Convolution Solutions for Drug-Driven Production

Why Convolution Describes a Delayed Response

A drug may change the rate at which a response is produced, while the response itself is removed continuously. In that situation, the response at a given time reflects not only the current drug concentration but also the production stimulated or inhibited earlier. Past production still contributes, but its contribution is reduced by loss over the time since it occurred. A convolution expresses this weighted accumulation directly.

This tutorial considers a response \(R(t)\) with constant first-order loss rate \(k_{\mathrm{out}}>0\). Drug concentration \(C(t)\) may vary over time and affect production through a function \(q(C(t))\). The production effect is assumed to act immediately on the production rate; the delay in \(R(t)\) arises from accumulation and loss, not from a separate effect compartment.

The integration variable \(s\) denotes the time at which a contribution to production occurs. The factor \(e^{-k_{\mathrm{out}}(t-s)}\) is the fraction of that contribution remaining at time \(t\), after loss over the elapsed time \(t-s\). Thus the convolution does not simply add all past production equally: recent production has had less time to be lost than production that occurred earlier.

Deriving the Convolution and Isolating the Drug Effect

The convolution follows from the integrating-factor solution for a linear differential equation. Multiply both sides of the model by \(e^{k_{\mathrm{out}}t}\). The product rule gives $$ \frac{d}{dt}\left[e^{k_{\mathrm{out}}t}R(t)\right] =e^{k_{\mathrm{out}}t}\left[k_{\mathrm{in},0}+q(C(t))\right]. $$

Integrating from 0 to \(t\), then multiplying by \(e^{-k_{\mathrm{out}}t}\), yields $$ e^{k_{\mathrm{out}}t}R(t)-R(0) =\int_0^t e^{k_{\mathrm{out}}s}\left[k_{\mathrm{in},0}+q(C(s))\right]ds, $$ and hence the stated convolution. This derivation requires constant \(k_{\mathrm{out}}\), a well-defined production function over the interval, and a response trajectory satisfying the differential equation. It does not require that \(q\) be linear in concentration.

With no drug effect, the baseline steady state is \(R_0=k_{\mathrm{in},0}/k_{\mathrm{out}}\). If the response starts at that baseline, the constant baseline production and baseline loss balance at every time, so the solution simplifies to $$ R(t)=R_0+\int_0^t e^{-k_{\mathrm{out}}(t-s)}q(C(s))\,ds. $$ If the initial response is not at baseline, the general expression is $$ R(t)=R_0+\left[R(0)-R_0\right]e^{-k_{\mathrm{out}}t} +\int_0^t e^{-k_{\mathrm{out}}(t-s)}q(C(s))\,ds. $$ The first term after \(R_0\) describes relaxation of the initial deviation; the integral describes the accumulated drug-driven change. A negative \(q\), as in production inhibition, contributes negatively to the response deviation.

Exponential Concentration Decline With Linear Production

Suppose concentration after a dose declines as \(C(t)=C_0e^{-k_et}\), where \(C_0\) is the concentration at time zero and \(k_e>0\) is the concentration decline rate. For a linear production increment \(q(C)=\alpha C\), the drug-driven response change from baseline is $$ \Delta R(t)=\alpha C_0\int_0^t e^{-k_{\mathrm{out}}(t-s)}e^{-k_es}\,ds. $$ When \(k_{\mathrm{out}}\ne k_e\), evaluating this integral gives $$ \Delta R(t)=\alpha C_0 \frac{e^{-k_et}-e^{-k_{\mathrm{out}}t}}{k_{\mathrm{out}}-k_e}. $$ The concentration decline rate and response loss rate enter symmetrically in the exponential terms, but they describe different processes: \(k_e\) governs the drug concentration, whereas \(k_{\mathrm{out}}\) governs response removal.

Saturable Production Requires Integrating the Actual Effect

For saturable production, a standard form is \(q(C)=E_{\max}C/(EC_{50}+C)\). Here \(E_{\max}\) is the maximum production increment and \(EC_{50}\) is the concentration giving half that increment. With exponential concentration decline, the convolution becomes $$ \Delta R(t)=\int_0^t e^{-k_{\mathrm{out}}(t-s)} \frac{E_{\max}C_0e^{-k_es}}{EC_{50}+C_0e^{-k_es}}\,ds. $$ In general this integral does not reduce to the same simple expression as the linear-effect case. It can be evaluated numerically, provided the integrand is computed using the concentration at each integration time \(s\), not just the concentration at the endpoint.

Equal Rates, Checks, and Failure Modes

The formula for linear production appears to have a zero denominator when \(k_{\mathrm{out}}=k_e\), but the convolution itself remains finite. When the rates are equal, the exponential factors in the integrand combine: $$ e^{-k_{\mathrm{out}}(t-s)}e^{-k_es} =e^{-k_et}, $$ which is constant with respect to \(s\). Therefore $$ \Delta R(t)=\alpha C_0t e^{-k_et} \qquad \text{when }k_{\mathrm{out}}=k_e. $$ This result also follows by taking the limit of the unequal-rate expression. It is important in computation: near-equal rates can make direct subtraction in the unequal-rate formula numerically unstable, so the limiting form or a stable numerical evaluation of the integral is preferable.

Several checks help identify an incorrect calculation or an unsuitable model. First, the exponent must be the loss rate multiplied by elapsed time since production, \(k_{\mathrm{out}}(t-s)\). Using \(k_{\mathrm{out}}s\) instead weights early and late production in the wrong direction. Second, the units of the integral must be response units: the kernel is dimensionless, while production rate times \(ds\) has response units. Third, if \(q(C)=0\) and \(R(0)=R_0\), the response must remain at baseline. If positive production stimulation is present and \(R(0)=R_0\), the integral is nonnegative.

A further check is the initial slope. From the differential equation, if \(R(0)=R_0\), then \(dR/dt\) at time zero equals \(q(C(0))\), because baseline production and loss cancel. A computed trajectory that initially moves in the opposite direction from the sign of \(q(C(0))\) suggests a sign, kernel, or indexing error. At late times after the drug effect has disappeared, the response deviation should decay with the constant loss rate.

Choosing and Interpreting a Calculation Method

For linear production and exponential concentration decline, the closed-form expression is efficient and makes the roles of the two rates transparent. For nonlinear production, the convolution is still exact as a mathematical representation, but its numerical evaluation may be more practical than seeking a closed form. Simpson’s rule is suitable for a smooth integrand on a finite interval; checking the estimate with a finer grid is a useful numerical diagnostic. A direct numerical solution of the differential equation is another option and can be convenient when the production function is complicated.

The earlier material on integrating-factor solutions and piecewise-constant exposure provides related ways to calculate trajectories. Convolution is particularly informative when the goal is to see how each past production increment contributes to the current response. Piecewise-constant methods approximate changing inputs by intervals, while direct differential-equation solvers propagate the state numerically. These approaches should agree to numerical precision when they use the same model and sufficiently accurate calculations.

The convolution does not establish that the chosen production mechanism is biologically correct. It assumes that loss is first-order with a constant rate, that the response is adequately described by one state variable, and that the specified production effect is the relevant drug action. If loss varies with time, this constant-loss kernel does not apply. If a response delay beyond turnover is needed, immediate production stimulation may also be inadequate. Model checking should therefore include the concentration and response time courses, the plausibility of the assumed production function, and residual patterns—not just agreement at one calculated time.