Worked High-Extraction Comparison Across Liver Models

Why Compare the Models at High Extraction?

When hepatic extraction is high, a compound is removed efficiently as blood passes through the liver. Consequently, predicted hepatic clearance is strongly constrained by hepatic blood flow. The three standard liver models express that constraint differently: the well-stirred model assumes complete mixing, the parallel-tube model assumes no axial mixing, and the dispersion model allows mixing along the flow path. Applying the same hypothetical inputs to all three makes the influence of those assumptions visible.

The calculations below use hepatic blood flow \(Q_H\), the unbound fraction in blood \(f_{u,B}\), and intrinsic clearance \(CL_{int}\). The product \(f_{u,B}CL_{int}\) is the effective intrinsic clearance in the blood-based hepatic equations. The extraction ratio \(E_H\) is the fraction of drug entering the liver that is removed during one pass; hepatic clearance is \(CL_H=Q_HE_H\). Clearances here are expressed on a blood basis.

The models assume steady hepatic blood flow, linear (nonsaturable) metabolism, and a fixed intrinsic clearance and unbound fraction during the calculation. They also describe a liver without an explicit shunt or separate parallel metabolic pathways. Model-specific assumptions about mixing are stated with each formulation. These are predictions under specified inputs, not universal rankings of the models.

The Three Formulations

For the well-stirred model, blood is treated as completely mixed within the liver. The outlet concentration is determined by a single well-mixed liver concentration. For the parallel-tube model, blood moves through equivalent tubes with no mixing along their length, so concentration declines continuously through each tube. Both models use the same dimensionless capacity-to-flow ratio:

With \(a\) defined this way, the well-stirred extraction and clearance are \(a/(1+a)\) and \(Q_Ha/(1+a)\). For the parallel-tube model, extraction is \(1-\exp(-a)\), and clearance is \(Q_H[1-\exp(-a)]\). The exponential form results from first-order loss as blood passes through a tube.

The dispersion model allows axial mixing as well as flow. One standard steady-state formulation uses the dimensionless dispersion number \(D_N\), with larger \(D_N\) representing more axial dispersion. Let \(s\) run from 0 at the inlet to 1 at the outlet, and let \(c(s)\) be the drug concentration divided by the inlet concentration. For this formulation, the concentration profile satisfies:

The inlet condition specifies the normalized total entering flux, and the outlet condition specifies zero concentration gradient at the exit. Once \(c(1)\) is found, extraction is \(E_H=1-c(1)\), and clearance is \(Q_HE_H\). For a numerical solution, the differential equation has roots \(r_+\) and \(r_-\), where \(r_{\pm}=[1\pm\sqrt{1+4D_Na}]/(2D_N)\). Writing \(c(s)=A\exp(r_+s)+B\exp(r_-s)\), the outlet condition determines \(R=B/A=-(r_+/r_-)\exp(r_+-r_-)\); the inlet condition then determines \(A\). This makes the dispersion calculation explicit rather than treating its output as an unexplained adjustment to another model.

Worked Comparison Using the Same Inputs

Why the Predictions Approach Flow

For the well-stirred model, dividing clearance by flow gives \(CL_H/Q_H=a/(1+a)\). The difference between flow and clearance, expressed as a fraction of flow, is therefore \(1-CL_H/Q_H=1/(1+a)\). As \(a\) increases, this fractional shortfall decreases and the clearance tends to \(Q_H\). The parallel-tube model instead has a fractional shortfall of \(\exp(-a)\), which also tends to zero as \(a\) increases. These limits explain why high extraction makes clearance flow-limited: once intrinsic capacity is ample, increasing capacity has little room to raise clearance above the amount of drug delivered to the liver by blood flow.

The dispersion model shares the physical upper bound \(CL_H\leq Q_H\) under the stated assumptions, because extraction cannot exceed the fraction of incoming drug removed in one pass. Its approach to that bound depends on both \(a\) and \(D_N\): the axial mixing assumption affects how concentrations and drug exposure are distributed within the liver. Thus, equal \(Q_H\), \(f_{u,B}\), and \(CL_{int}\) do not guarantee equal predictions when the models represent different concentration patterns.

Assumptions, Diagnostics, and Limits

A useful model comparison begins by checking the shared inputs before interpreting the differences. Flow must be in units compatible with clearance, \(f_{u,B}\) must be a fraction rather than a percentage, and \(CL_{int}\) and \(Q_H\) must use the same time and volume units. The product \(f_{u,B}CL_{int}\) must have clearance units, making \(a\) dimensionless. A value of \(a\) that is not dimensionless signals a unit or definition error.

Then check each output: \(0\leq E_H\leq1\), \(CL_H=Q_HE_H\), and \(CL_H\leq Q_H\). If any formulation yields extraction above 1 or clearance above flow under these assumptions, inspect the equation, units, and numerical calculation. At extreme extraction, compute the small residual fraction carefully: subtracting a number very close to 1 from 1 can lose numerical precision. In particular, retain \(\exp(-a)\) directly for the parallel-tube flow shortfall rather than rounding the extraction ratio prematurely.

The model choice is not resolved merely by selecting the largest or smallest clearance. The well-stirred model is a compact representation of complete mixing; the parallel-tube model represents flow without axial mixing; and the dispersion model adds a mixing parameter. Data or a justified modeling context are needed to decide which representation is suitable. The dispersion prediction also depends on the chosen \(D_N\), which must be defined consistently with the formulation being used.

The comparison has important limitations. It assumes constant flow, binding, and intrinsic clearance, and it does not represent saturation, time-varying physiology, shunting, or separate uptake and metabolism processes. A high extraction prediction does not by itself establish that real hepatic clearance is controlled only by flow: changes in binding, enzyme activity, or the model’s assumptions can matter. The calculations are conditional model predictions, not measurements of a real compound.

Common Pitfall

A tempting but incorrect argument is: “The drug has high extraction, so hepatic clearance equals hepatic blood flow in every model.” High extraction means clearance approaches flow, not that it must equal flow exactly. In the worked comparison, the well-stirred prediction is 9.00 L/h below flow, the dispersion prediction is 1.90 L/h below flow, and the parallel-tube prediction is 0.0111 L/h below flow. Reporting all three as exactly 90.0 L/h erases meaningful differences and contradicts the computed extraction ratios.

A second check is to keep extraction and clearance distinct: \(E_H\) is a dimensionless fraction, whereas \(CL_H\) has units of volume per time. Multiplying the extraction ratio by flow is required to obtain clearance.