Choosing a Complete Structure
A structural model is a linked account of how drug enters the body, distributes between modeled spaces, and leaves the body. Choosing each part separately can produce an internally inconsistent explanation: for example, an oral curve may rise because of absorption, while a declining terminal phase may reflect either elimination or slow absorption. The task is to select a complete structure that is compatible with the route, dose, sampling design, and observed concentration-time pattern, while avoiding complexity that the data cannot support.
The examples below use hypothetical concentration data and deterministic individual profiles. They illustrate structural reasoning, not proof that a particular biological mechanism is present. A concentration curve can be consistent with more than one mechanism, especially when observations are sparse or only one route is studied.
For linear elimination, the elimination rate is proportional to the amount in the eliminating compartment; in a one-compartment model, this gives a constant elimination rate constant \(k=CL/V\). A two-compartment model adds a peripheral space and distribution between it and the central space. Oral models must additionally specify how drug enters the systemic circulation, including the role of bioavailability \(F\). These familiar structures are used here as candidate components, not as conclusions to be imposed on every profile.
Worked Example: A Two-Phase IV Profile
Worked Example: Selecting a Compartment Structure for IV Data
Suppose a hypothetical IV bolus profile is well summarized by \(C(t)=8e^{-1.2t}+2e^{-0.12t}\), with concentration in mg/L and time in hours. Assume the dose is delivered at time zero, concentrations are measured after mixing, and there is no evidence in this example of time-varying or saturable elimination. Evaluate the predicted concentration at 0.5 hours and assess the structural implication.
Substitute \(t=0.5\) h, retaining precision in the exponential calculations:
The two positive exponential terms represent a faster and a slower declining component. In a conventional linear mammillary two-compartment model after an IV bolus, central concentration has a biexponential form, \(C(t)=Ae^{-\alpha t}+Be^{-\beta t}\), where the positive rate constants satisfy \(\alpha>\beta\). Thus this profile supports considering a central and peripheral compartment with first-order elimination from the central compartment and first-order intercompartmental exchange.
A one-compartment linear model predicts a single exponential decline. It is a reasonable simpler alternative if the apparent early curvature is explained by measurement noise or sampling, but it would be inadequate if systematic residuals show that it misses a reproducible distribution phase. The biexponential appearance alone does not establish the anatomical identity of a peripheral space; the structural interpretation remains a model of the observed kinetics.
From Oral Input to Concentration
For an oral dose with first-order absorption into a one-compartment body, suppose the amount remaining at the absorption site is \(A_g(t)=D e^{-k_a t}\). The systemic input rate is \(F k_a A_g(t)\). With first-order elimination from the central compartment, the amount \(A(t)\) satisfies the following model. Here \(D\) is the oral dose, \(k_a\) is the absorption rate constant, \(F\) is bioavailability, \(V\) is the apparent distribution volume, and \(k=CL/V\).
The model assumes a single well-mixed disposition compartment, first-order absorption and elimination, constant \(F\), and no lag time or presystemic process that changes the input over time. To solve the amount equation, multiply both sides by the integrating factor \(e^{kt}\). The left side becomes the derivative of \(e^{kt}A(t)\); integration from zero to \(t\) then gives:
When \(k_a>k\), absorption is faster than elimination and the late decline is governed by \(k\). When \(k_a<k\), absorption is slower and the late decline is governed by \(k_a\), a situation called flip-flop kinetics. This distinction matters because a terminal slope from oral data alone does not necessarily identify the elimination rate.
Worked Example: Oral Input with Faster Absorption
Worked Example: A One-Compartment Oral Model
Consider a hypothetical 100 mg oral dose with \(F=0.80\), \(V=20\) L, \(CL=4\) L/h, and \(k_a=1.0\) h\(^{-1}\). Assume constant bioavailability, first-order input and elimination, and a single disposition compartment. The elimination rate constant is \(k=CL/V=4/20=0.20\) h\(^{-1}\), so \(k_a>k\).
The concentration coefficient is:
At one hour, using unrounded exponential values:
The predicted peak time follows by setting the derivative of the concentration equation to zero: \(k_a e^{-k_a t}=k e^{-kt}\). Rearranging gives \(t_{\max}=\ln(k_a/k)/(k_a-k)\). Here \(t_{\max}=\ln(1.0/0.20)/(1.0-0.20)=\ln(5)/0.80=2.0118\) h. This rise and subsequent decline are compatible with first-order absorption and elimination. An instantaneous-input model would predict its maximum at time zero, so it would not explain a clearly observed post-dose rise under these assumptions.
The oral curve alone generally identifies combinations of parameters more readily than each parameter separately. In particular, \(F\) and \(V\) enter the concentration scale together, while \(CL\) and \(V\) determine \(k\). Without additional information, such as an IV profile or justified parameter constraints, a good fit does not guarantee that every parameter has been uniquely identified.
Worked Example: Slow Absorption and Flip-Flop
Worked Example: Why an Oral Terminal Slope Can Mislead
Now consider a different hypothetical oral case with \(D=100\) mg, \(F=0.60\), \(V=30\) L, \(CL=15\) L/h, and \(k_a=0.10\) h\(^{-1}\). Under the same one-compartment, first-order assumptions, \(k=CL/V=0.50\) h\(^{-1}\), so absorption is slower than elimination. The coefficient is \(F D k_a/[V(k_a-k)]=(0.60)(100)(0.10)/[30(0.10-0.50)]=-0.50\) mg/L. The negative coefficient is paired with a negative difference of exponentials, leaving positive concentrations.
