Terminal Samples and the Slow Phase

Why the Slow Phase Needs Late Samples

After an intravenous bolus, a multicomponent concentration-time profile can decline rapidly at first and more slowly later. The late portion matters because it carries information about the slowest decay rate and the concentration contribution associated with that rate. If observation ends too early, faster components may still contribute substantially to the measured concentration, so the final observed slope need not represent the slowest phase.

The issue is not simply how many samples were collected. A schedule can contain many early measurements yet provide little direct information about the terminal portion. Conversely, a few well-spaced, accurately measured late samples may help reveal whether the profile has reached a slow, approximately log-linear decline. Sparse late samples are still vulnerable to measurement error: with only two points, a small change in either concentration can change the estimated slope appreciably.

Definition: For a positive multiexponential profile \(C(t)=\sum_{j=1}^{m} A_j e^{-\lambda_j t}\), with positive amplitudes \(A_j\) and ordered decay rates \(\lambda_1>\lambda_2>\cdots>\lambda_m>0\), the slowest component is \(A_m e^{-\lambda_m t}\). Its decay rate \(\lambda_m\) is the slowest rate, and its terminal contribution at time \(t\) is its concentration value \(A_m e^{-\lambda_m t}\). The last measured concentration is generally the sum of all component contributions still present, not the terminal contribution alone.

The rates \(\lambda_j\) have units of inverse time; amplitudes and component concentrations have concentration units. In the positive-amplitude setting used here, each component contributes positively to total concentration. A component may be called “terminal” because it decays most slowly, even when faster components remain visible at the last observation.

Why the Apparent Slope Approaches the Slowest Rate

A useful way to see the role of late samples is to examine the instantaneous slope of the log concentration. This is a property of the profile, not a separate physiologic rate constant. Different exponential components contribute at different times, and the contribution of each faster component diminishes relative to the slowest one as time increases.

-\frac{d}{dt}\log C(t) = \frac{\sum_{j=1}^{m} A_j\lambda_j e^{-\lambda_j t}} {\sum_{j=1}^{m} A_j e^{-\lambda_j t}} = \sum_{j=1}^{m} w_j(t)\lambda_j, \qquad w_j(t)=\frac{A_j e^{-\lambda_j t}}{C(t)}.

Because the component concentrations are positive, the weights \(w_j(t)\) are positive and sum to one. Thus the apparent instantaneous decay rate is a weighted average of the component rates; while multiple components contribute, it need not equal \(\lambda_m\). To justify the terminal limit, divide the numerator and denominator by \(A_m e^{-\lambda_m t}\). For each faster component \(j<m\), its relative contribution contains the factor \(e^{-(\lambda_j-\lambda_m)t}\), which tends to zero because \(\lambda_j-\lambda_m>0\). The remaining ratio tends to \(\lambda_m\). Therefore, the log concentration becomes approximately linear with slope \(-\lambda_m\) only when the faster contributions have become small enough to be negligible for the intended precision.

\frac{A_j e^{-\lambda_j t}}{A_m e^{-\lambda_m t}} = \frac{A_j}{A_m}e^{-(\lambda_j-\lambda_m)t} \longrightarrow 0 \quad\text{as } t\longrightarrow\infty,\quad j<m.

This argument assumes a sum of positive exponential terms, constant rates over the observed interval, and no continuing drug input that changes the profile. The examples below use a hypothetical intravenous-bolus profile, so ongoing absorption is not part of the interpretation. In data, the practical question is whether late observations are consistent with the terminal approximation and are sufficiently precise to estimate its slope.

Worked Example: Finding the Terminal Contribution in a Three-Exponential Profile

Suppose a hypothetical bolus profile is described by \(C(t)=8e^{-1.20t}+3e^{-0.20t}+0.500e^{-0.0400t}\), where \(t\) is in hours and \(C(t)\) is in mg/L. The rates are ordered \(1.20>0.20>0.0400\ \text{h}^{-1}\), so the slowest component is \(0.500e^{-0.0400t}\).

At 12 hours, its terminal contribution is \(0.500e^{-0.0400(12)}=0.500e^{-0.48}=0.3094\ \text{mg/L}\). The middle component contributes \(3e^{-0.20(12)}=3e^{-2.4}=0.2722\ \text{mg/L}\). The fast component contributes \(8e^{-1.20(12)}=8e^{-14.4}=0.00000446\ \text{mg/L}\). Adding the unrounded contributions gives approximately \(0.5815500\ \text{mg/L}\), or \(0.5816\ \text{mg/L}\) to four significant figures.

At 12 hours the terminal component accounts for about \(0.3094/0.5815500=0.532\), or 53.2%, of total concentration. The middle component is still substantial, so the 12-hour concentration is not simply the terminal component. Even though 12 hours may sound late, whether it is late enough depends on the amplitudes and the separation between the decay rates.

