Lexical Matching

The simplest form of lexical matching that identifies mappings only when two concept labels are character-for-character identical after preprocessing. This technique is highly precise but suffers from low recall due to minor orthographic variations.

• Normalization required: Case folding, whitespace trimming, and punctuation removal
• Use case: Matching Diabetes Mellitus to Diabetes Mellitus across systems
• Limitation: Fails on Type 2 DM vs Diabetes Mellitus Type 2

Enhances exact matching by applying a standardization pipeline to both source and target labels before comparison. This addresses systematic differences in casing, punctuation, and stop word usage that otherwise prevent direct alignment.

• Steps: Lowercasing → diacritic removal → stop word filtering → punctuation stripping
• Example: Diabetes Mellitus, Type II normalizes to diabetes mellitus type ii
• Strength: Catches trivial formatting mismatches without semantic analysis

Quantifies string similarity by calculating the minimum number of single-character operations required to transform one string into another. The Levenshtein distance is the most common variant, counting insertions, deletions, and substitutions.

• Levenshtein: SNOMED → SNOMED CT has a distance of 3 (space, C, T)
• Damerau-Levenshtein: Adds transposition operations for common typos
• Thresholding: Mappings accepted only when normalized distance falls below a configured cutoff

Decomposes concept labels into individual word tokens and compares sets rather than character sequences. This approach is robust to word reordering and partial matches, making it suitable for multi-word clinical terms.

• Jaccard Similarity: Intersection over union of token sets
• Dice Coefficient: Twice the intersection divided by total token count
• Example: Acute Myocardial Infarction and Myocardial Infarction, Acute yield a Jaccard score of 1.0 after tokenization

Generates overlapping character subsequences of length n from each label and compares the resulting fingerprint sets. This technique captures sub-word similarities and is resilient to spelling variations and morphological differences.

• Common n: Bigrams (2) and trigrams (3) for clinical terminology
• Advantage: Detects similarity between Hypertension and Hypertensive via shared trigrams
• Application: Often used as a fast pre-filter before more expensive semantic matching

Encodes strings based on their pronunciation patterns to identify mappings between labels that sound alike but are spelled differently. This is particularly valuable for reconciling drug names and clinician dictation errors.

• Metaphone & Double Metaphone: Produce phonetic hashes for English-language terms
• Example: Furosemide and Frusemide generate identical phonetic codes
• Limitation: Language-specific; requires adaptation for multilingual terminology servers

An ontology alignment technique that leverages the formal semantics, hierarchical context, and logical axioms of concepts to determine similarity, rather than relying solely on string comparison.

• Uses description logic and OWL axioms to infer equivalence
• Considers subsumption relationships (parent-child hierarchies)
• Employs reasoners to detect logical inconsistencies in proposed mappings
• More computationally intensive but yields higher precision than lexical methods

A type of ontology alignment that asserts a relationship of logical equality or interchangeability between a concept in a source code system and a concept in a target code system.

• Defines mappings as equivalent, wider, narrower, or unmatched
• Critical for bidirectional mapping where round-trip translation must preserve meaning
• Represented in FHIR via the ConceptMap resource with equivalence property values
• Requires domain expert validation for clinical safety in high-stakes contexts

The task of linking disparate textual mentions of a clinical entity to a single, unique concept identifier in a standard terminology.

• Resolves variations like 'heart attack', 'MI', and 'myocardial infarction' to a single SNOMED CT code
• Often uses BERT-based alignment with contextual embeddings to capture semantic nuance
• Essential preprocessing step for clinical entity linking and cohort identification
• Outputs include a confidence score indicating mapping reliability

A software application providing a central repository and API for storing, querying, and distributing standardized medical code systems and value sets.

• Hosts terminologies like SNOMED CT, ICD-10-CM, LOINC, and RxNorm
• Exposes FHIR Terminology Service endpoints for code validation and translation
• Manages version migration when terminologies release updates with deprecated or retired concepts
• Enables real-time canonicalization of clinical data at enterprise scale

A computational measure of the closeness of meaning between two concepts, often calculated based on their distance and properties within an ontological graph.

• Path-based measures count edges between concepts in the hierarchy
• Information content measures weight concepts by their specificity and frequency
• Used to rank candidate mappings and set thresholds for automated alignment
• Complements lexical matching by adding structural context from the ontology graph

A workflow where a domain expert reviews, accepts, or rejects algorithmically generated ontology mappings to ensure final accuracy and clinical safety.

• Triggered when confidence scores fall below a defined threshold
• Generates mapping provenance metadata for audit trails and governance
• Essential for high-risk mappings involving medication codes (RxNorm) or diagnoses (ICD-10-CM)
• Balances automation efficiency with the clinical imperative of zero-error tolerance