Adverse Event Narrative Generator

Overview

Regulatory-grade narrative generation tool that transforms adverse event case data into CIOMS-compliant ICSR narratives suitable for submission to FDA, EMA, and other health authorities.

Key Capabilities:

• - CIOMS I Compliance: Standardized narrative structure per international guidelines
• ICH E2B Integration: Electronic submission format compatibility
• Temporal Analysis: Timeline reconstruction and causality assessment
• Medical Accuracy: Clinical terminology and MedDRA coding
• Multi-Case Processing: Batch narrative generation for periodic reporting
• Quality Validation: Automated checks for completeness and consistency

When to Use

✅ Use this skill when:

• - Drafting ICSR narratives for regulatory submissions
• Converting safety case data to standardized text
• Preparing adverse event reports for health authorities
• Generating case summaries for signal detection
• Creating pharmacovigilance documentation for clinical trials
• Standardizing narrative format across safety teams
• Training new drug safety associates on narrative writing

❌ Do NOT use when:

• - Case requires medical judgment or causality assessment → Use qualified safety physician
• Narrative for litigation or legal proceedings → Use legal documentation standards
• Patient-facing communications → Use INLINECODE0
• Aggregate safety summaries → Use INLINECODE1
• Coding MedDRA terms from verbatim → Use INLINECODE2

Integration:

• - Upstream: meddra-coder (MedDRA term coding), clinical-data-cleaner (data preparation)
• Downstream: safety-summary-reports (aggregate analysis), regulatory-submission-prep (FDA/EMA filing)

Core Capabilities

1. CIOMS I Narrative Structure

Generate standardized sections per CIOMS guidelines:

CODEBLOCK0

Standard Sections:

1. 1. Patient Demographics - Age, sex, weight, relevant characteristics
2. Medical History - Significant pre-existing conditions
3. Concomitant Medications - Other drugs at time of event
4. Suspect Drug(s) - Medication(s) in question with dosing
5. Adverse Event - Detailed reaction description with MedDRA terms
6. Diagnostic Results - Lab values, imaging, procedures
7. Treatment - Medical management of the event
8. Dechallenge/Rechallenge - Effect of drug withdrawal/reintroduction
9. Outcome - Final patient status and sequelae
10. Causality Assessment - Reporter's relationship evaluation

2. Temporal Relationship Analysis

Reconstruct timeline and assess temporal plausibility:

CODEBLOCK1

Assessments Generated:

• - Time to onset (latency period)
• Dechallenge response (positive/negative/unknown)
• Rechallenge response (if applicable)
• Temporal plausibility (consistent with known drug profile)

3. Causality Evaluation Support

Structure causality assessment per WHO-UMC criteria:

CODEBLOCK2

WHO-UMC Categories:

• - Certain - Event reproduced on rechallenge
• Probable/Likely - Reasonable time, positive dechallenge, alternative causes unlikely
• Possible - Compatible time, but alternative causes possible
• Unlikely - Incompatible time or alternative cause probable
• Conditional/Unclassified - Insufficient information
• Unassessable/Unclassifiable - Data contradictory or incomplete

4. Multi-Format Output

Generate narratives for different regulatory contexts:

CODEBLOCK3

Common Patterns

Pattern 1: Serious Adverse Event (Hospitalization)

Scenario: Patient hospitalized for severe drug reaction.

CODEBLOCK4

Narrative Emphasis:

• - Hospitalization details (admission/discharge dates)
• Severity markers (ICU stay, intubation)
• Lactate levels and trend
• Renal function status
• Complete recovery timeline

Pattern 2: Fatal Outcome Case

Scenario: Death suspected to be drug-related.

CODEBLOCK5

Critical Elements:

• - Complete medical history relevant to death
• Concomitant medications contributing
• Autopsy findings (if performed)
• Cause of death per death certificate
• Alternative causes ruled out
• Reporter's opinion on contribution to death

Pattern 3: Rechallenge Case

Scenario: Positive rechallenge confirms drug causation.

