Why Low Extraction Is a Useful Comparison
When hepatic extraction is small, most drug entering the liver leaves it in the outflow. The fractional amount removed is therefore small, and the three standard liver models approach a common clearance prediction. This gives a useful check on calculations and a way to understand which differences between models matter at low extraction. The models do not, however, become identical: their predictions begin to differ at higher order.
The comparison here holds the model inputs fixed: hepatic blood flow, \(Q\), and unbound intrinsic clearance, \(f_u CL_{int}\). Both have units of volume per time, so their ratio is dimensionless. The clearance quantities are on the same blood basis. The introductory material on extraction ratio defines extraction as the fraction removed in one passage; here, the focus is on how that fraction behaves in the low-extraction limit.
Model Equations and Their Shared First-Order Limit
Under the well-stirred model, the liver is treated as a single, uniformly mixed compartment. Under the parallel-tube model, blood passes through parallel channels with concentration declining along each channel. The dispersion model allows axial mixing as blood travels through the liver; its dimensionless dispersion number is denoted \(D_N\). For this comparison, all three models assume steady conditions, linear (unsaturated) elimination, the same \(Q\) and \(f_u CL_{int}\), and no change in these inputs across the liver passage. They describe different mixing patterns, not different clearance inputs.
The corresponding extraction ratios are as follows. The dispersion expression uses \(D_N>0\); its parameter \(a\) is defined in the equation.
In each case, \(CL_H=QE\). For small \(x\), expansion of the well-stirred expression gives \(E_{\mathrm{WS}}=x-x^2+O(x^3)\). Expansion of the exponential gives \(E_{\mathrm{PT}}=x-x^2/2+O(x^3)\). Thus both models have \(E\approx x\) at first order, but the second-order terms are different. Here \(O(x^3)\) denotes terms whose size is of third or higher order as \(x\) approaches zero.
Worked Example: First-Order Agreement and a Correctly Calculated Difference
Suppose, hypothetically, that \(Q=100\ \mathrm{L/h}\) and \(f_u CL_{int}=2.00\ \mathrm{L/h}\). Then \(x=2.00/100=0.0200\), a small dimensionless ratio. The common first-order estimate is \(CL_H\approx Qx=2.00\ \mathrm{L/h}\).
For the well-stirred model, \(CL_H=Qx/(1+x)=100(0.0200)/(1.0200)=1.960784\ \mathrm{L/h}\), or \(1.961\ \mathrm{L/h}\) to four significant figures.
For the parallel-tube model, \(CL_H=Q[1-\exp(-x)]=100[1-\exp(-0.0200)]=1.980133\ \mathrm{L/h}\), or \(1.980\ \mathrm{L/h}\).
Using the unrounded predictions, their difference is approximately \(1.9801327-1.9607843=0.01935\ \mathrm{L/h}\). This is about \(100(0.0193484/2.00)=0.9674\%\) of the shared first-order value. Both predictions are close to \(2.00\ \mathrm{L/h}\), while their small difference is consistent with their different second-order terms.
Dispersion Model: Low-Extraction Expansion
For the dispersion model, write the outflow-to-inflow concentration ratio as \(R=1-E_{\mathrm{D}}\). Rewriting the exact expression above gives a form convenient for expansion. The following steps show why the dispersion model has the same first-order limit and how its second-order term depends on \(D_N\).
Using \(a=\sqrt{1+4D_Nx}\), the exact expression becomes \(R=\dfrac{4a\exp((1-a)/(2D_N))}{(1+a)^2-(1-a)^2\exp(-a/D_N)}\).
For fixed \(D_N>0\) and small \(x\), \(a=1+2D_Nx-2D_N^2x^2+O(x^3)\). Substituting this into the logarithm of the ratio and retaining terms through \(x^2\) gives \(\log R=-x+[D_N-D_N^2(1-\exp(-1/D_N))]x^2+O(x^3)\).
Since \(R=\exp(\log R)\), use \(\exp(y)=1+y+y^2/2+O(y^3)\). The \(y^2/2\) term contributes \(x^2/2\), so \(E_{\mathrm{D}}=1-R=x-c(D_N)x^2+O(x^3)\), where \(c(D_N)=1/2+D_N-D_N^2(1-\exp(-1/D_N))\).
Consequently, the dispersion-model clearance is \(CL_H=Q[x-c(D_N)x^2+O(x^3)]\). Its first-order term is \(Qx=f_u CL_{int}\), the same as for the other two models. This expansion is for fixed \(D_N>0\) as \(x\) becomes small; it is not a claim that the truncated expression is exact at finite \(x\).
What the Second-Order Terms Show
For the well-stirred model, the coefficient of \(x^2\) in extraction is \(-1\); for the parallel-tube model it is \(-1/2\). In the dispersion model it is \(-c(D_N)\). As \(D_N\) approaches zero, \(c(D_N)\) approaches \(1/2\), matching the parallel-tube expansion. As \(D_N\) becomes large, \(c(D_N)\) approaches \(1\), matching the well-stirred expansion. These are limits of the second-order coefficient; at any fixed finite \(D_N\), the dispersion model retains its own prediction.
The coefficients also explain the direction of the leading difference. For small positive \(x\), the smaller second-order subtraction in the parallel-tube model gives a somewhat larger extraction than the well-stirred model. The dispersion model’s position at this order depends on \(D_N\). The differences are of order \(x^2\), whereas the shared clearance term is of order \(x\); their relative size therefore tends to zero as \(x\to0\).
