Methodology — From Public Drug Data to a Usable Interaction Tool

1. Primary Data Sources

The primary interaction source is official U.S. FDA drug labeling made available through openFDA . Labels can also be inspected through DailyMed. FDA labels contain Drug Interactions sections, boxed warnings, contraindications, pregnancy information, and other regulatory labeling text used by the pipeline.

Drug identities are normalized with NIH/NLM RxNorm. RxNorm supplies standardized identifiers (RxCUIs), allowing generic names, salts, precise ingredients, and naming variants to be matched more reliably.

2. Limited Curated Supplementation

A small number of well-established interactions may be supplemented from curated clinical references when FDA labeling describes the interaction only at a pharmacologic-class level or leaves a specific pair-level gap. Those records are provenance-tagged and are not silently represented as FDA-originated data.

This distinction matters: an interaction page should tell users where the underlying record came from. The site does not use educational blog text as a source for creating new database interactions.

3. Extraction and Normalization

FDA labels frequently express interactions in unstructured prose. The pipeline uses pattern matching and NLP assistance to identify drug names, interaction descriptions, mechanism language when present, and severity signals. Extracted names are then mapped to RxNorm concepts before pair records are stored.

The pipeline is intended to extract and organize source evidence, not to originate novel interaction theories. If a mechanism seems pharmacologically plausible but no source-of-record interaction exists, that plausibility alone does not create a pair in the database.

4. Severity Classification

Severity is an automated classification of source wording. It is useful for sorting and navigation, but it is not individualized clinical judgment.

Contraindicated

Source language indicates that the combination should not be used together, for example “contraindicated,” “do not use concomitantly,” or equivalent wording.

Major

Source language indicates serious, potentially dangerous, or life-threatening consequences.

Moderate

Source language indicates that monitoring, caution, or dose-management may be needed.

Minor

Source language indicates a limited or relatively small clinical effect.

Unknown

An interaction is documented in the source corpus, but the available wording is not sufficient to classify severity reliably.

5. Plain-Language Educational Content

Educational articles may use AI assistance to translate technical source material into ordinary language and to explain established basic science. That explanation layer has a strict boundary: it must not create a new interaction claim, contradict a source record, independently change severity, or fabricate a citation, statistic, or FDA statement.

Where an article uses a hypothetical example, the example is explanatory only. It does not establish that an interaction exists and should not be treated as personalized medical advice.

6. Update Process and Provenance

The data pipeline runs periodically to incorporate new or revised FDA labels and normalization corrections. Interaction records retain a source/provenance field so the interface can distinguish FDA-derived records from any explicitly curated supplement. Exact database counts may change as the source corpus is refreshed.

7. Known Limitations

• No interaction database is complete.
• FDA labeling may describe an effect at a drug-class level rather than as a specific pair.
• A “no documented interaction” result is not proof that a combination is safe.
• Automated extraction, normalization, and severity classification can produce edge-case errors.
• Individual factors such as dose, route, organ function, timing, and other medications can matter clinically.

Corrections and Feedback

If you identify an error, outdated source, provenance problem, or missing interaction, contact support@checkdruginteractions.com. Reports are checked against the underlying source before the database or page is changed.