At one hour:
At late times, \(e^{-0.50t}\) becomes negligible relative to \(e^{-0.10t}\), and the concentration is approximately \(0.50e^{-0.10t}\) mg/L. The terminal slope therefore reflects \(k_a=0.10\) h\(^{-1}\), not the elimination rate \(k=0.50\) h\(^{-1}\). The peak time is \(t_{\max}=\ln(k/k_a)/(k-k_a)=\ln(5)/0.40=4.0236\) h.
If IV data independently support \(k=0.50\) h\(^{-1}\), the slow oral terminal decline is coherent with flip-flop kinetics. If only oral data are available, a slow elimination model may also appear to fit the terminal phase. Treating that slope as proof of slow elimination would therefore be unjustified; the route-specific information is essential to distinguish the alternatives.
Integrating Evidence and Checking Alternatives
A defensible structural choice connects each feature of the profile to a model component and checks whether the full set of assumptions is credible. The following sequence keeps the reasoning explicit.
Identify whether the dose is IV or oral, whether the dose is a bolus or infusion, and whether observations cover distribution, absorption, and terminal phases. Sparse early samples cannot strongly establish a distribution phase.
For IV data, ask whether one exponential is adequate or whether reproducible curvature supports an additional compartment. Consider nonlinear or time-varying elimination only when the dose and time patterns provide a reason to do so.
Assess whether an immediate input, first-order absorption, or a more complex input is compatible with the observed rise and peak. Use IV information when available to separate absorption behavior from disposition.
Compare predicted and observed concentrations across the whole time course. Look for systematic residual patterns, implausible predictions, and phases the model consistently misses; do not judge structure from one point or the terminal slope alone.
Alternatives should be compared on the same data and with consistent assumptions about observations. A simpler model is attractive when it describes the relevant profile adequately; added compartments or specialized absorption processes are not automatically improvements merely because they can fit more shapes. Conversely, a systematically missed phase is a reason to reconsider the structure rather than to interpret every mismatch as random noise. Formal fit criteria can assist comparisons, but they do not determine whether a structural assumption is biologically or experimentally credible.
| Candidate structure | Evidence that may support it | Reason to reject or qualify it |
|---|---|---|
| One-compartment linear IV model | A single declining exponential adequately describes the observed IV profile. | Persistent early curvature or systematic phase-specific residuals may indicate unresolved distribution. |
| Two-compartment linear IV model | Reproducible fast and slow phases are supported by adequately timed IV samples. | Do not infer a second compartment from sparse or noisy curvature alone. |
| First-order oral absorption with linear elimination | A gradual rise and subsequent decline are compatible with the predicted profile. | A poor fit to delayed, multiple, or otherwise structured input may require an alternative; one oral curve may not identify all parameters. |
| Slow absorption with flip-flop kinetics | Oral terminal behavior is slower than independently supported IV disposition. | Without IV or other independent information, absorption and elimination explanations may be difficult to distinguish. |
Common Mistakes and Limitations
- Calling every terminal slope an elimination rate. That interpretation requires absorption to be faster than elimination. In flip-flop conditions, the oral terminal slope can instead reflect slow absorption.
- Adding compartments to explain every bend. A second compartment is not established by visual curvature alone. Sampling quality, assay error, and the timing of observations affect what the data can support.
- Choosing absorption and disposition independently. Oral concentration predictions depend on both input and disposition. A plausible absorption curve paired with an incompatible elimination model is not a coherent complete model.
- Treating a successful fit as mechanistic proof. Different structures can produce similar concentrations over a limited sampling window. A model is a useful representation conditional on its assumptions, not direct proof of a unique physiological explanation.
These cases also simplify real population analyses: they do not represent between-subject variability, residual error, covariates, or uncertainty in estimated parameters. Those features affect estimation and inference, but they do not remove the need to choose a structural model consistent with route, sampling, and concentration-time behavior.
Check Your Understanding
Answer each question, then compare your reasoning with the solutions.
- An IV profile has a reproducible fast decline followed by a slower decline. What structural alternative should be considered, and what evidence would still be needed?
- For oral dosing with \(k_a=1.2\) h\(^{-1}\) and \(k=0.3\) h\(^{-1}\), which rate governs the late terminal decline under the first-order model?
- An oral profile has a terminal slope consistent with \(0.08\) h\(^{-1}\), while an IV profile supports elimination at \(0.40\) h\(^{-1}\). Give one coherent interpretation.
- Why can an oral profile alone make it difficult to distinguish slow absorption from slow elimination?
- What is one diagnostic sign that a selected structural model may be inadequate?
Solutions. (1) Consider a two-compartment model, but require adequately timed, reproducible evidence of curvature and check that the more complex model improves the whole-profile description. (2) Since \(k_a>k\), elimination governs the late decline, at \(0.3\) h\(^{-1}\). (3) Slow absorption with \(k_a\) near \(0.08\) h\(^{-1}\) is a coherent flip-flop explanation, because IV data independently support faster elimination. (4) Both rates shape the oral curve, and when absorption is slow its terminal phase can resemble a slowly eliminated profile; independent disposition information helps separate them. (5) Systematic residual patterns across a phase, such as persistent early or terminal underprediction, suggest the model is missing structure.