SDOH Negated-Need Resolution¶

SDOH extractors can surface the same determinant in an asserted need and in a screening denial. openmed.clinical.sdoh_negated_need resolves each caller supplied finding in its own local sentence and clause so a denial for one determinant does not bleed into a neighboring finding.

This is a deterministic, offline review aid. It does not infer a need from a missing mention, assign a Z-code, or make a clinical decision.

Resolve findings¶

Pass the source text and findings with half-open start/end offsets. Existing SDOHFinding values and mappings are accepted. The returned records are safe metadata: they retain offsets and controlled labels, but not the finding value or cue text.

from openmed.clinical import resolve_sdoh_negated_needs

text = "No food insecurity, but reports a transportation barrier."
findings = [
    {
        "category": "food_insecurity",
        "start": text.index("food insecurity"),
        "end": text.index("food insecurity") + len("food insecurity"),
    },
    {
        "category": "transportation_barrier",
        "start": text.index("transportation barrier"),
        "end": text.index("transportation barrier")
        + len("transportation barrier"),
    },
]

results = resolve_sdoh_negated_needs(text, findings)
[(item.assertion, item.need_status) for item in results]
# [('negated', 'absent'), ('affirmed', 'present')]

Results are sorted by source offsets. input_index preserves the original finding position when a caller needs to join the safe metadata to a protected in-memory collection.

Assertion contract¶

The resolver keeps negation determinant-specific. Coordinated denials such as denies food insecurity and transportation barriers can cover both findings; an adversative or newly asserted clause such as but reports transportation starts a separate scope. Backward cues such as not present, absent, and ruled out are supported when they follow the finding.

Double negation is never reduced with even/odd parity. For example, not no is returned as assertion="unknown", double_negation=True, and review_required=True. Contradictory local cues such as a denied finding that is also explicitly present likewise become unknown and carry the controlled "contradictory_cues" review reason.

Pseudo-negation and incomplete documentation are not treated as a denial. Phrases such as cannot be excluded, not ruled out, not documented, and unknown produce an unresolved record for review.

Privacy and provenance¶

SDOHNegatedNeedEvidence.to_dict() and .to_json() contain only:

• source and cue offsets;
• canonical determinant and assertion labels;
• controlled review reasons and flags; and
• the caller input index.

They do not include text, value, raw cues, logs, or exception payloads. The original finding remains caller-owned and should stay in the authorized review workflow. The module uses only local regular-expression rules, performs no mandatory network call, and does not load restricted SDOH datasets.

All bundled examples and tests use synthetic wording. The output is not a compliance certification or an autonomous clinical decision guarantee; review the source span before downstream coding or referral action.