Comparing Concentration-Time Profiles And Summary Endpoints

Why Compare Both the Curve and Its Endpoints?

A PBPK prediction is a concentration-time profile, not just a single number. The profile describes how concentration changes through time, while summary endpoints such as area under the concentration-time curve (AUC) and maximum concentration (Cmax) compress aspects of that trajectory into quantities that are easier to compare or use in a decision. Neither view replaces the other: two profiles can have similar AUC but differ in peak height or timing, and similar Cmax does not ensure similar overall exposure.

A useful validation assessment therefore asks two related questions. How closely does the predicted profile follow the observed profile over the relevant time window? And how closely do prespecified summary endpoints agree? The first retains temporal detail; the second evaluates quantities tied to the intended application. Agreement on an endpoint should not be taken as evidence that every part of the profile is accurate.

Definition: A concentration-time profile is the concentration \(C(t)\) as a function of time \(t\), for a specified dose, route, formulation, population, and protocol. Over a specified interval \([t_0,t_K]\), AUC is the integral of concentration with respect to time. Cmax is the maximum concentration over a specified interval. When concentrations are available only at sampling times, the observed sampled Cmax is the largest recorded value; it need not equal the unobserved continuous-time peak.

Set Up a Comparable Assessment

Before calculating anything, confirm that the predicted and observed quantities refer to the same conditions and time frame. Concentrations must use compatible units, time must have the same origin, and the profiles must represent the same dose and relevant administration conditions. If measurements from multiple participants are being summarized, define whether the comparison concerns individual profiles or a group summary; comparing a predicted typical profile with pooled observed measurements can conceal individual variation.

AUC comparisons require a common integration window. For example, an observed AUC from 0 to 24 hours cannot be directly compared with a predicted AUC from 0 to 48 hours as if they were the same endpoint. Similarly, a Cmax comparison needs a clear definition of the search interval and of whether the model value is evaluated at observed sampling times or maximized over continuous time.

Working assumptions: The compared profiles represent the same target conditions and concentration scale; time points are correctly aligned; and the chosen interpolation or integration method is stated. A trapezoidal AUC calculated from discrete data approximates the integral by assuming a straight line between consecutive values. A sampled Cmax describes the largest value at the measured times, not necessarily the true peak between them. These are operational assumptions, not guarantees that the underlying concentration curve is linear or that sampling captured its maximum.

For observations at times \(t_0,\ldots,t_K\), with concentrations \(C_0,\ldots,C_K\), the linear trapezoidal estimate of AUC is

\operatorname{AUC}_{t_0-t_K}\approx \sum_{i=0}^{K-1}(t_{i+1}-t_i)\frac{C_i+C_{i+1}}{2}.

The same calculation is applied separately to observed and predicted concentrations using the same time points and window. Its units are concentration multiplied by time, such as mg·h/L when concentration is in mg/L and time is in hours.

Why the Trapezoidal Calculation Works

On one interval from \(t_i\) to \(t_{i+1}\), suppose concentration changes linearly from \(C_i\) to \(C_{i+1}\). The line has slope \((C_{i+1}-C_i)/(t_{i+1}-t_i)\). Integrating that line gives the interval area:

\begin{aligned} \int_{t_i}^{t_{i+1}} C(t)\,dt &= C_i(t_{i+1}-t_i) +\frac{C_{i+1}-C_i}{t_{i+1}-t_i} \frac{(t_{i+1}-t_i)^2}{2}\\ &=(t_{i+1}-t_i)\frac{C_i+C_{i+1}}{2}. \end{aligned}

Adding the interval areas yields the trapezoidal formula. The calculation is exact for a linear segment, but only an approximation when the true profile curves between samples. Long gaps around a rapid rise or peak can therefore produce substantial error, even when the arithmetic is correct.

Worked Example: Similar AUC, Different Peak

Suppose a hypothetical observed profile and PBPK prediction are compared at 0, 1, 2, 4, and 8 hours. Concentrations are in mg/L. The observed values are \(0, 8, 5, 2, 0.5\), and predicted values are \(0, 4, 6, 3, 0.5\). Calculate trapezoidal AUC and sampled Cmax for both.