Key Documentation:

• - First exposure dates and reaction
• Dechallenge response
• Rechallenge dates and circumstances
• Recurrence of same reaction
• Any differences in severity
• Conclusion on causality

Narrative Structure:
CODEBLOCK6

Pattern 4: Multi-Drug Reaction

Scenario: Multiple suspect drugs, need to identify most likely culprit.

Analysis Approach:

• - Temporal sequence of drug initiations
• Time to onset for each drug
• Known adverse reaction profiles
• Dechallenge attempts (if any)
• Rechallenge data (if any)
• Reporter's primary suspect

Narrative Organization:

1. 1. List all suspect drugs with indication and dates
2. Describe temporal relationship for each
3. Discuss dechallenge/rechallenge for each
4. Present reporter's causality assessment
5. Include alternative explanations

Complete Workflow Example

From case data to regulatory submission:

CODEBLOCK7

Python API:

CODEBLOCK8

Quality Checklist

Pre-Generation:

• - [ ] Case ID unique and formatted per SOP
• [ ] Patient age/sex complete
• [ ] Suspect drug(s) clearly identified
• [ ] Adverse event(s) coded with MedDRA PT
• [ ] Dates consistent (no future dates)
• [ ] Reporter information included

Narrative Content:

• - [ ] All CIOMS I sections present
• [ ] Temporal sequence clear and logical
• [ ] Dechallenge/rechallenge described (if applicable)
• [ ] Lab values with reference ranges
• [ ] Concomitant medications listed
• [ ] Medical history relevant to event
• [ ] Outcome clearly stated
• [ ] Causality assessment justified

Post-Generation:

• - [ ] MedDRA terms accurate and current
• [ ] No contradictory information
• [ ] Language objective and factual
• [ ] No speculation or opinion (except causality section)
• [ ] Patient identifiers removed or de-identified
• [ ] CRITICAL: Medical review completed
• [ ] CRITICAL: Causality assessment by qualified physician

Common Pitfalls

Completeness Issues:

• - ❌ Missing dechallenge information → Cannot assess causality

• - ❌ Vague temporal information → "Recently started" vs. specific dates

• - ❌ Incomplete concomitant medication list → Alternative causes missed

Medical Accuracy Issues:

• - ❌ Incorrect MedDRA coding → Wrong medical concept

• - ❌ Confusing correlation with causation → Temporal = causal

• - ❌ Omitting alternative diagnoses → Biased toward drug causation

Regulatory Issues:

• - ❌ Opinion in narrative body → "Clearly caused by drug"

• - ❌ Patient identifiers → HIPAA/privacy violation

• - ❌ Abbreviations not defined → Assumes reader knowledge

References

Available in references/ directory:

• - cioms_i_guidelines.pdf - CIOMS I international reporting standards
• INLINECODE9 - ICH E2B(R3) electronic format details
• INLINECODE10 - MedDRA terminology and coding principles
• INLINECODE11 - WHO causality assessment criteria
• INLINECODE12 - FDA Form 3500A instructions
• INLINECODE13 - EU Good Pharmacovigilance Practices
• INLINECODE14 - Example narratives by case type

Scripts

Located in scripts/ directory:

• - main.py - CLI interface for narrative generation
• INLINECODE17 - Core narrative composition engine
• INLINECODE18 - Timeline reconstruction and analysis
• INLINECODE19 - Causality evaluation support
• INLINECODE20 - Medical terminology and coding
• INLINECODE21 - Completeness and quality checks
• INLINECODE22 - Convert between CIOMS, E2B, MedWatch formats
• INLINECODE23 - Multi-case narrative generation

Limitations

• - Medical Review Required: Generates draft only; requires physician review before submission
• Causality Assessment: Structures reporter's assessment; does not perform independent causality evaluation
• MedDRA Version: Uses installed MedDRA version; may not have latest terms
• Language: Optimized for English; other languages may need translation
• Literature Integration: Does not automatically search literature for similar cases
• Signal Detection: Individual case narratives only; aggregate analysis requires other tools
• Legal Proceedings: Not suitable for litigation support or expert witness reports