Coding and Classification Schemes

Abstract

This chapter introduces a number of important coding and classification systems that have been and remain influential in healthcare. We briefly discuss the International Classification of Diseases (ICD), diagnosis-related groups (DRGs), the Read codes, SNOP and SNOMED, LOINC, and the Unified Medical Language System (UMLS).

This chapter introduces a number of important coding and classification systems that have been and remain influential in healthcare. We briefly discuss the International Classification of Diseases (ICD), diagnosis-related groups (DRGs), the Read codes, SNOP and SNOMED, LOINC, and the Unified Medical Language System (UMLS).

International Classification of Diseases

The ICD (International Classification of Diseases) is the standard international diagnostic classification for general epidemiological, health management, and clinical uses [1]. It is used to classify diseases and other problems for payment, management, and research purposes, recorded on many types of health records including medical records and death certificates. Note that this list excludes use at the point of care. It enables international comparisons of mortality and morbidity by WHO member states.

The origins of the ICD have been traced back to John Graunt’s London Bills of Mortality in the seventeenth century and the work of John Farr and Jacques Bertillon in the late nineteenth century to produce the International Classification of Causes of Death.

ICD-9 was published by the WHO in 1977, and ICD-9-CM (clinical modification) is still used in the USA for payment purposes with annual updates, extending the original ICD-9 with more morbidity and procedure codes.

ICD-10 was published in 1992 and is used by most countries outside USA. The full title isThe International Statistical Classification of Diseases and Related Health Problems, and it is published in three volumes. Volume 1 is a tabular list, Volume 2 is the instruction manual, and Volume 3 is an alphabetical list. ICD-10 has an alphanumeric coding scheme with one letter followed by three numbers at the four-character level. The first draft (alpha) of a new ICD-11 has been published and is expected to be finished in 2015.

There are 21 chapters in ICD-10, corresponding roughly to body systems. Medical conditions have been grouped in a way that was felt to be most suitable for general epidemiological purposes and evaluation of healthcare. Within each chapter, the various diseases are listed with three-digit codes with an optional fourth or fifth digit for additional detail. Tables14.1 to14.3 illustrate the hierarchical structure of ICD-10 showing the chapters, then the main blocks or sections in one chapter, and then the third-level codes within one block.

Table 14.1 ICD-10 chapters and code ranges

For example, Chapter X (“Diseases of the respiratory system”) is subdivided into ten blocks (Table14.2).

Table 14.2 Blocks of codes within ICD-10 Chapter X

At the third level, acute upper respiratory infections block has seven categories, which have three-character codes (Table14.3).

Table 14.3 Three-character codes within ICD-10 acute upper respiratory infections

Each category is specified with inclusion and exclusion criteria. Most groups have a further level of detail. For example, J04 (acute laryngitis and tracheitis) is divided into acute laryngitis (J04.0), acute tracheitis (J04.1), and acute laryngotracheitis (J04.2). An additional code from the section on bacterial, viral, and other infectious agents may be used to identify the infectious agent.

ICD is primarily used by professional coders working with clinical records to provide data to summarize and compare hospital caseloads. The terminology used is not intended or suitable for use directly by clinicians at the point of care and is not detailed enough to meet the needs of hospital specialists. For example, the most detailed description of a fractured tibia is that it is S82.2.0 (closed fracture of shaft of tibia). This says nothing about the type of fracture (simple, spiral, or compound), laterality, or whether the fibula is also fractured.

Diagnosis-Related Groups

DRGs were originally created as a research tool to help answer questions such as “why do some patients stay in hospital longer than others?” [2]. The product of any hospital is defined as the sum of the set of goods and services provided to individual patients. In other words, a hospital produces patient care, which involves treatment, tests, and other services. However, no two patients are ever exactly the same, and so a method was needed to classify similar patients such that the costs, process, and outcome of care could be studied systematically.