Observed AUC. The four interval areas are \(1(0+8)/2=4\), \(1(8+5)/2=6.5\), \(2(5+2)/2=7\), and \(4(2+0.5)/2=5\). Thus, \(\operatorname{AUC}_{0-8}=4+6.5+7+5=22.5\) mg·h/L.

Predicted AUC. The interval areas are \(1(0+4)/2=2\), \(1(4+6)/2=5\), \(2(6+3)/2=9\), and \(4(3+0.5)/2=7\). Thus, predicted \(\operatorname{AUC}_{0-8}=2+5+9+7=23.0\) mg·h/L.

The predicted AUC is \(0.5\) mg·h/L higher, a difference of \(0.5/22.5\times100=2.22\%\) relative to the observed AUC. But the observed sampled Cmax is 8 mg/L at 1 hour, while the predicted sampled Cmax is 6 mg/L at 2 hours. Both profiles are sampled over the same 0–8-hour interval, but their sampled Cmax times differ. The close AUC values do not show that peak magnitude or timing agrees.

Assess the Profile Alongside the Endpoints

A direct profile comparison keeps the time-specific values visible. At matched times, define the discrepancy as predicted concentration minus observed concentration. A descriptive root mean square error (RMSE) summarizes the squared discrepancies on the concentration scale, but it depends on which times are included and weights each listed time equally. It is not a universal measure of clinical acceptability. A time-weighted discrepancy can instead account for unequal gaps, but it introduces its own weighting choice.

A practical assessment reports the profile and endpoint results together. It should make clear the time grid, the integration window, the interpolation rule, and whether Cmax is sampled or continuous. The following example illustrates why a single matching endpoint cannot certify profile agreement.

Worked Example: Same Sampled Cmax, Different AUC

Use the same hypothetical sampling times, 0, 1, 2, 4, and 8 hours. The observed concentrations are \(0, 8, 6, 3, 1\) mg/L; predictions are \(0, 8, 7, 6, 4\) mg/L.

Both sampled profiles have Cmax \(=8\) mg/L at 1 hour. For the observed AUC, the interval areas are \(4\), \(7\), \(2(6+3)/2=9\), and \(4(3+1)/2=8\), giving \(28\) mg·h/L. For the predicted AUC, they are \(4\), \(7.5\), \(2(7+6)/2=13\), and \(4(6+4)/2=20\), giving \(44.5\) mg·h/L.

The peak values agree at the sampled time, but predicted AUC exceeds observed AUC by \(16.5\) mg·h/L, or \(16.5/28\times100=58.9\%\) relative to observed AUC. The later concentrations account for much of this difference. A Cmax-only comparison would miss this substantial exposure disagreement.

A compact profile summary can be useful when the full table is large, provided its definition is explicit. For \(m\) matched sampling times, one descriptive choice is

\operatorname{RMSE}= \sqrt{\frac{1}{m}\sum_{i=1}^{m} \left(C_{\mathrm{pred}}(t_i)-C_{\mathrm{obs}}(t_i)\right)^2}.

The RMSE has concentration units and emphasizes larger discrepancies because they are squared. It does not indicate when discrepancies occur, whether predicted concentrations tend to be above or below observations, or whether the size is acceptable for the intended use. Always inspect the time-specific comparisons as well as any scalar summary.

Worked Example: Summarizing Time-Specific Profile Differences

Consider hypothetical observed concentrations \(0, 8, 5, 2, 0.5\) mg/L and predictions \(0, 7, 6, 2.5, 0.5\) mg/L at 0, 1, 2, 4, and 8 hours. The predicted-minus-observed differences are \(0,-1,+1,+0.5,0\) mg/L.

There are five matched times, so \(\operatorname{RMSE}=\sqrt{(0^2+(-1)^2+1^2+0.5^2+0^2)/5} =\sqrt{2.25/5}=\sqrt{0.45}=0.671\) mg/L.

The absolute discrepancies at those times are \(0,1,1,0.5,0\) mg/L. If these absolute differences are linearly interpolated between samples, their trapezoidal area is \(0.5+1+1.5+1=4\) mg·h/L; dividing by the 8-hour window gives a time-averaged absolute discrepancy of \(0.5\) mg/L. The two summaries answer different questions: RMSE weights the five listed times equally, whereas the integrated absolute discrepancy weights by elapsed time under the stated interpolation assumption.

