Why Endpoint Choice Must Be Made Separately
Reference scaling changes the acceptance criterion for a specified pharmacokinetic endpoint; it does not automatically change the criterion for every endpoint measured in the same study. In particular, high within-subject variability in \(C_{\max}\) does not establish that AUC is eligible for scaling, and high variability in AUC does not establish that \(C_{\max}\) is eligible. Eligibility depends on the governing regulatory framework and product-specific instructions, as well as evidence about the variability of the endpoint to which scaling is proposed.
This distinction matters because \(C_{\max}\) and AUC represent different aspects of exposure and may have different within-subject variability. A study can therefore support a scaled analysis for one endpoint and a conventional analysis for the other. The choice should be documented endpoint by endpoint before interpreting the results, rather than inferred from whichever endpoint happens to have the largest observed variability.
Framework Rules Come Before Variability Calculations
The applicable framework determines which endpoints can be scaled and under what conditions. Under the commonly used EMA approach, scaling is available for \(C_{\max}\) when the within-subject variability of the reference product meets the framework’s high-variability criterion. AUC is not scaled under that approach; its conventional acceptance limits remain in force. Under FDA approaches, reference scaling may be allowed for \(C_{\max}\), AUC, or both, depending on the applicable recommendations and product-specific guidance. It is not safe to infer the permitted endpoints from the agency name alone.
Product-specific guidance can narrow or specify the general approach. It may identify the endpoint or endpoints eligible for scaling, the design and data needed to estimate reference variability, the decision criterion, and any additional constraints. The current guidance for the exact product, formulation, strength, route, and study condition must therefore be checked. If guidance is silent or ambiguous, the analyst should not treat silence as permission to scale; the appropriate regulatory interpretation should be established before analysis.
Variability is assessed on the log scale for these methods. If a within-subject log-scale standard deviation is available, its corresponding coefficient of variation is \(CV=\sqrt{\exp(\sigma^2)-1}\). Conversely, \( \sigma=\sqrt{\ln(1+CV^2)} \), where the CV is expressed as a proportion. These quantities describe variability for a particular endpoint and reference product; they do not, by themselves, decide whether a regulator permits scaling.
| Question | What must be established | Why it is endpoint-specific |
|---|---|---|
| Is scaling permitted? | The applicable framework and product-specific guidance allow scaling for this endpoint. | Rules may permit scaling for one metric but not another. |
| Is the variability criterion met? | The relevant within-subject variability estimate satisfies the specified threshold and method. | \(C_{\max}\) and AUC can have different reference variability. |
| Are the required data available? | The design and observations identify the reference variability required by the framework. | An endpoint may lack usable replicate reference observations even when another endpoint has them. |
| What criterion applies? | The endpoint-specific scaled or conventional acceptance criterion and any extra constraints are known. | One endpoint’s criterion cannot be substituted for another’s. |
A Decision Sequence for Endpoint Selection
Before fitting a scaled model or applying a scaled criterion, separate the regulatory decision from the statistical calculation. The following sequence keeps the two from being conflated.
Specify the jurisdiction, current general framework, and product-specific guidance for the product and study conditions.
For each required endpoint, record whether scaling is explicitly permitted, prohibited, or not clearly addressed.
Determine whether the reference variability estimate and study design meet the guidance requirements for that endpoint.
Use the scaled criterion only where permitted and justified. Retain the conventional criterion where scaling is unavailable or its conditions are unmet.
Report the framework, guidance version, evidence, and criterion selected for each endpoint so the analysis can be audited.
Worked Endpoint Decisions
Worked Example: EMA Approach With Different Variability by Endpoint
Hypothetical setting. An analyst is assessing a product under an applicable EMA approach. Assume the product guidance is consistent with the usual EMA rule: \(C_{\max}\) may be scaled if reference within-subject CV is at least 30%, whereas AUC remains subject to conventional limits. A suitable replicate study provides estimated reference CVs of 36% for \(C_{\max}\) and 24% for AUC. The design and estimates are assumed to meet the guidance’s data requirements.
For \(C_{\max}\), compare the estimate with the variability threshold:
Thus, \(C_{\max}\) meets the stipulated variability condition. Because the assumed guidance also permits scaling this endpoint, a reference-scaled analysis is eligible for \(C_{\max}\). This conclusion concerns eligibility; it does not establish bioequivalence.
For AUC, the estimated CV is \(24\%\), below 30%. More importantly, the stipulated EMA approach does not scale AUC even if its variability were higher. AUC therefore uses the conventional acceptance criterion. The result is endpoint-specific: scale may be considered for \(C_{\max}\), but not for AUC.
Worked Example: FDA Product Guidance Permits Only One Endpoint
Hypothetical setting. Suppose the current FDA product-specific guidance explicitly allows reference scaling for \(C_{\max}\) but specifies conventional bioequivalence for AUC. A replicate study estimates reference CVs of 33% for \(C_{\max}\) and 41% for AUC. Assume the guidance’s variability threshold is 30%, the appropriate reference contrasts are estimable, and all required study conditions are met.
For \(C_{\max}\), the endpoint is expressly eligible and its estimated variability exceeds the stipulated threshold:
The guidance and evidence therefore support applying the scaled criterion to \(C_{\max}\). For AUC, the observed CV also exceeds 30%:
However, the product-specific guidance specifies conventional bioequivalence for AUC. The variability result cannot override that instruction, so AUC is analyzed using the conventional criterion. High variability is evidence relevant to a permitted scaling decision; it is not an independent authorization to scale.