Bob Fetter and others at Yale University approached the problem by classifying patients into groups based primarily on diagnosis and the amount of resources usually required, in order to identify those patients who had an unusually long length of stay for their condition. They did this using retrospective statistical analysis of hospital in-patient returns to identify clinically relevant groups of patients with similar expected lengths of stay (iso-resource groups). A key principle of the original work on diagnosis-related groups (DRGs) was that it was based on analysis of what had actually happened, not on what people thought should happen.

DRGs can be used to define the products of hospitals, in order to compare activities, length of stay, services used, and so on. Using the DRG paradigm, each hospital is a producer of DRGs. In many cases, individual DRGs may be the appropriate units of attention, while in other cases, it is useful to work with strategic product line groups (SPGs), which are clusters of similar DRGs performed by the same specialists.

DRGs might have remained a useful research tool had it not been for the introduction of the Medicaid prospective payment scheme (PPS) in 1983. At that time, Medicaid stopped paying hospitals on the basis of cost incurred, which provided incentives for hospitals to keep patients in longer, and instead paid a fixed price per patient based on their DRG category.

DRGs have a hierarchical structure. The original DRGs had 23 major diagnostic categories, based on principal diagnosis, each of which is partitioned into medical and surgical groups according to whether the patient had an operation during their stay. The medical groups are further divided according to primary diagnosis, age, and the presence of complications and comorbidities (CC), which have an impact on length of stay of more than one day. Similarly, surgical groups are divided by type of operation, age, complications, comorbidities, and the presence of malignancy. After a number of refinements, this led to a classification of 467 groups such that patients in any one group might be expected to use broadly the same range of hospital resources.

Over the years, the basic DRG scheme has evolved into a family of systems, including Medicare DRGs, Refined DRGs, All Patient DRGs, Severity DRGs, International-Refined DRGs and Health Related Groups (HRGs).

Use of DRGs depends on high-quality coded data, with a high level of agreement between coded data and the real world health events they represent. Ultimately, this depends on the quality, clarity, and organization of the medical record.

The Read Codes

The Read codes are used in primary care. They are important for several reasons. The Read codes are one of the two direct predecessors of SNOMED CT. Without the Read codes, there would be no SNOMED CT as we now know it. The Read codes have been used by all GPs in the UK and New Zealand for the past 15 years. New Zealand still uses the original 4-byte codes, described here.

Health management relies on comparable coded data. It is hard to imagine that the government in England would entrust 80% of the healthcare budget to GPs to manage, without them having coded data to monitor what is going on. This is happening as a result of the Health and Social Care Act 2012. The Read codes are a good example of a successful clinical coding system which was fit for purpose, at least in general medical practice.

The development of the Read codes began in 1983, when, with colleagues Dr. James Read and Dr. David Markwell, the author helped design a new computer system for use in general practice. An early design decision was to use a development tool that used fixed-length fields, requiring all codes, terms, and lookup keys to have a predefined length.

The original design used alphanumeric codes with 4 characters (later extended to 5 characters) and terms up to 30 characters long (later extended). A key requirement was that the coding scheme should be as comprehensive as possible, covering everything that might be entered into a patient’s computerized record. No existing coding scheme could be found which met these criteria, so we chose to write one from scratch (as did many of our competitors during that period).

The motivation was commercial, a point of view shared by other GP suppliers. GPs did not want to do any extra work and were mildly computerphobic. They had little interest in developing their own local coding schemes and wanted a system that worked out of the box. We wanted a scheme that would allow one-finger typists to enter data in the consulting room, by typing in a few letters and the computer doing the rest, and that would work straight from the box. We also wanted a system that could generate reports almost instantly.

Our idea was to take existing classifications and convert these into the appropriate format. These included ICD-9 for diseases, the British National Formulary (BNF) for drugs and the International Classification of Procedures in Medicine (ICPM). Later this was extended to include the UK national coding scheme for operations OPCS-4.

Dr. James Read, a GP in Loughborough, undertook the editing task and developed new sections for examination findings, preventive care, administrative procedures, and other subjects for which no suitable model could be found. Dr. David Markwell developed the software. What was originally planned to take 3 months took almost 3 years, and the scheme was finally launched as the eponymous Read codes in 1986 [3].

