Terminal-Phase Samples and Elimination Rate

Why the End of the Concentration Profile Matters

Concentrations measured near the beginning of a profile can reflect several processes at once, including absorption and distribution. Later, once those processes no longer substantially shape the observed decline, concentrations may follow a simpler, approximately log-linear pattern. Samples in that part of the profile are called terminal-phase samples. Their timing matters: measurements taken too early can reflect a faster initial decline, while measurements taken too late may fall below the assay’s quantifiable range.

A useful terminal-phase slope summarizes how quickly concentrations decline during this late phase. Its corresponding half-life describes the time required for concentration to halve if the terminal pattern continues. These quantities help characterize the observed concentration-time profile, but they do not automatically identify the underlying elimination process. That distinction is especially important when distribution or absorption still affects the late observations.

Definition: If late concentrations are adequately described by \(C(t)=A_z e^{-\lambda_z t}\), where \(C(t)\) is concentration at time \(t\), \(A_z\) is a positive constant, and \(\lambda_z>0\), then \(\lambda_z\) is the terminal rate constant (the magnitude of the terminal log-linear slope). Its unit is inverse time, such as \(\mathrm{h}^{-1}\). The corresponding terminal half-life is \(t_{1/2,z}=\ln(2)/\lambda_z\), with units of time.

In a one-compartment model with first-order elimination and no continuing absorption, the terminal rate constant can equal the elimination rate constant. In a multicompartment model, the terminal slope can instead reflect the slowest disposition phase. For that reason, \(\lambda_z\) is a careful description of the observed terminal decline; it should not be casually equated with an elimination rate constant in every model.

From Late Concentrations to a Slope

Taking natural logarithms of the terminal model gives a straight-line relationship:

\ln C(t)=\ln A_z-\lambda_z t

The intercept is \(\ln A_z\), and the slope is \(-\lambda_z\). Thus, a negative fitted slope \(b\) corresponds to the positive terminal rate constant \(\lambda_z=-b\). The logarithm is applied to concentration values, not to time. If base-10 logarithms are used instead, the fitted slope must be multiplied by \(-2.303\) to obtain \(\lambda_z\).

The half-life formula follows directly from the definition of halving. Under the terminal model, advancing time by \(t_{1/2,z}\) multiplies concentration by \(e^{-\lambda_z t_{1/2,z}}\). Requiring that factor to equal one-half yields:

e^{-\lambda_z t_{1/2,z}}=\frac{1}{2} \quad\Longrightarrow\quad -\lambda_z t_{1/2,z}=\ln\left(\frac{1}{2}\right)=-\ln(2) \quad\Longrightarrow\quad t_{1/2,z}=\frac{\ln(2)}{\lambda_z}

This derivation assumes a positive, constant terminal rate constant over the interval being described. The result does not say that all concentrations, including those in earlier phases, must halve at that same rate.

Worked Example: Half-Life from Two Terminal Samples

Suppose two hypothetical samples are judged to be in the terminal phase: \(C(12)=5.00\ \mathrm{mg/L}\) and \(C(24)=1.25\ \mathrm{mg/L}\), with times in hours. Assume the terminal model is appropriate over this interval and the concentrations are positive and quantifiable.

The slope between the two log concentrations is:

b=\frac{\ln(1.25)-\ln(5.00)}{24-12} =\frac{\ln(1.25/5.00)}{12} =\frac{\ln(0.25)}{12} =\frac{-1.38629}{12} =-0.1155\ \mathrm{h}^{-1}

Therefore, \(\lambda_z=-b=0.1155\ \mathrm{h}^{-1}\). The terminal half-life is:

t_{1/2,z}=\frac{0.693147}{0.1155}=6.001\ \mathrm{h}

The sample concentrations have fallen to one-quarter over 12 hours, which is two halvings. A half-life of about 6 hours is consistent with that observation. With only two samples, however, the line is determined exactly and there is no additional point with which to check whether the decline was actually log-linear.

Using Several Terminal Samples

With three or more appropriately timed samples, a straight line can be fitted to log concentration against time. For ordinary least squares regression, the fitted slope is the sum of the products of each centered time and centered log concentration, divided by the sum of squared centered times. Centering means subtracting the corresponding sample mean. A negative slope gives the estimate \(\hat{\lambda}_z=-b\); the estimated half-life is \(\ln(2)/\hat{\lambda}_z\).

