Procedures

Goal of Extraction

The Core Clinical Procedures Library is a procedure-focused clinical data extraction tool designed to identify, extract, and ontologically map clinical and surgical procedures and imaging studies from unstructured clinical narratives and patient health record documentation. It extracts documented procedures, interventions, clinical treatments, and imaging studies across clinical notes, procedure reports, imaging reports, and assessment sections, standardizing representation through SNOMED CT concept mapping with associated confidence metrics.

Extraction Methodology

The extraction process operates at the sentence level, capturing each procedure and imaging study with precise character-level locations (start and end positions) that enable direct traceability back to the “document”. Each extracted procedure or imaging study record includes the procedure/imaging name, anatomical site, laterality, and temporal context. When a patient’s procedure or imaging study appears multiple times within a single document—whether from repeated documentation, procedural sequences, or documentation across different note sections—each instance is extracted and recorded separately to preserve the complete procedure and imaging history. The library performs extraction on a document-by-document basis; there is no automatic cross-document reconciliation or cross-patient data merging.

Note: A document is determined by our intelligent document splitting feature. Medical PDFs can contain a single document or multiple documents. A document’s type can provide clinical context important for the data extraction. This feature separates and categorizes an input PDF into individual documents by document type to prepare them for data extraction. An input PDF can remain one document or can be split into multiple documents through this process. A “document” in the context of this data extraction refers to the documents identified after the intelligent document splitting process.

Interpretation & Validation

Every extracted procedure and imaging study is assigned a confidence score (0.0–1.0 scale). Validation flags are applied to distinguish clinically documented procedures and imaging studies from non-actionable mentions, including those planned for future performance, mentioned in document headers, referenced hypothetically, or representing patient discussions or goals. These flags and scores guide appropriate reliance on extracted data and support informed clinical interpretation.

Clinical Considerations

Temporal context is critical for clinical decision-making; understanding when procedures were performed and when imaging studies were obtained directly impacts assessment of treatment sequencing and clinical outcomes. Anatomical specificity, including site and laterality information, has significant implications for clinical management and risk stratification, particularly for procedures and imaging studies affecting paired anatomical structures or specific anatomical regions. Serial procedure and imaging documentation provides essential context for understanding the procedural and diagnostic history and treatment progression. Clinical judgment should be applied when interpreting extracted results, particularly for entries with validation flags.

Target Concepts

Coding System: SNOMED CT

All extracted procedures and imaging studies are mapped to SNOMED CT (Systematized Nomenclature of Medicine Clinical Terms), with each procedure or imaging study assigned a valid, active SNOMED concept identifier (6–18 digit numeric codes), along with its associated full SNOMED concept description. This structure provides both the standardized data needed for system integration and the contextual information needed for informed clinical interpretation.

Clinical Concepts Extracted

Procedures are clinically meaningful interventions that have been performed, documented, or assessed by a clinician. This includes surgical procedures, clinical interventions, treatments, and imaging studies. Associated clinical attributes—including temporality and experiencer—are captured as they appear in source documentation.

We define procedures with SNOMED concepts. Only concepts that fall into one of these clinical categories are included:

• Therapeutic or Preventive Procedures
• Diagnostic Procedure

This means incidental findings, morphologic descriptors, and vague qualifier terms are excluded. Concepts in the following sections are also excluded: “Injection”, “Medications ordered this encounter”, “Orders”, “Medications”, “Orders (continued)”, “Result notes (continued)”. We are looking for identifiable, valid procedures.

The concept must be a specific, named procedure. A small number of overly broad or non-specific SNOMED concepts are excluded by design — for example, generic terms like “recommendation to (procedure)” or “discussion (procedure)” add no diagnostic specificity to a procedure list.

The extracted concept must meet a confidence threshold. The extraction engine assigns a confidence score to each identified concept. Only concepts where the system is greater than 90% confident in the mapping to a SNOMED concept are included.

Extracted Fields

Procedure Extraction Fields

Example of Extracted Output

• Note: This example demonstrates the content of the output, not the structure

Procedure Illustrative Output Example

• Example JSONL object showing a typical extraction result

1
{
2
  "filename": "patient_2.pdf",
3
  "subdoc_id": "patient_2.pdf-0003",
4
  "manifest_index": 2,
5
  "measure": "procedures",
6
  "chartdate": "2022-02-06",
7
  "subject_id": "11110001",
8
  "processing_date": "2024-03-08T00:00:00",
9
  "data": [
10
    {
11
      "sentence": "DICATIONS: |  | This 55-year-old gentleman has come to the office requesting sterilization via bilateral vasectomy. I | discussed the indications and the need for procedure with the patient in detail, and he has given consent to | proceed. He has been given ***prophylactic antibiotics***. |  | PROCEDURE NOTE: |  | Once satisfactory sedation have been obtained, the patient was placed in the supine position on the | operating table. Genitalia was shaved and then prepped with Betadine scrub and paint solution and were | dra",
12
      "section_name": "INDICATION",
13
      "entity_text": "prophylactic antibiotics",
14
      "entity_start": 832,
15
      "entity_end": 856,
16
      "concept_id": "422181004",
17
      "description": "Administration of prophylactic antibiotic (procedure)",
18
      "entity_confidence": 1.0,
19
      "temporality": "",
20
      "is_heading_title": false,
21
      "is_future_procedure": false,
22
      "is_hypothetical": false,
23
      "is_instruction_goal": false,
24
      "experiencer": "patient"
25
    }
26
  ]
27
}