Short Follow-Up Can Misrepresent the Slow Rate

If samples are collected only while faster components remain important, a slope calculated from the last few observations may be steeper than the slowest rate. In a log-concentration plot, “steeper” means a larger magnitude of the negative slope, corresponding to a larger apparent decay rate. Extending follow-up can show whether that slope becomes shallower and stabilizes, but observations must remain measurable and reliable.

Worked Example: Comparing Apparent Slopes Across Follow-Up

Use the profile from the previous example. At 24 hours, the middle contribution is \(3e^{-4.8}=0.02469\ \text{mg/L}\), and the slow contribution is \(0.500e^{-0.96}=0.1914\ \text{mg/L}\). The fast term is negligible, giving \(C(24)\approx0.2161\ \text{mg/L}\). The 12-hour concentration was approximately \(0.5815500\ \text{mg/L}\).

For two concentrations \(C(t_1)\) and \(C(t_2)\), the positive apparent decay rate from a log-linear line through the two points is \(\widehat{\lambda}=\log\{C(t_1)/C(t_2)\}/(t_2-t_1)\). Using the 12- and 24-hour concentrations gives \(\widehat{\lambda}=\log(0.5815500/0.2161357)/(24-12)\approx0.0825\ \text{h}^{-1}\). This is more than twice the true slowest rate, \(0.0400\ \text{h}^{-1}\), because the middle component still contributes at 12 hours and remains present at 24 hours.

At 36 hours, the middle contribution is \(3e^{-7.2}=0.002239\ \text{mg/L}\), while the slow contribution is \(0.500e^{-1.44}=0.1185\ \text{mg/L}\), giving \(C(36)\approx0.1207\ \text{mg/L}\). At 48 hours, the corresponding contributions are \(3e^{-9.6}=0.0002032\ \text{mg/L}\) and \(0.500e^{-1.92}=0.07330\ \text{mg/L}\), giving \(C(48)\approx0.07351\ \text{mg/L}\). The 36-to-48-hour slope is \(\log(0.1207/0.07351)/12\approx0.0413\ \text{h}^{-1}\), much closer to \(0.0400\ \text{h}^{-1}\). This illustrates why later observations can identify the slow phase more directly.

Two points always define a straight line on a log-concentration plot, so a line through two late observations is not, by itself, evidence that a stable terminal phase has been observed. More points across the late interval allow the analyst to check for curvature and to see whether one influential observation is determining the estimated slope.

Sparse Late Samples Make the Estimate Sensitive to Error

Late measurements can be less precise in practice because concentrations are low, assay limitations may become relevant, and few scheduled samples may fall in the terminal interval. When a rate is estimated from two observations, both measurement errors affect the ratio and therefore the log slope. A long interval between samples can help separate rates, but it does not remove the effect of concentration error or guarantee that both observations are truly terminal.

Worked Example: How Two Noisy Late Samples Change the Slope

Suppose hypothetical observed concentrations at 36 and 48 hours are \(0.121\) and \(0.0740\ \text{mg/L}\). The two-point estimate is \(\log(0.121/0.0740)/12=\log(1.6351)/12\approx0.0410\ \text{h}^{-1}\). This is close to the profile’s slowest rate of \(0.0400\ \text{h}^{-1}\).

Now consider a plausible pair in which the first observation is 10% low and the second is 10% high: \(0.1089\) and \(0.08140\ \text{mg/L}\). The estimated rate becomes \(\log(0.1089/0.08140)/12=\log(1.3378)/12\approx0.0243\ \text{h}^{-1}\). With the opposite deviations, \(0.1331\) and \(0.06660\ \text{mg/L}\), the estimate is \(\log(0.1331/0.06660)/12=\log(1.9985)/12\approx0.0577\ \text{h}^{-1}\).

These examples do not assign probabilities to the deviations; they show sensitivity to possible errors. A sparse late pair may produce a rate well below or above the actual terminal rate even when each concentration differs from its underlying value by only about 10%. Additional informative late observations can reveal such instability; simply repeating many early samples cannot substitute for terminal coverage.

Similar Early Profiles Can Hide Different Slow Phases

If follow-up ends before the terminal component dominates, more than one combination of terminal amplitude and rate can be compatible with the observed portion of a profile. The amplitude controls the component’s contribution, while its rate controls how quickly that contribution declines. This is a sampling-information issue: it does not establish that two structural parameterizations are equivalent, nor does it identify which model is correct.

Worked Example: Two Slow Components Diverge After Follow-Up Ends

For comparison, retain the fast and middle terms from the hypothetical profile and replace its slow term with \(0.586755e^{-0.0800t}\). This alternative was chosen so that its slow contribution matches \(0.500e^{-0.0400t}\) at 4 hours: \(0.500e^{-0.0400(4)}=0.4261\ \text{mg/L}\), and \(0.586755e^{-0.0800(4)}=0.4261\ \text{mg/L}\), to four significant figures.

