Adolescent Drug Interaction Signals in FAERS Data

ISEF Category: Biomedical and Health Sciences

Subcategory: Other  ·  Difficulty: Advanced  ·  Setup: Home Setup  ·  Time: Full Year

The Hook

A drug pair can look quiet in adults and noisy in teens. That gap matters because teen medication use is messy, uneven, and often under-studied. Your project can mine real adverse-event reports to find interaction signals that show up more in adolescents.

What Is It?

Think of FAERS like a giant rumor board for side effects. The FDA collects reports from doctors, patients, and companies, then stores them in a public database. DrugBank helps you match drug names to one shared label, so the same medicine does not get counted three different ways. Your project asks which drug pairs get reported more often in teens than in adults.

Disproportionality analysis is a comparison trick. You measure how often a drug pair appears, then compare that to how often you would expect it to appear if age did not matter. Shrinkage estimators pull extreme scores back toward the middle when the data are thin, which helps you avoid chasing a loud signal that came from only a few reports.

Why This Is a Good Topic

This is a strong science fair topic because you can use public data, ask a clear safety question, and test a real-world problem without a wet lab. It connects to pediatric drug safety, where teen data are often sparse and medication patterns differ from adults. You can learn data cleaning, statistical signal detection, and how to tell whether a pattern still holds after stricter checks.

Research Questions

• How does age group change the top-ranked drug-drug interaction signals in FAERS?
• What is the effect of applying a shrinkage estimator on interaction pairs with low report counts?
• Does restricting the analysis to serious reports change which adolescent signals stay significant?
• To what extent do opioid, antidepressant, and stimulant pairs differ between adolescents and adults?
• Which interaction pairs appear in adolescents but not adults after correcting for multiple testing?
• What is the effect of collapsing similar drugs into drug classes on signal stability?

Basic Materials

• Laptop or desktop computer with stable internet access and at least eight GB of RAM.
• Public FAERS quarterly files from the FDA website.
• Drug name mapping source, such as DrugBank or PubChem.
• Spreadsheet software for quick screening and notes.
• Python or R with pandas and statistics packages.

Advanced Materials

• High-memory workstation for large FAERS merges.
• Institutional access to DrugBank exports or equivalent curated interaction tables.
• Relational database software such as PostgreSQL or SQLite.
• R packages for disproportionality analysis and shrinkage estimation.
• Version control and reproducible notebooks, such as Git and Jupyter.

Software & Tools

• Python: Cleans FAERS tables, joins drug dictionaries, and runs signal calculations.
• R: Fits disproportionality models and makes comparison plots.
• Jupyter Notebook: Keeps code, notes, and outputs in one reproducible file.
• OpenRefine: Standardizes messy drug names before analysis.

Experiment Steps

1. Define the age bands, outcome window, and interaction rule you will test.
2. Map every drug name to one standard identifier so synonyms do not split your counts.
3. Choose one raw disproportionality score and one shrinkage-adjusted score to compare.
4. Plan the filters that will remove duplicates, obvious miscoding, and tiny cells.
5. Set up a way to rank the strongest signals separately for adolescents and adults.
6. Build a comparison plan that checks whether the same pair stays high after sensitivity tests.

Common Pitfalls

• Mixing brand names, generic names, and salt forms, which splits one drug pair into several records.
• Treating any co-listed pair as a true interaction, which confuses shared prescribing with a safety signal.
• Comparing adolescents and adults before checking report counts, which makes rare pairs look bigger than they are.
• Leaving duplicate FAERS cases in the dataset, which can count the same event several times.
• Filtering too hard before shrinkage, which can hide small but useful signals.

What Makes This Competitive

A class-level version stops at the top few signals. A stronger project checks whether those signals survive name cleanup, duplicate removal, age re-binning, and shrinkage. You can raise the level further by comparing several interaction metrics, testing stability across report years, and checking overlap with known DrugBank interactions. That gives you a project with a clear method and a clear argument for why the results matter.

Project Variations

• Restrict the study to one therapeutic class, such as antidepressants or stimulants, and compare teen versus adult interaction signals within that class.
• Compare raw disproportionality scores with shrinkage-adjusted scores to see which pairs stay stable when counts are sparse.
• Split FAERS by seriousness or reporter type, then test whether adolescent signals change across those subsets.

Learn More

• FDA FAERS Public Dashboard: Search the FDA site for adverse event reporting data and supporting documentation.
• PubMed: Search review articles on pharmacovigilance, disproportionality analysis, and pediatric adverse drug reactions.
• DrugBank: Use the free drug pages and interaction summaries on the DrugBank site to check pair names and mechanisms.
• NIH NCBI Bookshelf: Look for free chapters on drug safety, epidemiology, and signal detection methods.
• MIT OpenCourseWare: Search statistics and probability courses if you want a stronger base in inference and uncertainty.