Clinical Extractions

Overview

Clinical extraction is the process of identifying and pulling structured, computable data from unstructured clinical documents such as physician notes, discharge summaries, operative reports, and other free-text records. Instead of leaving this information buried in narrative text, extraction surfaces this information as discrete data points that applications can query, analyze, and act on.

Extraction by Clinical Domain

Extractions are organized by clinical domain, each targeting a specific category of clinical information. When configuring a pipeline, you select which domains to enable by setting the extraction_type parameter. Only the extraction types you enable will be run, so you can tailor the pipeline to your use case and reduce processing time.

The table below summarizes the available extraction types, their parameter names, and the clinical information each one targets.

Clinical DomainExtraction TypeDescription
ProblemsproblemsDiagnoses, clinical conditions, acute and chronic diseases, complications, and comorbidities
ProceduresproceduresClinical & surgical procedures and interventions, along with imaging studies across all medical disciplines
Vital SignsvitalsQuantitative vital sign measurements including blood pressure, heart rate, temperature, respiratory rate, oxygen saturation, and other vital sign parameters
Laboratory ResultslabsQuantitative laboratory test results from blood work, chemistry panels, hematology, immunology, and other laboratory disciplines, as documented in clinical notes or laboratory reports
MedicationsmedicationsMedication administration and prescription information
Family Historyfamily_historyFamily members’ diagnoses, health conditions, acute and chronic diseases, complications, and comorbidities
Social Historysocial_historySocial and behavioral health information, including alcohol use, nicotine use, other substance use, marital status, and employment history

Available Extraction Types

All sample examples provided in this section are from pseudonymized data and/or synthetic data sets.