Clinical NLP

<!--
# COPYRIGHT NOTICE
# This file is part of the "Universal Biomedical Skills" project.
# Copyright (c) 2026 MD BABU MIA, PhD <md.babu.mia@mssm.edu>
# All Rights Reserved.
#
# This code is proprietary and confidential.
# Unauthorized copying of this file, via any medium is strictly prohibited.
#
# Provenance: Authenticated by MD BABU MIA

-->

---
name: 'clinical-nlp-extractor'
description: 'Extracts medical entities (Diseases, Medications, Procedures), patient outcomes with temporal anchors, and registry-oriented real-world data from unstructured clinical text using regex and simple rules (or LLM wrappers).'
measurable_outcome: Execute skill workflow successfully with valid output within 15 minutes.
allowed-tools:
  - read_file
  - run_shell_command
---


# Clinical NLP Entity Extractor

The **Clinical NLP Skill** converts free-text clinical notes into structured data. It identifies key medical entities like problems/diagnoses, medications, and procedures.

## When to Use This Skill

*   When analyzing unstructured EHR notes.
*   To populate a patient's problem list or medication reconciliation.
*   To de-identify text (phi-removal) - *Basic version*.

## Core Capabilities

1.  **NER (Named Entity Recognition)**: Extracts Problems, Drugs, Procedures.
2.  **Negation Detection**: (Basic) Checks if a finding is denied ("No fever").
3.  **Structuring**: Returns JSON format compatible with FHIR/USDL.
4.  **Rare-neoplasm RWD abstraction with LLMs**: For registry-quality capture in rare neoplasms such as bone sarcoma, design entity and outcome schemas before extraction, map outputs to registry fields, anchor temporal events such as diagnosis, treatment, progression, recurrence, and follow-up, normalize low-prevalence entities to the cohort vocabulary, preserve criterion-level provenance and source-note evidence for each extracted item, label uncertain values, run abstraction QA against expected field formats and clinician-defined vocabularies, and route sarcoma-style sparse-cohort or low-confidence cases through clinician adjudication loops.
5.  **LLM-assisted rare-neoplasm RWD collection**: For bone sarcoma-like low-volume cohorts, use LLM-assisted abstraction from clinician notes to map extracted facts to registry variables, attach temporal anchors to diagnosis, treatment, response, progression, recurrence, and follow-up events, handle PHI through the project de-identification or secure processing workflow before secondary use, sample extractions for validation against manual review, and send ambiguous, discordant, or clinically critical fields to clinician adjudication before release.
6.  **Outcome extraction with temporal anchoring**: Identify patient outcome mentions in clinician notes, normalize outcome events and event timing, label temporal relations and current/past/future outcome status, infer explicit or relative dates/times plus event onset and resolution windows, distinguish baseline from follow-up events, classify uncertainty and negation, validate extracted outcomes against chart-review labels for RWE workflows, run longitudinal consistency checks across notes, retain audit-ready source-note evidence spans for each extracted outcome, and generate downstream registry/RWE-ready tables with outcome status, event timing, temporal anchor, uncertainty class, negation, provenance, and audit trails.
7.  **Sparse oncology registry extraction**: Extract entities and outcomes from oncology notes for rare neoplasms such as bone sarcoma, map each abstracted value to registry fields, anchor findings to explicit or relative clinical timelines, route missing or uncertain values to human validation queues, and retain audit-ready provenance linking extracted values to source notes, evidence spans, and validation status.
8.  **Rare-neoplasm RWD variable extraction**: Before extracting oncology rare-neoplasm real-world data, define registry variables for stage, treatments, outcomes, and follow-up; extract stage, treatment, and outcome fields from notes with explicit temporal anchors; support sparse and variable terminology common in rare cohorts; and validate extracted variables against manual abstraction before using them in downstream analyses.
9.  **Clinician-note outcome timing extraction**: Identify outcome mentions in clinician notes, normalize event dates and clinically described time windows, distinguish historical, current, and future outcomes, retain uncertainty labels, and emit evidence spans suitable for registry or real-world-data workflows.
10. **Outcome temporal anchoring from clinician notes**: Identify outcome mentions, normalize explicit dates and relative intervals, distinguish current versus historical events, attach evidence spans to each outcome-timing assertion, and flag uncertain timing for review.
11. **Criterion-level rare-neoplasm registry abstraction**: Use LLMs to abstract registry variables from oncology notes for rare neoplasms, preserve criterion-level provenance for each extracted value, handle sparse disease-specific vocabularies such as bone sarcoma terminology, and validate extracted variables against manual abstraction before downstream real-world-data use.
