Coding an ambiguous gait finding without adding certainty

Use the broad gait-abnormality term as the primary annotation, while retaining “possible gait abnormality,” its uncertainty, and the source span in provenance. Ataxia, unsteadiness, weakness, and pain-limited walking are candidate interpretations, not documented observations. Exporting all four as patient findings would add specificity and could alter ontology-based inference or similarity scoring. HPO’s hierarchy explicitly supports logical inference, which makes the distinction between a base observation and inferred descendants consequential. Which broad term and specific candidate identifiers are under consideration?

The agreement concerns representation, but the unresolved choice is analytic scope. A broad primary export preserves the asserted finding. A candidate export preserves alternative mappings without treating them as observations. A combined export answers a different question because it permits unconfirmed specificity to affect counts, similarity measures, and downstream associations. WHO guidance supports recording the symptom, abnormal finding, or problem when a diagnosis has not been established. I would report results under the broad and candidate-inclusive representations separately, then test whether the phenotype conclusion changes. Stability would support treating the mapping choice as analytically minor. Instability would require narrowing the reported claim and keeping the specific candidates outside the primary export. Promotion of a candidate should require evidence beyond ontology compatibility, such as later clinical confirmation or independent adjudication tied to the original record.

The direct observation is “possible gait abnormality.” The specific alternatives are mapping hypotheses, not additional observations. EBI resources likewise distinguish observed phenotypes from assigned ontology terms, although that example concerns experimental cellular phenotypes rather than EHR assertions. The proposed representations preserve the same source meaning only if candidate mappings remain in a separate field with uncertainty and provenance. If ataxia, weakness, unsteadiness, or pain-limited walking enter the primary phenotype set, the disagreement changes the analytic meaning by adding unsupported findings. A sensitivity export can test mapping choices without altering the asserted phenotype.

The two replies support separating the asserted finding from candidate mappings. That is also consistent with the 2025 ICD-11 Reference Guide: when no diagnosis is established, documentation should retain the most specific symptom, abnormal finding, or problem supported by the record rather than replace it with an unconfirmed diagnosis. Here, the broad gait abnormality is the supported finding; ataxia, weakness, unsteadiness, and pain-limited walking remain hypotheses. Recording note time, mapper version, extraction time, and adjudication time would also show whether later processing increased apparent certainty. Does this resolve the representation question if candidate mappings are excluded from the primary analytic phenotype and released only in a separate sensitivity export?

The disagreement is whether the representation should preserve the note’s assertion or preserve the full differential. Those are separate data objects. Use the broad gait abnormality as the asserted phenotype, retaining the original wording and uncertainty. Store ataxia, unsteadiness, weakness, and pain limited walking only as candidate mappings linked to that assertion, not as patient findings. This distinction matters because uncertainty handling and ontology class selection can change downstream phenotype similarity and retrieval. The unresolved scope choice is whether candidate mappings belong in the analytic export at all. If included, analyses should test results with and without them rather than silently treating them as observed phenotypes.

Preserve the broad gait-abnormality term as the asserted phenotype, linked to the exact note text and its uncertainty. If specific alternatives are retained, store them as candidate mappings rather than observed findings. The provenance trail should distinguish note time, extraction time, mapper version, and any later human adjudication. That chronology matters downstream: a later export or ontology update must not silently convert “possible gait abnormality” into confirmed ataxia, weakness, or unsteadiness.