Empirical Time Functions for Clearance

Why Use an Empirical Time Function?

Clearance is often treated as constant during a dosing interval or study. Yet concentration profiles may show a systematic change over time that a constant-clearance model does not capture. An empirical time function provides a compact way to represent that change when the data support a time trend but do not justify a more detailed mechanism. It describes how clearance varies; by itself, it does not explain why it varies.

The function must be connected to the pharmacokinetic model, not applied as an arbitrary adjustment to observed concentrations. In a one-compartment model with amount \(A(t)\), volume \(V\), concentration \(C(t)=A(t)/V\), and time-varying clearance \(CL(t)\), elimination is represented by \(dA/dt=-CL(t)C(t)\). This tutorial focuses on specifying and checking \(CL(t)\), rather than on identifying a biological mechanism for its change.

Definition: An empirical time function for clearance specifies clearance as a deterministic function of time, such as an exponential, linear, or piecewise function. A reference time \(t_{\mathrm{ref}}\) is the time at which the function is anchored; the clearance at that time is \(CL_{\mathrm{ref}}=CL(t_{\mathrm{ref}})\). Time units and clearance units must be specified so that every parameter has interpretable units.

Time should be defined in the context of the model: it might be time since the first dose, time since treatment began, or time since a dose. The choice matters because shifting the time origin changes the numerical value of a reference-time parameter. State it explicitly and use the same definition when fitting, predicting, and interpreting the model.

Exponential and Linear Functions

An exponential function is useful when a proportional rate of change is a reasonable empirical description. One parameterization is:

CL(t)=CL_{\mathrm{ref}}\exp\!\left[k(t-t_{\mathrm{ref}})\right].

Here \(CL_{\mathrm{ref}}>0\), and \(k\) has units of inverse time. If \(k>0\), clearance increases with time; if \(k<0\), it decreases. The function is positive at every finite time when \(CL_{\mathrm{ref}}>0\), so it avoids the negative predictions that an unconstrained straight line can produce. The exponential form assumes a constant proportional rate of change: its derivative satisfies \(dCL(t)/dt=kCL(t)\). That shape may be convenient, but it is not automatically a biological law.

Worked Example: Evaluating Exponential Clearance

Suppose a hypothetical model uses \(CL_{\mathrm{ref}}=3.2\) L/h at \(t_{\mathrm{ref}}=4\) h, with \(k=0.025\) h\(^{-1}\). Time is measured in hours since treatment began. Calculate clearance at 12 h and check its direction of change.

Substitute the values into the specified function:

CL(12)=3.2\exp[0.025(12-4)] =3.2\exp(0.2) \approx 3.2(1.2214) \approx 3.908\ \mathrm{L/h}.

Because \(k\) is positive, clearance increases over time. The 3.908 L/h value is the model prediction at 12 h, not an estimate of a biological cause. The parameter units are consistent: \(0.025\ \mathrm{h}^{-1}\times 8\ \mathrm{h}=0.2\), a dimensionless exponent.

A linear function is a direct description of an additive trend:

CL(t)=CL_{\mathrm{ref}}+s(t-t_{\mathrm{ref}}),

where the slope \(s\) has units of clearance per unit time, such as L/h². The reference time makes \(CL_{\mathrm{ref}}\) the clearance at a meaningful point rather than an extrapolated intercept at time zero. Unlike the exponential function, the linear form requires an explicit positivity check over the time interval in which it will be used.

Worked Example: Checking a Linear Function’s Range

Consider a hypothetical linear function with \(CL_{\mathrm{ref}}=5.0\) L/h at \(t_{\mathrm{ref}}=6\) h and slope \(s=0.18\) L/h². The model is intended for times from 0 to 24 h. Evaluate clearance at both endpoints and check the constraint.

At the beginning of the interval:

CL(0)=5.0+0.18(0-6)=5.0-1.08=3.92\ \mathrm{L/h}.

At the end:

CL(24)=5.0+0.18(24-6)=5.0+3.24=8.24\ \mathrm{L/h}.