Worked Example: A High Variability Estimate Does Not Establish Eligibility
Hypothetical setting. A team proposes scaling AUC because a pilot study reports an AUC reference within-subject CV of 35%. The governing product guidance specifies conventional AUC analysis and does not permit reference scaling for that endpoint. The team also reports a \(C_{\max}\) reference CV of 28%.
The AUC estimate exceeds a commonly used 30% threshold:
That arithmetic is correct, but it does not answer the eligibility question. The stated guidance does not permit AUC scaling, so the AUC endpoint remains conventional despite the estimate. For \(C_{\max}\), \(28\%<30\%\), so the stated variability threshold is not met; moreover, the supplied facts do not establish that scaling is permitted for \(C_{\max}\). The team has not justified scaling either endpoint.
The proper next step is not to select whichever CV is larger. It is to follow the stated endpoint rules and use conventional analysis where required. If the guidance’s status or interpretation is uncertain, resolve that regulatory question before choosing the analysis.
Why the Endpoint Decision Is Not a Statistical Shortcut
The logic can be summarized as a necessary-condition argument. Let \(G_j\) mean scaling is allowed by the applicable guidance for endpoint \(j\), and let \(V_j\) mean the endpoint-specific variability and data requirements are satisfied. Scaling is supportable only if both conditions hold:
A large estimated CV may support \(V_j\), but it cannot establish \(G_j\). Nor can \(G_j\) for one endpoint establish \(G_k\) for a different endpoint. This is why the decision cannot be made by applying one study-wide label such as “highly variable” to every pharmacokinetic measure.
The endpoint-specific distinction also follows from the meaning of the estimates. Reference variability for \(C_{\max}\) describes variation in peak concentrations under repeated reference treatment; AUC variability describes variation in total exposure. They are separate quantities, potentially based on different measurement patterns and subject to different guidance. Even when both are calculated from the same participants, they need not be similar.
Diagnostics, Alternatives, and Limitations
Before relying on an endpoint-specific estimate, check that it was produced from the required reference-treatment observations and the appropriate study design. Confirm the endpoint definition, units, sampling and analysis procedures, and the data used for the variability calculation. Check that the relevant reference contrasts can be estimated and that missing or unusable observations have not left the required evidence unavailable. The precise model, estimator, and diagnostic tools depend on the applicable framework; do not replace those requirements with a CV calculated by an unrelated shortcut.
Warning signs include using test-treatment variability in place of reference variability, copying a \(C_{\max}\) CV into the AUC decision, relying on an estimate from an ineligible study design, or overlooking a product-specific restriction. An unexpectedly large CV may also warrant a review of data quality and the study conditions before it is treated as evidence for scaling. Later model diagnostics can help evaluate the adequacy of a fitted analysis, but they cannot make an otherwise prohibited endpoint eligible.
The principal alternative to reference scaling is conventional average bioequivalence with fixed acceptance limits. Use that approach when it is required by the framework or guidance, or when scaling conditions are not met. Another possible response to inadequate evidence is to obtain or analyze data using a design that meets the applicable requirements, not to assume that another endpoint’s design or variability estimate supplies the missing information.
Endpoint selection has limits. General regulatory descriptions do not replace current product-specific instructions; criteria can differ by jurisdiction, product, or guidance revision. A variability estimate is also subject to sampling uncertainty, so an estimate near a threshold should not be treated as a precise classification without considering the prescribed analysis. Finally, eligibility for scaling is not a conclusion about treatment equivalence: the endpoint must still meet its assigned bioequivalence criterion, including any additional constraints.
Common Mistakes
- “The product is highly variable, so scale both endpoints.” The correction is to evaluate each endpoint under its own framework and evidence.
- “AUC variability exceeds the threshold, so AUC must be scaled.” A threshold does not supersede guidance that requires conventional AUC analysis.
- “Scaling is allowed for \(C_{\max}\), so it is allowed for AUC too.” Permission for one metric does not transfer to another.
- “An endpoint that is eligible for scaling has passed bioequivalence.” Eligibility selects a criterion; the endpoint still has to satisfy that criterion.
- “A variability estimate from any available study is sufficient.” The estimate must come from data and a design acceptable under the applicable instructions.
Check Your Understanding
Answer each question by separating regulatory permission from endpoint-specific evidence.
- Under a stipulated framework that permits scaling \(C_{\max}\) but not AUC, reference CV estimates are 34% for \(C_{\max}\) and 38% for AUC. Which endpoint may be considered for scaling?
- A product-specific instruction requires conventional AUC analysis. A valid reference AUC CV estimate is 42%. Does that estimate authorize scaling?
- Why is a high reference CV for \(C_{\max}\) insufficient to establish that AUC is eligible for scaling?
- Guidance permits scaling an endpoint only when its reference CV is at least 30%. Its estimate is 29%. What can be concluded about scaling under that stated rule?
Solutions. (1) \(C_{\max}\) may be considered for scaling because the stipulated framework permits it and 34% exceeds the stated threshold; AUC remains conventional despite its higher CV. (2) No. The product-specific instruction requires conventional analysis, and the variability estimate cannot override it. (3) The endpoints have distinct within-subject variability, and regulatory permission is endpoint-specific; \(C_{\max}\) evidence does not establish either AUC eligibility or AUC variability. (4) The stated variability condition is unmet because \(29\%<30\%\); scaling is not supported under that rule, even if other conditions might otherwise be satisfied.