fda-analyst

fda-analyst

A skill for analyzing FDA novel drug approvals. Powered by the trove project's curated index of recent FDA approvals plus the actual approval-package documents on accessdata.fda.gov.

When to use this skill

Invoke when the user asks about:

• A specific FDA-approved drug — what was the basis for approval, what trials supported it, what endpoints were used.
• Adverse events the FDA flagged at approval (read from the medical review, not the label).
• Regulatory pathway — was it accelerated approval, breakthrough, priority review, fast track.
• Approval-document content — anything in the medical review, statistical review, pharmacology review, chemistry review.
• Cross-drug comparisons along regulatory dimensions.

Don't use this skill for

• Clinical advice or dosing. This is a reference tool over public approval documents, not a substitute for the prescribing label or clinical judgment.
• Drugs approved before 2021. The current trove index covers 2021–2024 only. Earlier drugs aren't in the published bundle yet.
• Devices, biosimilars, generics, or supplemental approvals. Out of scope. Vaccines and blood products are also out of scope — trove ingests only the CBER cell & gene therapy page, not the broader CBER vaccines/blood list.
• Devices, biosimilars, or generic (ANDA) approvals. Out of scope — the trove index covers FDA's curated "Novel Drug Approvals" lists (NMEs and novel BLAs).
• Questions that don't require the actual approval package — labeling questions, prescribing information, formulary status, etc. Use FDA's public-facing tools or the prescribing label.

How to query

The data lives as a JSON/Parquet bundle served from troveproject.com. DuckDB can query the Parquet over HTTPS; you can also fetch the JSON directly.

Bash + DuckDB CLI for the index:

duckdb -c "SELECT drug_name, active_ingredient, approval_date,
          application_type, application_number, drugs_at_fda_url
   FROM read_parquet('https://troveproject.com/data/fda_approvals_nme_recent.parquet')
   WHERE LOWER(drug_name) LIKE '%[name]%'
   LIMIT 5"