The Interaction Checker: What It Covers and What It Does Not

What it does

You declare what you are taking — prescriptions, supplements, plant preparations, foods, seasonings, alcohol, and lifestyle factors such as smoking — and the checker resolves each declared item to the mechanism ROLES it plays: which enzymes and transporters it is a substrate of, which it inhibits, which it induces, and which pharmacodynamic axes it loads. It then looks for collisions between those roles, and reports each collision as a mechanism with a direction, a strength class, a plain-language consequence and the citations it rests on. Because the unit of reasoning is a mechanism rather than a pair, the same rule that catches grapefruit with simvastatin catches piperine with a kinase inhibitor, and the rule that catches St John wort with ciclosporin catches it with an oral contraceptive — without anyone having typed those specific pairs. The design consequence worth stating plainly is that the checker aims to be right about DIRECTION for a wide range of substances, and is deliberately modest about MAGNITUDE for any of them.

• Input: anything a person can declare, not just approved drug products.
• Resolution: substance to mechanism roles (substrate, inhibitor, inducer, provider, agonist).
• Output: mechanism, direction, strength class, plain consequence, citations.
• Every finding is traceable to a source you can read.
• Findings whose evidence is in-vitro, single-source or contested are labelled as such rather than presented at the same confidence as the rest.

Sources: U.S. Food 2023 · Flockhart DA 2021 · Zanger UM 2013

Mechanism axes it covers

The axes below are the ones modelled. Each corresponds to a page in this shelf or in the wider library, so a finding can be read back to the mechanism that generated it.

• CYP3A4 inhibition and induction, including the distinction between intestinal and hepatic effects.
• CYP2C9 inhibition and induction.
• CYP2C19 inhibition and induction.
• CYP2D6 inhibition — and the fact that this enzyme is not meaningfully inducible.
• CYP1A2 inhibition and induction, including induction by smoke and DE-induction on cessation.
• CYP2E1 inhibition and induction, including the acute-versus-chronic ethanol inversion.
• Mechanism-based versus reversible inhibition, where the sourced data distinguish them.
• Prodrug inversion: findings are directionally reversed for substrates whose activity lives in a metabolite.
• UGT glucuronidation inhibition and induction, to the extent the literature supports it.
• SULT sulfonation inhibition and competitive saturation, marked as mechanism-level rather than quantified.
• P-glycoprotein (ABCB1) inhibition and induction, with intestinal and blood-brain-barrier consequences distinguished.
• MAO-A and MAO-B inhibition, and dietary tyramine and L-dopa load.
• Serotonergic load: reuptake inhibition, release, and direct receptor agonism.
• QT prolongation as an additive pharmacodynamic axis.
• Seizure-threshold lowering as an additive axis.
• Additive hepatotoxicity as an organ-level rather than pharmacokinetic axis.
• 11-beta-HSD2 inhibition and the potassium-lowering axis that makes digoxin more toxic at an unchanged level.
• Narrow-therapeutic-index flagging on the substrate side, which is what converts a modest interaction into a serious finding.

Sources: U.S. Food 2023 · Flockhart DA 2021 · Woosley RL 2025 · Zanger UM 2013 · Rowland A 2013 · Coughtrie MWH 2016 · Schinkel AH 2003

What it does NOT cover — read this list generously

This list is long on purpose. Every item on it is a real way the checker can be silent about something that matters, and a tool that does not publish its blind spots is worse than no tool, because it converts ignorance into false reassurance.