At time zero, the original total is \(8+3+0.500=11.50\ \text{mg/L}\); the alternative total is \(8+3+0.586755=11.586755\ \text{mg/L}\), about 0.754% higher. At 4 hours the slow contributions match to the displayed precision, and the fast and middle terms are unchanged, so the total profiles also match there. Early measurements can therefore look similar despite different slow rates.

At 12 hours, the original total is approximately \(0.5815500\ \text{mg/L}\). The alternative slow contribution is \(0.586755e^{-0.0800(12)}=0.2247\ \text{mg/L}\), while its middle contribution remains \(0.2722\ \text{mg/L}\) and its fast contribution is \(0.00000446\ \text{mg/L}\). Its total is approximately \(0.4968\ \text{mg/L}\), about 14.6% below the original total. The two candidate slow phases have become distinguishable only as follow-up extends into the portion of the profile where the slow component matters more.

Planning and Checking Terminal Coverage

A useful sampling plan includes enough late observations to assess the declining profile after faster components have waned, while keeping concentrations above the assay’s quantification limit where possible. The timing should reflect the expected range of slow rates and amplitudes; there is no universal clock time that defines “late.” A pilot profile or prior knowledge can help plan the observation window, but the data should still be checked for evidence that the slow phase was actually observed.

Check whether the late data are measurable.
Confirm that observations used to characterize decline are quantifiable and are not dominated by censoring or assay noise.

Inspect the late log-concentration pattern.
Look for approximate linearity over multiple observations. Systematic curvature can indicate that faster components still contribute or that the assumed profile is inadequate.

Check sensitivity to individual samples.
Compare the estimated slope over plausible late intervals or assess the effect of omitting one late observation. Large changes signal weak support from the sampling schedule.

Distinguish absence of evidence from precise estimation.
A fitted slow rate is not automatically well determined. Broad uncertainty, strong dependence on the final sample, or a terminal window with few quantifiable points indicates limited information.

These checks are diagnostic, not proofs that the model is correct. A visually straight late segment may still be short, noisy, or based on very few observations. Conversely, if late observations are below quantification limits, the data may support only a bound or limited information about the slow decline rather than a precise rate estimate.

Common Pitfall and Practical Limits

Wrong reasoning: “The last measured concentration is the terminal component because it was measured last.” Correction: Each observed concentration is the sum of the components still contributing at that time. The slowest term dominates only when faster contributions have become sufficiently small.
Wrong reasoning: “A straight line through the last two log concentrations proves the terminal rate.” Correction: Any two positive observations define a line. Multiple late samples are needed to check whether the slope is stable and whether the points are consistent with a terminal segment.
Wrong reasoning: “More samples solve the problem, regardless of when they are taken.” Correction: Additional early samples can characterize early behavior but add little direct evidence about a slow component if the follow-up never reaches the terminal region.

The main practical alternatives to inadequate late coverage are to extend follow-up, increase the number of samples across the expected terminal interval, or improve low-concentration measurement capability. Each has limitations: longer follow-up may be burdensome, additional samples may be sparse or costly, and assay improvements cannot replace observations at informative times. If the available data do not contain a measurable terminal segment, the slowest rate and terminal contribution may remain weakly estimated; conclusions should reflect that limitation rather than treating a fitted value as definitive.

Key takeaway: The slowest decay rate is learned from the part of the profile where faster components have largely faded. Short follow-up can make a faster mixture look like the terminal phase, and sparse noisy late samples can make its estimated slope unstable.

Check Your Understanding

Answer each question, then compare with the concise solutions below.

In a positive sum of exponentials with ordered rates, why does the apparent log-concentration slope approach the slowest rate?
A final concentration is \(0.40\ \text{mg/L}\), and the slowest component contributes \(0.25\ \text{mg/L}\) at that time. Is the final concentration equal to the terminal contribution? What fraction is terminal?
Two late concentrations are \(0.200\) and \(0.100\ \text{mg/L}\), measured 10 hours apart. Calculate the two-point apparent decay rate.
Give two signs that an estimated slow rate may be weakly supported by the sampling schedule.

Solutions. 1. Dividing each faster component by the slowest leaves a factor proportional to \(e^{-(\lambda_j-\lambda_m)t}\), which tends to zero; the slowest component then dominates and the apparent rate tends to \(\lambda_m\). 2. No: the total includes all remaining components. The terminal fraction is \(0.25/0.40=0.625\), or 62.5%. 3. \(\widehat{\lambda}=\log(0.200/0.100)/10=\log(2)/10\approx0.0693\ \text{h}^{-1}\). 4. Examples include curvature in the late log-concentration pattern, only one or two quantifiable terminal observations, strong slope changes when one point is omitted, or substantial uncertainty in the fitted rate.