Intrinsic-Clearance Sensitivity Across Liver Models

Why Sensitivity to Intrinsic Clearance Matters

A hepatic clearance model does more than predict a clearance value: it also describes how that value responds when the liver’s intrinsic ability to eliminate a drug changes. This response is important when assessing an enzyme perturbation, interpreting uncertainty in an intrinsic-clearance estimate, or comparing model predictions. The response is not generally proportional across the full range of extraction. As extraction rises, hepatic clearance approaches hepatic blood flow, and further increases in intrinsic clearance have diminishing effects.

Here, the three model formulations are compared with hepatic blood flow \(Q\), unbound fraction \(f_u\), and intrinsic clearance \(CL_{int}\) held fixed except when explicitly varying \(CL_{int}\). The focus is the mathematical sensitivity to \(CL_{int}\), not the separate effects of changing \(f_u\) or \(Q\).

Model Inputs and Sensitivity Measures

Let \(x=f_uCL_{int}/Q\) be the dimensionless ratio of unbound intrinsic clearance to hepatic blood flow. Let \(E=CL_h/Q\) be the hepatic extraction ratio, where \(CL_h\) is hepatic clearance. Each model can be written \(CL_h=Q E(x)\), with a different function \(E(x)\). For the dispersion model, \(D_N\) denotes the dimensionless dispersion number.

Definition: The local sensitivity of hepatic clearance to intrinsic clearance is the derivative \(\partial CL_h/\partial CL_{int}\), evaluated with \(Q\), \(f_u\), and, for the dispersion model, \(D_N\) held constant. The elasticity is the proportional response \(S_{CL_{int}}=(CL_{int}/CL_h)(\partial CL_h/\partial CL_{int})\). For a small relative change in \(CL_{int}\), elasticity approximates the relative change in \(CL_h\): \(\Delta CL_h/CL_h \approx S_{CL_{int}}(\Delta CL_{int}/CL_{int})\).

The derivative is dimensionless when both clearances use the same volume-per-time units: it gives the change in hepatic clearance per equal absolute change in intrinsic clearance. Elasticity is also dimensionless, but has a different interpretation: it describes proportional change. A value of 0.7, for example, means that a small 1% increase in intrinsic clearance corresponds approximately to a 0.7% increase in hepatic clearance, under the stated fixed-input assumptions.

The formulas below assume steady-state liver-model relationships, positive flow and intrinsic clearance, constant \(f_u\), and linear elimination represented by the model. They do not describe time-varying enzyme activity, saturation, or a simultaneous change in flow or binding. For the dispersion model, \(D_N>0\); the limit \(D_N\to0\) recovers the parallel-tube relationship.

Deriving the Three Sensitivity Expressions

The well-stirred, parallel-tube, and dispersion models express extraction as follows. The dispersion expression uses the usual steady-state outflow-to-inflow relationship.

\begin{aligned} E_{WS}(x)&=\frac{x}{1+x},\\ E_{PT}(x)&=1-e^{-x},\\ E_{D}(x)&=1-\exp\left(\frac{1-\sqrt{1+4D_Nx}}{2D_N}\right). \end{aligned}

Because \(CL_h=QE(x)\) and \(x=f_uCL_{int}/Q\), the chain rule gives \(\partial CL_h/\partial CL_{int}=f_u E'(x)\). The factor \(Q\) cancels the \(1/Q\) in the derivative of \(x\). Elasticity then follows by multiplying this local sensitivity by \(CL_{int}/CL_h=x/(f_uE)\), giving \(S_{CL_{int}}=xE'(x)/E(x)\).

\begin{aligned} \frac{\partial CL_{h,WS}}{\partial CL_{int}} &=\frac{f_u}{(1+x)^2}, & S_{CL_{int},WS} &=\frac{1}{1+x},\\[4pt] \frac{\partial CL_{h,PT}}{\partial CL_{int}} &=f_u e^{-x}, & S_{CL_{int},PT} &=\frac{x}{e^x-1}. \end{aligned}

For the dispersion model, define \(g(x)=(1-\sqrt{1+4D_Nx})/(2D_N)\), so \(E_D=1-e^{g(x)}\). Differentiating the exponent gives \(g'(x)=-1/\sqrt{1+4D_Nx}\). Consequently, \(E_D'(x)=e^{g(x)}/\sqrt{1+4D_Nx}\), which yields:

\frac{\partial CL_{h,D}}{\partial CL_{int}} = f_u\frac{e^{g(x)}}{\sqrt{1+4D_Nx}}, \qquad S_{CL_{int},D} = \frac{x e^{g(x)}}{\sqrt{1+4D_Nx}\,[1-e^{g(x)}]}.

These are local measures. They describe the slope at the current parameter value, not the exact result of a large change. For finite changes, evaluate the clearance function at both the initial and changed values of \(CL_{int}\).

Worked Example: Calculating Sensitivity at Intermediate Extraction

Suppose a hypothetical liver has \(Q=90\) L/h, \(f_u=0.10\), and \(CL_{int}=450\) L/h. For the dispersion calculation, take \(D_N=0.50\). Calculate \(x\), hepatic clearance, local sensitivity, and elasticity under all three models.

