Why Structural Choice Matters
A model-based formulation comparison uses predicted concentrations, or summaries derived from those predictions, to describe exposure over time. Those predictions depend on the assumed structure of the drug’s movement through the body. A one-compartment model treats the body as one well-mixed distribution space; a two-compartment model allows a central space to exchange drug with a peripheral space. If the selected structure cannot represent the observed concentration-time shape, the resulting formulation comparison may reflect model misspecification rather than a meaningful formulation difference.
The goal is not to select the most elaborate model by default. It is to choose a structure that is pharmacologically interpretable, consistent with the sampling information, and adequate for the concentration profiles being compared. The equations below describe oral dosing with first-order absorption and linear disposition: concentration is proportional to dose, and disposition parameters do not change with concentration.
One-Compartment Disposition With First-Order Absorption
Assume an oral dose \(D\) is absorbed into a single well-mixed compartment. The fraction reaching systemic circulation is \(F\), the apparent distribution volume is \(V\), the first-order absorption-rate constant is \(k_a\), and the elimination-rate constant is \(k_e\). First-order absorption means that the rate of drug entry is proportional to the amount remaining at the absorption site. First-order elimination means that the rate of removal is proportional to the amount in the compartment. Assume \(k_a\ne k_e\), parameters are constant over time, and there is no lag time or other input.
The amount entering the compartment per unit time is \(F D k_a e^{-k_a s}\) at time \(s\). Drug entering at time \(s\) is then reduced by elimination over the remaining interval \(t-s\), giving a survival factor \(e^{-k_e(t-s)}\). Integrating contributions from all entry times between 0 and \(t\), then dividing the amount by \(V\), gives:
Here \(C(t)\) is the predicted concentration at time \(t\). The expression is positive for either ordering of \(k_a\) and \(k_e\): the sign of the denominator changes together with the sign of the difference of exponentials. The parameters \(k_a\) and \(k_e\) have units of inverse time, so the predicted concentration has units of dose divided by volume.
Worked Example: A One-Compartment Oral Profile
Suppose a hypothetical oral dose is \(D=100\) mg, with \(F=0.80\), \(V=20\) L, \(k_a=1.2\) h\(^{-1}\), and \(k_e=0.20\) h\(^{-1}\). Assume the model conditions above are reasonable and calculate the predicted concentration at \(t=2\) h.
The absorption and elimination rates are positive and unequal, and \(k_a>k_e\), so the displayed expression applies. Substitution gives:
The coefficient is \(4.8\) mg/L, and the exponential difference is \(e^{-0.4}-e^{-2.4}=0.670320-0.0907180=0.579602\). Therefore \(C(2)=4.8(0.579602)=2.78209\) mg/L, or approximately 2.782 mg/L.
As a shape check, the model’s peak time is \(t_{\max}=\ln(k_a/k_e)/(k_a-k_e)\). For these values, \(t_{\max}=\ln(6)/(1.0)=1.79176\) h, so a concentration measured at 2 h should be just past the predicted peak. This check is consistent with the calculation, but it does not validate the model against observed data.
Two-Compartment Disposition With Oral Absorption
A two-compartment disposition model allows drug in a central compartment to exchange with a peripheral compartment. After an instantaneous intravenous input, its central concentration can be represented by two exponential terms, \(A e^{-\alpha t}+B e^{-\beta t}\), where \(\alpha>\beta>0\) are disposition rate constants and \(A\) and \(B\) are coefficients with units of inverse volume per dose unit if the dose has not been included in the response. For the oral formula below, define \(A\) and \(B\) as the coefficients in the dose-normalized intravenous concentration response; they therefore have units of inverse volume. In the standard two-compartment model, the faster exponential is usually associated with distribution and the slower one with the terminal phase.
Assume the same linear, first-order absorption process as before. Each of the two disposition exponentials receives input over time. Convolving the absorption input with each exponential produces:
This formula assumes \(k_a\ne\alpha\) and \(k_a\ne\beta\). If \(k_a\) equals one of the disposition rates, the corresponding fraction has the continuous limit \(t e^{-k_a t}\) (with that shared rate in the exponential). For example, the full contribution associated with coefficient \(A\) when \(\alpha=k_a\) is \(F D k_a A\,t e^{-k_a t}\). The fraction itself has units of time; the outside factor supplies the remaining units so the contribution is a concentration.
