Overview

The Antibiotic Indications module extracts clinical syndromes from notes to meet Joint Commission requirements for antimicrobial stewardship documentation (MM.09.01.01 EP 13-15).

Key Point: Joint Commission requires documentation of the clinical syndrome (e.g., "CAP", "UTI", "sepsis"), NOT ICD-10 codes or billing constructs.

This module provides:

  1. Clinical Syndrome Extraction - AI extracts the indication from notes (e.g., "Community-Acquired Pneumonia")
  2. Guideline Matching - Maps syndrome to CCHMC treatment guidelines
  3. Agent Appropriateness - Checks if prescribed antibiotic matches guideline recommendations
  4. Human Review - ASP team verifies syndrome and agent appropriateness

How It Works

1
New Antibiotic Order

System detects new antibiotic orders from monitored medications list

2
Clinical Syndrome Extraction

LLM extracts the clinical syndrome from notes (48h window). Example: "CAP", "pyelonephritis", "sepsis"

3
Red Flag Detection

Identifies concerning patterns: viral illness, asymptomatic bacteriuria, no documented indication

4
Guideline Matching

Maps syndrome to CCHMC guidelines. Checks if prescribed agent is first-line, alternative, or off-guideline

5
ASP Review

Team verifies syndrome extraction and agent appropriateness. Reviews contribute to model training

Clinical Syndrome Categories

The system recognizes ~40 clinical syndromes across these categories:

Category Example Syndromes
Respiratory CAP, HAP, VAP, aspiration pneumonia, empyema
Urinary Simple UTI, pyelonephritis, CAUTI
Bloodstream Bacteremia, sepsis, line infection, endocarditis
Skin/Soft Tissue Cellulitis, abscess, wound infection
Intra-abdominal Appendicitis, peritonitis, C. diff
CNS Meningitis, VP shunt infection
Bone/Joint Osteomyelitis, septic arthritis
ENT AOM, sinusitis, strep pharyngitis
Febrile Neutropenia Neutropenic fever (oncology)
Prophylaxis Surgical prophylaxis, PCP prophylaxis

Red Flags (Triggers Alert)

The system flags these concerning patterns for ASP review:

Red Flag Meaning Example
Likely Viral Notes suggest viral illness but antibiotics prescribed "Viral URI" or "bronchiolitis" with antibiotic order
Asymptomatic Bacteriuria Positive UA/culture without UTI symptoms Incidental positive UA in patient without dysuria/frequency
No Indication Documented Notes don't explain why antibiotics started Antibiotic order with no infection-related notes
Never Appropriate Syndrome where antibiotics rarely/never indicated RSV bronchiolitis, common cold

Reviewing Extractions

When reviewing a case, you'll verify two things:

1. Syndrome Verification

Is the extracted clinical syndrome correct?

Decision When to Use
Confirm Syndrome LLM extraction is correct (e.g., correctly identified "CAP")
Correct Syndrome Change to different syndrome (e.g., "HAP" not "CAP")
No Indication No valid indication documented in notes
Viral Illness Viral illness - antibiotics not indicated
Asymptomatic Bacteriuria ASB - treatment not indicated

2. Agent Appropriateness (Optional)

Is the prescribed antibiotic appropriate for the syndrome?

Decision When to Use
Appropriate Good choice for this syndrome (matches guideline first-line)
Acceptable Not first-line but clinically reasonable (allergy, prior treatment)
Inappropriate Wrong antibiotic for this syndrome
Skip Don't assess agent appropriateness

Training Data: Your reviews are collected for model fine-tuning. After ~500 reviewed cases, we can fine-tune a faster, specialized model.

Guideline Matching

Each syndrome maps to CCHMC-specific treatment guidelines. The system checks if the prescribed antibiotic is:

Category Meaning Action
First Line Matches CCHMC guideline recommendation No intervention needed
Alternative Acceptable alternative per guideline Review if first-line was available
Off-Guideline Not recommended for this syndrome Consider intervention

Confidence Levels

The LLM reports confidence in its syndrome extraction:

Level Meaning Example
Definite Syndrome explicitly stated in notes "Started ceftriaxone for CAP"
Probable Strongly implied by clinical context Fever + infiltrate + antibiotic, but "CAP" not stated
Unclear Cannot determine from notes Antibiotic ordered but notes don't explain why

Three-Layer Appropriateness Model

Layer 1: Indication

What clinical syndrome is being treated?

Implemented

Layer 2: Agent

Is this antibiotic appropriate for the syndrome?

Implemented

Layer 3: Duration

Is the duration appropriate?

Phase 3

Fallback: ICD-10 Classification

When notes are unavailable, the system falls back to ICD-10 based classification (Chua et al.):

Category Code Meaning
Always A Antibiotic indicated per ICD-10 codes
Sometimes S May need antibiotics - clinical judgment required
Never N No antibiotic indication per ICD-10
Prophylaxis P Surgical or medical prophylaxis
Febrile Neutropenia FN Neutropenia + fever

Note: ICD-10 classification is a fallback. Clinical syndrome extraction from notes is preferred because it reflects real-time clinical reasoning and meets JC requirements.

Analytics

The module tracks patterns for stewardship reporting:

  • By Syndrome: Which infections are most common
  • By Antibiotic: Agent appropriateness by medication
  • By Location: Patterns by unit/ward
  • By Service: Ordering service patterns
  • Review Metrics: Correction rate, syndrome distribution

References

  • The Joint Commission. MM.09.01.01 EP 13-15 - Antimicrobial Stewardship Standard. 2024 revision.
  • CDC Core Elements of Hospital Antibiotic Stewardship Programs. 2019.
  • IDSA/SHEA. Implementing an Antibiotic Stewardship Program: Guidelines. 2016.
  • Chua KP, et al. Appropriateness of outpatient antibiotic prescribing. BMJ 2019;364:k5092.
Demo Environment: All patient data displayed is simulated. No actual patient data is available through this dashboard.