As the work evolved, we found that we had improved on earlier classification and coding systems in several respects. No paper version was ever published, facilitating regular updates and extensions. The simple position-dependent unidimensional hierarchy was easy to implement in software. The scheme was designed by GPs for use by GPs in their surgery (rather than for secondary use such as epidemiology and international comparisons).

The first publication was in theBritish Journal of Healthcare Computing in May 1986 [3]. The next section is based on this original paper. The number of codes in the original Version 1 (May 1986) are presented in Table14.4.

Table 14.4 Read codes Version 1, 1986

Later developments greatly increased the number of terms, but there is no evidence that this actually improved usability to a significant extent – probably the reverse.

Hierarchical Codes

The structure of the classification hierarchy is mapped directly by codes. In the same way as a map grid reference specifies a position on a map, each code specifies its position within the classification. The original Read Clinical Classification has four-digit alphanumeric codes using the numerals 0–9 and the letters A–Z. The first character relates to level 1, the second to level 2 and so on. Consider code B136; this is broken down in Table14.5.

Table 14.5 Example of Read code hierarchy

The four-digit codes increase in detail from left to right. The alphanumeric coding system using four-digit codes allows 1,679,6l6 possible entries (36 to the power four). The scheme was later extended to allow lowercase letters (a–z), with the exception of a couple of letters such as O and l, which can easily be confused, giving 60 options at each level, total 60⁴ (about 12 million options).

Automatic Encoding

The classification incorporates automatic encoding. Entry of the first few letters of any term displays a list beginning with those letters from which the user chooses by line number. Consider the term “rubella.” Entry of the letters “rub” triggers the list in Table14.6.

Table 14.6 Pick list for rubella

National and international medical classifications, such as ICD, have been developed to facilitate the production of statistics for epidemiology and research. None of these classifications covers the whole field of medicine, and none is suitable for clinical use because their coded content is not sufficiently specific.

James Read aimed to be comprehensive in both breadth of cover and also in the detail of the terms used in general practice. The Read Clinical Classification was based where possible on existing classifications, but large areas of medicine had not been classified before, and Read extended the areas covered by existing schemes to include history, symptoms, examination findings, prevention, and administration (and medication).

Diseases

At the time, the International Classification of Diseases Ninth Revision (ICD-9) was the standard statistical classification of diseases used by hospitals throughout the world. Sections of the Read Clinical Classification which deal with diagnoses, injuries, and death are directly based on ICD-9. The Read first-digit codes A to Q correspond directly to ICD chapters, with the exception of Chapter XVI (symptoms, signs and ill-defined conditions) which was covered in greater detail elsewhere. Each Read category was precisely cross-referenced to ICD.

This section of the Read Classification had 17 first-level codes, 115 two-digit codes, 728 three-digit codes, 2,598 four-digit codes, and 2,575 synonyms. The level of detail at each level is illustrated by the example in Table14.7.

Table 14.7 Hierarchy for cerebral hemorrhage

Procedures

The International Classification of Procedures in Medicine (ICPM) complemented ICD-9 as a standard classification of surgical, diagnostic, and therapeutic procedures, but never received international acceptance. The Read Clinical Classification covered the whole of ICPM with the exception of the section on drugs, medicaments, and biological agents.

In many cases, the content and detail was expanded to provide clinically specific rubrics. For example, the results of laboratory procedures are classified as in Table14.8.

Table 14.8 Urine test for glucose

The decision to include both the procedure (urine test for glucose) and the finding (urine glucose test negative) in the same structure was a mistake which has created problems ever since.

A change, made shortly after the publication of this paper, was to start sublists at 0 rather than 1. The lists shown here are those in the original paper, not those widely implemented.

In operative procedures, mastectomy, for example, is classified as shown in Table14.9. Two problems in the operations sections were the length of many operation names, which required awkward abbreviations that were sometimes hard to understand, and the need for further levels of detail.

Table 14.9 Mastec­tomy

Sections of the Read Clinical Classification covering diagnostic procedures (including laboratory and X-ray) and therapeutic procedures (including surgery) comprised 6,023 code categories and 2,483 synonyms.

