Estimating Absorption Duration From Concentration Profiles

Why the End of Input Is a Profile Feature

For zero-order absorption, drug enters the systemic circulation at a constant rate for a finite time. The absorption duration is the length of that input period. It is not generally observed directly: concentration measurements show the consequences of input and elimination together. The estimation problem is to locate the transition from ongoing input to decline after input has stopped.

A concentration profile can locate this transition most clearly when it has samples on both sides of the peak. Sparse sampling, measurement error, or uncertainty about the disposition rate can make the duration much less certain. The estimates below use a one-compartment model to make the connection precise; they are not claims that every real concentration profile has a sharp, easily observed peak.

Definition: Suppose systemic input begins at time zero and occurs at a constant rate $R_0$ until time $T$. The absorption duration is $T$. In a one-compartment model with volume $V$ and first-order elimination rate constant $k=CL/V$, write $q=R_0/V$. For an initial concentration of zero, the predicted concentration is $$ C(t)= \begin{cases} \dfrac{q}{k}\left(1-e^{-kt}\right), & 0\leq t\leq T,\\[6pt] \dfrac{q}{k}\left(1-e^{-kT}\right)e^{-k(t-T)}, & t>T. \end{cases} $$ Here $q$ has units of concentration per time, $k$ has units of inverse time, and $T$ and $t$ use the same time units.

This model assumes constant input over the stated interval, linear one-compartment disposition, a known input start time, and no other active dose or input affecting the profile. It also assumes that concentrations are measured on the model's concentration scale. Departures from these assumptions can shift or blur the apparent transition.

Why the Transition Occurs at the Peak

During input, the concentration rises toward the level at which elimination would balance the input if input continued indefinitely. Once input stops, concentration declines under elimination alone. To see why the transition occurs at $T$, differentiate the model on either side of that time.

\frac{dC(t)}{dt}= \begin{cases} q-kC(t)=q e^{-kt}>0, & 0\leq t<T,\\ -kC(t)<0, & t>T \text{ and } C(t)>0. \end{cases}

The concentration is continuous at $T$: both parts of the model give $C(T)=\dfrac{q}{k}(1-e^{-kT})$. Its slope changes from positive just before $T$ to negative just after $T$. Thus, under these assumptions, the continuous-time profile reaches its maximum at the end of input. The slope need not be continuous there; indeed, its change is the model's signature of input stopping.

This result gives a useful visual principle, but it does not mean that the largest observed concentration time is exactly $T$. Samples are taken at discrete times, and measurement error can move the observed maximum. A sampled peak is a landmark for the transition, not automatically a precise duration estimate.

Worked Example: Locating a Transition From Concentration Samples

Consider hypothetical concentrations generated by the model with $q=1.00\ \mathrm{mg/L/h}$, $k=0.200\ \mathrm{h^{-1}}$, and $T=5.00\ \mathrm{h}$. Suppose samples are collected at 2, 4, 6, and 8 hours.

At 2 and 4 hours, input is ongoing: $$ C(2)=\frac{1.00}{0.200}(1-e^{-0.200(2)}) =5(1-e^{-0.4})=1.648\ \mathrm{mg/L}, $$ $$ C(4)=5(1-e^{-0.8})=2.753\ \mathrm{mg/L}. $$ After input has stopped, the concentrations are $$ C(6)=5(1-e^{-1})e^{-0.200(6-5)} =2.588\ \mathrm{mg/L}, $$ $$ C(8)=5(1-e^{-1})e^{-0.200(8-5)} =1.735\ \mathrm{mg/L}. $$ The samples rise from 2 to 4 hours and then fall by 6 hours. The maximum observed concentration is at 4 hours, but the model's true transition is at 5 hours. These data indicate that the peak is near the interval spanning 4 to 6 hours; they do not justify reporting the sampled maximum time, 4 hours, as the absorption duration.

The example is deliberately noise-free, so the falling concentrations after input stops follow the model exactly. With real observations, small fluctuations can make adjacent increases and decreases unreliable evidence about the transition.

Estimating Duration When Input and Disposition Are Known

If $q$ and $k$ are known and a concentration $C(t)$ is measured after input has ended, the post-input equation can be rearranged to estimate $T$. The equation is applicable only when $t>T$, and the known parameters and model assumptions must be appropriate for the profile.

C(t)=\frac{q}{k}(1-e^{-kT})e^{-k(t-T)} =\frac{q}{k}(e^{kT}-1)e^{-kt}

Multiplying by $ke^{kt}/q$ and isolating $T$ gives

e^{kT}=1+\frac{kC(t)e^{kt}}{q}, \qquad T=\frac{1}{k}\ln\left(1+\frac{kC(t)e^{kt}}{q}\right).

The argument of the logarithm is dimensionless: $kC(t)e^{kt}/q$ has units $(1/\mathrm{time})(\mathrm{concentration})/(\mathrm{concentration}/\mathrm{time})$. A calculated value $T\geq t$ contradicts the assumption that the selected sample is post-input and signals that the model, inputs, or parameter values should be reconsidered.

Worked Example: Solving for Duration From a Post-Input Sample

Suppose the hypothetical model has $q=1.00\ \mathrm{mg/L/h}$ and $k=0.200\ \mathrm{h^{-1}}$. A sample at $t=7.00\ \mathrm{h}$, known to be after input ended, is $C(7)=2.119\ \mathrm{mg/L}$. Estimate $T$.

