Classifying DDI Magnitude Under Regulatory Frameworks

Why Classification Requires a Framework

A predicted exposure ratio is a quantitative result; “weak,” “moderate,” and “strong” are classifications that depend on a specified framework. The same ratio can be described accurately under one framework and misclassified under another if the analyst uses the wrong endpoint, direction, substrate, or threshold set. A category summarizes the magnitude of a defined exposure change. It does not, by itself, prescribe a dose change, monitoring plan, or avoidance recommendation.

For many conventional regulatory classifications, the relevant quantity is the effect of a perpetrator drug on the AUC of a sensitive substrate drug. The ratio is calculated as exposure with the perpetrator present divided by exposure in the corresponding control condition. An increase is consistent with inhibition of a pathway affecting substrate disposition; a decrease is consistent with induction. The classification depends on the specified framework and must be interpreted in the context of the model and endpoint that produced the prediction.

The ratio’s direction matters. For example, a ratio of 0.25 means AUC with the interacting drug is one-quarter of control AUC: a 75% decrease, not a 25% decrease. In contrast, a ratio of 4 means AUC is four times control, which is a 300% increase. These are not interchangeable descriptions.

Conventional AUC-Based Categories

A commonly used regulatory framework classifies inhibitor strength by the increase in AUC of a sensitive substrate. Under this convention, a strong inhibitor causes an increase of at least fivefold, a moderate inhibitor causes an increase of at least twofold but less than fivefold, and a weak inhibitor causes an increase of at least 1.25-fold but less than twofold. Expressed as ratios, these boundaries are \(R_{\mathrm{AUC}}\geq5\), \(2\leq R_{\mathrm{AUC}}<5\), and \(1.25\leq R_{\mathrm{AUC}}<2\), respectively.

For inducer strength, the corresponding convention is expressed as a percentage decrease in substrate AUC: strong is a decrease of at least 80%, moderate is a decrease of at least 50% but less than 80%, and weak is a decrease of at least 20% but less than 50%. Since percentage decrease is \(100(1-R_{\mathrm{AUC}})\), these categories can also be written as ratio ranges.

The boundary inclusions follow directly from the definitions. For example, a 50% decrease gives \(R_{\mathrm{AUC}}=1-0.50=0.50\), which is moderate, not weak; a 20% decrease gives \(R_{\mathrm{AUC}}=0.80\), which is weak. Under these criteria, ratios between 0.80 and 1.25 fall outside the weak-through-strong inducer and inhibitor ranges. That statement describes these categories only; it is not a universal declaration of “no effect.”

Worked Classifications

Deriving the Ratio Thresholds and Applying Them Correctly

The conversion between ratio and percentage change is simple but useful for checking classifications. If control AUC is \(A_0\) and AUC with the interacting drug is \(A_1\), then \(R_{\mathrm{AUC}}=A_1/A_0\). The relative change from control is \((A_1-A_0)/A_0\). Substituting \(A_1/A_0=R_{\mathrm{AUC}}\) gives the relative change \(R_{\mathrm{AUC}}-1\). Multiplying by 100 expresses that change as a percentage.

For a decrease, the positive percentage reduction is the negative of this relative change: \(1-R_{\mathrm{AUC}}\). Thus a threshold decrease \(d\), expressed as a proportion, corresponds to \(R_{\mathrm{AUC}}=1-d\). For example, an 80% reduction corresponds to \(1-0.80=0.20\). This derivation also explains why the inducer ratio inequalities run in the opposite direction from inhibitor inequalities: a larger decrease produces a smaller ratio.

The framework’s assumptions matter. A conventional perpetrator-strength category is tied to an effect on a specified sensitive substrate under defined conditions; it should not be assigned merely because a model labels a drug an “inhibitor” or “inducer.” The predicted ratio must represent comparable interacting and control conditions, use the intended exposure measure, and be suitable for comparison with that framework. If the prediction is for a different substrate, dose, population, or endpoint than the criterion presumes, the category may not be transferable without justification.

AUC-based thresholds also should not be silently applied to Cmax. A prediction might show a modest AUC change but a larger peak change, or the reverse. If a particular regulatory or product-specific document supplies a Cmax criterion, apply that criterion to Cmax; otherwise report the Cmax ratio separately rather than assigning it an AUC-based strength category. Similarly, when the relevant measure is an active moiety or another specially defined endpoint, use the endpoint specified by the governing criterion.

Frameworks, Label Criteria, and Common Failure Modes

Regulatory conventions provide a useful common language, but there is no single threshold scale that should be treated as universal for every drug, jurisdiction, endpoint, or decision. Regulatory guidance may define perpetrator categories using sensitive-substrate AUC changes. A product label may instead give a drug-specific criterion, name particular concomitant drugs, or link a category to a specific action. A product-specific instruction takes its meaning from the wording and evidence for that product; it should not be replaced by a general cutoff simply because the general scale is familiar.

For example, if a label says to avoid a drug classified as a strong inhibitor under a specified framework, a predicted ratio of 3.00 is in the moderate inhibitor range under the conventional thresholds above. That comparison alone does not establish that coadministration is safe, permitted, or advisable: the exact label language and relevant evidence still govern. Conversely, a category such as “strong” does not itself say whether the correct response is dose adjustment, monitoring, or avoidance. Those are clinical consequences and recommendations, not definitions of magnitude.

• Reversing the ratio. If the calculation is control divided by interacting condition, a reported ratio of 0.25 may be mislabeled as an increase. State the numerator and denominator explicitly.
• Confusing ratio with percent change. A ratio of 2.00 is a 100% increase, not a 200% increase. Compute change as \((R_{\mathrm{AUC}}-1)\times100\%\).
• Using the wrong endpoint. An AUC classification threshold does not automatically classify Cmax. Verify which metric the framework specifies.
• Treating categories as universal. Thresholds depend on the relevant regulatory or product-specific criteria. Name the criteria used and check their applicability.
• Turning magnitude into a clinical action. “Moderate” is not synonymous with “adjust the dose,” and “below weak” is not synonymous with “no risk.” Report the category separately from recommendations.

Diagnostics are therefore partly arithmetic and partly contextual. Check that the ratio can be reproduced from the stated exposures, verify that the direction agrees with the raw values, and recalculate percentage change. Then check whether the classification is based on the appropriate substrate, endpoint, and framework. If those checks fail, a neat category label can create false precision. Model uncertainty and the strength of evidence may also affect confidence in the prediction, but the point-estimate category should not be presented as a certainty about the true interaction.