History/Symptoms

The history and symptoms section of the Read Classification contains family, social, and medical history as well as presenting symptoms. The relevant section in ICD-9 (Chapter XVI symptoms, signs and ill-defined conditions) is incomplete, and reclassification was needed to provide adequate clinical detail. Where any history/symptom factor has gradable variables, each option is offered as a separate fourth-level category (e.g., see Table14.10).

Table 14.10 Smoking

Each term was defined. “Heavy smoker” is 12–24 cigarettes a day or 80–160 per week, and 20 cigarettes is equivalent to 2 large cigars, 5 medium cigars, 10 small cigars, or an ounce of tobacco.

The history/symptoms section had 1,299 codes and 901 synonyms. History data is of cardinal importance in diagnosis and the prevention of disease and disability. “Listen to the patient, he is trying to tell you the diagnosis.”

Occupations

The OPCS Classification of Occupations was the basis of this section of the Read Clinical Classification with 1,749 coded occupational categories and 936 synonyms. Occupation is an important part of any patient database used for prevention or epidemiology.

Examination/Signs

The classification of examination findings and signs is organized by systems. This part of the Read Clinical Classification was classified from scratch in the absence of any other recognized classification covering patient examination. This section comprised 19 second-level, 282 third-level, and 1,480 fourth-level categories with 890 synonyms.

For example, retinal examination is classified as shown in Table14.11.

Table 14.11 Retinal inspection

Prevention

Preventive procedures were classified from scratch. This was a key section of the classification for GPs, particularly as computer-based prevention records and protocols could lead to significant improvements in the quality of patient care and practice income. This section includes:

• Contraception

• Pregnancy care and birth details

• Child healthcare

• Vaccination and immunization

• Chronic disease monitoring

• Health education and counseling

• Screening, etc.

The level of detail provided for cervical smear screening is shown in Table14.12.

Table 14.12 Cervical screening

The preventive procedures section of the Read Clinical Classification had 1,279 categories and 460 synonyms.

Administration

This section covered all aspects of practice administration. Examples include the stages of patient registration and deregistration, administrative details of patient encounters, processing of claim forms, staff administration, practice finance, and audit reporting.

There are 696 coded categories and 416 synonyms in the administration section of the Read Classification. For example, contraception FPl001 claim status is classified as shown in Table14.13.

Table 14.13 Contraceptive FP1001 claims processing

Drugs

A significant extension, made later in 1986, shortly after the first paper was published, was to extend the scheme to cover medicines. Drugs were given codes starting with lowercase letters a–z, corresponding to the chapters in the first edition of the BNF.

Development

One of the reasons for changing the name from the Abies Medical Dictionary to the eponymous Read codes was to encourage other suppliers to use them too. One of the first to take up this offer was Dr. David Stables at EMIS. James Read set up an independent company known as CAMS (Computer Aided Medical Systems Ltd) to market the codes with royalties split equally between CAMS and Abies. Both recognized that it would make sense for this work to be centrally supported by the Department of Health, rather than by a privately owned computer software developer.

In 1987, the Department of Health commissioned the Joint Computing Group of the BMA’s General Medical Services Committee and the Royal College of General Practitioners to evaluate clinical coding systems for use by GPs. The working party considered the following morbidity coding schemes:

• • ICD-9

• • ICHPPC-2

• • ICPC

• • OXMIS

• • Read Clinical Classification

• • RCGP classification

• • Update morbidity dictionary

• • SNOMED

The final report (August 1988) listed the most important requirements to be:

• • Comprehensive in breadth and depth

• • Appropriate for GP usage

• • Provision for central maintenance

• • Amenable to statistical analysis

• • Compatibility with ICD-9

• • A hierarchical structure (second-level requirement)

• • Accessibility of coding structure to the user (third-level requirement)

The working group recommended the Read codes, with some qualifications:

• • Longer rubrics were needed for operations.

• • Align to national coding schemes (ICD-9, OPCS-4, PPA Drug Index, SOC (Standard Occupational Classification)).

