Medication-status contradiction precheck¶

medication_status_contradiction_precheck() compares explicit medication status and event-time evidence before a clinical NLI model runs. It is a deterministic local consistency check, not a medication recommendation engine.

from openmed.clinical.nli_medication_status import (
    medication_status_contradiction_precheck,
)

result = medication_status_contradiction_precheck(
    {
        "medication_key": "synthetic-medication",
        "status": "continued",
        "event_time": "2026-01-02",
    },
    {
        "medication_key": "synthetic-medication",
        "status": "stopped",
        "event_time": "2026-01-02",
    },
)

assert result.contradiction
assert result.inference_allowed is False

Status and time contract¶

The controlled statuses are started, continued, held, changed, stopped, historical, and uncertain. Common explicit aliases are normalized. Missing status becomes uncertain; it is never defaulted to an active or recommended state.

Absolute event times use the existing medication reconciliation timestamp contract. Relative expressions must first be normalized by a timeline layer with an explicit anchor.

The result status is one of:

• compatible: controlled statuses agree without a time conflict.
• contradiction: explicit active and inactive states conflict at the same normalized event time.
• review_required: a state is uncertain or changed, a status transition has different times, or the times needed to order a disagreement are missing.
• not_applicable: both records identify different medications.

Ordered transitions are escalated even when they appear plausible. The precheck does not infer which regimen is current, preferred, or recommended.

Privacy and scope¶

Reports contain controlled status labels, a coarse time relation, source offsets, and domain-separated SHA-256 fingerprints. They do not contain raw medication identities, raw event times, source text, or arbitrary status strings. Errors are value-free. The implementation performs no network call, uses no wall clock, and loads no terminology data.

The result is an assistive consistency signal for human review. It is not a prescription, medication order, treatment recommendation, or autonomous clinical decision.