This calculation uses a specific working model: the selected observations come from one approximately log-linear terminal phase, the times are correctly recorded, and the measured concentrations are suitable for analysis on the log scale. Ordinary least squares also treats the log-concentration deviations as having mean zero and constant variance, with no systematic dependence among residuals for the usual regression-based uncertainty calculations. These assumptions are not guaranteed just because a regression line can be drawn.

Worked Example: Regression on Four Terminal Samples

Consider hypothetical terminal samples at 4, 8, 12, and 16 hours. Their natural log concentrations are 2.015732, 1.575732, 1.205732, and 0.785732, respectively. The corresponding concentrations are approximately 7.506, 4.834, 3.339, and 2.194 \(\mathrm{mg/L}\). Assume, for this calculation, that the four observations belong to the same terminal log-linear phase and that a straight-line fit is reasonable.

The mean time is 10 hours, and the mean log concentration is 1.395732. Therefore, the centered times are \(-6,-2,2,6\) hours, and the centered log concentrations are \(0.620000,0.180000,-0.190000,-0.610000\). The regression slope is:

b= \frac{(-6)(0.620000)+(-2)(0.180000)+(2)(-0.190000)+(6)(-0.610000)} {(-6)^2+(-2)^2+(2)^2+(6)^2} = \frac{-3.720-0.360-0.380-3.660}{36+4+4+36} = \frac{-8.120}{80} =-0.1015\ \mathrm{h}^{-1}

Thus, the estimated terminal rate constant is \(\hat{\lambda}_z=0.1015\ \mathrm{h}^{-1}\), and the estimated terminal half-life is:

\hat{t}_{1/2,z}=\frac{0.693147}{0.1015}=6.829\ \mathrm{h}

The arithmetic supports an estimated terminal half-life of about 6.83 hours for this hypothetical profile. It does not by itself establish that the selected points are genuinely terminal; that judgment depends on the study context and on examining the concentration-time pattern.

Why a Short Sampling Window Can Mislead

A concentration profile need not be a single exponential from the first observed time onward. For example, an illustrative two-component profile can be written as \(C(t)=8e^{-0.30t}+2e^{-0.08t}\ \mathrm{mg/L}\), with \(t\) in hours. The faster component contributes substantially early on. At sufficiently late times, it becomes small relative to the slower component, and the slope approaches \(-0.08\ \mathrm{h}^{-1}\). If sampling stops early, an estimated slope can therefore describe a mixture of phases rather than the terminal decline.

Worked Example: Comparing an Early Window with a Late Window

Use the hypothetical two-component profile \(C(t)=8e^{-0.30t}+2e^{-0.08t}\ \mathrm{mg/L}\). This example assumes the stated model generates the concentrations without measurement error; it illustrates how time-window choice changes the slope calculated from the observations.

At 2 and 4 hours, the model gives:

C(2)=8e^{-0.60}+2e^{-0.16}\approx6.0948\ \mathrm{mg/L}, \qquad C(4)=8e^{-1.20}+2e^{-0.32}\approx3.8619\ \mathrm{mg/L}

The two-point slope over this early window is:

b_{\mathrm{early}}= \frac{\ln(3.8619/6.0948)}{4-2} \approx -0.2280\ \mathrm{h}^{-1}, \qquad t_{1/2,\mathrm{early}}\approx\frac{0.693147}{0.2280}=3.040\ \mathrm{h}

At 32 and 40 hours, the model gives:

C(32)\approx0.15515\ \mathrm{mg/L}, \qquad C(40)\approx0.08157\ \mathrm{mg/L}

The late-window slope and associated half-life are approximately:

b_{\mathrm{late}}= \frac{\ln(0.08157/0.15515)}{40-32} \approx-0.0803618\ \mathrm{h}^{-1}, \qquad t_{1/2,\mathrm{late}}\approx\frac{0.693147}{0.0803618}=8.6253\ \mathrm{h}

The model’s slow component has a rate constant of \(0.08\ \mathrm{h}^{-1}\), with a half-life of \(0.693147/0.08=8.664\) hours. The late estimate is close to those values, while the early estimate is much faster because the early observations still include a substantial fast component. A short window that ends at 4 hours would not reveal the slower terminal behavior.

