improve the availability of computer-readable information sources (retrieval and integration levels). Terminology and concepts from multiple vocabularies and classifications are linked through a "metathesaurus" (automated lexical matching techniques and structured knowledge embedded in existing information sources as MEDLINE and SNOMED). Concepts in the metathesaurus are assigned semantic types. The UMLS electronically links vocabularies and classification systems through a system of four knowledge sources (http://www.nlm.nih.gov/pubs/factsheets/umls.html): Metathesaurus, Semantic Network, Specialist Lexicon, and Information Sources Map. The Metathesaurus is organized by concept or meaning, and contains semantic information on approximately 476,322 biomedical and related concepts with 1,051,903 different names. The Metathesaurus contains vocabulary terms, classifications, coding systems, and thesauri developed and maintained by various professional organizations, such as the American Nurses Association, and identifies alternate names for the same concept and relationships between different concepts.

The Specialist Lexicon contains syntactic information about health care-related terms and concept names from the Metathesaurus, as well as other non-health related English words used in communication that are not necessarily included in the scope of the Metathesaurus. The Semantic Network is comprised of a network of general categories or classifications which consistently categorize all concepts from the Metathesaurus, and identifies allowable relationships between terms. The Information Sources Map contains information on the available sources of the machine-readable health related information. Each term or concept is defined and cross-mapped to terms or concepts within other classification systems or vocabularies.

McCormick and al. discussed the fact that all vocabularies and classifications for nursing that have been approved by the American Nurses Association are incorporated into the UMLS. These nursing vocabularies and classifications in UMLS can be extrapolated, resulting in what could be described as a Unified Nursing Language System (UNLS) (McCormick, et al, 1994; McCormick & Zielstorff, 1995). This new system can be extrapolated and tested against large scale nursing data repositories to determine if it is also representative of vocabularies such as acute care, primary care, long-term care, outpatient, community, school health nursing, occupational health, and the many realms where nursing care is delivered.

To summarise, a classification system is valid if it answered at the following requirements:

completeness

use of synonyms

use of preferred terms

use of lexical variations

insensitive to spelling errors

non-overlapping classes

well-structured

sufficient granularity

coding attributes of concepts

When evaluating a classification, all these important characteristics are analysed. Moreover, it is important to note that if a proposed classification has all the above-mentioned characteristics, a new one emerged, i.e. the classification can be computerised, that is a today information society condition.

The model presented by McCormick assumes that there are three types of vocabularies needed in health care. The point of care is an "interface" vocabulary that occurs at the individual and practice level, and includes terms that are used between clinician and patient, and/or clinician and clinician to describe and convey related patient and clinical information. Vocabularies at the point of service level emphasise the settings where care is delivered, and are more specialty focused.

If we consider the continuum of care, from primary care to acute care, it is clear that there is a need for different classifications based on different vocabularies.

Today there are information systems that provide interface vocabulary to support decision-making at the patient care level as Oceania (http://www.oceania.com) and a Canadian information system called Purkinje (http://www.purkinje.com).

At the network level, the vocabulary must enable the links between the different strata of healthcare delivery, from primary care to hospital clinics. Also, the vocabulary must support the link of healthcare professionals documentation to health administration. In this sense, it can be viewed as an "integrator" vocabulary. In her paper, McCormick uses the term "reference" instead of integrator but the mean is the same. At this level, the vocabulary has a high degree of abstraction, synthesising knowledge from multiple sources (different settings, different health fields, different levels of care) and offering to the decision makers the necessary items to describe their actions. SNOMED, MEDCIN, MEDICOMP and Dr. Elmer Gabrieli's natural language processing (Gabrieli, 1993) are recognised network vocabularies.