Sampling-Time Sensitivity in Transit Absorption

Why Sampling Time Matters

A transit absorption model describes drug entering the systemic circulation over a distribution of times rather than at one fixed delay. A concentration profile can therefore contain information about both the timing and spread of input. But that information is not present equally at every sampling time. A sample taken well before appreciable input may be close to baseline; a sample taken after nearly all input has occurred may be insensitive to the timing of absorption. Samples around the rising and peak portions of the input profile are often more useful for resolving its shape.

The useful sampling window depends on the transit parameters and on disposition. In particular, the time of maximum absorption rate is not generally the time of maximum plasma concentration: elimination and distribution alter the concentration profile. The calculations below use the transit-time distribution to assess when input occurs, then distinguish that input information from the concentration measurements used to estimate a model.

Transit Input and Timing Quantities

Definition: For $N$ transit compartments with mean transit time $MTT$, let $k_{tr}=N/MTT$. The time from dose to systemic input has a gamma distribution with shape $N$ and rate $k_{tr}$. Its density, cumulative fraction, and instantaneous input rate per unit dose are $$ g(t)=\frac{k_{tr}^{N}t^{N-1}e^{-k_{tr}t}}{(N-1)!}, \qquad G(t)=1-e^{-x}\sum_{j=0}^{N-1}\frac{x^j}{j!}, \qquad x=k_{tr}t. $$ Here $G(t)$ is the fraction of the dose that has entered by time $t$, and $g(t)$ has units of inverse time. The density has mean $MTT$. For $N>1$, its mode is $(N-1)/k_{tr}=MTT(N-1)/N$.

The cumulative fraction $G(t)$ answers a timing question: how much input has occurred by a particular time? The density $g(t)$ answers a different question: how rapidly is input occurring at that time? Neither quantity is a plasma concentration. Under a one-compartment disposition model with first-order elimination, for example, concentration after a single dose is proportional to $$ C(t)=\frac{FD}{V}\int_0^t g(s)e^{-k_{el}(t-s)}\,ds, $$ where $F$ is bioavailability, $D$ is dose, $V$ is volume of distribution, and $k_{el}$ is the elimination rate constant. This convolution shows why concentration depends on both prior input and subsequent elimination.

How Input Sensitivity Changes Over Time

A useful first check is how the cumulative input fraction changes when $MTT$ changes. Holding $N$ fixed, write $G(t)=G_N(x)$, where $x=Nt/MTT$. Differentiating the gamma cumulative distribution gives $$ \frac{dG_N(x)}{dx}=\frac{e^{-x}x^{N-1}}{(N-1)!}. $$ Because $dx/dMTT=-x/MTT$, the sensitivity of cumulative input to mean transit time is $$ \frac{\partial G(t)}{\partial MTT} =-\frac{x}{MTT}\frac{e^{-x}x^{N-1}}{(N-1)!} =-\frac{x}{MTT}g_{\Gamma}(x), $$ where $g_{\Gamma}(x)$ is the gamma density expressed on the dimensionless $x$-scale. This derivative is negative: increasing $MTT$ delays input, so the fraction absorbed by a fixed time cannot increase. Its magnitude is small when $x$ is very small or very large, and is generally larger over the transition where the cumulative fraction is changing appreciably. Thus, very early and very late samples can carry little information about timing even if they are measured precisely.

The input-rate mode provides a practical guide to the transition region. For a given $N$ and $MTT$, samples around and on both sides of that mode can help reveal the rising and falling portions of input. This is a guide, not a rule that a sample exactly at the mode is optimal: measured concentrations are shaped by disposition, and a design must also consider residual error, feasible sampling times, and the range of plausible parameter values.

Worked Example: Locating the Main Input Window

Suppose a hypothetical single-dose model has $N=3$ transit compartments and $MTT=4$ h. Assume these values are fixed for this calculation. Then $k_{tr}=3/4=0.75\ \mathrm{h}^{-1}$, and the input-rate mode is $MTT(N-1)/N=4(2)/3=2.667$ h.

At 4 h, $x=k_{tr}t=0.75(4)=3$. The instantaneous input rate per unit dose is $$ g(4)=0.75e^{-3}\frac{3^2}{2} =0.75(0.0497871)(4.5) =0.1680\ \mathrm{h}^{-1}. $$ At the same time, the cumulative fraction is $$ G(4)=1-e^{-3}(1+3+3^2/2) =1-(0.0497871)(8.5) =0.5768. $$ So about 57.68% of the dose has entered by 4 h, while input is still occurring at a substantial rate. A schedule with samples both before and after this period is more informative about timing than one consisting only of very early and very late samples.

Worked Comparisons of Sampling Times

Worked Example: Comparing Transit Numbers at Selected Times

Consider two hypothetical transit models with the same $MTT=4$ h, one with $N=2$ and one with $N=3$. Compare their cumulative input fractions at 2, 4, and 8 h. The rate is $k_{tr}=N/4$ in each model, so the dimensionless time $x$ differs between them.

At 2 h, the $N=2$ model has $x=1$, giving $G=1-e^{-1}(1+1)=0.2642$. The $N=3$ model has $x=1.5$, giving $G=1-e^{-1.5}(1+1.5+1.5^2/2)=0.1912$. The difference $G_{N=2}-G_{N=3}$ is $0.07309$.

At 4 h, the fractions are $1-e^{-2}(1+2)=0.5940$ for $N=2$ and $1-e^{-3}(1+3+3^2/2)=0.5768$ for $N=3$. Their difference is $0.01718$. At 8 h, the fractions are $1-e^{-4}(1+4)=0.9084$ and $1-e^{-6}(1+6+6^2/2)=0.9380$, respectively, so the difference is $-0.02961$.