• • A fully resourced UK standing professional committee should be established to maintain and control the classification.

• • Guidance should be provided on usage.

The Department of Health set out to implement these recommendations and after almost 2 years of tortuous negotiations purchased the Read codes for £1.25 million in April 1990, leading to the establishment of the NHS Centre for Coding and Classification [4].

Why Read Codes Were Successful

Features of the first-generation Read codes that made them successful were:

• • A single responsible author/editor

• • Fit for purpose, written by a GP for GPs

• • Comprehensive (examination findings, history, administration, etc.)

• • Modest evolutionary step (built on ICD-9, etc.)

• • Easy to implement in software and on screen

• • Understandable by users

The Read codes improved on earlier classification and coding systems in several respects:

• • They were designed specifically for use by GPs in their surgery, not for epidemiology and international comparisons.

• • The simple position-dependent unidimensional hierarchy was easy to understand by users.

• • No paper version was ever published, facilitating multiple updates and extensions.

• • Easy to implement in software.

Problems

One of the problems with Read coded data has been the quality of information. It is easy to make a mistake when entering data, which seriously impacts data quality. For instance, entry of the termphysio will give a list of options, the first being the occupation[03J1. physiotherapist]. It was easy to choose an occupation when what should have been chosen was[8H77. refers to physiotherapist].

The Read codes combine the features of a classification and a coding scheme. However, no hierarchical coding scheme can ever be multipurpose, because they are built around a single hierarchical axis and each code is classified in one way only. The Read codes proved highly successful in general practice, for which they were designed. However, attempts to use the original versions in hospitals proved impracticable, primarily because the simple hierarchical scheme could reflect only one view, namely, the general practice perspective. Hospital doctors did not understand why information retrieval in one dimension was easy, but in another dimension was difficult and slow.

Position-dependent coding schemes cannot be updated. Once a concept has been placed in the classification, it is not practicable to move it, even if it has been placed in a location that is later regarded as wrong. It is not possible to add in new codes in the middle of a sequence.

Another problem is the inherent multidimensionality of medicine. For example, tuberculosis meningitis is a type of tuberculosis, which is an infectious disease (and is given codeA130.), but it is also an inflammatory disease of the central nervous system and has another codeF004.. Having two separate codes creates code redundancy, which can cause inaccuracies in hierarchy-based analysis of clinical data stored using the codes.

Being restricted to only four levels (later extended to five levels) in the hierarchy causes another problem. Consider the five level hierarchy for mastectomy in Table14.14.

Table 14.14 Mastectomy hierarchy

It is not possible to add a more detailed variant of this operation, such as subcutaneous mastectomy for gynecomastia in the appropriate position because there is no sixth level. A possible solution is to add it as a sibling alongside subcutaneous mastectomy in the fifth level with a code such as 71307. However, this creates the danger that when retrieving cases of subcutaneous mastectomy (71304), those recorded using 71307 would be missed.

The NHS Clinical Terms Project was started in 1992 as a major attempt to address all of the issues listed above. Expenditure on the Read codes between 1990 and 1998 was £32 million [5]. The resulting scheme, which is known as Clinical Terms Version 3 (CTV3), was merged with the College of American Pathologists’ SNOMED RT during 1999–2002 to create SNOMED CT (seeChap. 15). First, we consider the early history of SNOMED.

SNOP and SNOMED

SNOMED has a long history. Back in 1955, the College of American Pathologists (CAP) established a committee to develop a nomenclature for anatomic pathology. In 1965, they published theSystematized Nomenclature of Pathology (SNOP), which describes pathology findings using four axes:

• Topography (anatomic site affected)

• Morphology (structural changes associated with disease)

• Etiology (the cause of disease) including organisms

• Function (physiologic alterations associated with disease)

SNOP was the first multiaxial coding system used in healthcare. By 1975, Roger Côté and colleagues had extended SNOP by adding additional dimensions covering diseases and procedures to give it a broader scope with the nameSystematized Nomenclature of Medicine (SNOMED).

