Terminal-Phase Samples for Two-Compartment Parameters

What Late Samples Can—and Cannot—Tell Us

In a two-compartment concentration profile, the early distribution phase and the later terminal phase contain different information. Late samples can be especially useful for estimating the terminal rate of decline and its associated amplitude. They do not, by themselves, usually reveal the full set of compartmental parameters. This distinction matters when a sampling schedule is intended to characterize a patient’s disposition: an apparently precise estimate of terminal half-life is not the same as precise estimates of central volume, intercompartmental clearance, and peripheral volume.

The introductory material on distribution-phase sampling concerns the contribution of earlier observations to the fast component. Here the focus is the complementary question: when does the slow component dominate, what does it identify, and what remains ambiguous if observations are concentrated in that phase?

The parameters \(\alpha\) and \(\beta\) are hybrid, or macro-, rate constants. They are not generally equal to individual compartmental rate constants. To make that relationship explicit, let \(CL\) denote clearance from the central compartment, \(V_c\) the central volume, \(Q\) intercompartmental clearance, and \(V_p\) peripheral volume. Define \(k_{10}=CL/V_c\), \(k_{12}=Q/V_c\), and \(k_{21}=Q/V_p\). Under this standard linear model, the macro-rate constants are the two positive roots of the characteristic equation below.

These results assume positive compartment volumes and clearances, linear and time-invariant disposition, an IV bolus with known dose, and a correctly specified two-compartment model. They also assume that concentration observations are made in the central compartment and that there is no additional process, such as time-varying clearance, changing the profile over the sampling interval. The characteristic equation follows from the two-compartment amount-balance equations; its roots determine the two exponential rates in the central concentration solution.

Estimating the Terminal Rate and Amplitude

If the fast term is negligible, taking logarithms gives a straight-line relationship between log concentration and time:

Thus the slope of a line fitted to late log concentrations estimates \(-\beta\), and its intercept estimates \(\log B\). With two ideal, error-free terminal observations at \(t_1<t_2\), the corresponding estimates are \(\widehat{\beta}=[\log C(t_1)-\log C(t_2)]/(t_2-t_1)\) and \(\widehat{B}=C(t_1)e^{\widehat{\beta}t_1}\). In actual data, measurement error makes a regression using several appropriately selected terminal observations preferable to relying on a single pair.

That regression has its own assumptions. The log-linear approximation must be adequate over the selected times; observations must be independent, and a conventional unweighted log-linear fit assumes approximately constant error variance on the log-concentration scale. If the assay error is better represented as additive on the original concentration scale, or if precision changes substantially with concentration, a suitable observation model and weighting are needed instead. A straight-looking segment does not establish that the compartmental model is correct, but curvature or systematic residual patterns can warn that the selected observations are not consistently terminal.

A useful way to assess the terminal approximation is to compare the fast and slow contributions directly. Their ratio at time \(t\) is \((A/B)e^{-(\alpha-\beta)t}\). The fast term becomes small relative to the slow term only as this ratio declines. The time at which that happens depends not only on the separation between \(\alpha\) and \(\beta\), but also on the ratio \(A/B\). Consequently, “late” is not a universal clock time: it depends on the individual profile and the intended approximation tolerance.

Why Terminal-Only Data Leave Parameters Ambiguous

The terminal rate is a composite of the compartmental rates. Although \(\beta\) is informative about disposition, its value alone cannot be inverted to obtain all four underlying quantities \(CL,V_c,Q,V_p\). Even adding the terminal amplitude \(B\) does not generally resolve the ambiguity. The amplitude depends on the dose, central volume, and the relationship among the macro-rates and \(k_{21}\). Distinct parameter sets can therefore produce the same terminal rate and amplitude.

The following construction shows this directly. For an IV bolus, the coefficient of the slow exponential is \(B=(D/V_c)(k_{21}-\beta)/(\alpha-\beta)\). If the dose is known, distinct choices of \(\alpha\), \(k_{21}\), and \(V_c\) can preserve both \(\beta\) and \(B\), while the rate-constant identities determine compatible \(k_{10}\) and \(k_{12}\). The example uses positive values throughout, so the ambiguity is not caused by a boundary or an invalid parameter set.

Diagnostics, Comparisons, and Practical Limits

Before interpreting a fitted terminal slope, check that the selected samples support the approximation and that their concentrations are measured reliably. A practical assessment includes plotting log concentration against time, examining whether late points are approximately linear, and inspecting residuals from the terminal fit for systematic curvature. Pairwise slopes between successive late samples can be compared: large, patterned differences may indicate that the fast component remains material, that measurement error is influential, or that the assumed model does not describe the data over that interval. A single nearly straight segment is not a guarantee of adequate information about all compartmental parameters.

Terminal sampling also faces a precision trade-off. Later observations may be more nearly governed by the slow component, yet concentrations fall and may approach the assay’s lower limit of quantification. Censoring, rounding, or relatively large assay error can then make the slope unstable. If the useful late interval contains too few quantifiable observations, adding still later samples may contribute little reliable information. Conversely, moving all samples earlier can mix the fast and slow terms and bias a simple terminal regression. The response model and assay characteristics therefore matter alongside the choice of sampling times.

Compared with terminal-only sampling, a schedule that includes distribution-phase observations can represent both exponential components and provide information unavailable from the slow component alone. A terminal-only schedule is narrower in purpose: it may be suitable when the main target is terminal decline and late concentrations are measurable, but it should not be presented as a complete design for estimating every two-compartment parameter. Model-based estimation from the full profile is another alternative; it uses all observations through the biexponential model rather than treating selected points as an isolated log-linear segment. Its estimates still depend on adequate data, model assumptions, and the chosen error structure.

The main pitfall is to treat the terminal slope as though it were a direct measurement of one peripheral rate constant or to infer that a stable terminal fit proves central parameters are well estimated. The terminal rate is a composite eigenvalue, and the example above shows that different underlying parameters can yield the same terminal behavior. Report the target precisely—for example, an estimated terminal rate in \(\text{h}^{-1}\) or terminal half-life in hours—and distinguish that target from the broader question of identifying the compartmental model.