Excellent question. Yes, it is appropriate and sometimes even advisable to conduct disproportionality analysis (DA) on stratified data—such as focusing on a specific drug class—but with important caveats and considerations.

✅ When Stratified Disproportionality Analysis is Appropriate ✔️ 1. Class-Specific Signal Detection

When your interest is in detecting signals within a therapeutic or chemical class, it can be more meaningful to compare a drug to its peer drugs than to all drugs in the database.

Example: If you're studying dopamine agonists, it might make more sense to compare pramipexole vs. other dopamine agonists, not vs. all drugs (e.g., antibiotics, antihypertensives).

📈 Benefits of Stratified DA

Reduces confounding:

Drugs in the same class often treat the same conditions, reducing indication bias.

Enhances specificity:

Helps identify whether a signal is unique to a drug or a class effect.

Clinical relevance:

Physicians often decide between drugs within the same class.

⚠️ Limitations and Risks of Stratified DA Issue Description Reduced power Smaller data pool may reduce statistical power to detect rare events. Overlook broader signals A drug might show no signal compared to peers, but still high vs. all drugs. Bias from class composition If the class is not homogeneous (e.g., old vs. new drugs), bias may be introduced. False negatives Subtle signals might get "masked" if peers have similar safety profiles. 🧪 Best Practices 1. Justify the Stratification

Define and document your drug class criteria clearly (e.g., ATC codes, pharmacological class).

2. Compare Both Ways (Dual Strategy)

Do two DAs:

Drug vs. all other drugs (broad signal check)

Drug vs. other drugs in class (class-specific check)

This allows differentiation of drug-specific vs. class-wide signals.

3. Check Signal Stability

Use statistical shrinkage methods (e.g., Bayesian approaches) when working with small sample sizes.

4. Adjust for Indications or Age/Gender

Stratification or multivariate models (logistic regression, Bayesian regression) can reduce indication or demographic bias.

🧮 Example: Drug vs Class vs All

Let’s say:

You observe an ADR with Drug A (a proton pump inhibitor).

You compare:

Drug A vs. All Drugs → Signal

Drug A vs. Other PPIs → No signal

Interpretation: The ADR might be a class effect, not unique to Drug A.

🧰 Conclusion Question Answer Can you stratify DA by drug class? ✅ Yes, it's valid. Should you rely only on class-based DA? ⚠️ No, use both views. Does it help reduce confounding from indications? ✅ Often yes. Are there trade-offs? ✅ Yes, lower power and potential for false negatives.

All the best 👍