Worked Example: Comparing All Three Models at Low Extraction
Consider hypothetical inputs \(Q=100\ \mathrm{L/h}\), \(f_u CL_{int}=2.00\ \mathrm{L/h}\), and \(D_N=0.500\). Then \(x=0.0200\). For the dispersion expansion, \(c(0.500)=0.5+0.5-0.25(1-\exp(-2))=0.783834\).
Using terms through \(x^2\), the dispersion extraction is \(E_{\mathrm{D}}\approx0.0200-0.783834(0.0200)^2=0.0196865\), giving \(CL_H\approx100(0.0196865)=1.96865\ \mathrm{L/h}\). This is a second-order approximation, not the exact evaluation of the dispersion equation.
The first-order estimate is \(2.00\ \mathrm{L/h}\). The well-stirred and parallel-tube predictions calculated in the previous example are \(1.96078\ \mathrm{L/h}\) and \(1.98013\ \mathrm{L/h}\), respectively. The second-order dispersion estimate falls between them for this \(D_N\). The three are close because \(x\) is small, but their differences are visible at order \(x^2\).
Worked Example: How Smaller Extraction Improves Convergence
Suppose a second hypothetical case has \(Q=80.0\ \mathrm{L/h}\), \(f_u CL_{int}=0.400\ \mathrm{L/h}\), and \(D_N=0.200\). Then \(x=0.400/80.0=0.00500\), smaller than in the preceding example. The shared first-order clearance is \(Qx=0.400\ \mathrm{L/h}\).
The well-stirred prediction is \(80.0(0.00500)/(1.00500)=0.398010\ \mathrm{L/h}\). The parallel-tube prediction is \(80.0[1-\exp(-0.00500)]=0.399002\ \mathrm{L/h}\).
For the dispersion model, \(c(0.200)=0.5+0.2-0.04(1-\exp(-5))=0.660270\). Its second-order estimate is \(80.0[0.00500-0.660270(0.00500)^2]=0.398679\ \mathrm{L/h}\). All three estimates lie close to \(0.400\ \mathrm{L/h}\). The expansions show why: reducing \(x\) reduces the second-order corrections relative to the common first-order term.
Checking Whether the Approximation Is Appropriate
A low-extraction approximation is an asymptotic statement, not a universal numerical cutoff. Check the dimensionless ratio \(x=f_u CL_{int}/Q\) before treating \(CL_H\approx f_u CL_{int}\) as adequate. Then compare the approximation with the exact model equation if the desired precision requires it. A discrepancy larger than the acceptable modelling tolerance is a practical sign that first order alone is insufficient.
Second-order terms help diagnose the likely scale of the discrepancy: they are proportional to \(x^2\), and their coefficients differ by model. But a second-order estimate is not a substitute for the exact model when \(x\) is not small or when higher-order terms matter. The dispersion expansion also presumes a specified, fixed \(D_N\); changing \(D_N\) while taking the low-\(x\) limit is a different limiting problem.
The main alternatives are to use the first-order common approximation for a quick low-extraction comparison, use a second-order expansion to understand leading model differences, or evaluate each model’s exact expression for quantitative predictions. The first is simplest but omits model-specific corrections; the second clarifies those corrections but truncates higher-order terms; the exact expressions avoid that truncation but do not remove the assumptions built into each model.
Common Mistake and Practical Limits
- Using the common limit as if it were exact. The approximation \(CL_H\approx f_u CL_{int}\) is first order. At finite \(x\), each model has a different second-order term, so exact predictions need not agree.
- Comparing rounded values to report a more precise difference. Subtracting rounded clearances may not reproduce the difference calculated from unrounded predictions. Keep sufficient precision through the calculation, then round the reported result.
- Mixing inputs or clearance bases. A comparison of model equations is informative only when \(Q\), \(f_u CL_{int}\), units, and blood basis are consistent.
- Ignoring model assumptions. The formulas rely on steady conditions and linear elimination. If elimination is saturated or inputs vary during the passage, these expressions may not describe the situation.
The low-extraction comparison is therefore best used as a controlled limit and a calculation check. It establishes convergence when the shared ratio \(x\) is small; it does not establish which model is most suitable for a particular liver or drug, nor does it justify using a truncated approximation outside its regime.
Check Your Understanding
Answer each question, then compare your reasoning with the concise solutions below.
- For \(Q=50.0\ \mathrm{L/h}\) and \(f_u CL_{int}=0.250\ \mathrm{L/h}\), calculate \(x\) and the common first-order clearance.
- Why do the well-stirred and parallel-tube models agree to first order but differ at second order?
- What is the second-order coefficient \(c(D_N)\) in the dispersion extraction expansion when \(D_N=0.500\), to four significant figures?
- At fixed \(D_N\), what happens to the relative importance of model-specific corrections as \(x\to0\)?
- A learner reports a clearance difference by subtracting two values rounded to three significant figures, then gives more digits for the difference. What should be done instead?
Solutions. (1) \(x=0.250/50.0=0.00500\), and \(Qx=0.250\ \mathrm{L/h}\). (2) Both extraction expansions begin with \(x\), but their \(x^2\) coefficients are \(-1\) and \(-1/2\), respectively. (3) \(c(0.500)=0.5+0.5-0.25(1-\exp(-2))=0.7838\). (4) The leading model-specific corrections are order \(x^2\), while the common term is order \(x\), so their relative importance tends to zero. (5) Subtract unrounded predictions, retain adequate precision in intermediate calculations, and round the difference only for reporting.