AUC and Cmax as Joint Regulatory Criteria

Why AUC and Cmax Are Both Considered

A bioequivalence comparison commonly evaluates both the extent of exposure and the peak concentration. AUC summarizes exposure over a specified time interval, while Cmax is the highest observed concentration in the concentration–time profile. These measures describe related but distinct features: similar total exposure does not guarantee a similar peak, and a similar peak does not guarantee similar total exposure.

Regulatory decisions therefore generally apply the relevant acceptance criterion to each required endpoint. For a conventional average bioequivalence assessment, the estimated test-to-reference geometric mean ratio and its confidence interval are evaluated on the ratio scale for AUC and for Cmax. The precise AUC definition, the required endpoints, and any product-specific criteria must be specified by the applicable regulatory guidance or protocol; they should not be assumed to be identical for every product.

Definition: A joint AUC–Cmax assessment evaluates the prespecified bioequivalence criterion separately for AUC and Cmax. Under conventional criteria, each endpoint’s confidence interval for the test-to-reference geometric mean ratio must lie within its applicable lower and upper acceptance limits. Passing one endpoint does not establish that the other endpoint passes.

The conventional limits discussed in the earlier material on acceptance limits are often 0.80 and 1.25 on the ratio scale. They are applied to the confidence interval, not merely to the point estimate. If AUC meets its criterion but Cmax does not, the conventional joint assessment has not shown bioequivalence on both required endpoints. The reverse is also true.

Applying the Criterion to Each Endpoint

A usual analysis works with log-transformed exposure measures. For a given endpoint, let \(\hat{\theta}\) denote the estimated mean difference between test and reference on the log scale, and let \(SE\) be its standard error. With the model and degrees of freedom appropriate to the study, a two-sided confidence interval on the log scale has the form \(\hat{\theta}\pm t^*SE\), where \(t^*\) is the relevant critical value. Exponentiating both limits gives the confidence interval for the test-to-reference geometric mean ratio.

\left[ \exp(\hat{\theta}-t^*SE),\; \exp(\hat{\theta}+t^*SE) \right]

For conventional limits of 0.80 and 1.25, the interval passes when its lower limit is greater than 0.80 and its upper limit is less than 1.25, using unrounded values for the decision. The equivalent condition on the log scale is that both confidence limits lie between \(\ln(0.80)\) and \(\ln(1.25)\). This follows because the exponential function is strictly increasing: exponentiating preserves the ordering of the limits.

0.80 < \exp(L) \quad \text{and} \quad \exp(U) < 1.25 \quad\Longleftrightarrow\quad \ln(0.80) < L \quad \text{and} \quad U < \ln(1.25)

The conventional 90% confidence interval corresponds to the two one-sided tests approach at a 5% significance level under the usual model. For this tutorial’s calculations, assume that the endpoint-specific log-scale estimates have the stated standard errors, that the stated \(t^*\) is appropriate for the fitted model, and that the analysis is valid for the prespecified study design. The examples use hypothetical summary values; they do not describe actual studies.

Worked Example: Both AUC and Cmax Meet the Criterion

In a hypothetical study, the estimated test-to-reference geometric mean ratio for AUC is 1.020, with log-scale \(SE=0.050\). For Cmax, the ratio is 1.040, with log-scale \(SE=0.040\). Suppose the appropriate 90% critical value is \(t^*=1.725\) for both calculations. Assess each endpoint using limits of 0.80 and 1.25.

AUC. The log of the ratio estimate is \(\ln(1.020)=0.019803\). Its log-scale interval is \(0.019803\pm 1.725(0.050)=0.019803\pm0.086250\), or \([-0.066447,\;0.106053]\). Exponentiating gives \([0.9357,\;1.1119]\), approximately.

Cmax. The log of the ratio estimate is \(\ln(1.040)=0.039221\). Its log-scale interval is \(0.039221\pm1.725(0.040)=0.039221\pm0.069000\), or \([-0.029779,\;0.108221]\). Exponentiating gives \([0.9707,\;1.1143]\), approximately.

Both intervals lie entirely within \([0.80,\;1.25]\). Thus both endpoints meet the stated conventional criterion. The conclusion is limited to these endpoints and assumptions; any other prespecified regulatory requirements would also have to be satisfied.

Why One Passing Endpoint Is Not Enough

The requirement is not that the endpoints have similar numerical ratios, nor that their results be averaged together. Each endpoint answers a different exposure question. AUC may indicate comparable overall exposure while Cmax indicates that peak concentrations are not sufficiently well bounded by the acceptance interval. Conversely, similar peak concentrations do not ensure comparable total exposure. A passing result for one cannot compensate for a failing result for the other.

Worked Example: AUC Passes but Cmax Does Not

Consider a second hypothetical study with the same AUC result as in the previous example: AUC ratio 1.020 and \(SE=0.050\), giving an interval of \([0.9357,\;1.1119]\). The Cmax ratio is 1.080 with log-scale \(SE=0.090\). Let \(t^*=1.725\). Determine the joint conclusion.