12. **Rare-neoplasm registry workflow fields**: For rare-neoplasm real-world-data extraction, define registry fields for tumor subtype, treatment lines, response, progression, and follow-up dates before running LLM extraction; require uncertainty labels for extracted values and manual QA before releasing sparse rare cancer cohort datasets.
13. **Rare-neoplasm clinician-note RWD extraction mode**: Use LLM-assisted abstraction from clinician notes for sparse rare-neoplasm cohorts such as bone sarcoma, predefine registry-style fields and allowed values, validate temporal consistency across diagnosis, treatment, response, progression, recurrence, and follow-up dates, and route ambiguous or low-evidence abstractions to manual adjudication before registry use.
14. **Real-world oncology abstraction audit workflow**: For rare-neoplasm clinical notes, use LLMs to extract registry variables, temporal outcomes, treatment lines, and evidence snippets, then run audit sampling against source-note snippets and send uncertain, discordant, or sampled fields to human adjudication before registry release.
15. **Rare-neoplasm clinician-note RWD extraction pattern**: For oncology real-world-data studies in rare neoplasms, start from registry variable dictionaries, normalize tumor-specific entities such as bone sarcoma terms to study vocabularies, attach temporal anchors to diagnosis, treatment, response, progression, recurrence, and follow-up events, emit missingness flags for absent or indeterminate variables, run human abstraction audit on extracted fields and evidence spans, and export structured registry-ready tables for downstream oncology RWD analysis.
16. **Clinician-note outcome identification and timing validation**: Use LLM-assisted extraction to identify patient outcomes in clinician notes, anchor each outcome to event time, handle negated or uncertain outcome mentions, reconcile outcome status across longitudinal notes, and validate structured outputs against chart-reviewed labels before downstream use.
17. **LLM rare-neoplasm registry abstraction pattern**: For rare-neoplasm RWD collection from clinician notes, extract registry fields with bone sarcoma example variables such as tumor subtype, anatomic site, stage or extent, treatment lines, response, progression, recurrence, and follow-up dates; attach source-span evidence to each field, emit uncertainty and missingness flags, route low-confidence or discordant fields to adjudication queues, and validate extracted registry values against manual abstraction before analysis.
18. **Clinician-note outcome event timing aggregation**: Extract patient outcome events from clinician notes with timing normalization, source-span evidence, and uncertainty labels; aggregate event status and timing longitudinally at the patient level; and validate extracted event/timing outputs against chart review before registry or real-world-evidence use.
19. **Bone sarcoma rare-neoplasm RWD extraction validation**: For registry-oriented rare-neoplasm extraction, use bone sarcoma as the exemplar for predefined entity schemas, temporal anchors, and outcome/event tables; route ambiguous, discordant, or clinically critical fields through adjudication loops; and validate precision and recall against manually abstracted charts before downstream registry or real-world-data analysis.
20. **Patient outcome extraction with temporal anchoring**: Identify outcome mentions in clinician notes, normalize event timing, classify outcome status and uncertainty, link each structured assertion to source-note evidence spans, reconcile timing across longitudinal clinical datasets, and validate extracted outcome/timing labels against chart-review labels before downstream use.
21. **Rare-neoplasm real-world-data extraction from clinician notes**: Define registry variables before abstraction, extract treatment and outcome fields from clinician notes, handle sparse bone sarcoma examples with explicit missingness and uncertainty labels, and require abstraction QA with clinician adjudication for ambiguous, discordant, or clinically critical fields.
22. **Patient outcome timing tables for RWE**: Extract patient outcomes from clinician notes with temporal anchoring, normalize event times from explicit and relative note context, assign uncertainty labels, route ambiguous outcome status or timing to adjudication queues, and emit longitudinal outcome tables for real-world evidence workflows.
23. **Bone sarcoma rare-neoplasm RWD extraction pattern**: Use bone sarcoma as the rare-neoplasm example for LLM-assisted real-world data collection by predefining registry variables, extracting temporally anchored clinical events, preserving evidence spans for each value, assigning uncertainty labels, and sending uncertain or clinically important abstractions through clinician adjudication loops.
24. **Patient outcome identification with adjudicated timelines**: Use LLM-assisted review of clinician notes to identify patient outcomes, anchor outcome onset and resolution timing, preserve source-note evidence spans for each timing assertion, assign uncertainty classes, and validate extracted outcomes against adjudicated outcome timelines before downstream use.
