Distribution-Phase Misfit In Multicompartment Profiles

Why the Early and Intermediate Profile Matters

A model can follow the late decline in concentration while missing the shape of the earlier profile. In a multicompartment profile, that mismatch may arise because the model does not represent a distribution phase: concentrations can fall relatively rapidly at first, then decline more slowly as the terminal phase dominates. A terminal fit alone can therefore conceal important errors in predicted early and intermediate concentrations.

The diagnostic question here is specific: do overlays and residuals show a systematic pattern during the early-to-intermediate part of the profile that is consistent with an incorrect distribution phase? Such a pattern is evidence to investigate, not proof of a particular mechanism. It must be interpreted with the sampling times, model predictions, and plausible competing explanations in view.

Expected Shapes Under Simple Compartment Models

For a linear one-compartment model after an IV bolus, concentration follows a single exponential. Here \(C(t)\) is concentration at time \(t\), \(D\) is dose, \(V\) is volume of distribution, and \(k\) is the first-order elimination rate constant. The model assumes an instantaneous bolus, linear time-invariant disposition, and a single exponential decline over the interval considered.

Taking logarithms gives \(\log C(t)=\log(D/V)-kt\). Thus, on a plot of log concentration against time, this idealized one-compartment profile is a straight line with constant slope \(-k\). This is a useful shape reference, not a general rule that every observed log profile must be linear.

For a classical linear two-compartment IV bolus model, the concentration can be written as the sum of two exponentials:

The first, faster term is commonly called the distribution component; the second, slower term is the terminal component. Their relative contributions change with time. The fast term does not abruptly stop at a boundary: it becomes progressively smaller relative to the slow term. This representation presumes linear disposition and the specified IV bolus setting. Other routes, inputs, or processes can create early curvature too, so a biexponential-looking profile is not by itself proof of multicompartment distribution.

Reading Overlays and Residual Patterns

An overlay places observed concentrations and model-predicted concentrations against time on a common scale. A linear concentration scale shows absolute differences clearly; a log concentration scale can make relative profile shape and the terminal decline easier to inspect. Use the same time and concentration units and an appropriate prediction for the observations being displayed. For an individual profile, compare observations with predictions for that individual; do not mistake a population-level prediction for an individual-specific fit.

Define the raw residual at observation \(j\) for a given profile as observed minus predicted concentration:

A positive residual means the observation is above the prediction; a negative residual means it is below. For distribution-phase diagnosis, the order of residuals across time matters. A run of positive early-to-intermediate residuals followed by smaller residuals or residuals centered around zero later can be consistent with a model that underpredicts the early contribution while describing the terminal decline. The reverse sign pattern can indicate early overprediction. Isolated residuals or irregular sign changes are weaker evidence of a systematic phase mismatch.

Residuals should be considered on a scale appropriate to the observation model. Under a constant-variance additive error model, raw residuals are in concentration units and their spread is expected to be roughly comparable across the range. Under proportional error, the relative discrepancy may be more informative than the absolute one. Residual scaling is not a substitute for looking at the underlying overlay, and a particular residual sequence does not uniquely identify its cause.

Why the Log-Profile Slope Changes

The one-compartment reference has a constant log-concentration slope. A two-exponential profile has a changing slope because the relative weights of its components change. The derivative makes the reason explicit. For \(C(t)=Ae^{-\alpha t}+Be^{-\beta t}\), differentiating and dividing by \(C(t)\) gives:

The positive fraction on the right is a weighted average of \(\alpha\) and \(\beta\), with weights proportional to the two component concentrations. Since \(\alpha>\beta\), the apparent decline rate lies between them while both components contribute. As time increases, the fast component's weight decreases; the weighted rate moves toward \(\beta\). This establishes why curvature can appear on a log concentration-time overlay even when the terminal portion looks nearly linear.

Assumptions, Diagnostics, and Failure Modes

A phase interpretation is most defensible when the model and data are comparable and the design can resolve the relevant shape. Check that the plotted times are actual times since the relevant input, units and dose records are correct, and the observations cover both the early-to-intermediate interval and enough of the later decline to assess whether the terminal profile is better described. Consider whether the selected prediction is appropriate for the individual or population summary being examined. The simple biexponential interpretation also relies on the IV bolus and linear disposition assumptions stated above.

When the model is missing a distribution contribution, an overlay may show a coherent early-to-intermediate departure that diminishes later; residuals may show a corresponding run of one sign. In contrast, sparse early samples may leave too little information to reveal curvature. A few irregular residuals may reflect ordinary variability rather than a phase pattern. On a log plot, values near or below quantification limits can also make the apparent shape difficult to judge; do not treat a displayed log pattern as reliable when its values are not meaningfully observed.

A one-compartment versus two-compartment contrast is useful for understanding profile shapes, but real analyses may require other explanations to be considered. Incorrect input timing or absorption can affect early concentrations; an incorrect elimination description can affect the terminal region; covariates, dose records, and assay or study differences can also warrant checking. A visible pattern should prompt a targeted model review rather than an automatic change to compartment count.

Common Mistake and Practical Limits

A common mistake is to extend a terminal straight line backward and treat the difference between that extrapolation and early observations as a direct measurement of distribution. The correction is that a terminal component is a late-time approximation to the total profile, not generally a valid prediction at early times. In the worked contrast, the terminal-only curve underpredicts the full profile because the fast component is still present; the residual difference illustrates the miss, not a separately observed concentration compartment.

Another pitfall is to call every early residual run a distribution problem. If sampling is sparse, an early point can have strong influence on the visual impression; if the time origin or input is wrong, the apparent phase may be shifted. Also, a model can show a satisfactory population-level overlay while some individual profiles retain phase-specific discrepancies, or the reverse. Examine the unit of analysis and sampling distribution before drawing a broad conclusion. These plots are diagnostic evidence, not a definitive structural selection procedure.