Skip to main content

Overview

AI-powered clinical documentation is rapidly transforming healthcare, from ambient scribes that generate SOAP notes to summarization tools that distill complex records. Rubric helps ensure these systems produce accurate, complete, and compliant documentation.

Use Cases

Ambient Documentation

AI scribes that generate notes from provider-patient conversations

Visit Summarization

Condensing lengthy records into actionable summaries

Discharge Summaries

Automated generation of discharge documentation

Chart Abstraction

Extracting structured data from unstructured notes

What Rubric Evaluates

Clinical Accuracy

Does the generated note accurately reflect the patient encounter?
  • Factual correctness - Are symptoms, findings, and diagnoses correct?
  • No hallucinations - Is everything supported by the source material?
  • Appropriate terminology - Are medical terms used correctly?

Completeness

Does the note include all required elements?
  • Required sections - SOAP components, ROS, PE findings
  • Critical information - Allergies, medications, red flags
  • Billing support - HCC codes, quality measures

Compliance

Does the note meet regulatory requirements?
  • Documentation standards - CMS guidelines, specialty requirements
  • Copy-forward detection - Inappropriate duplication from prior notes
  • Attribution - Proper sourcing of historical information

Quick Start

from rubric import Rubric

client = Rubric()

# Log a generated clinical note
client.notes.log(
    project="visit-summarizer",
    
    # Source input (transcript, prior notes, etc.)
    input_text="""
    Patient is a 52-year-old male presenting for follow-up of type 2 diabetes.
    Reports good compliance with metformin. Checking blood sugars at home, 
    running 120-140 fasting. Denies hypoglycemia. No polyuria or polydipsia.
    Last A1c was 7.2% three months ago.
    """,
    
    # AI-generated output
    output={
        "soap_note": {
            "subjective": "52 y/o male with T2DM presents for routine follow-up. Reports good medication compliance with metformin. Home glucose monitoring shows fasting values 120-140 mg/dL. Denies hypoglycemic episodes, polyuria, or polydipsia.",
            "objective": "Vitals pending. Labs: Last A1c 7.2% (3 months ago).",
            "assessment": "1. Type 2 diabetes mellitus - at goal (A1c < 7.5%)",
            "plan": "1. Continue metformin 1000mg BID\n2. Repeat A1c today\n3. Annual diabetic eye exam referral\n4. Follow-up 3 months"
        },
        "icd_codes": ["E11.9"],
        "cpt_codes": ["99214"]
    },
    
    # Expected output (if available)
    expected={
        "required_elements": ["medication_review", "home_monitoring", "a1c_discussion"],
        "icd_codes": ["E11.65"]  # DM with hyperglycemia might be more specific
    }
)

Evaluation Metrics

MetricDescriptionTarget
Factual AccuracyPercentage of claims supported by source> 95%
Completeness ScoreRequired elements present> 90%
Hallucination RateUnsupported claims< 2%
Code AccuracyCorrect ICD/CPT codes> 85%

Next Steps