• MAGNITUDE in a specific person. The checker gives direction and a strength class. It does not predict your concentration, your INR, or your level.
• Your genotype. CYP2D6, CYP2C19 and CYP2C9 phenotype changes outcomes by more than most interactions do, and the checker does not know yours.
• DOSE and TIME. Most findings do not scale with the amount you took or how long you have been taking it, and induction takes days to appear while inhibition can be immediate.
• Transporters other than P-glycoprotein. OATP1B1 and 1B3 (statin uptake — the mechanism behind several real statin interactions), BCRP, OCT, MATE, BSEP and the rest are not modelled.
• Renal elimination interactions: competition for tubular secretion, changes in urine pH, the NSAID and ACE-inhibitor and diuretic triple that damages kidneys, lithium handling.
• Absorption-level interactions: chelation and complexation (calcium, iron, magnesium, aluminium with tetracyclines, quinolones, levothyroxine and bisphosphonates), bile-acid sequestrants, adsorbents, gastric pH changes affecting dissolution, and dose-separation strategies generally.
• Protein-binding displacement, and drugs with high protein binding where a total concentration misleads.
• Phase-2 pathways beyond the UGT and SULT material here: glutathione S-transferase, N-acetyltransferase (NAT2 acetylator status), thiopurine methyltransferase, methylation and COMT.
• Non-CYP phase-1 enzymes: aldehyde oxidase, xanthine oxidase (the azathioprine and allopurinol interaction), flavin monooxygenase, carboxylesterase, alcohol and aldehyde dehydrogenase.
• Hepatic and renal impairment, which change every clearance assumption underneath every finding.
• Age, pregnancy, lactation, paediatric dosing, and drug transfer into breast milk.
• Pharmacodynamic additivity outside the modelled axes: additive CNS and respiratory depression, anticholinergic burden, additive bleeding risk, additive hypotension, hypoglycaemia, hyponatraemia, serotonin-independent neurotoxicity.
• Product identity and standardisation. An unstandardised botanical has an unknown quantity of any given constituent, and the same plant name covers preparations with very different constituent profiles — cultivar, plant part and extraction solvent all change the answer.
• Adulteration and contamination. Substituted species, undeclared pharmaceuticals in supplements, heavy metals, pesticides and residual solvents are undetectable by any interaction table.
• Novel and semi-synthetic cannabinoids, and any substance for which no pharmacokinetic literature exists. The checker cannot reason about an enzyme relationship nobody has measured. This gap is largest exactly where the market moves fastest.
• Food effects beyond the ones explicitly modelled: fat content and its effect on lipophilic absorption, meal timing, grapefruit-adjacent citrus not individually listed, high-protein and high-carbohydrate effects on some clearances.
• Allergy, idiosyncratic and immune-mediated reactions, which are not dose- or mechanism-predictable at all.
• Withdrawal and discontinuation effects, and the reciprocal interaction of STOPPING an inhibitor or inducer — which is a real event the checker will not prompt you about.
• Anything that is simply not in the dataset yet. The dataset is curated by hand and its coverage is uneven by design, deepest where this library has done the reading.

Sources: U.S. Food 2023 · Flockhart DA 2021 · Holbrook AM 2005 · Teschke R 2011 · Rowland A 2013

How to read a clean result

A clean result means one thing only: no documented interaction in this dataset for the items you declared, on the mechanism axes it models. It is not a clearance, not a safety finding, and not an opinion about whether what you are doing is wise. There are at least five reasons a genuinely dangerous combination can return clean. The interaction may not be documented anywhere, which is the ordinary situation for most plant preparations and every novel compound. It may be documented in a source not yet incorporated here. The mechanism may be one the checker does not model — a transporter, a renal interaction, an additive pharmacodynamic effect outside the listed axes. One of the substances may not be what its label says. Or the relevant variable may be you: your genotype, your liver, your kidneys, your other conditions. The reverse caution applies with equal force. A reported finding is not a prediction that you will be harmed; it is a statement that a mechanism exists and points in a direction, at a strength the sources support. When a finding is marked contested, that marking is the actual state of the evidence and not a disclaimer bolted on afterwards.

• Clean equals undocumented here, never safe.
• A finding equals a documented mechanism with a direction, not a prophecy.
• Contested markings are load-bearing information, not legal padding.
• The strongest use of the tool is to generate a specific question to take to a pharmacist or prescriber, with the mechanism and the citation in hand.

Sources: U.S. Food 2023 · Flockhart DA 2021 · Teschke R 2011

Provenance and limits of the dataset

The corpus behind the checker is hand-curated from primary literature, FDA labelling and public reference tables, and every finding carries the citations it rests on so that a reader can check the source rather than trust the tool. Citations whose identifier has been resolved are marked accordingly; citations whose identifier has not been resolved are marked unverified and rendered with that marker rather than quietly presented as equivalent. Where a claim is in-vitro only, animal only, single-source, or derived from anecdotal and forum material, that grade is attached to the claim. Where sources conflict — the kavalactone CYP question on this shelf is the clearest instance, with potent in-vitro inhibition and a null in-vivo human probe result — the conflict is shown rather than averaged into a single confident answer. There is no comprehensive, openly licensed, redistributable drug-interaction dataset available to a public repository: the retired public interaction API, the commercially licensed comprehensive databases and the approved-products-only scope of regulatory labelling each rule themselves out, and none of them covers the plants this library exists to serve. A curated mechanism table with visible provenance and published blind spots is the honest architecture available, and its coverage is uneven by construction — deepest where the reading has been done, thin elsewhere, and silent where nobody has measured anything.

Sources: U.S. Food 2023 · Flockhart DA 2021 · Gurley BJ 2005 · Mathews JM 2002 · Teschke R 2011

Not advice

This is an education and harm-reduction reference. It does not diagnose, does not prescribe, does not tell anyone to take or to stop taking anything, and cannot see the things that decide the answer in a real person. Its purpose is to make a mechanism legible so that a person can ask a better question and recognise a hazard they would otherwise not have known existed. Withholding mechanism from someone who is going to proceed anyway is the harm this library exists to prevent; pretending that a mechanism table can substitute for a clinician who can see the whole picture would be a different harm, and both are refused here.

Sources: U.S. Food 2023 · Flockhart DA 2021