25. **Bone sarcoma-style LLM registry abstraction**: For rare-neoplasm real-world-data extraction, predefine structured oncology variables before running LLM abstraction, require source-span evidence for each extracted value, explicitly separate missing values from negative findings, validate outputs against manual abstraction, and track temporal provenance from source notes for diagnosis, treatment, response, progression, recurrence, and follow-up events.
26. **Outcome extraction with structured variable linkage**: Use LLM-assisted extraction to identify patient outcome mentions in clinician notes, normalize event timing, distinguish historical, current, and future events, link extracted spans to structured variables, and validate extracted outcomes against adjudicated chart-review labels.
27. **Oncology rare-neoplasm RWD extraction workflow**: For rare neoplasms such as bone sarcoma, define registry fields before abstraction, extract candidate values from clinical notes with LLM-assisted adjudication, preserve temporal provenance for each registry value and event, audit missingness separately from negative findings, and require clinician validation before registry release or downstream analysis.
28. **Patient outcome event extraction for RWE**: Define outcome event schemas before extraction, identify patient outcomes in clinician notes, normalize explicit and relative dates/times into structured temporal anchors, label uncertainty for outcome status and timing, preserve source-note evidence for each event assertion, and validate extracted events and timing against chart review before real-world-evidence use.
29. **Rare-neoplasm small-cohort RWD extraction pattern**: For unstructured oncology notes in small rare-neoplasm cohorts, map each extracted item to a predefined registry schema, extract temporally anchored events such as diagnosis, treatment, recurrence, progression, and follow-up, emit missingness flags that separate absent documentation from negative findings or indeterminate values, sample abstractions for clinician validation, and use bone sarcoma-style examples such as tumor subtype, anatomic site, treatment lines, recurrence, progression, and follow-up dates to guide schema-level abstraction.
30. **Clinician-note patient outcome ontology extraction**: Extract patient outcomes from clinician notes by mapping mentions to a study outcome ontology, normalizing event times into explicit temporal anchors, labeling outcome and timing uncertainty, sampling extracted cases for clinician adjudication, and generating downstream real-world-evidence tables with outcome, event time, uncertainty, provenance, and adjudication status fields.
31. **Clinician-note outcome extraction with temporal gold-standard validation**: Use LLM-assisted extraction to identify and time patient outcomes in clinician notes, normalize event times from explicit dates and relative note context, assign uncertainty labels for outcome status and timing, aggregate outcome mentions across longitudinal notes into patient-level timelines, and validate event and timing outputs against chart-review gold standards before downstream use.
32. **Rare-neoplasm RWD extraction with abstention**: For rare-neoplasm registries inspired by the bone sarcoma example, define registry schemas before extraction, extract candidate variables from clinical notes with source evidence spans, preserve provenance for each value and source note, abstain rather than infer when required fields are missing or unsupported, and validate structured outputs against chart abstraction before downstream analysis.
33. **Clinician-note outcome extraction with mention and patient-level evaluation**: Identify outcome mentions in clinician notes, normalize event timing, distinguish historical from future outcome events, retain text evidence spans for each extracted assertion, and evaluate extraction accuracy at both mention and patient levels before downstream use.
34. **Bone sarcoma rare-neoplasm registry extraction**: For low-prevalence oncology cohorts, use bone sarcoma as the rare-neoplasm example by defining registry schemas before abstraction, extracting outcome and treatment events from clinical notes, validating event order and dates across the patient timeline, and requiring clinician audit before registry release or downstream real-world-data use.
35. **Rare-neoplasm chart-note RWD abstraction QA**: For LLM-assisted rare-neoplasm data collection from clinician notes, map each extracted value to predefined registry fields, use bone sarcoma examples such as subtype, site, stage or extent, treatment lines, response, progression, recurrence, and follow-up dates, anchor events to explicit or relative clinical timelines, distinguish missing documentation from negative or indeterminate findings, validate extracted fields against chart review, and report abstraction QA metrics at the field and patient levels without inventing unsupported values.
36. **Longitudinal outcome timing extraction for RWE**: Extract patient outcomes and outcome timing from clinician notes, normalize explicit and relative temporal expressions, assign event-status labels, preserve evidence spans for each outcome-time assertion, route ambiguous or discordant events through adjudication workflows, and report validation metrics against adjudicated longitudinal real-world-evidence datasets.
