Interpreting Delayed Exposure-Response With an Effect Compartment

Why Plasma Exposure and Biomarker Response May Not Move Together

An exposure-response analysis can appear to show little or no pharmacologic effect when biomarker measurements are made soon after plasma concentrations rise. The mismatch may reflect a delay between plasma exposure and the site or process governing the response. In that situation, plotting the biomarker against plasma concentration over time can produce a loop: the response at a given plasma concentration differs between the rising and falling portions of the exposure profile.

An effect compartment is a parsimonious way to represent this delay. It does not require that the effect site be a separately measured anatomical compartment. Rather, it assigns a latent concentration to the biophase or response-driving process and allows that concentration to equilibrate gradually with measured plasma concentration. The model can help distinguish delayed pharmacology from a genuinely weak exposure-response relationship, but it cannot do so without assumptions and informative data.

Definition: An effect-compartment model introduces an unobserved effect-site concentration $C_e(t)$ that approaches plasma concentration $C_p(t)$ at a finite rate: $$ \frac{dC_e(t)}{dt}=k_{e0}\bigl(C_p(t)-C_e(t)\bigr), $$ where $k_{e0}>0$ is the equilibration-rate constant, usually expressed in inverse time units. The effect-site concentration is then linked to the response, rather than linking the response directly to $C_p(t)$.

A larger $k_{e0}$ means faster equilibration and a shorter lag; a smaller value means slower equilibration. The corresponding equilibration half-time is $\ln(2)/k_{e0}$. This is a model-based timescale, not necessarily a directly measured physiological transit time.

Deriving the Effect-Site Concentration

If plasma concentration is constant at $C_p$ over an interval, the differential equation can be solved by separating the deviation from equilibrium. Let $D(t)=C_e(t)-C_p$. Since $C_p$ is constant, $dD/dt=-k_{e0}D$, whose solution is $D(t)=D(0)e^{-k_{e0}t}$. Therefore:

C_e(t)=C_p+\bigl(C_e(0)-C_p\bigr)e^{-k_{e0}t}.

Starting from $C_e(0)=0$, the effect-site concentration rises toward a constant plasma concentration as $C_e(t)=C_p(1-e^{-k_{e0}t})$. It reaches half of that constant concentration after $\ln(2)/k_{e0}$. For a changing plasma profile, the effect compartment smooths and delays the input; it does not simply equal the current plasma concentration.

A common direct-effect model for a fractional inhibitory biomarker response is:

I(t)=I_{\max}\frac{C_e(t)}{IC_{50}+C_e(t)}, \qquad B(t)=B_0\bigl(1-I(t)\bigr),

where $I(t)$ is fractional inhibition, $I_{\max}$ is the maximum possible inhibition, $IC_{50}$ is the effect-site concentration producing half of $I_{\max}$, and $B_0$ is the baseline biomarker value. This direct link assumes the biomarker responds promptly to $C_e(t)$. If the biomarker itself turns over slowly, additional response dynamics may be needed; an effect compartment alone may not represent both delays.

Worked Example: Calculating the Effect-Site Concentration

Suppose a hypothetical drug produces a constant plasma concentration of $2.00\ \mathrm{mg/L}$ for one hour. The effect-site concentration starts at zero, and $k_{e0}=0.200\ \mathrm{h^{-1}}$. Calculate $C_e$ after one hour and the fractional inhibition if $I_{\max}=1$ and $IC_{50}=2.00\ \mathrm{mg/L}$.

Using the constant-input solution: $$ C_e(1)=2.00\bigl(1-e^{-0.200(1)}\bigr) =2.00(1-0.818731) =0.362539\ \mathrm{mg/L}. $$

Using the displayed concentration rounded to four significant figures, $C_e(1)=0.3625\ \mathrm{mg/L}$, the predicted inhibition is: $$ I(1)=\frac{0.3625}{2.00+0.3625} =\frac{0.3625}{2.3625} =0.1534. $$

Thus, the model predicts about 15.34% inhibition after one hour, even though the plasma concentration has already reached $2.00\ \mathrm{mg/L}$. The lag arises from the modeled delay in equilibration, not from a change in the concentration-response function.

