Worked Low-Extraction Comparison Across Liver Models

Why Compare the Models at Low Extraction?

The three common liver models can produce different clearance predictions even when they use the same hepatic blood flow, blood unbound fraction, and intrinsic clearance. For a low-extraction compound, those predictions are often close in relative terms, but they are not identical. A worked comparison makes clear which differences arise from the model formulation, how large they are for a specified input set, and why reporting the inputs and units matters.

The comparison below uses a hypothetical compound and hypothetical liver parameters. The models are evaluated at the same conditions; differences are not attributed to different assays or subjects. The earlier material on low-extraction limits establishes that the predictions approach one another as extraction becomes small. Here the focus is the numerical comparison and interpretation of the remaining differences.

Definition: Let \(Q_h\) be hepatic blood flow in L/h, \(f_{u,b}\) the fraction of drug unbound in blood, and \(CL_{int}\) the intrinsic clearance referenced to unbound drug in blood, in L/h. Define the dimensionless capacity-to-flow ratio \(k=f_{u,b}CL_{int}/Q_h\). Hepatic extraction is \(E_h=CL_h/Q_h\), where \(CL_h\) is hepatic clearance in L/h. For a model that removes drug only in the liver, hepatic availability is \(F_h=1-E_h\).

Model Equations and Conditions

For a fair comparison, all three equations use the same \(Q_h\), \(f_{u,b}\), and \(CL_{int}\). The standard formulations assume linear metabolism over the concentration range considered, steady conditions, and no additional process such as active uptake or efflux that would change the effective availability of drug to the metabolizing tissue. In this comparison, \(CL_{int}\) is treated as an input rather than inferred separately for each model. The dispersion model also requires a specified dimensionless dispersion number, denoted \(D_N\).

\begin{aligned} CL_{h,\mathrm{WS}} &= \frac{Q_h f_{u,b}CL_{int}}{Q_h+f_{u,b}CL_{int}}, & E_{h,\mathrm{WS}} &= \frac{k}{1+k},\\ CL_{h,\mathrm{PT}} &= Q_h\left(1-e^{-k}\right), & E_{h,\mathrm{PT}} &= 1-e^{-k}. \end{aligned}

The subscripts WS and PT denote the well-stirred and parallel-tube models. In the well-stirred model, drug is assumed to mix rapidly within the liver, so the liver is represented by a uniform concentration. In the parallel-tube model, blood passes through parallel pathways with a concentration decline along each pathway. The dispersion model allows axial mixing during passage through the liver. One standard form for its extraction is

\begin{aligned} a &= \sqrt{1+4D_Nk},\\ E_{h,\mathrm{D}} &= 1- \frac{4a\exp\left(\frac{1-a}{2D_N}\right)} {(1+a)^2-(1-a)^2\exp\left(-\frac{a}{D_N}\right)},\\ CL_{h,\mathrm{D}} &= Q_h E_{h,\mathrm{D}}. \end{aligned}

Here \(D\) labels the dispersion formulation. \(D_N\) is dimensionless, not a clearance or a flow. The equation requires \(D_N>0\). In the limit of negligible dispersion, the dispersion prediction approaches the parallel-tube prediction; a finite \(D_N\) can leave a small difference. The precise value depends on both \(k\) and \(D_N\), so the dispersion number must be reported when presenting a dispersion-model result.

The link between extraction and clearance is \(CL_h=Q_hE_h\). Thus extraction is a fraction, whereas clearance has units of volume per time. Mixing the two quantities, or comparing clearances calculated with different flows without accounting for that change, can obscure what the models predict.

Worked Comparison With a Low-Extraction Compound

Worked Example: Calculate All Three Predictions

Suppose a hypothetical compound has \(Q_h=90.0\) L/h, \(f_{u,b}=0.100\), and \(CL_{int}=45.0\) L/h. For the dispersion calculation, use \(D_N=0.0500\). These are stipulated inputs for illustration, not observations from a real study.

First calculate the effective intrinsic clearance and dimensionless ratio:

f_{u,b}CL_{int}=0.100(45.0)=4.50\ \mathrm{L/h}, \qquad k=\frac{4.50\ \mathrm{L/h}}{90.0\ \mathrm{L/h}}=0.0500.