SNOMED was developed around a model of illness that started with normal structure (topography) and function. Sickness typically involves some abnormal function and abnormal structure (morphology). This has some cause (etiology), which may be internal or external. Medicine seeks to reverse the process from the sick state to the healthy state by using administrative, diagnostic, and therapeutic procedures, which act on function or body structure. Disease was added to give easy mapping to ICD. Occupations and organisms were added later. By 1998, SNOMED 3.5 had expanded to 11 axes and 157,000 records (Table14.15).

Table 14.15 SNOMED 3.5 axes

The next generation of SNOMED, SNOMED RT (Reference Terminology), involved a major change. Spackman, Cambell, and Côté [6] describe the need for a reference terminology as follows:

The need for a reference terminology can be illustrated by the situation facing many managed care organizations. They may have several different hospitals and clinics, each with an existing set of information systems. These organizations need to aggregate data from several systems in order to manage the quality and cost of care across the entire organization. Rather than totally replacing their existing information systems with one common system, they have a need to record data from each system using or referring to a common reference terminology. Aggregate data can then be grouped and analyzed using the various hierarchies of the reference terminology [6].

In developing SNOMED RT, the decision was made to adopt the KRSS (Knowledge Representation System Specification) as the description logic syntax for SNOMED RT [7].

SNOMED RT gave each concept code a semantic definition stated in description logic. SNOMED RT was completed in 2000 and provided one of the two key sources for SNOMED CT (seeChap. 15).

LOINC

Laboratory and clinical systems need to merge data for a variety of purposes, including clinical care, quality improvement and reporting, public health reporting, and research. While many systems use electronic messages to transmit results, most systems use their own internal, idiosyncratic codes to identify the results inside those messages. As a result, receiving systems cannot understand their contents without mapping every item to their own codes. This problem would be solved if each system used the same set of universal identifiers for clinical and laboratory observations. LOINC provides such a set of identifiers [8]. Note that LOINC provides codes for the observation names (e.g., eye color), not the observation finding (e.g., blue eyes).

LOINC is a community-built code system that facilitates the exchange, pooling, and processing of laboratory and other clinical observations. It is a controlled terminology that contains unique identifiers and fully specified names built using a ­formal structure, which distinguishes among tests and observations that are clinically different.

The LOINC Committee and development of the database were organized by researchers at the Regenstrief Institute in 1994. LOINC is updated regularly and now contains about 60,000 terms. The Regenstrief Institute serves as the overall steward for the LOINC development effort and works together with the LOINC Committee to define the overall naming conventions and policies for the development process, which is deliberately empirical, nimble and open.

LOINC has been widely adopted. In April 2012 there were more than 17,000 users from 145 countries. LOINC terms have been translated into more than a dozen other languages, with several other translation efforts currently underway. LOINC has been adopted in both the public and private sector by government agencies, laboratories, care delivery organizations, health information exchange efforts, healthcare payers, research organizations, and within many exchange standards.

LOINC creates codes and a formal name for each concept that corresponds to a single kind of observation measurement or test result. The formal LOINC name is fully specified in the sense that it contains the features necessary to disambiguate among similar clinically distinct observations. The fully specified name is constructed using a six-part semantic model to produce a precoordinated expression. This does not capture all possible information about the procedure or result – just enough to unambiguously identify it.

The six parts of the model are:

• • Component or analyte (e.g., potassium)

• • Kind of property (e.g., mass concentration g/L)

• • Time aspect (e.g., point in time or interval)

• • System or sample type (e.g., urine or blood)

• • Type of scale (e.g., quantitative, coded, or narrative)

• • Type of method

In addition to the fully specified name, LOINC also provides alternate names for use in other contexts.

The LOINC structure also contains codes for the atomic elements (parts) that make up the fully specified names. LOINC parts are used to construct hierarchies that organize LOINC terms, link synonyms, and descriptions across many terms and are the basis for an efficient mechanism for translating LOINC names. LOINC also contains a robust model for representing enumerated collections of observations (panels), which captures the hierarchical structure of the elements, attributes of individual data elements, value sets, and panel-specific attributes of data elements.

