药品不良事件信号检测

QUICK_START.md

What This Skill Does

Detects and quantifies adverse drug event signals using FDA FAERS disproportionality analysis (PRR, ROR, IC), FDA label mining, mechanism-based prediction, and literature evidence. Produces a Safety Signal Score (0-100) for clinical and regulatory decision-making.

Quick Examples

Example 1: Full Safety Signal Report

"Detect adverse event signals for atorvastatin"

The skill will:

1. Resolve atorvastatin to CHEMBL1487 / DB01076
2. Query FAERS for top adverse events (fatigue, diarrhoea, myalgia, rhabdomyolysis, ...)
3. Calculate PRR/ROR/IC for each significant event with 95% CI
4. Extract FDA label warnings and contraindications
5. Compare to statin class (simvastatin, rosuvastatin, pravastatin)
6. Check PharmGKB for pharmacogenomic risk factors (SLCO1B1)
7. Search PubMed for safety literature
8. Calculate Safety Signal Score (e.g., 62/100 = Moderate Concern)
9. Generate comprehensive markdown report

Example 2: Specific Adverse Event Investigation

"Is pembrolizumab associated with myocarditis? Calculate the disproportionality."

The skill will focus on:

1. Resolve pembrolizumab to ChEMBL ID
2. Run FAERS_calculate_disproportionality(drug_name="PEMBROLIZUMAB", adverse_event="Myocarditis")
3. Get PRR, ROR, IC with 95% CI and signal strength
4. Stratify by demographics (age, sex)
5. Check FDA label for myocarditis warnings
6. Search literature for immune checkpoint inhibitor cardiotoxicity

Example 3: Drug Class Comparison

"Compare safety of apixaban vs warfarin for bleeding events"

The skill will:

1. Run FAERS_compare_drugs(drug1="APIXABAN", drug2="WARFARIN", adverse_event="Haemorrhage")
2. Compare PRR/ROR/IC side-by-side
3. Repeat for related events: GI haemorrhage, intracranial haemorrhage, epistaxis
4. Generate comparative safety table

Example 4: Emerging Signal Detection

"Are there emerging safety signals for semaglutide not in the label?"

The skill will:

1. Get all FAERS signals for semaglutide
2. Calculate disproportionality for top 20 events
3. Extract all FDA label warnings
4. Identify signals in FAERS that are NOT in the label = potential emerging signals
5. Search recent literature and preprints for confirmation
6. Rank by signal strength and case count

Example 5: Post-Market Surveillance Report

"Generate a pharmacovigilance signal detection report for methotrexate"

Produces full 10-section report with:

• Executive summary
• FAERS profile with >15 adverse events
• PRR/ROR/IC for all significant events
• FDA label safety extraction
• Target safety profile
• Drug class comparison
• PGx risk factors
• Literature evidence
• Safety Signal Score
• Clinical recommendations

Key Outputs

1. Safety Signal Score (0-100): Quantitative risk metric

   • 75-100: High concern
   • 50-74: Moderate concern
   • 25-49: Low-moderate concern
   • 0-24: Low concern

2. Disproportionality Table: PRR, ROR, IC with 95% CI for each AE

3. Signal Strength Classification: Strong/Moderate/Weak/None for each event

4. Evidence Grading: T1 (regulatory) to T4 (computational) for each finding

5. Markdown Report: Saved to [DRUG]_adverse_event_report.md

Common Drug Names to Try

Drug	Class	Expected Key Signals
Atorvastatin	Statin	Rhabdomyolysis, myalgia, hepatotoxicity
Metformin	Biguanide	Lactic acidosis
Warfarin	Anticoagulant	Hemorrhage
Pembrolizumab	Anti-PD-1	Immune-mediated AEs (colitis, pneumonitis)
Doxorubicin	Anthracycline	Cardiotoxicity
Methotrexate	Antimetabolite	Hepatotoxicity, pancytopenia
Ibuprofen	NSAID	GI bleeding, renal injury

Tips

• Use generic names in UPPERCASE for FAERS queries (e.g., "ATORVASTATIN")
• MedDRA terms are required for specific event queries (e.g., "Rhabdomyolysis", not "muscle breakdown")
• PRR >= 2.0 with lower CI > 1.0 and N >= 3 indicates a signal
• PRR >= 5.0 indicates a strong signal requiring investigation
• Check FDA label to distinguish known vs novel signals
• Drug class comparison helps determine if signal is drug-specific or class-wide