Truncated AUC Analysis for Long-Half-Life Drugs

Why Truncate AUC?

For a drug with a long elimination half-life, estimating exposure all the way to infinity may require substantial extrapolation from the observed concentration profile. A truncated area under the concentration–time curve (AUC) instead measures exposure from dosing to a fixed, prespecified time, such as 72 hours. The aim is to compare formulations over a defined, observed portion of the profile—not to estimate the entire exposure by extending the curve beyond the sampling period.

Truncation is a change in the exposure endpoint, not merely a calculation shortcut. A bioequivalence conclusion for AUC from zero to 72 hours concerns exposure through 72 hours. By itself, it does not establish that the formulations have equivalent exposure after that time or equivalent AUC to infinity. The truncation time and the analysis must therefore be justified and specified before the results are examined.

A fixed cutoff makes the comparison interpretable only when it is applied consistently to both formulations and all subjects. The sampling schedule should support estimation of the area through the cutoff. If no concentration is measured at \(T\), the protocol must specify how the concentration at that boundary will be obtained; an analyst should not choose a convenient extension after seeing the data. Long half-life alone does not make every truncated endpoint appropriate. The endpoint should reflect the study question and applicable regulatory guidance.

Calculating the Prespecified Endpoint

A common noncompartmental calculation uses the linear trapezoidal rule between consecutive observed concentrations. For measurements \(C_i\) and \(C_{i+1}\) at times \(t_i\) and \(t_{i+1}\), the interval area is the average concentration multiplied by the interval duration. Summing the interval areas up to \(T\) gives the truncated AUC.

Concentration may be reported in ng/mL and time in hours, giving AUC units of ng·h/mL. The calculation must use the same stated integration convention for test and reference. If a protocol specifies a linear-up/log-down method or another convention, use that convention rather than silently substituting a different one. The earlier material on deriving AUC from concentration data provides the broader noncompartmental context; here the key is that integration stops at the prespecified boundary.

Comparing Formulations on the Log Scale

For a standard average bioequivalence assessment, the truncated AUC values are generally analyzed after natural-log transformation. In a 2×2 crossover, the model accounts for the design terms, including treatment, period, and sequence, with subjects nested within sequence. A simplified statement of the model is that log AUC is explained by these design effects plus a residual error. The treatment contrast estimates the difference between test and reference mean log AUC.

This analysis relies on an appropriate crossover design and analysis set; complete and correctly assigned endpoint data; independent residual errors with a common variance; and residuals that are reasonably consistent with the model’s normal-error assumption. The log transformation also requires positive AUC values. Diagnostics should assess data integrity, influential profiles, residual patterns, and sequence or period issues that could compromise interpretation. The endpoint’s cutoff and calculation method must be common to both treatments.

Exponentiating the estimated log treatment difference gives the test-to-reference geometric mean ratio. A 90% confidence interval on the log scale is also exponentiated to obtain an interval for that ratio. Under the usual average bioequivalence limits of 80% to 125%, bioequivalence is concluded for this endpoint when the entire 90% interval lies within those bounds, subject to applicable requirements and the prespecified analysis.

Here \(L\) and \(U\) are the lower and upper log-scale confidence limits. The usual equivalence rule is \(0.80 \leq \exp(L)\) and \(\exp(U) \leq 1.25\). This is an intersection requirement: both bounds must be met. A point estimate inside the limits is not enough if the confidence interval extends beyond either limit. The interval describes uncertainty about the treatment ratio under the model; it is not a probability statement that the fixed true ratio has a particular chance of lying inside this realized interval.

What the Truncation Changes

A truncated endpoint gives less weight to exposure after the cutoff because that exposure is excluded, not estimated. Consequently, two formulations can have similar truncated AUCs but different late concentration profiles. Conversely, a difference that emerges late may influence \(AUC_{0-\infty}\) while having little influence on \(AUC_{0-T}\). Neither result is inherently contradictory: the endpoints summarize different time intervals.

The practical interpretation depends on why the cutoff was chosen. If the prespecified question concerns early or intermediate exposure and the cutoff captures that interval reliably, a truncated AUC can avoid dependence on a long and uncertain extrapolation. But if differences after the cutoff matter to the intended use or safety question, truncation alone cannot answer that question. The endpoint should not be selected after comparing which definition produces a more favorable result.

Diagnostics, Alternatives, and Limitations

Before interpreting a truncated AUC comparison, check both the endpoint construction and the inferential model. Confirm that the time units, concentration units, formulation labels, and cutoff are consistent; that the measured profile covers the intended interval; and that the same integration rule was applied to each formulation. Review missing or BLQ concentrations and protocol deviations using prespecified handling rules. Assess the log-scale model’s residuals and influential observations, and examine period or sequence patterns that could signal design or data problems. A sensitivity analysis may be useful when a justified protocol deviation affects the endpoint, but it should not replace the prespecified primary analysis.

The main alternatives answer different questions. AUC to the last quantifiable concentration uses the observed profile to its endpoint, which can vary across subjects; \(AUC_{0-\infty}\) adds an estimated terminal contribution and can be sensitive when little terminal data are available. A truncated AUC uses a common fixed boundary and avoids extrapolating beyond it, but deliberately omits later exposure. \(C_{\max}\) summarizes peak concentration rather than integrated exposure. These endpoints are not interchangeable, and a favorable result for one does not erase an unfavorable result for another.

Common warning signs include sparse observations near the cutoff, substantial differences in sampling times between treatments, implausible concentration profiles, or a result driven by one influential subject. These do not automatically invalidate an analysis, but they call for investigation and transparent reporting. If the endpoint cannot be estimated reliably at the specified boundary, a nominally precise model output does not repair the underlying measurement problem.

The central reporting discipline is to name the endpoint, its cutoff, its units, and the interval used for the decision. A truncated AUC can be a useful exposure measure for a long-half-life drug when it is scientifically justified and reliably measured. Its value depends on preserving the boundary between what the data support through \(T\) and what remains unknown after \(T\).