Deriving Overshoot From Persistent Compensation

Why a Compensatory State Can Outlast the Drug

A rebound endpoint need not rise because the drug concentration rises again. During treatment, a counter-regulatory response may develop and oppose the drug’s direct action. After withdrawal, the direct action and the compensatory response can both fade, but at different rates. If the direct action fades faster, compensation can temporarily dominate and push the endpoint beyond its untreated baseline.

The derivation below isolates that timing mechanism. It describes a single endpoint for which the drug directly lowers the measured value and the compensatory state raises it. The signs would be reversed for an endpoint with the opposite orientation; what matters is that the two components oppose one another.

Definition: Let $t=0$ be drug withdrawal, and let $Y_0$ be the constant untreated baseline of the endpoint. Write the post-withdrawal deviation from baseline as $\Delta Y(t)=Y(t)-Y_0$. The direct drug effect at withdrawal has magnitude $D>0$ endpoint units and is suppressive; the opposing compensatory effect has magnitude $H>0$ endpoint units. Suppose these components decay exponentially: $$ \Delta Y(t)=-D e^{-k_e t}+H e^{-k_a t}, \qquad t\geq 0, $$ where $k_e>0$ is the decay rate of the direct effect and $k_a>0$ is the decay rate of compensation. Both rates have units of inverse time. Overshoot relative to the untreated baseline means $\Delta Y(t)>0$.

This is a simplified post-withdrawal model, not a general description of every rebound trajectory. It assumes that the direct effect follows a first-order exponential decline after withdrawal, that the compensatory state also declines exponentially, and that their effects add linearly on the endpoint scale. The baseline remains constant, there is no further drug input, and $D$ and $H$ are the magnitudes of the two effects at withdrawal. The rate $k_e$ may correspond to drug elimination only when the direct pharmacodynamic effect tracks concentration proportionally over the range considered.

Deriving the Crossing Condition

Set the deviation equal to zero to find when the endpoint crosses its untreated baseline. Rearranging the two opposing terms gives

H e^{-k_a t}=D e^{-k_e t} \quad\Longleftrightarrow\quad e^{(k_e-k_a)t}=\frac{D}{H}.

If $k_e\ne k_a$ and $D,H>0$, the algebraic crossing time is therefore

t_{\mathrm{cross}}=\frac{\ln(D/H)}{k_e-k_a}.

A positive crossing time from an initially suppressed endpoint requires $D>H$ and $k_e>k_a$. The first condition means the endpoint is below baseline at withdrawal, since $\Delta Y(0)=H-D<0$. The second means the direct drug effect decays faster than compensation. Together they make the crossing time positive. To see that the crossing is from below to above, factor the deviation as

\Delta Y(t)=e^{-k_a t}\left[H-D e^{-(k_e-k_a)t}\right].

When $k_e>k_a$, the exponential inside the brackets decreases from 1 toward 0. If $D>H$, the bracket starts negative and eventually becomes positive, with exactly one zero. The outside factor is always positive, so it does not change the sign. Thus compensation eventually outweighs the direct effect, and the endpoint crosses baseline once.

The condition $D>H$ is specific to a later crossing from below, not to overshoot in every sense. If $H>D$, the endpoint is already above baseline at withdrawal. If $H=D$ and $k_e>k_a$, it is at baseline at withdrawal and above it for every $t>0$. In both cases, $\Delta Y(t)>0$ can occur without a later crossing from below.

Worked Example: A Later Crossing From Below

Consider hypothetical endpoint effects at withdrawal of $D=2.720$ units and $H=1.000$ unit. Let $k_e=0.8000$ per day and $k_a=0.2000$ per day. The endpoint starts below baseline because $\Delta Y(0)=1.000-2.720=-1.720$ units, and the direct effect decays faster because $0.8000>0.2000$.

The crossing time is

t_{\mathrm{cross}} =\frac{\ln(2.720/1.000)}{0.8000-0.2000} =\frac{1.0006}{0.6000} \approx 1.668\ \text{days}.

