Why Clearance Patterns Need a Model
A hepatic clearance change can be more responsive to hepatic blood flow or to the liver’s intrinsic ability to eliminate drug. These are often described as flow-limited and capacity-limited patterns. They are not fixed labels attached to a drug in every circumstance: the pattern depends on extraction, on which input changes, and on the liver model used to translate those inputs into clearance.
The key comparison is between the effect of changing hepatic blood flow \(Q\) and the effect of changing intrinsic clearance \(CL_{int}\), with other inputs held fixed. Because the unbound fraction in blood \(f_u\) scales the intrinsic clearance available to the liver, it is useful to define the effective capacity \(a=f_uCL_{int}\). The extraction ratio is \(E_H=CL_H/Q\), where \(CL_H\) is hepatic clearance. This tutorial uses these quantities to interpret patterns, not to infer a biological mechanism from clearance alone.
Model Assumptions and Sensitivity
The classical liver models considered here assume a fixed liver model and a specified physiological flow, unbound fraction, and intrinsic clearance. They describe hepatic elimination under their respective assumptions; they do not automatically account for changing protein binding, transporter activity, extrahepatic clearance, or changes in liver physiology. A comparison that varies one input should therefore hold the others fixed, unless the intended question explicitly includes their joint change.
In the well-stirred model, blood is treated as mixed within the liver before leaving it. Its clearance is \(CL_H=Q a/(Q+a)\), and its extraction ratio is \(E_H=a/(Q+a)\). In the parallel-tube model, drug concentration falls along parallel hepatic paths; its clearance is \(CL_H=Q(1-e^{-x})\), where \(x=a/Q\), and its extraction ratio is \(E_H=1-e^{-x}\). In the dispersion model, axial mixing is represented by a dispersion number \(D\), so extraction is a model-specific function \(E_D(x,D)\). The exact response must be evaluated under that model and its specified \(D\); its limiting behaviour connects the parallel-tube and well-stirred descriptions.
These formulas make \(x=a/Q\) a useful comparison scale. Small \(x\) corresponds to low extraction, where clearance is close to \(a\) in the well-stirred and parallel-tube models. Large \(x\) corresponds to high extraction, where clearance approaches \(Q\). At intermediate extraction, model predictions and sensitivities need not agree.
Deriving the Sensitivity Measures
An elasticity is a derivative of the logarithm of clearance with respect to the logarithm of an input, holding the other specified model inputs fixed. A value near 1 indicates approximately proportional response locally; a value near 0 indicates little local response. Elasticities are local summaries, so they do not promise that a large finite change will have the same percentage effect.
For either model written as \(CL_H=Q E_H(x)\), with \(x=a/Q\), and with any model-specific parameter such as \(D\) held fixed, the chain rule gives a concise result. When \(Q\) increases at fixed \(a\), \(x\) decreases in the same proportion. When \(a\) increases at fixed \(Q\), \(x\) increases in the same proportion.
The final equality follows because \(\ln(CL_H)=\ln Q+\ln E_H(x)\), while \(\partial\ln x/\partial\ln Q=-1\) at fixed \(a\), and \(\partial\ln x/\partial\ln a=1\) at fixed \(Q\). Thus, for these fixed-parameter formulations, the two elasticities sum to one. If a model parameter such as \(D\) also changes with flow, this simple relation does not apply without including that parameter’s change.
For the well-stirred model, differentiating \(CL_H=Qa/(Q+a)\) gives \(\varepsilon_Q=a/(Q+a)=E_H\) and \(\varepsilon_a=Q/(Q+a)=1-E_H\). For the parallel-tube model, differentiating \(CL_H=Q(1-e^{-x})\) gives \(\varepsilon_a=x/(e^x-1)\) and \(\varepsilon_Q=1-x/(e^x-1)\). These different expressions are why a diagnosis should name the chosen model rather than treating extraction ratio as a universal sensitivity measure.
Worked Example: Interpreting a Well-Stirred Model at Intermediate Extraction
Suppose, hypothetically, that \(Q=90\) L/h and \(a=90\) L/h. Assume the well-stirred model applies and that unbound fraction and the remaining inputs stay fixed when either flow or effective capacity is changed. Calculate clearance and the local sensitivities.
The clearance is
The extraction ratio is \(E_H=CL_H/Q=45/90=0.500\). For the well-stirred model, \(\varepsilon_Q=E_H=0.500\) and \(\varepsilon_a=1-E_H=0.500\). At these conditions, clearance has equal local proportional sensitivity to flow and effective capacity. This is an intermediate pattern, not evidence that either input is irrelevant.
Worked Example: A Parallel-Tube Capacity Change at High Extraction
Consider a hypothetical parallel-tube liver with \(Q=90\) L/h and \(a=360\) L/h, so \(x=4\). Assume \(Q\) and the parallel-tube model remain fixed while capacity doubles from \(a=360\) to \(a=720\) L/h.
Initially, \(CL_H=90(1-e^{-4})=88.35\) L/h. After the capacity change, \(x=720/90=8\), and
Using unrounded model values, the percentage increase is
The change in clearance is small relative to the doubling of effective capacity. The initial capacity elasticity is \(4/(e^4-1)\approx0.07463\), so the parallel-tube model predicts low local capacity sensitivity at these high-extraction conditions. The finite-change result is consistent with that interpretation.
Worked Example: Comparing Flow and Capacity Changes Under Two Models
Use the same hypothetical baseline inputs, \(Q=90\) L/h and \(a=360\) L/h, but compare a 20% increase in flow with a doubling of capacity. In each calculation, change only the named input. This checks how model choice affects the pattern.