Choosing and Checking Terminal Samples

A useful sampling schedule must provide observations late enough to represent the terminal phase, but not so late that concentrations are mostly unquantifiable. There is no universal clock time that marks the beginning of the terminal phase: it depends on the drug, dose, route, participants, and processes shaping the profile. Timing decisions should therefore be guided by the expected profile and, when available, by observed concentration-time data.

Several checks help reveal when a terminal estimate may be unreliable:

Inspect the log-concentration pattern. The selected observations should be reasonably compatible with a straight-line decline against time. Curvature suggests that more than one phase may still be contributing, or that another process is affecting the observations.
Check the number and spread of times. Two points always define a line, even if the underlying profile is curved. Three or more points allow some assessment of linearity; observations spread over a meaningful time range help distinguish a slope from measurement noise.
Examine residuals and influential observations. A systematic pattern in residuals can indicate that a straight line is inadequate. A single late observation can strongly affect the slope, particularly when the other samples are close together.
Check quantifiability and data quality. Very low concentrations may be below the assay’s quantifiable range; they should not be treated as precise numerical values. Confirm sample times, units, and concentration records before fitting a slope.
Consider whether absorption or distribution may still matter. If the late profile is not governed by elimination alone, the terminal slope may reflect another process. For example, slow absorption can control the observed terminal decline after some dosing routes.

Ordinary least squares on log concentrations is a transparent way to estimate a line when its assumptions are reasonable. A two-point calculation is simpler but cannot assess linearity. Regression with more points offers residual checks, while alternative regression methods may be appropriate if the log-scale errors are not roughly constant in variance. No method can recover a reliable terminal slope if the sampling window misses the terminal phase or provides too little measurable information.

Common pitfall: “The last two samples have a slope, so that slope must be the elimination rate.” Any two positive concentrations at distinct times define a log slope, but they do not prove that the samples are terminal or that elimination alone governs the decline. Check the broader pattern, quantifiability, and plausible competing processes; describe the result as a terminal slope unless the model justifies a stronger interpretation.

Limits of the Estimate

A terminal half-life is derived from a fitted slope, so uncertainty in the selected observations and sampling times affects the result. It is also sensitive to which observations are chosen: including a point from an earlier phase can steepen the estimated decline, while a poorly measured late point can change it substantially. Reporting a half-life with many decimal places does not remove those limitations.

The terminal half-life summarizes the late exponential decline under the fitted model. It is not automatically the half-life of every part of the concentration profile, a guarantee about concentrations beyond the observed window, or a direct measure of the time needed for all drug in the body to disappear. The interpretation should match both the model and the sampled time range.

Check Your Understanding

Try each problem, then compare your reasoning with the solutions below.

A hypothetical terminal rate constant is \(0.20\ \mathrm{h}^{-1}\). Calculate the terminal half-life in hours.
Two hypothetical terminal concentrations are 8.0 \(\mathrm{mg/L}\) at 10 hours and 2.0 \(\mathrm{mg/L}\) at 16 hours. Calculate the terminal rate constant and half-life.
Why can two samples define a slope but still be insufficient to establish that a terminal phase has been observed?
A log-linear regression has slope \(+0.04\ \mathrm{h}^{-1}\) for concentration against time. Is this a positive terminal rate constant? Explain what should be checked.

Solutions. 1. \(t_{1/2,z}=0.693147/0.20=3.466\ \mathrm{h}\). 2. The slope is \(b=\ln(2.0/8.0)/(16-10)=\ln(0.25)/6=-0.2310\ \mathrm{h}^{-1}\). Thus, \(\lambda_z=0.2310\ \mathrm{h}^{-1}\) and \(t_{1/2,z}=0.693147/0.2310=3.001\ \mathrm{h}\). 3. Any two distinct times and positive concentrations determine a line, so two points cannot show whether the log-concentration pattern is straight or whether both points belong to the terminal phase. Additional appropriately timed, quantifiable samples help assess the pattern. 4. No. A positive fitted slope means concentration increases over the selected observations, whereas a terminal decline requires a negative slope and \(\lambda_z=-b>0\). Check the time ordering, concentration records, phase selection, and whether a log-linear terminal decline is appropriate.