Reading Hysteresis as a Timing Pattern

Hysteresis means that response depends on the exposure history as well as on the current plasma concentration. In an effect-compartment model, two observations with the same $C_p$ can have different $C_e$ values: during increasing exposure, $C_e$ tends to lag below $C_p$; after plasma concentration begins to fall, $C_e$ can remain above $C_p$. If response rises with $C_e$, the response can consequently be lower on the rising-exposure limb and higher on the falling-exposure limb at the same $C_p$.

The most useful interpretation is this same-exposure comparison, not a loop-direction label alone. The visual direction of a plotted loop depends on which variable is on each axis and whether the response is coded as inhibition, biomarker concentration, or another outcome. A loop is consistent with delayed equilibration, but is not proof of it: other changing processes can also create time-dependent exposure-response patterns.

Worked Example: A Hypothetical Hysteresis Loop

Consider a hypothetical concentration profile represented as constant over successive one-hour intervals. The concentrations over hours 0–1 through 5–6 are $2, 4, 6, 4, 2,$ and $0\ \mathrm{mg/L}$, respectively. Let $k_{e0}=0.200\ \mathrm{h^{-1}}$, $C_e(0)=0$, and use the direct inhibitory model with $I_{\max}=1$ and $IC_{50}=2.00\ \mathrm{mg/L}$. For each interval, use the exact constant-input recurrence with $q=e^{-0.200}=0.818731$: $$ C_{e,j+1}=C_{p,j}+\bigl(C_{e,j}-C_{p,j}\bigr)q. $$

For the first interval, $C_e(1)=2.00+(0-2.00)(0.818731)=0.3625\ \mathrm{mg/L}$. Continuing the same calculation gives the following endpoint concentrations:

Time (h)	Plasma concentration over preceding interval (mg/L)	Effect-site concentration at endpoint (mg/L)
1	2	0.3625
2	4	1.022
3	6	1.924
4	4	2.301
5	2	2.246
6	0	1.839

At hour 1, plasma concentration is $2.00\ \mathrm{mg/L}$, and inhibition is $0.3625/(2.00+0.3625)=0.1534$. At hour 5, plasma concentration is again $2.00\ \mathrm{mg/L}$, but inhibition is $2.246/(2.00+2.246)=0.5290$. Thus, at the same plasma concentration, the modeled response is about 15.34% on the rising limb and 52.90% on the falling limb. The effect-site concentrations differ because they summarize different exposure histories.

This constructed loop shows why a single exposure-response comparison that ignores time may be misleading. The values are generated under a specified hypothetical model; real data would need to support the parameter estimates and the assumptions behind that model.

When an Early Weak Response Is Not Evidence of No Efficacy

Suppose a biomarker is measured shortly after plasma concentration rises. A direct model using $C_p$ may predict a substantial immediate effect, while observed inhibition is still small. If the effect-site concentration has not equilibrated, the early observation can be compatible with a delayed effect. The relevant question is whether a plausible lag model predicts later response at the exposures actually observed, and whether later measurements agree with that prediction.

Worked Example: Distinguishing Delay From a Weak Maximum Effect

In a hypothetical constant-exposure experiment, plasma concentration is maintained at $5.00\ \mathrm{mg/L}$, starting with $C_e(0)=0$. Assume $k_{e0}=0.200\ \mathrm{h^{-1}}$, $I_{\max}=1$, and $IC_{50}=2.00\ \mathrm{mg/L}$. Calculate the predicted inhibition at one and four hours, and compare it with the eventual steady-state inhibition.

