From Concentration-Time Profiles to Candidate Models
An intravenous concentration-time profile is a record of how drug concentration changes after a known input. The curve can suggest whether a one-, two-, or three-compartment structural model is a useful candidate, but a curve should not be assigned a model merely because it bends on a graph. The key questions are whether the candidate equation describes the observed profile, whether its distinct exponential phases are visible in the sampling window, and whether the observations contain enough information to estimate the parameters.
The profiles below are hypothetical and are constructed to make these questions concrete. Concentration is denoted by \(C(t)\), time by \(t\), and an exponential rate by \(\lambda\), in inverse hours. An amplitude \(A\) has concentration units, such as mg/L. The exponential terms describe the observed concentration curve; their rate constants are not automatically the physiologic transfer rates of the compartments.
These candidate forms are appropriate as disposition profiles under assumptions such as a bolus-like IV input at \(t=0\), no later input, linear time-invariant disposition over the observed range, and concentration measured in a specified compartment (usually the central compartment). The model also assumes that the relevant rates do not change during the profile. Real observations include measurement error, and a proposed model must describe the data to an adequate degree without implying that every parameter is precisely estimable.
Why Exponential Phases Appear
In a one-compartment model after an IV bolus, let \(A(t)\) denote the amount of drug in the compartment, \(V\) its volume, and \(k\) its elimination rate constant. The mass-balance equation is \(dA/dt=-kA\), with initial amount \(A(0)=D\), where \(D\) is the dose. Separation and integration give the amount and concentration profiles:
Thus \(A=D/V\) and \(\lambda=k\) for this particular model. For a linear multi-compartment model, the coupled amount equations produce a concentration profile that can be expressed as a sum of exponential terms when the relevant rates are distinct. The coefficients and rates in that sum are macroparameters: they describe the concentration curve. Recovering compartment volumes and transfer rates from them requires a specified structural model and adequate information; a visible exponential term alone does not supply that interpretation.
At early times, a fast exponential may contribute substantially and then decay quickly. At sufficiently late times, the slowest exponential can dominate because the faster terms have become small. This explains why early samples are informative about fast phases and terminal samples about slow phases. The transition between dominance regions is also important: without samples near a transition, a phase can be difficult to separate from its neighbors.
Worked Profile Analyses
Worked Example: A Profile Consistent With One Exponential
Suppose a hypothetical 100 mg IV bolus is followed by concentrations of 5.000 mg/L at 0 h, 1.839 mg/L at 5 h, and 0.6767 mg/L at 10 h. Consider the one-exponential candidate \(C(t)=A e^{-\lambda t}\).
At time zero, the candidate gives \(C(0)=A\), so the observed initial concentration suggests \(A=5.000\) mg/L. Using the 5 h observation, the estimated rate under this candidate is
The resulting profile predicts \(C(10)=5e^{-0.2(10)}=5e^{-2}=0.6767\) mg/L, agreeing with the supplied rounded value. Equivalently, the concentration ratio over each 5 h interval is about \(e^{-1}=0.3679\), consistent with a constant exponential rate. With the additional assumptions of an instantaneous bolus and complete initial mixing, the initial concentration implies \(V=D/C(0)=100\text{ mg}/(5.000\text{ mg/L})=20.00\) L. The corresponding clearance in this one-compartment model is \(CL=kV=(0.2000\ \mathrm{h}^{-1})(20.00\ \mathrm{L})=4.000\) L/h.
These observations support a simple one-exponential description over the sampled times; they do not prove that no faster process occurred before the first observation. The volume and clearance calculations also rely on the stated model and dose assumptions, not on the curve shape alone.
Worked Example: A Profile With Two Visible Exponential Terms
Consider the hypothetical profile \(C(t)=4e^{-1.2t}+1.5e^{-0.12t}\) mg/L, with \(t\) in hours. Its candidate has a fast component with amplitude 4 mg/L and rate 1.2 h\(^{-1}\), and a slower component with amplitude 1.5 mg/L and rate 0.12 h\(^{-1}\). The implied observations are:
The fast contribution falls from 4 mg/L at time zero to about 0.000271 mg/L at 8 h. The slow contribution remains about 0.574339 mg/L at 8 h, so the late profile is nearly governed by the slower term. Early observations are needed to detect the rapid change; later observations help characterize the slower decline. For example, the 8 h concentration alone cannot reveal the fast amplitude or rate, because that component is almost absent by then.
If only late samples were collected, a one-exponential curve might adequately describe the observed window even though the constructed profile has two terms. Conversely, closely spaced early samples without sufficiently late observations might reveal curvature but give weak information about the slow rate. The distinction is between how many terms generated a hypothetical curve and how many phases a particular sampling design can support.
Worked Example: A Three-Exponential Candidate and Its Visible Phases
Now consider \(C(t)=6e^{-2t}+3e^{-0.4t}+e^{-0.05t}\) mg/L, with \(t\) in hours. The rates are ordered from fastest to slowest. The following substitutions show how the components contribute at selected times:
The 2 h observation still contains a substantial fast term, while by 8 h that term is negligible and the intermediate term remains visible alongside the slow term. At 20 h, the slow term dominates. These observations illustrate how early, intermediate, and late portions of a profile can each contain different information. However, the sampling schedule shown is only illustrative: the profile’s three terms are specified in advance, whereas real data do not label their components.