37. **LLM rare-neoplasm RWD extraction workflow**: Structure clinician notes into registry-ready real-world-data fields for low-prevalence cancer cohorts, using bone sarcoma examples to map tumor, treatment, response, progression, recurrence, and follow-up details to a predefined registry dictionary; retain source-note evidence and temporal anchors, run abstraction QA for missing, indeterminate, or discordant fields, and send low-confidence or clinically important abstractions to human adjudication.
38. **Outcome extraction with patient-level temporal aggregation**: Use LLM-assisted extraction to identify patient outcomes in clinician notes, normalize each outcome to event time, assign uncertainty labels for outcome status and timing, aggregate mention-level events into patient-level outcome timelines, and validate extracted outcomes against manually abstracted outcomes before downstream use.
39. **Bone sarcoma rare-neoplasm RWD extraction guidance**: Use bone sarcoma as the rare-neoplasm example for LLM-assisted real-world data collection by mapping extracted findings to predefined registry variables, abstracting longitudinally across clinician notes, anchoring diagnosis, treatment, response, progression, recurrence, and follow-up events in time, emitting missingness flags for undocumented, negative, or indeterminate values, sampling extracted records for manual validation, and preserving audit-ready provenance linking each value to source notes, evidence spans, temporal anchors, missingness status, and validation decisions.
40. **Patient outcome extraction with timing labels**: Identify outcome mentions in clinician notes, normalize event times, distinguish historical, current, and future outcomes, preserve evidence spans for each assertion, and emit uncertainty labels for registry or real-world-data use.
41. **Registry-oriented LLM abstraction for rare neoplasms**: For low-prevalence cancer cohorts such as bone sarcoma, define registry-oriented extraction schemas before LLM abstraction, capture provenance from each extracted value back to source notes, support temporal outcome fields, and require clinician validation before cohort release or downstream real-world-data use.
42. **Registry-oriented rare-cancer RWD extraction workflow**: For rare cancers such as bone sarcoma, define registry data dictionaries before abstraction, extract structured variables from clinical notes, timestamp diagnosis, treatment, response, progression, recurrence, follow-up, and outcome events, adjudicate uncertain or unsupported fields before release, and prepare audit-ready exports for oncology registries with provenance, evidence spans, uncertainty labels, and adjudication status.
43. **Oncology consultation-document prognostic modeling**: When predicting cancer survival from initial oncology consultation documents, compare zero-shot prompting with task-specific fine-tuning under the same cohort definition; define censoring-aware survival labels and prediction horizons before modeling; check for leakage from follow-up text, outcome dates, post-consultation data, or fields unavailable at the initial consultation; evaluate calibration of predicted risks or survival probabilities; and limit outputs to research, audit, or decision-support exploration unless prospective validation, governance, and clinician oversight are in place.
44. **Schema-first LLM abstraction for rare-neoplasm RWD**: For rare-neoplasm real-world-data collection from clinician notes, define the registry schema before extraction, map bone sarcoma registry variables into structured fields, capture temporal clinical events, place uncertain or unsupported abstractions into adjudication queues, and validate extracted variables against manual abstraction before downstream use.
45. **Clinician-note patient outcome extraction with temporal anchoring**: Use LLM-assisted extraction to identify patient outcomes in clinician notes, normalize outcomes and event timing, preserve source evidence and uncertainty spans, reconcile event status across longitudinal notes, and report manual adjudication metrics for uncertain, discordant, or reviewed outcome-time assertions before downstream use.
46. **Clinician-note outcome extraction with adjudicated temporal labels**: Extract patient outcomes from clinician notes with event-time normalization, temporal anchors, uncertainty labels, negation handling, longitudinal consistency checks across notes, and validation against clinician-adjudicated outcome and timing labels before downstream use.
47. **Bone sarcoma real-world data extraction for rare neoplasms**: For LLM-assisted rare-neoplasm abstraction, use bone sarcoma as the example for registry-oriented schema design, require note-level evidence spans for each extracted value, capture temporal fields for diagnosis, treatment, response, progression, recurrence, and follow-up, assign uncertainty labels, sample records for validation, and require clinician adjudication before downstream analytics.
48. **Sarcoma registry RWD collection pattern**: For rare-neoplasm oncology RWD studies, define sarcoma registry entity schemas before extraction, extract temporally anchored outcomes from clinical notes, attach evidence spans to each value, send uncertain or unsupported fields to human validation queues, and export audit-ready tables with provenance, temporal anchors, uncertainty, and validation status.
