Why Qualification Is a Separate Decision
Conventional average bioequivalence uses fixed limits. As discussed in the earlier material on why high variability undermines conventional average bioequivalence, a highly variable drug can make it difficult to place the confidence interval for the test-to-reference difference inside those limits, even when the estimated difference is small. Reference scaling provides a possible alternative in certain regulatory approaches: the criterion can account for variability observed when the reference product is given repeatedly.
That possibility is conditional. A study must provide suitable information about within-subject variability of the reference product, and the estimate must meet the applicable jurisdiction’s threshold. Meeting that threshold qualifies the endpoint for consideration of a scaled approach; it does not show that the test and reference products are bioequivalent. The final analysis still has its own requirements and can fail.
What Is Compared With the Threshold?
The relevant quantity is reference-product variability within the same subjects across repeated administrations. A replicate crossover design, in which the reference is administered more than once to at least some subjects, supplies the repeated reference observations needed for this assessment. A conventional two-period crossover with one reference administration per subject does not provide repeated reference administrations and cannot directly estimate this within-subject reference variability in the required way.
The estimate is endpoint-specific. For example, within-subject variability for \(C_{\max}\) need not equal that for AUC. It is also distinct from variability between subjects and from the variability of a test-minus-reference contrast. A large overall coefficient of variation across participants is not, by itself, evidence that the reference-variability threshold has been met.
Regulatory analyses commonly express the relevant log-scale within-subject standard deviation as \(s_{wR}\), or express the corresponding within-subject coefficient of variation as \(CV_{wR}\). For orientation, the 30% CV threshold corresponds to a log-scale standard deviation of approximately 0.294. The applicable analysis plan and regulatory guidance determine how the estimate is obtained; the examples below treat it as a reported result rather than estimating it from replicate data.
| Approach | Variability threshold | Scope to keep in view |
|---|---|---|
| FDA reference-scaled approach | \(CV_{wR} \geq 30\%\), equivalently \(s_{wR}\) approximately at least 0.294 | Assess the relevant PK endpoint under the applicable FDA recommendations. |
| EMA scaled limits | \(CV_{wR} > 30\%\) | Scaling applies to \(C_{\max}\); AUC remains subject to conventional limits. |
The boundary wording matters. FDA uses a threshold of at least 30%; EMA describes the condition as exceeding 30%. Thus, an unrounded estimate exactly equal to 30.0% meets the stated FDA threshold but not the EMA threshold. Product-specific guidance and the current jurisdictional requirements must also be checked before selecting an approach.
A Decision Logic for Qualification
The qualification decision can be written as a sequence of necessary checks. First, the study and analysis must be appropriate for estimating within-subject reference variability. Second, the endpoint-specific estimate is compared with the jurisdiction-specific threshold. Only if those checks are satisfied is the endpoint eligible for the relevant scaled approach.
Here \(Q_j=1\) means that endpoint \(j\) qualifies for consideration of scaling; it does not denote bioequivalence. If \(B_j\) denotes the final bioequivalence conclusion for that endpoint, then qualification is a prerequisite for using a scaled route, not a sufficient condition for \(B_j=1\). In particular, the final assessment must still apply all criteria of the chosen approach to the test-reference results.
Worked Qualification Decisions
Worked Example: An Estimate Above 30%
In a hypothetical replicate study, the reported within-subject reference CV for \(C_{\max}\) is 33.2%. Determine whether this estimate meets the variability threshold for FDA and EMA scaling.
FDA. Compare the estimate with the FDA threshold: \(33.2\% \geq 30.0\%\). The \(C_{\max}\) endpoint meets the variability threshold for consideration of the FDA reference-scaled approach.
EMA. Compare the estimate with the EMA condition: \(33.2\% > 30.0\%\). The endpoint also meets the variability condition for EMA scaling of \(C_{\max}\).
Conclusion. This is a qualification result only. It does not establish bioequivalence; the test-reference analysis must still meet the final criteria. The result also does not imply that EMA scales AUC.
Worked Example: A Value at the Boundary
A hypothetical replicate study reports an unrounded within-subject reference CV of exactly 30.0% for an endpoint. Apply the FDA and EMA threshold wording.
FDA. The condition is at least 30%: \(30.0\% \geq 30.0\%\). The estimate meets the stated FDA variability threshold.
EMA. The condition is greater than 30%: \(30.0\% > 30.0\%\) is false. The estimate does not meet the stated EMA variability condition.
Conclusion. The approaches differ at this boundary. Use the unrounded analysis result when applying a threshold; a rounded display of 30% may conceal whether the underlying estimate is just below, exactly at, or just above 30%. The result still says nothing by itself about final bioequivalence.