For Cmax, \(\ln(1.080)=0.076961\), and the margin is \(1.725(0.090)=0.155250\). The log-scale interval is therefore \(0.076961\pm0.155250=[-0.078289,\;0.232211]\). Exponentiating yields approximately \([0.9247,\;1.2614]\).

The AUC interval lies within the limits. The Cmax interval has an upper limit above 1.25, so Cmax does not meet the criterion. Although its point estimate, 1.080, lies within the limits, that fact is not sufficient: the interval also represents uncertainty in the estimated ratio. The joint assessment does not establish bioequivalence on both endpoints.

Worked Example: Cmax Passes but AUC Does Not

In a third hypothetical study, the AUC ratio is 0.820 with log-scale \(SE=0.035\), and the Cmax ratio is 1.000 with log-scale \(SE=0.040\). Suppose \(t^*=1.725\) and conventional limits apply to both endpoints.

AUC. The log estimate is \(\ln(0.820)=-0.198451\). The margin is \(1.725(0.035)=0.060375\), so the log-scale interval is \([-0.258826,\;-0.138076]\). Exponentiating gives approximately \([0.7720,\;0.8710]\). The lower limit is below 0.80, so the AUC criterion is not met.

Cmax. The log estimate is \(\ln(1.000)=0\). Its margin is \(1.725(0.040)=0.069000\), giving a log-scale interval of \([-0.069000,\;0.069000]\). Exponentiating gives approximately \([0.9333,\;1.0714]\), which lies within the limits.

Cmax meets the criterion, but AUC does not. Therefore, the joint assessment does not establish bioequivalence on both endpoints. The point estimate for AUC is within the limits, yet its lower confidence limit is not; again, the interval rather than the point estimate determines the conclusion.

Assumptions, Diagnostics, and Limitations

The interval calculations are meaningful only if they come from an appropriate, prespecified analysis. In particular, the endpoints and their measurement windows should be chosen before examining the results, the exposure measurements should be constructed and checked appropriately, and the statistical model should reflect the study design. The estimates and standard errors must refer to the same endpoint, analysis population, and model. The examples assume those conditions and do not substitute for evaluating a real study’s design or data.

Useful checks include confirming that the test-to-reference ratio is oriented correctly, verifying that the AUC definition and Cmax measure match the prespecified endpoints, and checking calculations on the log scale before exponentiation. Inspect the precision of each interval: an interval extending beyond a limit may reflect an estimated ratio near a limit, substantial uncertainty, or both. Rounded output can conceal a value just outside a boundary, so regulatory decisions should use unrounded values. A data or model problem—such as an incorrect endpoint, an inconsistent analysis set, or a standard error from a different model—can make a carefully computed interval misleading.

The 80–125% limits are conventional, not universal. Product-specific guidance may specify a different endpoint, analysis, or criterion. This tutorial does not replace those requirements. It also does not treat AUC and Cmax as interchangeable measures or as a single combined numerical score. Their separate interpretation is the reason the joint assessment is informative.

A common mistake is to conclude, “AUC passed, and the Cmax point estimate is within 80–125%, so the products are bioequivalent.” The reasoning fails because the Cmax confidence interval may extend outside the applicable limits, as the second worked example shows. The correction is to assess the full required interval for every prespecified endpoint and make the overall conclusion only after those separate assessments.

Key takeaway: For a conventional joint AUC–Cmax assessment, apply the applicable acceptance criterion independently to each required endpoint. A result on one endpoint cannot offset failure on the other; bioequivalence is not established unless both meet their criteria and all other applicable requirements are satisfied.

Check Your Understanding

Answer each question using the endpoint-specific interval, not just its point estimate.

AUC has a 90% interval of \([0.91,\;1.13]\), and Cmax has an interval of \([0.79,\;1.10]\). With limits of 0.80 and 1.25, what is the joint conclusion?
Why does a Cmax ratio estimate of 1.08 not establish that Cmax meets the criterion?
If Cmax passes but the AUC interval extends below 0.80, can the joint assessment establish bioequivalence on both endpoints? Explain.
Why are the ratio limits applied to each endpoint’s interval rather than to an average of the AUC and Cmax ratios?

Solutions. (1) AUC passes, but Cmax does not because its lower limit, 0.79, is below 0.80; the joint assessment does not establish bioequivalence on both endpoints. (2) The point estimate does not show where the full confidence interval lies. The upper confidence limit could exceed 1.25, or the lower limit could fall below 0.80. (3) No. AUC and Cmax are separately required measures in this assessment, so the passing Cmax result cannot compensate for the AUC interval failing its criterion. (4) They represent different exposure characteristics; averaging their ratios would obscure endpoint-specific results and is not the stated regulatory criterion.