Under the model, the endpoint is below baseline before approximately 1.668 days and above baseline afterward. For example, at 3.978 days the predicted deviation is

\begin{aligned} \Delta Y(3.978) &=-(2.720)e^{-(0.8000)(3.978)} +(1.000)e^{-(0.2000)(3.978)}\\ &\approx 0.3385\ \text{endpoint units}. \end{aligned}

This is a positive deviation, as expected after the crossing. The calculation illustrates a model-based consequence of the stated hypothetical parameters; it does not by itself establish that compensation caused an observed rebound in any real setting.

When Is the Rebound Largest?

A positive deviation does not grow indefinitely. With $k_a>0$, the compensatory component also decays, so after reaching a maximum the predicted deviation returns toward zero. Differentiate the model:

\frac{d\Delta Y}{dt}=Dk_e e^{-k_e t}-Hk_a e^{-k_a t}.

Setting the derivative to zero and rearranging yields a candidate time for the maximum:

t_{\mathrm{peak}} =\frac{\ln(Dk_e/(Hk_a))}{k_e-k_a}.

For the later-crossing case $D>H>0$ and $k_e>k_a>0$, this time is positive: $Dk_e/(Hk_a)>1$. The derivative is initially positive because $Dk_e>Hk_a$, and its sign changes once at $t_{\mathrm{peak}}$, from positive to negative. Consequently, this is the unique maximum. It occurs after the baseline crossing because

t_{\mathrm{peak}}-t_{\mathrm{cross}} =\frac{\ln(k_e/k_a)}{k_e-k_a}>0.

The height of the maximum can also be written in a useful form. At the peak, the derivative equation implies $D e^{-k_e t_{\mathrm{peak}}}=(Hk_a/k_e)e^{-k_a t_{\mathrm{peak}}}$. Substituting into $\Delta Y$ gives

\Delta Y(t_{\mathrm{peak}}) =H e^{-k_a t_{\mathrm{peak}}}\left(1-\frac{k_a}{k_e}\right)>0.

This confirms that the peak lies above baseline under the later-crossing conditions. As $t$ tends to infinity, both exponential terms tend to zero, so the deviation eventually returns toward baseline.

Worked Example: Calculating the Peak Time

Use hypothetical withdrawal magnitudes $D=6.000$ endpoint units and $H=2.000$ endpoint units, with $k_e=0.7000$ per day and $k_a=0.2000$ per day. Since $D>H$ and $k_e>k_a$, the endpoint starts below baseline, crosses it, and has a later positive peak.

The peak time is

\begin{aligned} t_{\mathrm{peak}} &=\frac{\ln((6.000)(0.7000)/((2.000)(0.2000)))}{0.7000-0.2000}\\ &=\frac{\ln(10.5)}{0.5000} \approx 4.703\ \text{days}. \end{aligned}

The baseline crossing occurs earlier:

t_{\mathrm{cross}} =\frac{\ln(6.000/2.000)}{0.5000} =\frac{\ln(3.000)}{0.5000} \approx 2.197\ \text{days}.

The peak follows the crossing, consistent with $t_{\mathrm{peak}}-t_{\mathrm{cross}}=\ln(0.7000/0.2000)/0.5000>0$. The model therefore predicts a period above baseline after about 2.197 days, a maximum at about 4.703 days, and eventual return toward baseline.

Edge Cases and Model Checks

Worked Example: Equal Decay Rates Do Not Produce a Later Crossing

Suppose the hypothetical initial magnitudes are $D=2.000$ and $H=1.000$ endpoint units, but both components decay at $0.3000$ per day. Then

\Delta Y(t)=-2.000e^{-0.3000t}+1.000e^{-0.3000t} =-1.000e^{-0.3000t}<0 \qquad (t\geq 0).

At 3.000 days, $\Delta Y(3.000)=-e^{-0.9000}\approx-0.4066$ endpoint units. The relative balance never changes because both terms share the same decay factor. There is no later overshoot in this model, even though a compensatory effect is present.

A fitted or proposed trajectory should be checked against the model’s qualitative implications, not just its calculated crossing time. In the later-crossing case, the deviation starts negative, crosses once, reaches one positive maximum, and approaches zero from above. A trajectory with repeated crossings, persistent elevation far beyond the decay timescale, or an endpoint that continues moving away from baseline would not follow these two decaying exponentials without additional inputs or processes.