Worked Example: Different Endpoints, Different Outcomes
A hypothetical replicate study reports a reference within-subject CV of 34.1% for \(C_{\max}\) and 27.8% for AUC. Apply the thresholds, assuming the study and analysis are otherwise appropriate.
\(C_{\max}\). Since \(34.1\% \geq 30.0\%\), this endpoint meets the FDA threshold. Since \(34.1\% > 30.0\%\), it also meets the EMA variability condition for possible scaling.
AUC. Since \(27.8\% < 30.0\%\), this endpoint does not meet the FDA 30% threshold. It also does not exceed the EMA threshold. In addition, EMA scaled limits are not applied to AUC; AUC is assessed using conventional limits under that approach.
Conclusion. Do not treat the drug as having one undifferentiated variability result. Qualification is evaluated for each endpoint, and the regulatory scope differs by approach. Even for \(C_{\max}\), meeting the threshold does not settle the bioequivalence decision.
Assumptions, Diagnostics, and Alternatives
A threshold comparison is meaningful only if its input is the appropriate estimate. Before interpreting a result, check that the study includes repeated reference administrations and that the reported estimate comes from the specified endpoint and an appropriate analysis. Also verify the jurisdiction, the applicable product-specific requirements, and whether the reported figure is an unrounded estimate or a rounded summary.
Several patterns should prompt review rather than an automatic qualification statement:
- No repeated reference administration. A design without replicate reference dosing does not supply the repeated reference observations needed for this assessment.
- The wrong variability quantity. Between-subject variability, total variability, or test-reference contrast variability is not interchangeable with within-subject reference variability.
- Endpoints are pooled or conflated. A qualifying \(C_{\max}\) estimate does not qualify AUC, and an AUC result does not determine the \(C_{\max}\) result.
- Rounding obscures the boundary. An estimate displayed as 30% may fall on either side of the EMA strict threshold, or below the FDA threshold. Decisions should follow the unrounded analysis result and specified reporting rules.
- Qualification is reported as a success claim. A threshold result alone does not establish that the test-reference comparison meets the final equivalence criteria.
The main alternative is conventional average bioequivalence using fixed limits, when that is the applicable and prespecified approach. If a suitable estimate is unavailable or does not meet the threshold, one cannot justify reference scaling merely because conventional limits appear difficult to meet. A study may need a design capable of estimating reference variability, or the endpoint may need to be assessed through the applicable non-scaled method. The choice must be prespecified and consistent with the relevant regulatory guidance, not selected after seeing which analysis produces a preferred result.
Qualification also has limits. The threshold is an operational regulatory condition, not evidence that variability is harmless, that the reference product is intrinsically unstable, or that the test product matches it. Estimates from a finite study are subject to sampling variation, and a value close to 30% should be interpreted with care. The threshold comparison itself is not a confidence-interval statement about the true variability; uncertainty assessment and the final scaled procedure are governed by the applicable method.
Common Mistake
Incorrect reasoning: “The reference CV is above 30%, so the products are bioequivalent under the scaled method.”
Correction: An estimate above the threshold can qualify an endpoint for consideration of scaling, provided the design and analysis are appropriate. It does not establish the test-reference relationship. The final bioequivalence decision requires the complete specified analysis, including every applicable criterion. Report the qualification result and final decision as separate statements.
Check Your Understanding
For each question, identify the relevant threshold and distinguish qualification from the final bioequivalence conclusion. Solutions follow the questions.
- A hypothetical \(C_{\max}\) estimate is \(29.6\%\). Does it meet the FDA threshold? Does it meet the EMA condition?
- A hypothetical estimate is exactly \(30.0\%\). Why can the FDA and EMA qualification outcomes differ?
- A conventional two-period crossover study administers the reference once per subject. Can its overall CV be substituted for within-subject reference CV to qualify scaling? Explain.
- In a replicate study, \(C_{\max}\) has a reference within-subject CV of \(36\%\), while AUC has a value of \(25\%\). Does the \(C_{\max}\) result establish bioequivalence or qualify AUC?
Solutions. (1) No. \(29.6\% < 30.0\%\), so it meets neither the FDA at-least-30% threshold nor the EMA greater-than-30% condition. (2) FDA uses \(\geq 30\%\), so the boundary qualifies; EMA requires \(>30\%\), so exact equality does not. (3) No. The design does not provide repeated reference administrations, and overall CV is not a substitute for the endpoint’s within-subject reference variability. (4) No. The \(C_{\max}\) estimate meets both variability thresholds, but that only permits consideration of the relevant scaled approach; final bioequivalence remains undecided. The AUC estimate is below 30%, so it does not meet the variability threshold, and EMA does not scale AUC in any case.