The ratio is small, so extraction is expected to be low. For the well-stirred model,

E_{h,\mathrm{WS}}=\frac{0.0500}{1+0.0500}=0.04762, \qquad CL_{h,\mathrm{WS}}=90.0(0.047619)=4.286\ \mathrm{L/h}.

For the parallel-tube model,

E_{h,\mathrm{PT}}=1-e^{-0.0500}=0.04877, \qquad CL_{h,\mathrm{PT}}=90.0(0.0487706)=4.389\ \mathrm{L/h}.

For the dispersion model, \(a=\sqrt{1+4(0.0500)(0.0500)}=\sqrt{1.0100}=1.00499\). Substitution into the dispersion equation gives \(E_{h,\mathrm{D}}\) approximately \(0.04866\), and therefore

CL_{h,\mathrm{D}}=90.0(0.048658)\approx4.379\ \mathrm{L/h}.

The extraction predictions range from about 0.04762 to 0.04877, or 4.762% to 4.877%. The corresponding clearances range from 4.286 to 4.389 L/h. These are all low-extraction predictions, but the maximum clearance difference is about \(4.3893518-4.2857143=0.1036375\) L/h, which rounds to 0.104 L/h. The difference is small relative to the flow of 90.0 L/h, yet it is not zero and may matter if model differences are being evaluated precisely.

Why the Predictions Converge, but Do Not Match Exactly

At low extraction, the effective intrinsic clearance is small relative to flow, so \(k\) is small. A first-order expansion makes the shared leading behavior explicit. For the well-stirred model, expanding \(1/(1+k)\) near \(k=0\) gives \(E_{h,\mathrm{WS}}=k+O(k^2)\). For the parallel-tube model, expanding the exponential gives \(1-e^{-k}=k+O(k^2)\). Substitution into the dispersion equation likewise gives \(E_{h,\mathrm{D}}=k+O(k^2)\) for fixed positive \(D_N\). Thus all three have the same leading low-extraction term, \(f_{u,b}CL_{int}/Q_h\); the distinctions appear in higher-order terms.

\begin{aligned} \frac{k}{1+k} &= k-k^2+O(k^3),\\ 1-e^{-k} &= k-\frac{k^2}{2}+O(k^3),\\ E_{h,\mathrm{D}} &= k+O(k^2). \end{aligned}

The notation \(O(k^2)\) indicates terms whose magnitude is of second order or smaller as \(k\) approaches zero. It does not say that the models are exactly equal at a nonzero \(k\). For the worked inputs, \(k=0.0500\), so higher-order terms are small but visible in the results. The well-stirred prediction is lower than the other two for this input set; the parallel-tube prediction is highest, and the dispersion prediction lies between them.

Justification: Because \(CL_h=Q_hE_h\), the shared first-order extraction term gives \(CL_h\approx Q_hk=f_{u,b}CL_{int}\) in the low-extraction limit. The leading-order clearance is therefore independent of flow, while model-specific differences remain in the higher-order terms. This limit is an approximation, not a replacement for evaluating the chosen model at the actual inputs.

Availability and Interpretation

Worked Example: Translate Extraction Into Hepatic Availability

Using the same hypothetical inputs, calculate \(F_h=1-E_h\) for each model. This interpretation assumes that hepatic availability is being considered in isolation: it does not include incomplete absorption or intestinal first-pass loss.

For the well-stirred prediction, \(F_{h,\mathrm{WS}}=1-0.047619=0.9524\). For the parallel-tube prediction, \(F_{h,\mathrm{PT}}=1-0.0487706=0.9512\). For the dispersion prediction, \(F_{h,\mathrm{D}}=1-0.048658\approx0.9513\).

The model-specific hepatic availabilities are therefore approximately 95.24%, 95.12%, and 95.13%, respectively. Their differences are the complements of the extraction differences: a model with greater extraction predicts lower hepatic availability. These values should not be described as total oral bioavailability unless absorption and other first-pass contributions are also accounted for.

Worked Example: Quantify Dispersion-Number Sensitivity

Keep \(Q_h=90.0\) L/h and \(k=0.0500\) fixed, but evaluate the dispersion model at \(D_N=0.0100\) and \(D_N=0.0500\). For \(D_N=0.0100\), the dispersion equation gives \(CL_{h,\mathrm{D}}\approx4.38724\) L/h. For \(D_N=0.0500\), it gives \(CL_{h,\mathrm{D}}\approx4.37923\) L/h.

