parsing-lab-values
GitHub解析临床文本中的实验室数值及参考范围,判定结果为低、正常、高或危急。支持多种格式(如区间、不等式),优先尊重原实验室标记,返回结构化异常标志,用于辅助解读化验单。
Trigger Scenarios
Install
npx skills add maziyarpanahi/openmed --skill parsing-lab-values -g -y
SKILL.md
Frontmatter
{
"name": "parsing-lab-values",
"license": "Apache-2.0",
"metadata": {
"pairs": "after",
"project": "OpenMed",
"version": "1.0",
"category": "clinical-nlp"
},
"description": "Parse laboratory values and reference ranges from clinical text and flag results as low, normal, high, or critical with OpenMed. Use when the user needs to interpret lab results, compute abnormal flags, parse reference ranges like \"135-145\" or \"<5\", honor an originating-lab flag (H\/L\/critical), or turn extracted lab entities into structured high\/low\/critical signals. Covers openmed.clinical.parse_reference_range, derive_abnormal_flag, ReferenceRange, and AbnormalFlag, with UCUM\/LOINC framing. Unit-agnostic — it does not convert units. Pairs after extracting-clinical-entities (lab entities from analyze_text)."
}
Parsing lab values
Lab results in clinical text arrive as a value, a unit, and a reference range
("Sodium 132 mmol/L (135–145)"). To act on them you need a structured
abnormal flag — is 132 low, normal, high, or critical? OpenMed's
openmed.clinical lab helpers parse the reference range deterministically and
derive the flag, honoring any explicit flag the originating lab already supplied.
The helpers are unit-agnostic by design: they compare numbers within a stated
range and never convert units, so a mmol/L value is never silently compared
against a mg/dL range.
When to use
- After
extracting-clinical-entitiessurfaces lab/measurement entities and you need to classify each as low / normal / high / critical. - The user asks to parse reference ranges, flag abnormal labs, build a flagged
labs table, or interpret values like
<5,>=10,0.5 - 1.2. - You have an originating-lab flag (
H,L,C,HH) and want it honored over a derived comparison.
Quick start
from openmed.clinical import (
parse_reference_range, derive_abnormal_flag, LAB_FLAG_ADVISORY,
)
# Closed range
rng = parse_reference_range("135-145")
# -> {"low": 135.0, "high": 145.0, "low_inclusive": True, "high_inclusive": True}
derive_abnormal_flag(132, rng) # "low"
derive_abnormal_flag(140, "135-145") # "normal" (raw range string accepted)
derive_abnormal_flag(150, "135 to 145") # "high"
# One-sided bounds
derive_abnormal_flag(7, parse_reference_range("<5")) # "high" (above the cap)
derive_abnormal_flag(3, parse_reference_range(">=10")) # "low"
# Honor the lab's own explicit flag (takes precedence over derived comparison)
derive_abnormal_flag(132, "135-145", explicit_flag="C") # "critical"
derive_abnormal_flag(132, "135-145", explicit_flag="HH") # "critical"
# Unparseable / non-numeric inputs fail safe rather than guessing
derive_abnormal_flag("pending", "135-145") # "unknown"
derive_abnormal_flag(132, "see report") # "unknown"
print(LAB_FLAG_ADVISORY) # surface this disclaimer with derived flags
AbnormalFlag is one of "low" | "normal" | "high" | "critical" | "unknown".
ReferenceRange is a typed mapping of low, high, low_inclusive,
high_inclusive.
Workflow
- Get value + range + (optional) lab flag from extracted lab entities. The value should be numeric; the range may be a raw string or a parsed mapping.
- Parse the reference range with
parse_reference_range. It handles closed ranges ("135-145","0.5 - 1.2","135 to 145", en/em dashes) and one-sided bounds ("<5","<=5",">10",">=10"). Contradictory or unparseable ranges return empty bounds rather than a guess — by design. - Derive the flag with
derive_abnormal_flag(value, range, explicit_flag=). Resolution order: an explicit lab flag wins first (H/HIGH,L/LOW,C/CRIT/CRITICAL,HH/LL→ critical,N/NORMAL); an unknown explicit flag returns"unknown"instead of being silently ignored. With no explicit flag, it compares the numeric value against the parsed bounds, respecting inclusive vs. exclusive edges. - Handle
"unknown"explicitly. Non-numeric values, empty/unparseable ranges, or unrecognized explicit flags yield"unknown". Treat it as "needs review," not "normal." - Attach the advisory. Surface
LAB_FLAG_ADVISORYwherever derived flags are shown — derived flags are heuristic and do not replace the originating laboratory's own diagnostic flagging.
Hand-off to / from OpenMed
- From
extracting-clinical-entities:analyze_textlab/measurement entities give you the value text, unit, and often the reference range; this skill turns them into structured flags. Parse the numeric value out of the entity surface before callingderive_abnormal_flag. - OpenMed calls:
from openmed.clinical import parse_reference_range, derive_abnormal_flag, ReferenceRange, AbnormalFlag, LAB_FLAG_ADVISORY. - To
reconciling-problem-lists/ FHIR grounding: acritical/high/lowflag becomes a FHIRObservation.interpretationcode (HL7 v3 ObservationInterpretation:H,L,HH,LL,N). Ground the LOINC code and UCUM unit out-of-process; OpenMed emits the flag, not the terminology binding.
Edge cases & gotchas
- Unit-agnostic — convert before comparing. The helpers ignore units entirely. If the value and the range are in different units (mg/dL vs mmol/L), the flag is wrong. Normalize units before calling, or only compare value and range that share a unit.
- Inclusive vs. exclusive edges.
"<5"makes 5 the high bound exclusive; a value of exactly 5 flagshigh.parse_reference_rangerecordshigh_inclusive=Falsefor<andTruefor<=— respect it. - Critical needs an explicit flag. Derived comparison yields only low/normal/
high.
"critical"comes from the lab's explicit flag (C,HH,LL); the helpers do not infer critical thresholds beyond the reference range. - Empty bounds are intentional. A range with both bounds missing returns
"unknown"fromderive_abnormal_flag, not"normal". Don't treat unknown as in-range. - Local-first, advisory-only. Runs on-device; flags are decision support, not
a diagnosis. Always carry
LAB_FLAG_ADVISORY.
Standards & references
- LOINC — universal codes for laboratory observations: https://loinc.org/
- UCUM — Unified Code for Units of Measure: https://ucum.org/
- HL7 v3 ObservationInterpretation (H/L/HH/LL/N abnormal flags): https://terminology.hl7.org/CodeSystem-v3-ObservationInterpretation.html
- HL7 FHIR R4 Observation —
referenceRangeandinterpretation: https://hl7.org/fhir/R4/observation.html
Version History
- f213557 Current 2026-07-23 00:45