First, \(x=(0.10)(450)/90=0.5000\). Under the well-stirred model, \(E=0.5/1.5=0.3333\), so \(CL_h=90(0.3333)=30.00\) L/h. Its local sensitivity is \(0.10/(1.5)^2=0.04444\), and its elasticity is \(1/(1+0.5)=1/1.5=0.6667\). Thus, near these inputs, a 1% increase in \(CL_{int}\) corresponds to about a 0.667% increase in \(CL_h\).

Under the parallel-tube model, \(E=1-e^{-0.5}=0.3935\), giving \(CL_h=90(0.3935)=35.41\) L/h. The local sensitivity is \(0.10e^{-0.5}=0.06065\). The elasticity is \(0.5/(e^{0.5}-1)=0.7707\).

For the dispersion model, \(g=(1-\sqrt{1+4(0.50)(0.50)})/[2(0.50)]=1-\sqrt{2}=-0.4142\). Thus \(e^g=0.6609\), \(E=1-0.6609=0.3391\), and \(CL_h=90(0.3391)=30.52\) L/h. The local sensitivity is \(0.10(0.6609)/\sqrt{2}=0.04673\). Its elasticity is \(0.5(0.6609)/[\sqrt{2}(0.3391)]=0.6890\).

At this common value of \(x\), the parallel-tube model predicts the largest clearance and the largest local and proportional response. The well-stirred and dispersion predictions are closer, but not identical. The model-specific equations, rather than a single universal sensitivity, determine the results.

How Sensitivity Changes Across Extraction Regimes

At low extraction, \(x\) is small. Expanding the extraction functions around zero shows that each has the same leading term, \(E(x)\approx x\). Therefore \(CL_h=QE(x)\approx Qx=f_uCL_{int}\): hepatic clearance is approximately proportional to intrinsic clearance, and elasticity approaches 1. The models differ in how quickly their slopes depart from this low-extraction limit. For example, the local sensitivity factors \(E'(x)\) begin as \(1-2x\) for the well-stirred model, \(1-x\) for the parallel-tube model, and \(1-(1+2D_N)x\) for the dispersion model. These first-order expressions are approximations for small \(x\), not exact formulas at moderate extraction.

As \(x\) increases, extraction and clearance rise, but the clearance approaches the flow ceiling \(Q\). The remaining possible increase in clearance becomes smaller, so the local sensitivity and elasticity decline toward zero. Their rates of decline are model-specific. The parallel-tube model’s local sensitivity contains \(e^{-x}\), while the well-stirred model’s contains \((1+x)^{-2}\). The dispersion slope is determined both by its exponential term and by \(\sqrt{1+4D_Nx}\). Thus, a statement such as “high extraction means zero sensitivity” is only a limiting statement; at any finite \(x\), the sensitivity is generally positive.

Worked Example: Comparing Responses at Low Extraction

Use \(Q=100\) L/h, \(f_u=0.20\), and \(CL_{int}=50\) L/h, giving \(x=(0.20)(50)/100=0.1000\). Let \(D_N=0.50\) for the dispersion model. This is a low-extraction setting because each model predicts an extraction ratio near 0.1.

For the well-stirred model, \(E=0.1/1.1=0.09091\), so \(CL_h=9.091\) L/h. Its local sensitivity is \(0.20/(1.1)^2=0.1653\), and elasticity is \(1/1.1=0.9091\).

For the parallel-tube model, \(E=1-e^{-0.1}=0.09516\), so \(CL_h=9.516\) L/h. The local sensitivity is \(0.20e^{-0.1}=0.1810\), and elasticity is \(0.1/(e^{0.1}-1)=0.9508\).

For dispersion, \(g=1-\sqrt{1.2}=-0.09545\), so \(e^g\approx0.9090\), \(E\approx0.09103\), and \(CL_h\approx9.103\) L/h. Its local sensitivity is \(0.20(0.9090)/\sqrt{1.2}\approx0.1660\), and elasticity is approximately \(0.1(0.8298)/(0.09103)=0.9115\). Rechecking the calculation, the local sensitivity factor is \(E'(x)=e^g/\sqrt{1+4D_Nx}\approx0.8298\), so the local sensitivity is \(0.1660\); elasticity is \(xE'(x)/E(x)=0.1(0.8298)/0.09103\approx0.9115\). This corrected value is close to 1, as expected for low extraction.

All three models therefore give similar clearances and elasticities near 1 in this regime. Their small differences are visible, but they are less pronounced than in a setting where extraction is appreciable.

Worked Example: Comparing Responses at High Extraction

Consider a different hypothetical setting with \(Q=100\) L/h, \(f_u=0.10\), and \(CL_{int}=5000\) L/h, so \(x=5.000\). Take \(D_N=0.50\). The purpose is to compare the model responses at the same normalized input; the models need not predict the same extraction ratio.

The well-stirred model gives \(E=5/6=0.8333\), hence \(CL_h=83.33\) L/h. Its local sensitivity is \(0.10/36=0.002778\), and elasticity is \(1/6=0.1667\).