49. **Rare-neoplasm LLM registry abstraction with schema validation**: For low-prevalence cohorts, use LLMs to abstract registry-oriented fields from clinician notes with bone sarcoma examples such as pathology type, tumor size, localization, grade, and primary resection; capture source-note evidence and temporal anchors for each extracted value, validate outputs against expected formats and clinician-defined vocabularies, and send unsupported, low-confidence, or discordant fields to human review before registry export.
50. **Patient outcome extraction with chart-review validation**: Use LLM-assisted review of clinician notes to classify patient outcome mentions, normalize outcome timing into explicit temporal anchors, attach evidence spans to each outcome-time assertion, and validate extracted outcome and timing labels against chart-review labels before downstream use.
51. **Bone sarcoma RWD extraction logs for rare neoplasms**: For clinician-note real-world-data extraction in rare neoplasms, use bone sarcoma as the worked example to define registry-oriented fields before abstraction, extract temporally anchored outcomes, track missing, negative, and indeterminate values separately, validate extracted fields against chart review, and emit audit-ready extraction logs with source note, evidence span, temporal anchor, missingness status, validation decision, and adjudication status.
52. **Patient outcome identification and timing extraction from clinician notes**: Use LLM-assisted extraction to identify patient outcomes in clinician notes, normalize outcome events and event timing to explicit temporal anchors, assign confidence scores and uncertainty labels for outcome status and timing, preserve source-note evidence spans, validate outputs against annotated notes before downstream use, and export registry/RWD-ready tables with normalized outcome, event time, temporal anchor, confidence, provenance, and validation status.
53. **Bone sarcoma rare-neoplasm RWD agreement workflow**: For LLM-assisted real-world-data collection in rare neoplasms, use the bone sarcoma example to design registry schemas before abstraction, require note-level evidence spans and temporal anchors for extracted values, track missing, negative, and indeterminate fields separately, compare human inter-rater agreement with LLM-human agreement during validation, route discordant or unsupported items to human adjudication, and preserve audit trails linking source notes, evidence spans, temporal anchors, missingness status, adjudication decisions, and final registry values.
54. **Clinician-note outcome ontology and timeline extraction**: Map outcome mentions from clinician notes to a study outcome ontology, normalize events and times into explicit temporal anchors, assign certainty labels for outcome status and timing, aggregate extracted events longitudinally at the patient level, preserve source evidence spans for each assertion, and validate event and timing outputs against annotated clinical notes before downstream registry or RWE use.
55. **Registry-oriented rare-neoplasm RWD extraction pattern**: Use bone sarcoma as the exemplar for LLM-assisted real-world-data collection by defining registry fields before extraction, extracting values from clinical notes with evidence spans, normalizing diagnosis, treatment, response, progression, recurrence, and follow-up timelines, flagging uncertain or unsupported values, and requiring clinical validation before analysis.
56. **Bone sarcoma LLM extraction pattern for rare-neoplasm registries**: For rare-neoplasm RWD collection, use LLMs to extract registry-style fields from clinical notes with bone sarcoma examples such as pathology type, tumor size, localization, grade, and primary resection; normalize dates and relative timing for diagnosis, treatment, progression, recurrence, and follow-up; emit uncertainty, missingness, and unsupported-value flags; and keep clinician validation loops for ambiguous fields before registry export.
57. **Adjudicated rare-neoplasm registry extraction workflow**: Using bone sarcoma as the example, define the registry schema and allowed values before extraction; extract diagnosis, treatment, response, progression, recurrence, follow-up, and outcome events with temporal anchors; distinguish documented negatives, undocumented fields, and indeterminate values; draw an adjudicated sample for manual review; report precision and recall against the adjudicated sample without inferring unsupported values; and export real-world datasets with field-level links to source notes, evidence spans, temporal anchors, missingness status, and adjudication decisions.
58. **Rare-neoplasm registry extraction and audit workflow**: Extract structured real-world data from clinical notes by normalizing diagnoses to the registry vocabulary, constructing temporally anchored treatment and outcome timelines, separating documented negatives from undocumented or indeterminate values, routing ambiguous and sampled records to clinician validation, preserving source-note evidence for every field, and reporting extraction audit metrics such as field completion, missingness, clinician agreement, corrections, and unresolved items without inventing unsupported values.
