Source: https://github.com/aipoch/medical-research-skills

When to Use

1. Pharmacovigilance / safety signal screening when you need adverse event counts, common reactions, or serious-event rates for a drug.
2. Medical device regulatory research when you need 510(k)/PMA context, device classification, UDI lookups, or device adverse events/recalls.
3. Recall and enforcement monitoring when you need to track Class I/II/III recalls across drugs, devices, or foods.
4. Substance identity resolution when you need UNII/CAS/name-based lookups and basic substance relationship/structure retrieval.
5. Veterinary safety analysis when you need animal adverse events filtered by species/breed and product.

Key Features

• Unified Python interface (FDAQuery) for multiple openFDA domains (drug, device, food, animalandveterinary, other).
• Convenience helpers for common tasks:
  • Drug events, labels, recalls, shortages
  • Device events, classification, 510(k), PMA, UDI
  • Food events and recalls
  • Animal/veterinary adverse events
  • Substance (UNII/name) lookups
• Supports openFDA query patterns:
  • Fielded search strings, date ranges, wildcards
  • Aggregations via count_by_field(...) (with .exact support)
  • Pagination via skip/limit and bulk retrieval via query_all(...)
• Operational safeguards:
  • Optional API key support for higher daily limits
  • Built-in caching (TTL) and rate limiting (as implemented in scripts/fda_query.py)
  • Basic error handling patterns

Additional endpoint notes and query syntax are typically documented in: references/api_basics.md, references/drugs.md, references/devices.md, references/foods.md, references/animal_veterinary.md, references/other.md.

Dependencies

• Python 3.9+
• openFDA API access (public)
• Optional: openFDA API key (recommended for higher daily quota)

Package-level dependencies (e.g., requests) are defined by the repository implementation in scripts/fda_query.py. If you maintain this skill, pin them in requirements.txt (for example, requests==2.31.0) to ensure reproducibility.

Example Usage

The following example is designed to be runnable in a repository that contains scripts/fda_query.py and the FDAQuery class.

1) Set an API key (optional, recommended)

export FDA_API_KEY="your_key_here"

2) Run a complete script

import os
from datetime import datetime, timedelta

from scripts.fda_query import FDAQuery


def drug_safety_profile(fda: FDAQuery, drug_name: str):
    # Total adverse events (meta.total)
    events = fda.query_drug_events(drug_name, limit=1)
    total = events.get("meta", {}).get("results", {}).get("total", 0)

    # Top reactions (aggregation)
    reactions = fda.count_by_field(
        "drug",
        "event",
        search=f"patient.drug.medicinalproduct:*{drug_name}*",
        field="patient.reaction.reactionmeddrapt",
        exact=True,
    )
    top_reactions = reactions.get("results", [])[:10]

    # Serious events
    serious = fda.query(
        "drug",
        "event",
        search=f"patient.drug.medicinalproduct:*{drug_name}*+AND+serious:1",
        limit=1,
    )
    serious_total = serious.get("meta", {}).get("results", {}).get("total", 0)

    # Recent recalls
    recalls = fda.query_drug_recalls(drug_name=drug_name)
    recall_results = recalls.get("results", [])

    return {
        "drug": drug_name,
        "total_events": total,
        "serious_events": serious_total,
        "serious_rate_pct": (serious_total / total * 100.0) if total else 0.0,
        "top_reactions": top_reactions,
        "recalls_sample": recall_results[:5],
    }


def monthly_event_trend(fda: FDAQuery, drug_name: str, months: int = 6):
    trends = []
    for i in range(months):
        end = datetime.now() - timedelta(days=30 * i)
        start = end - timedelta(days=30)
        date_range = f"[{start.strftime('%Y%m%d')}+TO+{end.strftime('%Y%m%d')}]"

        search = (
            f"patient.drug.medicinalproduct:*{drug_name}*"
            f"+AND+receivedate:{date_range}"
        )
        result = fda.query("drug", "event", search=search, limit=1)
        count = result.get("meta", {}).get("results", {}).get("total", 0)

        trends.append({"month": start.strftime("%Y-%m"), "events": count})

    return list(reversed(trends))


def main():
    fda = FDAQuery(api_key=os.getenv("FDA_API_KEY"))

    # Drug: safety profile + trend
    profile = drug_safety_profile(fda, "aspirin")
    trend = monthly_event_trend(fda, "aspirin", months=6)

    # Device: quick cross-database lookup
    device_lookup = {
        "adverse_events": fda.query_device_events("pacemaker", limit=10),
        "classification": fda.query_device_classification("DQY"),
        "510k": fda.query_device_510k(applicant="Medtronic"),
        "udi": fda.query("device", "udi", search="brand_name:*pacemaker*", limit=5),
    }

    # Food: recall monitoring
    food_recalls = fda.query_food_recalls(reason="undeclared peanut", limit=10)

    # Substance: UNII lookup
    substance = fda.query_substance_by_unii("R16CO5Y76E")

    print({"drug_profile": profile, "drug_trend": trend})
    print({"device_lookup_keys": list(device_lookup.keys())})
    print({"food_recalls_count": len(food_recalls.get("results", []))})
    print({"substance_keys": list(substance.keys())})


if __name__ == "__main__":
    main()

3) Run the repository examples (if provided)

python scripts/fda_examples.py

Implementation Details

API domains and endpoints

This skill is a thin client over openFDA endpoints, typically accessed as:

• Drugs: drug/event, drug/label, drug/ndc, drug/enforcement, drug/drugsfda, drug/drugshortages
• Devices: device/event, device/510k, device/classification, device/enforcement, device/recall, device/pma, device/registrationlisting, device/udi, device/covid19serology
• Foods: food/event, food/enforcement
• Animal/Veterinary: animalandveterinary/event
• Other/Substances: other/substance, other/nsde

Exact helper method names (e.g., query_drug_events, query_device_510k) are implemented in scripts/fda_query.py.

Query construction

• Searches are passed as openFDA query strings (Lucene-like), e.g.:
  • Field match: patient.drug.medicinalproduct:aspirin
  • Wildcards: *aspirin* (use sparingly)
  • Boolean: A+AND+B
  • Date range: receivedate:[20240101+TO+20241231]
• Pagination uses:
  • limit (page size)
  • skip (offset)
• Aggregations use count_by_field(domain, endpoint, search, field, exact=True):
  • When exact=True, the implementation typically appends .exact to the aggregation field to avoid tokenization issues.

Rate limits and authentication

• openFDA supports unauthenticated access with lower daily quotas; an API key increases the daily request limit.
• The client is expected to:
  • Attach the API key when provided
  • Apply rate limiting and retries (per FDAQuery implementation)

Result handling and robustness

• Responses generally follow:

{
  "meta": { "results": { "skip": 0, "limit": 100, "total": 12345 } },
  "results": []
}

• Always guard for:
  • Missing results
  • Empty result sets
  • error objects returned by the API

Caching

• If enabled in FDAQuery, caching reduces repeated calls for identical queries.
• Typical parameters (implementation-dependent):
  • use_cache=True
  • cache_ttl=<seconds>