At one hour: $$ C_e(1)=5.00(1-e^{-0.200})=0.9063\ \mathrm{mg/L}, \qquad I(1)=\frac{5.00(1-e^{-0.200})}{2.00+5.00(1-e^{-0.200})}=0.3119. $$

At four hours: $$ C_e(4)=5.00(1-e^{-0.800})=2.753\ \mathrm{mg/L}, \qquad I(4)=\frac{2.753}{2.00+2.753}=0.5792. $$

At steady state, $C_e$ approaches $5.00\ \mathrm{mg/L}$, so: $$ I(\infty)=\frac{5.00}{2.00+5.00}=0.7143. $$

The model predicts 31.19% inhibition at one hour, increasing to 57.92% at four hours and approaching 71.43% at steady state. Therefore, a weak early response could be consistent with delayed pharmacology under these assumptions. These calculations do not establish that the model is correct: the observed time course, uncertainty, and competing explanations must be assessed.

Assumptions, Diagnostics, and Alternatives

Interpreting a fitted effect compartment requires more than a visible loop. The basic model assumes that measured $C_p(t)$ adequately represents the driving input, $k_{e0}$ is positive and constant over the analyzed period, and $C_e(t)$ follows first-order equilibration. The direct response model further assumes an appropriate concentration-response shape and a stable baseline or a separately modeled baseline. Observations should be timed densely enough around both increasing and decreasing exposure to distinguish a lag from a weak or noisy relationship.

Useful checks include overlaying observed biomarker data and model predictions against time, plasma concentration, and predicted effect-site concentration. Examine whether the model systematically misses early or late observations, whether the same-exposure rising/falling pattern is reproduced, and whether estimated parameters are stable when a few observations are removed or sampling periods are changed. A fitted $k_{e0}$ with very wide uncertainty, a value pushed toward an effectively instantaneous response, or a value so small that the observed study never approaches equilibrium may indicate that the data do not identify the lag well. Sparse sampling can make $k_{e0}$, $IC_{50}$, and $I_{\max}$ difficult to distinguish.

The effect compartment is not the only possible explanation for delay. A direct response model links the response to current $C_p$ and is suitable when no meaningful lag is evident. An indirect-response or turnover model represents delay through production or loss of the biomarker itself; it may be more appropriate when the biomarker has slow turnover. A time-varying baseline or another changing influence can also produce a loop. The model should be chosen based on the scientific mechanism and the pattern of data, not solely on which curve fits best.

A central limitation is that $C_e(t)$ is latent: it is inferred from the assumed dynamics, not observed directly. A loop can support the presence of temporal dependence, but by itself cannot identify its biological cause. Nor does a good fit prove efficacy. The model’s estimate of effect and its uncertainty must be considered alongside study design, biomarker validity, and the exposure range. Do not extrapolate confidently beyond the observed exposures or times when the model has little supporting data.

Common pitfall: “The biomarker barely changed at the first post-dose measurement, so the drug has no effect.” This conclusion treats plasma exposure as if it were the immediate driver and ignores the possibility of delayed equilibration. The correction is to inspect the exposure and response time courses, evaluate a justified lag model, and check whether later measurements support the predicted effect. Conversely, a fitted effect compartment should not be used to dismiss a weak response unless its assumptions and predictions are supported by the data.

Check Your Understanding

Answer each question, then compare your reasoning with the solutions.

If $k_{e0}$ decreases while the plasma profile stays the same, what happens to the effect-site equilibration half-time?
With $C_p=4.00\ \mathrm{mg/L}$, $k_{e0}=0.100\ \mathrm{h^{-1}}$, and $C_e(0)=0$, calculate $C_e$ after one hour under constant plasma concentration.
For an inhibitory response driven by $C_e$, why can inhibition differ at the same $C_p$ on rising and falling exposure limbs?
Name one observation pattern that would raise concern that $k_{e0}$ is poorly identified.

Solutions. (1) The half-time $\ln(2)/k_{e0}$ increases, so equilibration is slower. (2) $C_e(1)=4.00(1-e^{-0.100})=4.00(1-0.904837)=0.3807\ \mathrm{mg/L}$. (3) $C_e$ reflects prior exposure: it tends to lag below $C_p$ while exposure rises and can remain above $C_p$ after exposure falls, giving different responses at the same current plasma concentration. (4) Examples include very wide uncertainty in $k_{e0}$, strong sensitivity of its estimate to a few data points, or sampling that never captures the response changing during exposure increase or decline.