The choice between one and two compartments is a choice about disposition shape, not absorption order. Both structures can have first-order absorption. A two-compartment profile can describe an early rapid change followed by a slower decline; a one-compartment profile has only one disposition exponential after input. The observed oral curve can nevertheless look more complex than either description if the absorption process is misspecified or if the data do not cover the relevant time regions.
Worked Example: Concentration in a Two-Compartment Model
Consider a hypothetical model with \(D=4\) mg, \(F=0.50\), \(k_a=1.2\) h\(^{-1}\), \(\alpha=2.0\) h\(^{-1}\), \(\beta=0.10\) h\(^{-1}\), \(A=0.30\) L\(^{-1}\), and \(B=0.70\) L\(^{-1}\). The rate constants are positive, \(\alpha>\beta\), and the absorption rate differs from both disposition rates. Calculate \(C(12)\).
The common multiplier is \(F D k_a=(0.50)(4\ \mathrm{mg})(1.2\ \mathrm{h}^{-1})=2.4\) mg/h. The faster disposition contribution inside the brackets is:
The slower contribution is:
Thus the bracketed sum is approximately \(0.191668898\ \mathrm{L}^{-1}\mathrm{h}\), and \(C(12)\approx 0.460005356\) mg/L, or approximately 0.4600 mg/L. The slow term dominates at 12 h because the fast exponential has almost vanished. This is a feature of the chosen parameter values, not a general guarantee that every two-compartment oral profile has a visible terminal phase in the sampled data.
Choosing Between Structures
Structural choice should be guided by the intended use of the model and by the information in the observed concentration-time data. A two-compartment model may be warranted when the data show systematic curvature that a one-compartment model cannot capture, particularly when early and later disposition phases are both sampled. A more complex model also estimates more parameters, so apparent improvement in fit is not sufficient by itself. The model should have a coherent pharmacological interpretation, and the available sampling should provide information about the features it is intended to represent.
| Feature | One-compartment structure | Two-compartment structure |
|---|---|---|
| Disposition shape | One exponential disposition phase | Two exponential disposition phases |
| Parameter burden | Fewer disposition parameters | More disposition parameters and exchange structure |
| Useful when | A single disposition phase adequately describes the sampled profile | Data support distinct rapid and slow disposition behavior |
| Main risk | Systematic curvature may be left in residuals | Additional parameters may be poorly supported or overfit sparse data |
A practical review examines plots of observed concentrations and fitted predictions on both the concentration and log-concentration scales, residuals over time, and whether the early, peak, and terminal regions are represented. Residuals should not show a persistent time-related pattern. A curved trend in residuals, such as repeated underprediction early and overprediction later, suggests that the structural shape may be inadequate. A few large residuals may instead reflect measurement error, an unusual observation, or a data problem; they do not automatically justify another compartment.
The assumptions used for formulation comparisons also matter. The model assumes the same basic disposition structure is appropriate for the formulations under comparison, and that concentrations arise from the specified absorption and disposition processes. The comparison may later represent formulation-related parameters or effects, but structural choice alone does not define those effects. In particular, choosing a two-compartment structure does not establish that its parameters can all be reliably separated from one another using the available oral data.
Worked Example: Checking Residual Patterns Before Adding a Compartment
Suppose a hypothetical concentration dataset is fitted with both structures. The table gives observed concentrations and predictions from the respective fitted models. Concentrations are in mg/L; the fitted predictions are illustrative outputs of the fits.