In a real analysis, the evidence for three phases would come from the pattern across all observations, including whether a two-term candidate leaves systematic residual structure and whether the third term is estimated with useful precision. A numerical fit that returns three amplitudes and three rates is not, by itself, evidence that the data support all six quantities.
What the Observations Can Support
A useful structural analysis separates three questions. First, does a candidate equation represent the observed profile over the study’s sampling window? Second, are the phases distinguishable in that window? Third, which parameters are sufficiently informed by the actual observations to be reported with confidence? These questions are related, but they are not interchangeable.
Confirm that the profile is from the stated IV input, that there is no unmodeled input during the interval, and that concentration and time units are consistent. A misspecified input or changing disposition can create curvature that a more complex disposition model would incorrectly absorb.
Early samples may reveal a fast phase, samples around transitions can distinguish adjacent terms, and late samples may reveal the terminal slope. A phase that has already decayed before sampling, or is below measurement precision, is not well supported by the observed data.
Assess whether adding a term improves the description in a meaningful, systematic way, rather than relying on a single bend or a lower residual sum alone. Inspect residual patterns, parameter uncertainty, and sensitivity to removing or changing influential observations.
Report the profile features the observations support. Do not interpret a weakly estimated amplitude or rate as established simply because software returned a value. A structurally identifiable parameter combination may still be practically uncertain for a particular sampling design.
The earlier material on structural and practical identifiability distinguishes whether a parameter can, in principle, be uniquely recovered from an ideal profile from whether finite, noisy observations estimate it precisely. Here, the practical issue is visible in the time coverage: a fast phase can be lost before the first sample, while a slow phase can be poorly estimated if follow-up ends too soon. If two rates are close, their exponential terms can look similar over a limited range, making their amplitudes and rates difficult to separate.
A common diagnostic is the semilog profile, where concentration is plotted on a logarithmic vertical scale against linear time. A single exponential gives a straight line on this scale, provided concentrations are positive and the model is appropriate. Curvature can signal multiple phases, but it can also arise from error, an inappropriate observation model, or other model misspecification. A visually straight terminal segment supports a terminal exponential description over that segment; it does not establish that the full profile is one-compartment.
Choosing Among Candidate Models
| Candidate | What it can describe | What the data need to show | Main limitation |
|---|---|---|---|
| One exponential | A single exponential decline over the observed interval | Consistent exponential decay, with no meaningful systematic curvature left unexplained | A fast phase before the first sample may be missed |
| Two exponentials | A fast and a slow contribution to concentration | Sampling that covers both phases and enough of their transition to distinguish them | One amplitude or rate may be weakly estimated with sparse or narrow-range data |
| Three exponentials | Fast, intermediate, and slow contributions | Early, intermediate, and terminal coverage with enough precision to separate all three contributions | Extra terms can be poorly determined or fit noise when the data do not resolve them |
The simpler model is not automatically correct, and the most complex model is not automatically more informative. The appropriate candidate is the simplest one that adequately describes the observations for the analysis purpose, while retaining parameters supported by the design. A comparison should consider residual behavior and parameter uncertainty, not just the apparent fit. Equally, visual phase counting is a diagnostic, not a substitute for a specified model and a careful assessment of the data.
The limitations are important. A concentration-time profile does not, by itself, determine a unique physiologic story. Multiple structural parameter combinations can yield similar curves, and finite sampling or assay limits can hide phases. The profiles in the worked examples are exact hypothetical curves; actual measurements scatter around a model, and uncertainty in dose timing, concentration measurement, and model assumptions further limits conclusions. Structural analysis identifies plausible descriptions and information boundaries; it does not guarantee that every fitted parameter has a stable physiologic interpretation.
Common Mistake
Check Your Understanding
Answer each question using the distinction between a candidate curve and the information available from the observations.
- A profile is sampled only from 6 h onward, and its late concentrations are nearly linear on a semilog plot. What can this support, and what does it not establish?
- In the two-exponential example, which term dominates at 8 h? Show the comparison using the computed contributions.
- For the three-exponential profile, explain why the 20 h observation is informative about the slow term but weakly informative about the fast term.
- Why is a lower residual sum after adding a third exponential not sufficient evidence that three phases are reliably supported?
- Under the one-compartment assumptions in the first example, calculate the implied volume and clearance from the stated dose and initial concentration and estimated rate.
Solutions. (1) The late observations support an approximately exponential description of the sampled terminal interval. They do not establish that the entire profile has only one phase; faster phases may have occurred before 6 h. (2) At 8 h, the fast contribution is about 0.000271 mg/L and the slow contribution is about 0.574339 mg/L. The slow term dominates. (3) At 20 h, the slow term contributes about 0.367879 mg/L, while the fast term \(6e^{-40}\) is negligible. The measurement therefore describes the slow tail but contains almost no information about the fast rate or amplitude. (4) Additional parameters can reduce residual error even when they are weakly identified or fit noise. Support also depends on systematic improvement, time coverage, residual behavior, and the precision and stability of the estimates. (5) \(V=D/C(0)=100\text{ mg}/(5.000\text{ mg/L})=20.00\) L, and \(CL=kV=(0.2000\ \mathrm{h}^{-1})(20.00\ \mathrm{L})=4.000\) L/h, conditional on the one-compartment IV bolus assumptions.