A Regulatory Review Starts With the Applicable Standard
A BE review is not a search for one favorable number. The reviewer must first determine which regulatory standard applies, then check whether the study evaluated the required endpoints using the required margins, and finally apply the stated decision rule to every required criterion. A favorable AUC result, for example, cannot compensate for a failed primary Cmax criterion when both are required.
The governing standard depends on the jurisdiction, product, dosage form, and regulatory question. Begin with the applicable regional framework and current product-specific guidance, if available. Check that the reference product or comparator is the one specified for that jurisdiction and application. The general conventional criteria are a starting point, not permission to disregard product-specific requirements. If guidance is ambiguous, conflicting, or silent on an endpoint, document the issue and seek the appropriate regulatory clarification rather than selecting whichever rule gives the more favorable result.
For a conventional average BE assessment, the usual primary exposure measures are AUC and Cmax. The conventional acceptance limits are 80.00% to 125.00% for the test-to-reference geometric mean ratio, with the 90% confidence interval required to lie entirely within those limits. This is a standard starting point, not a universal rule for every product. A product-specific recommendation may narrow, widen, or supplement criteria, and a regional standard may differ. Earlier material on regulatory use of partial AUC criteria and supportive endpoints is also relevant when a product’s guidance explicitly requires those measures.
A Compact Review Workflow
For each case file, write down the governing document or standard, the population and comparison to which it applies, each required endpoint, that endpoint’s margin, and whether it is primary or supportive. Then compare the reported interval with the relevant limits. Keep the calculation separate from the regulatory conclusion: an endpoint can be statistically assessed correctly while the submission still fails an overall required criterion.
Record the jurisdiction and application, then check the current product-specific guidance and the specified comparator. Note any conditions or product-specific exceptions.
Separate required primary endpoints from supportive or descriptive measures. For each required endpoint, record the margin and confidence-interval rule.
Verify that the reported comparison, scale, confidence level, and study conditions match the applicable criterion. Compare both interval endpoints with the acceptance limits.
If the standard requires all primary criteria to pass, the submission passes that BE decision only if every required primary criterion passes.
Name the governing criteria, state which required endpoints met or failed them, and avoid claiming a broader conclusion than the review supports.
Why the Confidence-Interval Rule Works
Let \(R\) denote the test-to-reference geometric mean ratio for an endpoint, and let \(L\) and \(U\) be its lower and upper acceptance limits. The endpoint meets the usual average BE criterion when its specified confidence interval for \(R\) is entirely between \(L\) and \(U\). For conventional limits, \(L=0.80\) and \(U=1.25\). The ratio is often estimated on the log scale and then exponentiated so the reported interval is on the ratio scale.
This interval rule corresponds to two one-sided tests (TOST) at significance level \(\alpha=0.05\), giving a 90% confidence interval. The equivalence hypotheses for one endpoint can be expressed as \(H_0: R \leq L\) or \(R \geq U\), against the alternative \(H_a: L < R < U\). Both one-sided null components must be rejected. On the log scale, taking logarithms turns the ratio limits into lower and upper bounds for the log contrast; exponentiating the confidence limits returns the interval for \(R\).
When several primary endpoints are jointly required, the overall alternative is that all endpoints meet their respective criteria. The overall null is that at least one required endpoint does not meet its criterion. Therefore the decision is an intersection-union rule: declare overall BE only if every required endpoint passes its own test. No averaging of endpoint results can rescue a failed criterion. This logic also means the reviewer should not treat a supportive endpoint as a substitute for a required primary endpoint.
Worked Example: Conventional AUC and Cmax Criteria
Hypothetical case file. A conventional immediate-release product is being reviewed under a regional standard that specifies AUC and Cmax with 80.00% to 125.00% limits. The analysis reports test-to-reference geometric mean ratios and 90% confidence intervals. Assume the submitted study and analysis are otherwise consistent with the applicable requirements.
For AUC, the ratio is 96.0%, with a 90% confidence interval of 90.0% to 103.0%. The lower limit comparison is \(90.0\% \geq 80.0\%\), and the upper limit comparison is \(103.0\% \leq 125.0\%\). AUC passes.
For Cmax, the ratio is 103.0%, with a 90% confidence interval of 97.0% to 110.0%. Here \(97.0\% \geq 80.0\%\) and \(110.0\% \leq 125.0\%\). Cmax passes.
Both required primary criteria pass, so the all-criteria rule supports a BE conclusion for these criteria. This conclusion does not mean the products are identical; it means the specified intervals meet the applicable equivalence limits.
Case Files With Tailored Requirements
A reviewer should not assume every product uses the same endpoints or margins. Product-specific requirements can reflect the product’s characteristics and the regulatory question. The relevant review task is to apply the actual applicable recommendation, not to infer a special criterion from a product label or to import a requirement from a different jurisdiction.
Worked Example: A Product-Specific Narrower AUC Margin
Hypothetical case file. A product-specific guidance document for a fictional drug requires AUC and Cmax. It specifies 90.00% to 111.11% for AUC and 80.00% to 125.00% for Cmax. These limits are stipulated for this example only; they are not a general rule for every drug described as narrow therapeutic index.