The equal-rate example is one important boundary case. If $k_e<k_a$ and $D>H$, compensation fades faster than the direct effect; the deviation stays negative rather than crossing above baseline. For example, since $e^{-k_a t}<e^{-k_e t}$ when $k_a>k_e$ and $t>0$, $H<D$ implies $H e^{-k_a t}<D e^{-k_e t}$. The net deviation remains negative. More generally, if the endpoint is already above baseline at withdrawal, its later direction depends on the relative magnitudes and rates, so the conditions for crossing from below should not be applied without checking the initial state.

Interpretation, Alternatives, and Limitations

Persistent compensation is one explanation for a post-withdrawal rise beyond baseline, but the observed shape alone does not identify its cause. A return from a drug-suppressed value to baseline is different from overshoot: the former has $\Delta Y(t)<0$ while approaching zero, whereas overshoot requires $\Delta Y(t)>0$ at some time. A changing untreated disease trajectory can also alter the reference level, and continued exposure or another process can produce a similar observed curve. The earlier material on rebound versus return of the untreated disease addresses that distinction; here the baseline is explicitly held constant to isolate the compensation mechanism.

Compared with a model in which the endpoint simply returns to baseline as drug effect disappears, the present model adds a positive, persistent state. Compared with a model in which only the direct drug effect decays, it permits an initial counter-regulatory response to outlast that effect. The two-component form is analytically convenient and makes the crossing condition explicit, but it is not a substitute for a mechanistic model when the endpoint response is nonlinear, compensation has a delay, the baseline changes, or either state has more complex dynamics.

In practice, the assumed exponential components and fixed baseline should be treated as testable approximations. Check whether post-withdrawal measurements are consistent with the predicted initial direction, one crossing, and eventual decay; examine whether the time scale is plausible for both components; and compare the trajectory with appropriate information about the untreated baseline. Sparse measurements may miss the crossing or peak, while noisy measurements may make a transient deviation appear more certain than it is. Parameter values obtained by fitting this model are not automatically evidence that the compensatory mechanism is uniquely responsible.

Common pitfall: “The drug concentration is falling, so any later rise above baseline must be caused by compensation.” That conclusion does not follow from the trajectory alone. In this model, a later crossing from below requires $D>H>0$ and $k_e>k_a>0$; other mechanisms or a changing baseline may also explain an observed rise. If overshoot means simply $\Delta Y(t)>0$, these conditions are not necessary when the endpoint is already at or above baseline at withdrawal.

Check Your Understanding

Work through the conditions and calculations before reading the solutions.

For $D=3.000$ endpoint units, $H=1.200$ endpoint units, $k_e=0.9000$ per day, and $k_a=0.3000$ per day, calculate the later baseline-crossing time.
If $H=D$ and $k_e>k_a>0$, is there a later crossing from below, and when is the endpoint above baseline?
For $D>H>0$, what happens to the sign of $\Delta Y(t)$ when $k_e=k_a$?
For $D=2.400$, $H=1.000$, $k_e=0.8000$ per day, and $k_a=0.2000$ per day, calculate the crossing time and the peak time.
Why does observing an endpoint above baseline after withdrawal not, by itself, prove persistent compensation caused the overshoot?

Solutions. (1) The conditions for a later crossing from below hold, and $t_{\mathrm{cross}}=\ln(3.000/1.200)/(0.9000-0.3000)=\ln(2.500)/0.6000\approx1.527$ days. (2) There is no crossing from below because $\Delta Y(0)=0$; for every $t>0$, the faster-decaying direct effect is smaller in magnitude than compensation, so $\Delta Y(t)>0$. (3) The deviation is $(H-D)e^{-k_e t}<0$ for all $t\geq0$, so it never crosses baseline. (4) $t_{\mathrm{cross}}=\ln(2.400)/(0.6000)\approx1.459$ days, and $t_{\mathrm{peak}}=\ln((2.400)(0.8000)/(1.000)(0.2000))/(0.6000)=\ln(9.600)/0.6000\approx3.770$ days. (5) A changing untreated baseline or another process can produce a similar trajectory; the observed rise alone does not establish the mechanism.