| Time (h) | Observed | One-compartment prediction | Two-compartment prediction |
|---|---|---|---|
| 1 | 2.30 | 2.45 | 2.32 |
| 4 | 1.20 | 0.91 | 1.16 |
| 8 | 0.65 | 0.25 | 0.68 |
| 12 | 0.46 | 0.07 | 0.46 |
For a simple numerical check, compute the residual sum of squares, \(RSS=\sum_i (C_i-\widehat{C}_i)^2\), where \(C_i\) is observed concentration and \(\widehat{C}_i\) is its fitted prediction. The one-compartment residuals are \(-0.15,\ 0.29,\ 0.40,\ 0.39\) mg/L, so:
The two-compartment residuals are \(-0.02,\ 0.04,\ -0.03,\ 0.00\) mg/L, giving:
Here the one-compartment fit underpredicts all three later concentrations, with the largest deficit at 8 h, while the two-compartment predictions track the four values more closely. This is evidence to investigate the two-compartment structure, not proof that it is the correct model: the more flexible structure can reduce residual sum of squares simply by using additional parameters. A defensible choice also considers residual diagnostics, parameter plausibility, sampling adequacy, and the purpose of the comparison.
Derivation and Shape Checks
The one-compartment peak-time formula follows directly from the concentration equation. Let \(K=F D k_a/[V(k_a-k_e)]\), so \(C(t)=K(e^{-k_e t}-e^{-k_a t})\). Differentiating and setting the derivative equal to zero at an interior peak gives:
Taking logarithms yields \(\ln(k_a/k_e)=(k_a-k_e)t\), and therefore \(t_{\max}=\ln(k_a/k_e)/(k_a-k_e)\). For positive, unequal rates, this time is positive. This result is a consequence of the assumed one-compartment, first-order model; it is not a general formula for the peak time of a two-compartment oral profile.
Several checks help catch errors in calculations and model use. At \(t=0\), both oral model expressions predict \(C(0)=0\) when there is no lag time or instantaneous input. The concentration should remain nonnegative for physically meaningful parameters and coefficients. The units must also be consistent: rate constants are inverse time, and the final concentration must have concentration units. A negative predicted concentration or a time unit mismatch commonly signals a sign, denominator, or unit-conversion error.
Common Pitfalls and Limitations
- Treating every curved profile as evidence for two compartments. Curvature can arise from absorption behavior, sparse or noisy measurements, or other departures from assumptions. Check the data and residual patterns before changing disposition structure.
- Comparing models only by residual sum of squares. Adding parameters generally improves fit. Evaluate whether the added structure is supported and useful, rather than treating a lower RSS as decisive.
- Confusing absorption and disposition rates. \(k_a\) describes entry from the absorption site under this model; \(\alpha\) and \(\beta\) describe disposition phases. A terminal slope in oral data does not by itself reveal which process governs the observed decline.
- Using the oral model outside its assumptions. Nonlinear disposition, delayed or multiple absorption inputs, and time-varying processes are not represented by these equations. A richer compartment structure does not automatically repair an incorrect absorption assumption.
- Assuming a structural model proves a formulation conclusion. The model specifies a concentration-time shape. Equivalence conclusions require appropriately defined formulation comparisons and uncertainty assessment, not merely a plausible fitted curve.
These models are useful because they make assumptions explicit and generate interpretable concentration-time profiles. Their limitations are equally important: a good visual fit does not guarantee that each parameter is well determined, and the chosen structure can influence predicted concentrations outside the times that were observed. Structural choice should therefore be made with the comparison’s intended exposure summaries in mind, without treating a more complicated model as inherently more realistic.
Check Your Understanding
Answer each question using the model assumptions and equations in this tutorial. Brief solutions follow.
- In the one-compartment model, what does \(k_a\) describe, and what does \(k_e\) describe?
- For \(k_a=0.8\) h\(^{-1}\) and \(k_e=0.2\) h\(^{-1}\), calculate the one-compartment peak time.
- Why can a two-compartment model have a smaller residual sum of squares without being the preferred structure?
- What does \(C(0)=0\) represent in the oral equations, and which assumption could make that statement inappropriate?
Solutions. (1) \(k_a\) describes first-order absorption into the compartment; \(k_e\) describes first-order elimination from it. (2) \(t_{\max}=\ln(0.8/0.2)/(0.8-0.2)=\ln(4)/0.6=2.310\) h. (3) The two-compartment structure has additional parameters and can improve fit through flexibility; residuals, plausibility, sampling support, and purpose must also be considered. (4) With no lag time or instantaneous input, no drug has entered the systemic compartment at time zero. A model that includes an instantaneous input or uses a different time origin need not predict zero at the stated time zero.