The laboratory section of LOINC covers the observations that can be made on a specimen and includes most clinical laboratory testing, including chemistry, urinalysis, hematology, serology, microbiology (including parasitology and virology), toxicology, and molecular genetics. Presently, about 70% of the terms in LOINC are for laboratory observations.

The clinical section of LOINC covers the observations that can be made on a whole person or population. Some of the domains covered by Clinical LOINC include vital signs, hemodynamic measurements, anthropomorphic measures, patient assessments, obstetrical ultrasound, radiology reports, and clinical documents and sections.

LOINC is distributed and made freely available from the LOINC website, with new releases published twice yearly. The main LOINC database is distributed in several file formats. In addition, Regenstrief develops and distributes a software program called theRegenstriefLOINCMappingAssistant (RELMA) that provides tools for browsing the database and mapping local terminology to the LOINC terms. Additionally, Regenstrief develops the accompanying documentation and also distributes several accessory files to the main distribution, such as a file containing the full representation of enumerated collections and a hierarchy file.

LOINC terms and other resources have been translated by volunteers into many languages from its native English that are also made available from its website. Regenstrief has developed a refined process for enabling translation of the LOINC terms based on the parts from which they are constructed. By translating a list of parts for the terms of interest, translators only have to translate an element like “glucose” once, and it can then be applied to all of the terms that contain it.

Mapping local observation codes to LOINC provides a bridge across the islands of data that reside in isolated electronic systems. However, mapping is a resource-intensive and often rate-limiting step in interoperable data exchange. RELMA (Regenstrief LOINC Mapping Assistant) helps with this process.

By design, LOINC covers a circumscribed content domain (observation identifiers) and is used in conjunction with other terminology standards. LOINC is included as a source vocabulary in the National Library of Medicine’s Unified Medical Language System (UMLS). LOINC has historically been used in conjunction with the Systematized Nomenclature of Medicine (SNOMED), with LOINC providing codes for the question and SNOMED providing codes for the answer or value. Recently, SNOMED has developed a model for representing observable entities based on LOINC’s semantic structure, and work is underway to more closely coordinate terminology development between LOINC and SNOMED.

The Unified Code for Units of Measure (UCUM) is a standard for electronic communication of units of measure with growing adoption that Regenstrief has included as way to represent units used for a LOINC term within the LOINC database.

LOINC has been implemented in many settings and the adopting communities have uncovered different approaches that address unique features of those contexts. Lessons from nationwide and regional health information exchanges, public health reporting, and implementation in resource-constrained settings continue to inform ongoing developments to improve the effectiveness and efficiency of care delivery.

UMLS

The National Library of Medicine has developed the Unified Medical Language System (UMLS) as a terminology resource, intended for use mainly by developers of health information systems [9]. The UMLS contains a number of knowledge sources and tools, including the UMLS Metathesaurus and Semantic Network.

The UMLS Metathesaurus is a large multipurpose multilingual vocabulary database that contains over five million terms and one million concepts covering information about health and biomedical concepts, their names, and relationships between them. The Metathesaurus is built from over 100 different source vocabularies and seeks to reflect and preserve the meanings concept names and relationships from these sources. For example, if two source vocabularies use the same name for different concepts or define the same concept in different ways, the Metathesaurus represents both sets of meanings and relationships and indicates which meaning is used in which source vocabulary. The Metathesaurus preserves the many views of the world present in its source vocabularies because these views may be useful for different tasks.

The UMLS Semantic Network consists of a set of broad subject categories, referred to as Semantic Types, and a set of relationships (Semantic Relations), which exist between them. The Semantic Types are similar to, but not the same as the SNOMED CT hierarchies. There are 133 Semantic Types.

The Semantic Relations are similar to Relationships in SNOMED CT and include the “isa” subtype relationship for hierarchies and other nonhierarchical relationships grouped according to: physically related to, spatially related to, temporally related to, functionally related to, and conceptually related to.

Another UMLS product is the CORE (Clinical Observations Recording and Encoding) Problem List Subset of SNOMED CT, which is a common list of SNOMED CT concepts for use in problem lists and patient summaries [10].