The parallel-tube model gives \(E=1-e^{-5}=0.9933\), hence \(CL_h=99.33\) L/h. Its local sensitivity is \(0.10e^{-5}=0.0006738\), and elasticity is \(5/(e^5-1)=0.03392\).

For dispersion, \(g=1-\sqrt{11}=-2.3166\), \(e^g\approx0.0986\), and \(E\approx0.9014\), so \(CL_h\approx90.14\) L/h. The local sensitivity is \(0.10(0.0986)/\sqrt{11}\approx0.002973\), and elasticity is approximately \(5(0.0986)/[\sqrt{11}(0.9014)]=0.1648\).

Each model predicts a clearance below \(Q=100\) L/h, and each has a smaller elasticity than in the low-extraction example. The parallel-tube model is closest to the flow ceiling here and has the smallest sensitivity. The other two models retain a larger response at these inputs, illustrating why the extraction regime alone does not determine the exact model-specific response.

Interpreting and Checking a Sensitivity Calculation

Local sensitivity and elasticity answer related but distinct questions. A derivative is most useful for an absolute, small perturbation: \(\Delta CL_h\approx(\partial CL_h/\partial CL_{int})\Delta CL_{int}\). Elasticity is often more interpretable for a small percentage perturbation. Neither should be treated as an exact prediction for a large enzyme change. For that, substitute the changed \(CL_{int}\) into the model and calculate the new clearance directly.

A useful basic check is monotonicity. With positive \(Q\), \(f_u\), and \(CL_{int}\), the three expressions for \(E'(x)\) are positive, so increasing \(CL_{int}\) increases \(CL_h\) in each model. Also, \(E(x)\) lies between 0 and 1 for positive \(x\), so \(CL_h\) remains below \(Q\). A calculated negative derivative or a clearance above \(Q\) indicates a formula, sign, or unit error under these assumptions.

Compare sensitivity only after specifying what is held fixed. If \(Q\), \(f_u\), or \(D_N\) also changes, the calculation is a different perturbation and requires additional terms or a direct recalculation. This is especially important when translating a mathematical sensitivity into a biological interpretation: the equations alone do not establish that an enzyme change leaves other determinants unchanged.

Model Comparison, Limitations, and a Common Pitfall

Model	Local sensitivity to \(CL_{int}\)	Elasticity	Response pattern
Well-stirred	\(f_u/(1+x)^2\)	\(1/(1+x)\)	Declines algebraically as \(x\) rises.
Parallel-tube	\(f_u e^{-x}\)	\(x/(e^x-1)\)	Declines rapidly as extraction approaches its high limit.
Dispersion	\(f_u e^{g(x)}/\sqrt{1+4D_Nx}\)	\(xe^{g(x)}/[\sqrt{1+4D_Nx}(1-e^{g(x)})]\)	Depends on both \(x\) and \(D_N\); tends to zero at high \(x\).

These comparisons are conditional on the chosen model and inputs. The well-stirred formulation is a compact representation; the parallel-tube formulation represents concentration change along parallel flow paths; the dispersion formulation includes axial dispersion. The expressions should not be combined or averaged without a stated modeling rationale. Differences between predictions are consequences of the different assumed liver representations, not calculation noise.

Common pitfall: confusing the elasticity \(1/(1+x)\) with its numerical value. At \(x=0.5\), this is \(1/(1+0.5)=1/1.5=0.6667\), not 0.3333. The denominator is 1.5, and its reciprocal is approximately 0.6667. A second common error is to treat elasticity as the exact percent change for any finite perturbation. It is a local proportional response; for a large change, calculate the model at both input values.

Key takeaway: All three models approach proportionality between hepatic and intrinsic clearance at low extraction, and all become less sensitive as clearance approaches the flow ceiling. The amount and rate of that decline differ by model, so report the model, the operating point, and whether the reported response is a derivative or an elasticity.

Check Your Understanding

Answer each question, then compare your reasoning with the concise solutions below.

For the well-stirred model at \(x=1\), calculate elasticity and interpret it.
For the parallel-tube model at \(x=0.2\), state the local sensitivity in terms of \(f_u\), and calculate elasticity to three significant figures.
Why does the clearance response to a small intrinsic-clearance increase become limited at high extraction?
For the dispersion model, which additional parameter is needed to calculate sensitivity, and what should be held fixed in the derivative?

Solutions. (1) \(S_{CL_{int}}=1/(1+1)=0.5000\); near that operating point, a 1% increase in intrinsic clearance corresponds to approximately a 0.5% increase in hepatic clearance. (2) The local sensitivity is \(f_u e^{-0.2}=0.8187f_u\). Elasticity is \(0.2/(e^{0.2}-1)=0.903\). (3) Clearance approaches the upper limit \(Q\); with less remaining room below that ceiling, a further increase in intrinsic clearance produces a smaller clearance change. (4) The dispersion number \(D_N\) is required. In the stated derivative, hold \(Q\), \(f_u\), and \(D_N\) fixed while varying \(CL_{int}\).