For context, observed AUC is \(22.5\) mg·h/L. Predicted AUC is \(3.5+6.5+8.5+6=24.5\) mg·h/L. Observed sampled Cmax is 8 mg/L at 1 hour; predicted sampled Cmax is 7 mg/L at 1 hour. None of these summaries alone conveys the complete pattern: the prediction is lower at 1 hour and higher at 2 and 4 hours.

Choose the Summary That Matches the Question

Assessment	What it describes	What it can conceal
Time-aligned profile	Agreement or discrepancy at particular times	Can be hard to reduce to one number; sparse sampling leaves between-time behavior uncertain
AUC over a specified window	Integrated exposure over that window	Different peak heights, timing, and curve shapes can produce similar areas
Sampled Cmax	Largest concentration among the observed sampling times	A peak between samples may be higher; equal peaks do not ensure equal exposure
Continuous-time model Cmax	Maximum of the simulated concentration over a defined interval	Not directly equivalent to observed sampled Cmax unless the observation schedule captures the peak adequately

An alternative to trapezoidal integration is to integrate the simulated PBPK concentration function numerically on a sufficiently fine grid. That estimates the model’s continuous-time AUC, whereas applying the trapezoidal rule to model values at the observation times makes a more like-for-like comparison with observed trapezoidal AUC. The choice depends on the question, and the two calculations should not be silently mixed. Likewise, reporting the model’s continuous-time maximum may be scientifically useful, but it should be distinguished from the largest measured concentration.

There is no single acceptable numerical discrepancy for every application. A difference that is unimportant for one intended use may matter for another. The assessment should therefore report the actual profile and endpoint differences and interpret them against criteria appropriate to the prediction question, rather than inventing a universal threshold after seeing the results.

Common Pitfalls and Checks

Equating endpoint agreement with profile agreement. Similar AUC can hide different peaks or timing; matching Cmax can hide different later exposure. Review both endpoints and time-specific concentrations.
Calling sampled Cmax the true peak. The largest recorded value is only the maximum among measured times. State whether Cmax is sampled or obtained by maximizing the continuous simulated curve.
Comparing different windows. AUC depends on its integration limits. Use the same window or clearly label the different quantities.
Ignoring the time grid. A mean of discrepancies at irregularly spaced times weights each sample equally, not each hour equally. State the metric and its weighting.
Over-interpreting trapezoidal AUC. The calculation assumes linear interpolation between adjacent points. Sparse samples near a rapid rise or peak may make the approximation unreliable; report the sampling schedule and consider whether the window is adequately covered.
Reporting a scalar without its units or definition. AUC has concentration-times-time units, whereas Cmax and concentration-scale profile errors have concentration units. Include the window and calculation method.

Key takeaway: Evaluate the predicted concentration-time profile and the relevant summary endpoints as complementary evidence. AUC describes integrated exposure over a stated interval, Cmax describes a peak under a stated operational definition, and neither alone establishes agreement across the full trajectory.

Check Your Understanding

Answer each question, then compare with the worked solutions below.

At 0, 2, and 5 hours, hypothetical concentrations are 0, 6, and 3 mg/L. Calculate trapezoidal AUC from 0 to 5 hours.
Observed concentrations at three sampling times have a maximum of 5.8 mg/L at 2 hours. What does this establish, and what does it not establish about the continuous-time peak?
Two predicted profiles have the same AUC but different peak times. Is this contradictory? Explain.
Why should an observed AUC from 0 to 24 hours not be compared as the same endpoint to a predicted AUC from 0 to 48 hours?

Solutions. (1) The interval areas are \(2(0+6)/2=6\) and \(3(6+3)/2=13.5\), so AUC is \(19.5\) mg·h/L. (2) The observed sampled Cmax is 5.8 mg/L at 2 hours. A higher peak could have occurred between samples, so the continuous-time maximum is not established. (3) No contradiction: integration summarizes total area, and profiles with different shapes or peak times can have equal area. The profiles still differ over time. (4) The integration limits differ, so the quantities cover different exposure windows. A like-for-like comparison needs a common window or explicit labeling and justification for comparing different windows.