Medical Subject Headings contain two operationally distinct record classes. Descriptor Records define the formal subject vocabulary. Supplementary Concept Records represent specific chemicals, drugs, rare diseases, and other narrow concepts that map to formal descriptors. The distinction determines which records should serve as primary retrieval answers and which should supply supporting detail.
Record classes: Formal headings and mapped concepts
A Descriptor Record is a stable indexing heading with one or more positions in the MeSH hierarchy. A Supplementary Concept Record does not establish a parallel hierarchy. It connects a specific concept to one or more descriptors through mapped headings.
| Property | Descriptor Record | Supplementary Concept Record |
|---|---|---|
| Identifier pattern | D009203 | C550102 |
| Example | Myocardial Infarction | xinkeshu |
| Primary function | Stable subject heading for indexing and retrieval | Specific chemical, drug, rare disease, or narrow concept |
| Hierarchy behavior | Occupies one or more MeSH tree positions | Maps to one or more formal descriptors |
| Retrieval role | Primary answer vocabulary | Supporting detail vocabulary |
Vocabulary composition: Supplementary concepts dominate the candidate set
The measured 2026 vocabulary contains substantially more supplementary concepts than descriptors. A single mixed vector index therefore gives narrow C records many more opportunities to occupy early ranking positions.
Mapped headings: Supplementary concepts resolve to descriptors
Mapped headings state which formal descriptor represents a supplementary concept during indexing and retrieval. The mapping retains the specific concept while placing it inside the formal subject vocabulary.
| C identifier | Supplementary concept | Mapping | D identifier | Formal descriptor |
|---|---|---|---|---|
| C550102 | xinkeshu · specific Chinese herbal preparation | → | D004365 | Drugs, Chinese Herbal |
| C536782 | T-cell lymphoma 1A · narrow gene or protein concept | → | D016399 | Lymphoma, T-Cell |
| C000608195 | GSK1070806 · humanized monoclonal antibody | → | D061067 | Antibodies, Monoclonal, Humanized |
Mixed-index retrieval: Candidate volume changes the answer space
A search across all 355,144 D and C vectors ranks both record classes inside one candidate pool. For a general query such as “Blood thinner for irregular heartbeat,” narrow supplementary concepts can displace the formal headings expected as primary answers.
| Retrieval lane | Representative results | Interpretation |
|---|---|---|
| Mixed D and C index | C550102 · xinkeshu; C566055 · bleeding disorder concept; C566237 · atrial tachyarrhythmia concept | Specific concepts can displace the intended formal heading. |
| Descriptor-first index | D000925 · Anticoagulants; D001281 · Atrial Fibrillation; D001145 · Arrhythmias, Cardiac | The result vocabulary matches the formal answer contract. |
Retrieval architectures: Three candidate structures
The architecture determines the vocabulary returned to the reader. The same embeddings can support each structure, but the result contract differs.
| Architecture | Retrieval path | Strength | Boundary |
|---|---|---|---|
| One mixed ranking | Query → 355,144 D and C vectors → one ranking | Maximum concept specificity | Obscure concepts frequently displace general headings. |
| Descriptor-only ranking | Query → 31,110 D vectors → formal headings | Clean primary answer vocabulary | Exact drugs and emerging concepts receive less coverage. |
| Two lanes, one answer | Query → D ranking; query → C evidence → mapped D; merge → formal answer with supporting terms | Formal answers retain specific evidence | Requires explicit lane merging and mapping logic. |
Implementation consequences
- Store the record class and mapped descriptor identifiers with every embedded concept.
- Rank descriptor and supplementary vectors in separate retrieval lanes.
- Project supplementary evidence onto mapped descriptors before final answer selection.
- Return the formal descriptor as the primary result and retain matched supplementary concepts as supporting evidence.
References
- MeSH XML data elements: descriptor and supplementary concept record definitions National Library of Medicine grounded
- Supplementary Concept Records: mapping and indexing guidance National Library of Medicine grounded