These calculations show that the models do not have a fixed ordering at every time and that the difference in cumulative input can be modest at particular times. They do not establish that any one time is useless for estimating $N$: plasma concentrations, multiple measurements, residual error, and other model parameters also matter. They do show why choosing samples only by intuition about a single “best” time can miss useful differences in profile shape.

Worked Example: Late Samples Can Miss Timing Differences

Compare hypothetical $N=3$ models with $MTT=3$ h and $MTT=5$ h. At 12 h, the dimensionless times are $x=12$ and $x=7.2$. The cumulative fractions are $$ G_{MTT=3}(12)=1-e^{-12}(1+12+12^2/2)=0.9995, $$ $$ G_{MTT=5}(12)=1-e^{-7.2}(1+7.2+7.2^2/2)=0.9745. $$ At 16 h, the respective dimensionless times are $x=16$ and $x=9.6$, giving cumulative fractions $1-e^{-16}(1+16+16^2/2)=1.000$ and $1-e^{-9.6}(1+9.6+9.6^2/2)=0.9962$, to four significant figures.

The models differ in mean transit time by 2 h, but by 16 h both have delivered nearly all of the dose. A schedule dominated by these late times would therefore provide little direct evidence about when input occurred. The small remaining difference at 16 h is not a concentration difference; its practical value also depends on how the input history combines with elimination and on measurement precision.

Worked Example: Identifying a Poorly Resolved Schedule

A hypothetical study uses $N=3$, $MTT=4$ h as a planning model, but the true values are not known. The proposed schedule collects concentrations at 0.5, 1, 12, and 24 h. Assume the model's dose time is known, and there are no additional doses over this interval. At 1 h, $x=0.75$, so $G(1)=1-e^{-0.75}(1+0.75+0.75^2/2)=0.0405$, approximately. At 12 h, $x=9$, and $G(12)=1-e^{-9}(1+9+9^2/2)=0.9938$, approximately.

The early samples are taken when little input has occurred, and the later samples are taken after almost all input. There is no observation around the predicted input-rate mode of 2.667 h or across the subsequent transition. This schedule is therefore poorly positioned to describe the absorption phase. A more informative candidate might include times such as 1, 2, 3, 4, and 6 h, alongside appropriate later samples for disposition. Those times bracket the predicted input window; they are not guaranteed to be optimal, and their value should be checked using the full concentration model.

Design Checks, Alternatives, and Limitations

The transit fractions provide a transparent timing diagnostic, but a final design check should use predicted concentrations under the complete model. For plausible values of $MTT$, $N$, and disposition parameters, compare predicted concentration profiles at candidate times. One useful local check is a finite-difference sensitivity such as $\{C(t;MTT+\delta)-C(t;MTT-\delta)\}/(2\delta)$, with a small positive $\delta$ chosen so both parameter values are admissible. A time is potentially informative about $MTT$ when predictions change appreciably across plausible values, relative to the expected measurement error. This check should be repeated over a plausible parameter range rather than relying on one nominal value.

Several signs suggest that a schedule is not resolving absorption: fitted transit parameters change markedly when one early sample is removed; different plausible transit profiles predict nearly identical values at all observed times; estimates depend strongly on assumptions about disposition; or the observed profile has no measurements spanning the rise and decline. These are warnings to examine the model and design, not proof that the transit structure is wrong. Sparse early sampling is a particular concern because a low early concentration can reflect delayed input, slow input, or a combination of input and disposition.

Transit absorption is one way to represent distributed input. A lag-time model places input after a delay; first-order absorption describes an exponential input process; zero-order absorption describes input at a constant rate over a duration. These alternatives imply different input shapes, so a design suited to one may not distinguish another. The appropriate comparison is based on predicted profiles at the actual sampling times, not on the names of the models alone.

No sampling schedule can compensate for a misspecified disposition model, unrecorded dose timing, or poor measurement quality. In addition, population variability can make individuals differ in absorption timing, while a design assessed only at one typical parameter set may understate that variation. The cumulative fraction and mode calculations are therefore planning tools: they identify where timing information may lie, but do not by themselves quantify precision or guarantee parameter identifiability.

Common Pitfall

Common mistake: “The sample at the input-rate mode is the concentration peak, so that one sample identifies absorption timing.”

Correction: The input-rate mode is the peak of $g(t)$, not generally the peak of plasma concentration. Concentration integrates prior input while elimination removes drug. Timing is usually informed by a sequence of concentrations spanning the rising and changing portions of the profile, interpreted with the disposition model.

Check Your Understanding

Answer each question, then compare with the solutions below.

For $N=4$ and $MTT=8$ h, calculate $k_{tr}$ and the input-rate mode.
For the model in Question 1, what dimensionless time $x$ corresponds to a sample at 4 h?
Why can samples at 20 and 24 h provide little information about transit timing if nearly all input has occurred by then?
Does sampling exactly at the input-rate mode guarantee a sample at the plasma concentration maximum? Explain.

Solutions. (1) $k_{tr}=N/MTT=4/8=0.5\ \mathrm{h}^{-1}$. The mode is $MTT(N-1)/N=8(3)/4=6$ h. (2) $x=k_{tr}t=0.5(4)=2$. (3) When cumulative input is near one, later changes in $MTT$ alter the fraction already input only slightly; the remaining input-time differences may be difficult to resolve, depending on the concentration model and measurement error. (4) No. The input-rate mode maximizes $g(t)$, whereas plasma concentration also depends on earlier input and elimination, so the concentration maximum can occur at a different time.