Substitute into the rearranged equation: $$ T=\frac{1}{0.200}\ln\left(1+ \frac{(0.200)(2.118615)e^{(0.200)(7)}}{1.00}\right)\ \mathrm{h}. $$

Using $e^{1.4}=4.05520$, the term inside the logarithm is approximately $$ 1+(0.200)(2.118615)(4.05520)=2.7183. $$ Therefore, $$ T=\frac{\ln(2.7183)}{0.200}=5.000\ \mathrm{h}. $$ The result is consistent with $t=7.00\ \mathrm{h}>T$. This estimate relies on $q$ and $k$ being known and correct. If either is uncertain, their uncertainty also affects the duration estimate.

A post-input profile alone does not usually determine duration if $q$ is unknown. Its equation contains the combination $q(e^{kT}-1)$, as well as $k$, rather than separately revealing $q$ and $T$. Likewise, uncertainty in $k$ can be important: the observed decline reflects elimination after input stops, and an incorrect elimination rate can distort a duration inferred by back-calculation. Information from the rising part of the profile or externally supported parameter values can help constrain the estimate.

Sampling Times and the Precision of the Estimate

The most useful samples for locating $T$ are those that capture the profile near its change from rise to decline. Samples only far before the transition document ongoing input but do not locate when it stops. Samples only well after the transition document decline but may provide little direct evidence about the exact transition time.

Under the ideal model, if a sample at $t_a$ is followed by a higher concentration at $t_b$, then $T>t_a$: input cannot already have stopped before $t_a$ while the model concentration is declining. If a sample at $t_c$ is followed by a lower concentration at $t_d$, then $T<t_d$. These inequalities can give a broad interval for the transition when the increasing and decreasing observations are separated. They do not locate $T$ exactly, and with noisy observations they are only clues rather than strict bounds.

Worked Example: Comparing Dense and Sparse Sampling

Consider a second hypothetical profile with $q=1.00\ \mathrm{mg/L/h}$, $k=0.200\ \mathrm{h^{-1}}$, and $T=5.20\ \mathrm{h}$. Its peak concentration is $$ C(5.20)=5(1-e^{-1.04})=3.233\ \mathrm{mg/L}. $$ At 4 hours the concentration is still on the rising segment: $$ C(4)=5(1-e^{-0.8})=2.753\ \mathrm{mg/L}. $$ At 5 hours it is $C(5)=5(1-e^{-1})=3.161\ \mathrm{mg/L}$, and at 6 hours it is post-input: $$ C(6)=3.2327e^{-0.200(6-5.20)} =3.2327e^{-0.16}=2.755\ \mathrm{mg/L}. $$

With samples at 4, 5, and 6 hours, the largest sampled value is at 5 hours. Because the profile is rising from 4 to 5 and falling from 5 to 6, the transition lies between 4 and 6 hours under the ideal unimodal model. The sampled maximum at 5 hours is close to the true duration of 5.20 hours, but does not identify it exactly.

Now suppose only 4- and 8-hour samples are available. The 4-hour value is $2.753\ \mathrm{mg/L}$, whereas $$ C(8)=3.2327e^{-0.200(8-5.20)} =3.2327e^{-0.56}=1.847\ \mathrm{mg/L}. $$ The largest sampled value is now at 4 hours, a full 1.20 hours before the true end of input. The coarse schedule misses the peak region. Calling the observed maximum time the duration would therefore be a poor estimate. Fitting the full model can use the concentration values and assumed parameter information, but sparse sampling still leaves less direct information about where the transition occurred.

Model-Based Estimation, Diagnostics, and Limitations

A sampled maximum is simple to report but depends strongly on the sampling grid. A model-based estimate instead uses the predicted rise and post-input decline together, with $T$ as the transition parameter. When $q$ and $k$ are not known, they must be estimated or otherwise constrained along with $T$; the observed profile may not distinguish their effects well. In nonlinear mixed-effects analyses, individual durations may also vary around a population pattern, while sparse individual sampling can make those durations difficult to estimate. A population estimate should not be mistaken for an exactly observed transition in every individual.

Useful checks include plotting each concentration against its actual sampling time, checking whether the data show evidence of both rising and declining phases, and inspecting whether the estimated transition is supported by observations near the peak. After the estimated $T$, log concentration should decline with slope $-k$ under this model. Large systematic departures from that pattern, multiple apparent peaks, or a fitted transition outside the observed time range may indicate that the assumptions or model structure are inadequate.

A common mistake is to treat $T_{\max}$, the time of the largest sampled concentration, as the exact absorption duration. The correction is that $T$ is the model's continuous-time transition, while the sampled maximum is only observed at scheduled times. Another mistake is to use the post-input formula before verifying that the sample is after input ended; doing so can produce an algebraically calculated value that has no valid post-input interpretation.

Key takeaway: In the one-compartment zero-order model, absorption ends at the continuous-time peak, where the profile changes from rising to declining. The duration is estimated most credibly when sampling captures that transition and the disposition and input assumptions are defensible; a sampled peak alone is not an exact estimate.

Check Your Understanding

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

Under the model in this tutorial, why does concentration reach its maximum at $T$?
A sample at 3 hours is higher than a sample at 2 hours, and a sample at 7 hours is lower than a sample at 6 hours. What can these comparisons imply about $T$ under the ideal model?
Why can a post-input concentration not generally identify $T$ when both $q$ and $T$ are unknown?
What does a sampled maximum at 4 hours establish about the actual absorption duration?

Solutions. (1) Before $T$, $dC/dt=q e^{-kt}>0$; after $T$, $dC/dt=-kC(t)<0$, and concentration is continuous at $T$. (2) The increase implies $T>2$ hours, while the decrease implies $T<7$ hours; together they give the broad model-based interval $2<T<7$ hours, not an exact estimate. (3) The post-input equation depends on the combination $q(e^{kT}-1)$, so one concentration does not generally separate $q$ from $T$. (4) It identifies the largest value among the sampled times, not the exact continuous-time peak or duration; the sampling schedule and other profile information are needed to assess $T$.