The change between these two dispersion predictions is

4.38724-4.37923=0.00801\ \mathrm{L/h}.

At \(D_N=0.0500\), the difference from the parallel-tube prediction of approximately 4.38935 L/h is

4.38935-4.37923=0.01012\ \mathrm{L/h}.

The dispersion prediction at \(D_N=0.0500\) is about 0.00801 L/h below its prediction at \(D_N=0.0100\), and about 0.01012 L/h below the parallel-tube prediction. These are different comparisons: the first changes the dispersion number while keeping the model form, and the second compares two model formulations at the same \(k\) and flow. The values are close, but rounding them too early or conflating the comparisons can lead to an incorrect stated difference.

Checks, Comparisons, and Limitations

A model comparison is informative only if its inputs and assumptions are controlled. Before interpreting a difference, verify that \(Q_h\) is blood flow, \(f_{u,b}\) is the blood unbound fraction, and the intrinsic clearance uses a compatible blood-based convention. Confirm that the same values are used in each equation and that \(D_N\) is stated for the dispersion model. Recalculate \(E_h=CL_h/Q_h\) as a consistency check: an extraction estimate should be dimensionless and ordinarily lie between zero and one for a physically meaningful clearance between zero and flow.

A useful comparison is not a claim that one model is universally correct. The well-stirred model represents rapid mixing; the parallel-tube model represents flow through pathways with longitudinal concentration decline; and the dispersion model represents axial mixing. Their differences reflect structural assumptions. The example also fixes all inputs, so it does not incorporate parameter uncertainty or establish which model best describes a particular liver.

There are practical limits to the interpretation. If the input values are rounded, the last reported digits of a small model difference may not be meaningful. If \(f_{u,b}\), \(CL_{int}\), or flow is uncertain, that uncertainty may be as important as or larger than the structural difference calculated here. A comparison at a single low-extraction input set also does not establish the ordering or relative differences at other extraction levels.

A common mistake is to see that all three low-extraction equations have the same first-order term and conclude that their predictions are identical. The correct conclusion is narrower: the models converge to the same leading approximation as \(k\) tends to zero, while finite \(k\) leaves model-specific differences. Another mistake is to report a clearance difference as though it were an extraction difference. For example, 0.104 L/h is a difference in clearance; the corresponding extraction difference is \(0.104/90.0\), or about 0.00116, which is dimensionless.

Check the common inputs.
Confirm the flow, blood unbound fraction, intrinsic clearance, and their units before calculating any model prediction.

Calculate the shared ratio.
Find \(k=f_{u,b}CL_{int}/Q_h\), then use the model-specific equation without changing the inputs.

Compare like with like.
Report extraction as a fraction or percentage and clearance in L/h; identify separately any change in \(D_N\).

Interpret within scope.
Describe differences as consequences of the model formulations under the stated assumptions, not as proof that a model is correct for a given application.

Key takeaway: At low extraction, the three liver models share the leading approximation \(CL_h\approx f_{u,b}CL_{int}\), but finite extraction leaves small, calculable differences. Report the model, common inputs, dispersion number when relevant, and units so the comparison can be checked.

Check Your Understanding

Answer each question using the definitions and equations above. Full solutions follow the questions.

For \(Q_h=100\) L/h, \(f_{u,b}=0.0800\), and \(CL_{int}=50.0\) L/h, calculate \(k\) and the well-stirred extraction.
For the same inputs, calculate parallel-tube clearance to three significant figures.
If a model predicts \(CL_h=6.00\) L/h at \(Q_h=100\) L/h, what are extraction and hepatic availability?
Explain why a dispersion-model clearance difference should not be reported without its dispersion number.

Solutions. (1) \(f_{u,b}CL_{int}=0.0800(50.0)=4.00\) L/h, so \(k=4.00/100=0.0400\). The well-stirred extraction is \(0.0400/(1.0400)=0.03846\). (2) \(CL_{h,\mathrm{PT}}=100(1-e^{-0.0400})=100(0.03921)=3.921\) L/h, or 3.92 L/h to three significant figures. (3) \(E_h=6.00/100=0.0600\), and \(F_h=1-0.0600=0.940\), or 94.0%. (4) The dispersion equation depends on \(D_N\) as well as \(k\); different dispersion numbers can produce different predictions at otherwise identical inputs, so the number is required to identify and reproduce the calculation.