A linear function reaches its minimum and maximum over a closed interval at an endpoint (or is constant). Both endpoint predictions are positive, so this function remains positive throughout 0–24 h. This check is specific to that interval; extrapolating the line far enough backward would eventually predict negative clearance.

Constraint for a linear function: On a modeled interval \([t_{\min},t_{\max}]\), require \(CL_{\mathrm{ref}}+s(t-t_{\mathrm{ref}})>0\) at both endpoints. This guarantees positivity at every time in the interval because the function is linear. If the intended model extends over a broader interval, check that broader interval instead.

Piecewise Functions and Their Transitions

A piecewise function assigns different expressions to different time segments. It can describe a change in trend, or an abrupt transition, without requiring a single curve to fit the entire observation window. The knots, or boundaries between segments, should be stated along with the function and the time scale.

For example, a piecewise-constant clearance function can represent different levels before and after a prespecified time \(\tau\):

CL(t)= \begin{cases} CL_1, & t<\tau,\\ CL_2, & t\geq \tau. \end{cases}

This specification has a jump at \(\tau\) if \(CL_1\ne CL_2\). A jump may be a useful approximation when the modeled change is rapid relative to the sampling schedule, but it should not be introduced merely because it improves fit. If an abrupt change is not defensible, a continuous piecewise-linear function can instead join segment-specific slopes at a knot. Continuity requires that the adjacent segment formulas give the same clearance at that knot. Positivity must still be checked on every segment.

Worked Example: Amount Under Piecewise Clearance

Suppose a hypothetical one-compartment model has constant volume \(V=40\) L and initial amount \(A(0)=100\) mg. Clearance is 4 L/h from 0 to 5 h, then 6 L/h from 5 to 8 h. Treat the change at 5 h as an abrupt jump, and calculate the amount at 8 h.

For constant \(V\), concentration is \(A(t)/V\), so the amount equation is \(dA/dt=-CL(t)A(t)/V\). The cumulative elimination exponent over the two segments is:

\frac{1}{V}\int_0^8 CL(t)\,dt =\frac{(4\ \mathrm{L/h})(5\ \mathrm{h})+(6\ \mathrm{L/h})(3\ \mathrm{h})}{40\ \mathrm{L}} =\frac{38\ \mathrm{L}}{40\ \mathrm{L}} =0.95.

Thus:

A(8)=100\exp(-0.95)\ \mathrm{mg} \approx 100(0.38674)\ \mathrm{mg} \approx 38.67\ \mathrm{mg}.

The corresponding concentration is \(A(8)/V=38.67/40\approx0.9668\) mg/L. The calculation assumes the stated clearance values apply throughout their respective segments, volume is constant, and there is no input of drug during this interval. It illustrates the model’s consequence; it does not establish that a real clearance change was abrupt.

Justification: How Clearance Shapes the Amount Curve

For a one-compartment model with constant volume and no drug input over the interval, substituting \(C(t)=A(t)/V\) into the elimination equation gives \(dA(t)/dt=-CL(t)A(t)/V\). Separating variables and integrating from \(t_0\) to \(t\) yields the exact relationship:

\frac{dA(u)}{A(u)}=-\frac{CL(u)}{V}\,du, \qquad \ln\!\left(\frac{A(t)}{A(t_0)}\right) =-\frac{1}{V}\int_{t_0}^{t}CL(u)\,du,

A(t)=A(t_0)\exp\!\left[-\frac{1}{V}\int_{t_0}^{t}CL(u)\,du\right].

The integrated clearance, not just clearance at the final observation, determines the amount decline over the interval. This is why a time function must be evaluated consistently across time. The derivation assumes constant \(V\), positive amount, and elimination as the only process affecting amount in the interval. With input, additional terms enter the differential equation; with changing volume, the concentration–amount relationship also changes.

Choosing a Function and Checking Its Fit

These functions differ in shape and constraints. The right choice depends on the time pattern supported by the data, the observation window, and the intended interpretation.