59. **Registry-oriented rare-cancer RWD extraction**: Define the registry schema and allowed values before extracting rare-cancer data from clinical notes; link each note-derived field to its supporting evidence span; distinguish documented negatives, undocumented fields, and indeterminate values; normalize explicit dates and relative clinical times; require human validation for ambiguous, sampled, or clinically important fields; and monitor extraction quality through field-level completion, missingness, validation decisions, corrections, and unresolved-item tracking.
60. **Outcome-event identification with temporal anchoring**: Identify outcome events in clinician notes; assign event status, onset and resolution dates, and uncertainty intervals; separate note time from event time; deduplicate repeated mentions across longitudinal notes; and retain clinician-auditable evidence spans for every extracted event and temporal assertion.
61. **Ontology-constrained rare-cancer registry abstraction**: Extract registry-oriented real-world data for rare cancers using predefined, ontology-constrained fields and allowed values; distinguish documented negatives, undocumented fields, and indeterminate values; preserve field-level provenance to source notes and evidence spans; compare human inter-rater and LLM-human abstraction results; route discordant, ambiguous, or unsupported values to human adjudication; and validate final outputs against manually curated records before registry use.
62. **Longitudinal clinician-note temporal outcome extraction**: Identify and normalize patient outcome events across longitudinal clinician notes; distinguish document time from event time; represent dates as interval-censored ranges when only partial or relative timing is supported; preserve negation and uncertainty for both outcome status and timing; and evaluate at the patient level with patient-disjoint data splits so notes from the same patient cannot leak across training, validation, or test sets.
63. **Bone sarcoma rare-cancer longitudinal RWD workflow**: Define registry variables before extraction; normalize pathology and treatment terminology to the registry vocabulary; extract each value with its supporting evidence span and temporal anchor; measure field completeness and extraction precision; route uncertain, unsupported, or discordant fields to reviewers; and preserve field-level source, evidence, timing, and review provenance across longitudinal datasets.
64. **Outcome-event extraction with temporal anchoring**: Distinguish outcome occurrence from negation, plans, historical mentions, and uncertainty; normalize event dates and intervals; link each outcome to supporting note spans; score event identity and timing separately; and reconcile conflicting outcome assertions across longitudinal notes while preserving their source evidence.
65. **Rare-cancer longitudinal registry extraction workflow**: Define ontology-backed registry field schemas and allowed values before extracting data from longitudinal notes; resolve repeated, partial, relative, or conflicting diagnosis, treatment, response, progression, recurrence, and follow-up events into evidence-linked timelines; distinguish documented negatives, undocumented fields, and indeterminate values; assign field-level confidence scores without filling unsupported values; route low-confidence, discordant, sampled, or clinically important fields to clinician validation; and compare extracted fields and temporal events against manual abstraction before registry release or downstream analysis.
66. **Bone sarcoma registry abstraction pattern**: For rare-cancer registries, use bone sarcoma as a worked pattern by defining the extraction schema before abstraction, linking diagnosis, pathology, treatment, progression, recurrence, and follow-up events across time, normalizing pathology and treatment terms to registry vocabularies, attaching uncertainty flags without inferring unsupported values, validating on low-prevalence cohorts, and manually auditing a sample of extracted records.
67. **Outcome-event extraction with longitudinal temporal reconciliation**: Extract explicit outcome statements separately from outcomes inferred from clinical context; represent onset and resolution as dates or supported intervals; preserve negation and uncertain timing; reconcile repeated, updated, or conflicting events across longitudinal notes without discarding source evidence; and evaluate event extraction with event-level precision and recall for both outcome identity and temporal anchoring.

## Workflow

1.  **Input**: A string of clinical text or a text file.
2.  **Process**: Tokenizes and matches against patterns/dictionaries.
3.  **Output**: JSON list of entities with spans and types.

## Example Usage

**User**: "Extract entities from this note."

**Agent Action**:
```bash
python3 Skills/Clinical/Clinical_NLP/entity_extractor.py \
    --text "Patient has diabetes type 2. Prescribed Metformin 500mg. No chest pain." \
    --output entities.json
```

```

## References

*   https://pubmed.ncbi.nlm.nih.gov/42021926/
*   https://pubmed.ncbi.nlm.nih.gov/41886942/
*   https://pubmed.ncbi.nlm.nih.gov/42004490/

<!-- AUTHOR_SIGNATURE: 9a7f3c2e-MD-BABU-MIA-2026-MSSM-SECURE -->