The AUC ratio is 101.4%, with a 90% confidence interval of 91.2% to 109.0%. Both comparisons pass: \(91.2\% \geq 90.0\%\) and \(109.0\% \leq 111.11\%\).
The Cmax ratio is 96.0%, with a 90% confidence interval of 78.5% to 105.0%. Its upper limit is within range, but its lower limit is not: \(78.5\% < 80.0\%\). Cmax fails the specified criterion.
Because the guidance requires both AUC and Cmax to pass, the overall BE decision fails, even though AUC passes and the Cmax point estimate is inside its margin. The interval, not the point estimate alone, determines whether the criterion is met.
Worked Example: Modified-Release Product With a Partial AUC Criterion
Hypothetical case file. A fictional modified-release product guidance requires AUC, Cmax, and an early partial AUC, each assessed using a 90% confidence interval and 80.00% to 125.00% limits. It identifies Tmax as supportive and descriptive, not as a replacement primary criterion. Assume the guidance and the analysis plan specify the partial-AUC interval used here.
The reported intervals are AUC, 92.0% to 108.0%; Cmax, 88.0% to 119.0%; and early partial AUC, 76.0% to 104.0%. AUC passes because \(92.0\% \geq 80.0\%\) and \(108.0\% \leq 125.0\%\). Cmax passes because \(88.0\% \geq 80.0\%\) and \(119.0\% \leq 125.0\%\). The early partial AUC fails because \(76.0\% < 80.0\%\), although its upper confidence limit is within range.
The overall decision does not meet the specified BE criteria because one required primary endpoint fails. A descriptive Tmax comparison cannot reverse that result under the stipulated guidance. The review conclusion should state that AUC and Cmax passed but the required early partial AUC did not; it should not report an overall pass based on the two favorable intervals.
Diagnostics, Alternatives, and Common Pitfalls
Before accepting a numerical comparison, check that it is the correct comparison for the regulatory question. Confirm the test and reference identities, endpoint definitions, units, confidence level, and margins. For a ratio criterion, verify that results are expressed on the ratio scale (or correctly transformed back from the log scale). Check that an interval is not being compared with margins intended for another endpoint, strength, dosage form, jurisdiction, or product-specific condition.
A result close to a boundary is not automatically a failure or a pass. Apply the stated inclusive limits to the reported interval at the precision and rounding rules required by the governing standard. If the unrounded analysis is available, use it rather than drawing a decision from rounded values that straddle a boundary. When essential information is missing—for example, the guidance is not identified, a required endpoint is absent, or the interval’s confidence level is unclear—the appropriate conclusion is that the criterion cannot be verified, not that it passed.
- Using only the point estimate. A ratio inside the limits does not pass if its confidence interval extends outside them.
- Using the broad conventional limits by default. A tailored margin must be applied when it governs the product and question.
- Letting one endpoint compensate for another. If all listed primary criteria are required, every one must pass.
- Treating supportive results as primary. A supportive endpoint can add context, but it cannot replace a required criterion unless the applicable standard says it can.
- Choosing criteria after seeing results. Post hoc selection undermines the prespecified regulatory decision rule.
The main alternative to a product-specific review is to apply the general regional standard where no applicable product-specific requirement changes it. That is appropriate only after checking that the general standard truly governs the product and application. Likewise, comparisons of different agencies’ standards are useful for understanding regulatory differences, but criteria from one jurisdiction should not be silently substituted for those governing another.
A concise conclusion can follow this pattern: “Under [identified guidance], the required endpoints were [list], with acceptance limits of [limits]. The 90% confidence intervals for [endpoints] were within the specified limits; [endpoint] did not meet its criterion. Because all listed primary criteria are required, the overall BE criteria [were/were not] met.” If a criterion is not assessable, say so explicitly. This format makes the decision traceable without implying that a statistical result answers regulatory questions beyond the criteria reviewed.
Check Your Understanding
Answer each question, then compare your reasoning with the solutions below.
- A required endpoint has a ratio of 99% and a 90% confidence interval of 79% to 113%. Under 80.00% to 125.00% limits, does it pass? Why?
- A product-specific guidance requires AUC and Cmax. AUC passes, but Cmax fails. What is the overall conclusion under an all-criteria rule?
- A submission uses conventional limits, but the reviewer finds current product-specific guidance with a different AUC margin. What should the reviewer do?
- A required partial AUC fails, while AUC and Cmax pass. Tmax is reported as supportive and appears similar between products. Can the reviewer conclude that all required criteria pass?
Solutions. (1) No. The lower confidence limit is \(79\% < 80\%\), so the interval is not entirely within the acceptance limits even though the point estimate is 99%. (2) The overall BE criteria are not met, because every required primary endpoint must pass. (3) Determine whether that product-specific guidance applies to the product, jurisdiction, and application, and apply its specified margin if it governs. Do not retain the conventional margin merely because it is more favorable. (4) No. The required partial AUC criterion fails, and a supportive Tmax result cannot replace it under the stated requirements.