Under the well-stirred model, the baseline clearance is \(90(360)/(90+360)=72.0\) L/h. With flow raised to \(108\) L/h, clearance is \(108(360)/(108+360)=83.08\) L/h, a \(15.38\%\) increase from the unrounded values. With capacity doubled to \(720\) L/h while flow remains \(90\) L/h, clearance is \(90(720)/(90+720)=80.00\) L/h, an \(11.11\%\) increase. The well-stirred model therefore predicts a larger response to the specified flow change, although both inputs matter.
Under the parallel-tube model, baseline clearance is \(90(1-e^{-4})=88.35\) L/h. At \(Q=108\) L/h and unchanged capacity, \(x=360/108=3.3333\), so clearance is \(108(1-e^{-3.3333})=104.1\) L/h, approximately \(17.88\%\) above baseline. With capacity doubled at baseline flow, clearance is \(90(1-e^{-8})=89.97\) L/h, approximately \(1.832\%\) above baseline. Here too the specified flow change has the larger effect, but its predicted magnitude—and the capacity response—depends on the chosen liver model.
The percentage changes above are calculated from unrounded model expressions and then rounded for display. Subtracting rounded endpoint clearances can produce a slightly different percentage; that is a rounding discrepancy, not a different model prediction.
Diagnosing Patterns in Practice
A defensible diagnosis starts by naming the model and the input being perturbed. Calculate the baseline extraction ratio and, when useful, the local elasticities. Then calculate finite changes directly from the model if the proposed perturbation is not small. For the dispersion model, evaluate \(E_D(x,D)\) and its sensitivity at the relevant \(x\) and \(D\); do not substitute an elasticity formula derived for another model.
State whether the change is in \(Q\), \(CL_{int}\), \(f_u\), or a combination. For an isolated comparison, hold the other model inputs fixed.
Use the corresponding clearance equation and state its assumptions. For a dispersion model, specify the dispersion number and whether it is held fixed.
Use local elasticities for small changes or calculate new clearances for finite changes. Describe the larger response as relative flow or capacity limitation under the stated conditions.
Confirm units, extraction ratio, changed inputs, and rounding. Do not treat model predictions as proof that a particular biological input changed in a patient.
Several diagnostics can reveal a weak or inconsistent interpretation. If a high-extraction case is described as strongly capacity-sensitive under the parallel-tube model, check the calculation: \(x/(e^x-1)\) becomes small as \(x\) grows. If the stated flow and capacity elasticities do not sum to one, verify the derivatives and confirm that all other model parameters were held fixed. If observed clearance changes are being used to assign a unique cause, remember that different combinations of \(Q\), \(f_u\), and \(CL_{int}\) can produce similar clearances. Identifiability is not supplied by a sensitivity calculation alone.
Model Comparisons, Limitations, and a Common Pitfall
| Model | Clearance form | Pattern to check | Important qualification |
|---|---|---|---|
| Well-stirred | \(Qa/(Q+a)\) | \(\varepsilon_Q=E_H\); \(\varepsilon_a=1-E_H\) | At high extraction, clearance is relatively flow-sensitive; at low extraction, it is relatively capacity-sensitive. |
| Parallel-tube | \(Q(1-e^{-x})\) | \(\varepsilon_a=x/(e^x-1)\); \(\varepsilon_Q=1-\varepsilon_a\) | At high extraction, capacity sensitivity can be especially small; intermediate predictions differ from the well-stirred model. |
| Dispersion | \(Q E_D(x,D)\) | Evaluate sensitivity using the specified \(D\) and the dispersion-model extraction function. | A flow-related change in \(D\) must be accounted for if \(D\) is not held fixed. |
The well-stirred and parallel-tube models share low- and high-extraction limiting behaviour, but their intermediate predictions need not match. The dispersion model provides another representation of intrahepatic mixing, with predictions depending on its dispersion number. These approaches are alternatives for describing liver behaviour, not interchangeable calculation shortcuts. Choosing among them requires biological and application-specific justification; agreement in one condition does not establish that their predictions will agree under a perturbation.
A common mistake is to say, “High extraction means capacity cannot matter,” or the reverse, “Low extraction means flow has no effect,” as if either statement were exact. The correction is to quantify sensitivity under the selected model and state it as relative, local, and conditional. For example, in the well-stirred model, high extraction makes \(\varepsilon_Q\) approach 1 and \(\varepsilon_a\) approach 0, but at finite extraction neither elasticity is generally exactly zero. Finite changes, changing binding, extrahepatic elimination, and model misspecification can also alter the observed pattern.
Check Your Understanding
Answer each question, then compare with the brief solutions below.
- In the well-stirred model, \(Q=80\) L/h and \(a=80\) L/h. Find \(CL_H\), \(E_H\), and both local elasticities.
- For a parallel-tube liver with \(x=1\), write the expression for capacity elasticity. Is the local pattern more capacity-sensitive or flow-sensitive?
- Why should a dispersion-model sensitivity calculation state whether \(D\) changes with flow?
- A clearance rise follows a simultaneous increase in \(Q\) and \(CL_{int}\). Can that observation alone establish which input caused most of the change? Explain.
Solutions. (1) \(CL_H=80(80)/(80+80)=40.0\) L/h and \(E_H=40/80=0.500\); hence \(\varepsilon_Q=0.500\) and \(\varepsilon_a=0.500\). (2) \(\varepsilon_a=1/(e-1)\approx0.5820\), while \(\varepsilon_Q\approx0.4180\); locally, capacity sensitivity is larger. (3) Because \(D\) affects the dispersion-model extraction function; if it changes with \(Q\), the flow response includes that additional change and the fixed-\(D\) sensitivity is incomplete. (4) No. Both inputs changed, and clearance alone generally does not separate their contributions without additional information or a stated model-based counterfactual comparison.