Function	Shape and constraint	Useful when
Exponential	Positive for \(CL_{\mathrm{ref}}>0\); constant proportional rate of change	A smooth multiplicative trend is a parsimonious empirical description
Linear	Constant additive slope; positivity must be checked over the modeled interval	A smooth additive trend is adequate over a limited time range
Piecewise	Segment-specific levels or slopes; check positivity and continuity as intended	The data support distinct time segments or a specified transition
Constant clearance	No time trend	A time-dependent function is not justified by the data

Before interpreting the fitted function, inspect whether it describes the data across time rather than merely improving a single overall fit statistic. Useful checks include plotting observed concentrations and model predictions against time, examining residuals for remaining time patterns, and checking whether parameter estimates and predictions are sensitive to the chosen time origin, knot locations, or observation window. Clearance itself may not be observed directly, so concentration residuals are evidence about the whole pharmacokinetic model, not a direct measurement of the clearance function.

Warning signs include implausible clearance magnitudes, a linear function approaching zero at an interval boundary, a large fitted jump at a knot with little data nearby, and substantial changes in fitted parameters when the reference time is shifted or a few observations are removed. Sparse sampling around a transition can make its timing and magnitude difficult to distinguish. A flexible function can also absorb misspecification elsewhere in the pharmacokinetic model, so a better fit alone is not proof that clearance truly changes in the proposed way.

A constant-clearance model is the simpler alternative and should remain under consideration if it adequately describes the data. Mechanistic induction or turnover models are alternatives when the scientific question concerns a biological process and the data can support its parameters. The earlier material on enzyme induction develops such mechanistic descriptions. An empirical time function may be useful when a time trend needs to be represented without committing to that mechanism, but its coefficients should not be interpreted as induction rates or causal effects unless the model explicitly supports that interpretation.

The function also has limits beyond its fitted window. An exponential trend can grow or decay without bound under extrapolation; a linear trend can eventually become negative; and a piecewise function may behave abruptly at a knot. These are mathematical consequences of the chosen form, not necessarily plausible long-term physiology. State the interval over which predictions are intended and avoid treating extrapolated values as reliable without additional justification.

Common Mistake

A frequent error is to fit a linear clearance trend over the observed times, see positive fitted values at most observations, and conclude that the model is physically valid. That is insufficient: clearance must remain positive throughout the entire interval where the model will be used, including the endpoints. Another error is to interpret an exponential slope \(k\) as a constant additive increase. In fact, \(k\) is a proportional rate: \(dCL(t)/dt=kCL(t)\), so the absolute change depends on the current clearance. Check units, reference time, full-range positivity, and the intended meaning of the parameter before interpreting the fitted curve.

Key takeaway: Empirical clearance functions are compact descriptions of time trends, not mechanisms. Define the time origin and reference point, use units-consistent parameters, enforce positive clearance over the prediction interval, and assess whether the chosen shape is supported by the data.

Check Your Understanding

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

For an exponential clearance function, what are the units of \(k\), and what does a negative \(k\) imply?
A linear function has \(CL_{\mathrm{ref}}=4.0\) L/h at \(t_{\mathrm{ref}}=2\) h and slope \(s=-0.30\) L/h². Is it positive throughout 0–10 h?
What does a piecewise-constant function with unequal values on either side of a knot imply at the knot?
Why does the amount decline under time-varying clearance depend on an integral over time?

Solutions. (1) \(k\) has units of inverse time, so its product with elapsed time is dimensionless. A negative \(k\) means clearance decreases proportionally over time. (2) The endpoint values are \(CL(0)=4.0+(-0.30)(-2)=4.6\) L/h and \(CL(10)=4.0+(-0.30)(8)=1.6\) L/h. Both are positive, so the linear function is positive throughout the interval. (3) It implies an abrupt jump in the modeled clearance at the knot. (4) The one-compartment amount equation gives \(A(t)=A(t_0)\exp[-\int_{t_0}^{t}CL(u)\,du/V]\) when volume is constant and there is no input; therefore the accumulated clearance over the interval determines the amount ratio.