Identifiability of Flow and Intrinsic Clearance

Why One Clearance Value May Not Tell You Which Parameter Is Wrong

A hepatic clearance estimate is often used to learn about both hepatic blood flow and metabolic capacity. But a measured clearance is an outcome of their joint action: a low value could reflect low flow, low intrinsic clearance, or some combination of the two. The central question here is not how to calculate a clearance from chosen inputs, but whether the observed clearance contains enough independent information to estimate those inputs separately.

The discussion uses the well-stirred liver model. Its inverse calculation was the subject of the earlier material on recovering intrinsic clearance; the key issue here is what can and cannot be inferred when flow is also unknown. Identifiability is always conditional on a specified model and on what is measured. A parameter may be identifiable under one set of measurements and assumptions but not under another.

Model and Conditions for the Identification Question

Let \(Q_h\) be hepatic blood flow, \(f_{u,b}\) the unbound fraction in blood, \(\mathrm{CL}_{\mathrm{int}}\) intrinsic hepatic clearance, and \(\mathrm{CL}_h\) hepatic blood clearance. For the well-stirred model,

The model treats the liver as a well-mixed compartment supplied by blood flow. The equation assumes linear, non-saturating elimination over the concentrations considered; stable flow, binding, and intrinsic clearance; and that the clearance observation is hepatic blood clearance, not an unseparated total systemic clearance. It also assumes that the measured or specified \(f_{u,b}\) applies to the relevant blood concentrations. If renal clearance or another elimination route contributes to a systemic estimate, that estimate cannot simply be inserted as \(\mathrm{CL}_h\).

For the identification argument, assume \(Q_h>0\), \(f_{u,b}>0\), and \(\mathrm{CL}_{\mathrm{int}}>0\). If the unbound fraction is known, the two unknowns of interest are \(Q_h\) and \(\mathrm{CL}_{\mathrm{int}}\). Define the effective unbound intrinsic capacity \(A=f_{u,b}\mathrm{CL}_{\mathrm{int}}\). Then the observation satisfies \(\mathrm{CL}_h=Q_h A/(Q_h+A)\). This compact form makes the ambiguity visible: one clearance value is a single constraint on two unknown quantities.

The Family of Parameter Values Consistent With One Observation

Suppose \(\mathrm{CL}_h\) and \(f_{u,b}\) are known but \(Q_h\) is not. Solving the well-stirred equation for intrinsic clearance gives

This inverse is defined for finite positive intrinsic clearance only when \(Q_h>\mathrm{CL}_h\). For every admissible choice of \(Q_h\) above the observed clearance, the equation gives a corresponding \(\mathrm{CL}_{\mathrm{int}}\) that reproduces that same clearance. Thus a single observation does not select one point; it selects a curve of parameter pairs. If \(f_{u,b}\) is also unknown, there is still less information: the observation constrains combinations of flow, binding, and intrinsic clearance, not all three values individually.

The extraction ratio \(E_h=\mathrm{CL}_h/Q_h\) does not remove the ambiguity when flow is unknown, because calculating \(E_h\) already requires \(Q_h\). Even if \(E_h\) is measured directly, the model gives

This identifies a ratio of effective capacity to flow, not their absolute scale. Multiplying both \(Q_h\) and \(f_{u,b}\mathrm{CL}_{\mathrm{int}}\) by the same positive factor leaves \(E_h\) unchanged.

Why the Ambiguity Is Structural

The ambiguity is not cured by collecting many exact replicate measurements of the same clearance under unchanged conditions. Replicates can reduce measurement error and improve the estimate of \(\mathrm{CL}_h\), but they still provide repeated information about the same model output. They do not add an independent constraint on \(Q_h\) or \(\mathrm{CL}_{\mathrm{int}}\).

One local way to express this is through the sensitivity of the predicted clearance to each parameter. With \(A=f_{u,b}\mathrm{CL}_{\mathrm{int}}\) and \(f_{u,b}\) treated as known, the logarithmic sensitivities are

These are two sensitivities of one predicted quantity, not two independent measurements. In a local sensitivity matrix with one clearance observation and two parameters, there is one row and therefore the matrix cannot have rank two. The data cannot locally distinguish arbitrary changes in both parameters from one another. This rank argument supports the explicit family of solutions derived above; it is not a claim that every parameter-identification problem can be decided by counting observations alone.

Measurements That Can Add Information

The most direct strategy is to measure one parameter independently, then infer the other from clearance. Independent hepatic blood-flow information can identify \(\mathrm{CL}_{\mathrm{int}}\) when \(f_{u,b}\) is known. Conversely, an independent estimate of intrinsic clearance, together with \(f_{u,b}\), can identify flow if it implies a finite positive solution. A measurement of \(f_{u,b}\) is also important: without it, even known flow only identifies the product \(f_{u,b}\mathrm{CL}_{\mathrm{int}}\), rather than separating binding from intrinsic clearance.

A further option is to report the effective capacity \(A=f_{u,b}\mathrm{CL}_{\mathrm{int}}\), or a model-derived quantity such as extraction, when the available observations support that quantity but not its underlying components. This avoids presenting an arbitrary choice of flow as if it were learned from the clearance data. Any use of an external assay or physiological measurement must still account for whether it represents the same biological conditions and scale as the in vivo model.

Feasibility Checks, Diagnostics, and Alternatives

The equation also provides useful consistency checks. For positive finite \(Q_h\) and \(A\), predicted clearance must satisfy \(0<\mathrm{CL}_h<Q_h\) and \(0<\mathrm{CL}_h<A\). Thus, if an independently measured flow is less than or equal to the observed clearance, the standard well-stirred equation cannot reproduce that clearance with finite positive intrinsic clearance. Likewise, if independently specified \(A\) is less than or equal to the observed clearance, no positive finite \(Q_h\) solves the equation. Such a mismatch calls for checking units, blood-versus-plasma scaling, the clearance measurement, and model assumptions; it is not a reason to force a negative or infinite parameter estimate into interpretation.

In practical estimation, warning signs include very wide uncertainty intervals, large parameter correlations, estimates that move substantially when bounds or starting values change, and a flat profile of fit across different flow-clearance combinations. These signs indicate that the data do not sharply distinguish the parameters, even if software returns a single numerical optimum. Adding subjects or replicates can improve precision in the observed clearance, but will not resolve structural ambiguity unless the added data contain new information.

The main alternatives are to add an independent physiological or assay measurement; to estimate only an identifiable composite such as \(A\) or a clearance predicted under a stated fixed flow; or to use a richer data design that supplies distinct information. Fixing flow can make the inverse equation calculable, but the result then depends on that external flow value and its uncertainty. Jointly fitting several observations is useful only when their measurement relationships add independent constraints; simply repeating the same endpoint is not enough. Different hepatic models also imply different parameter-to-clearance relationships, so identifiability under the well-stirred equation should not be assumed to transfer unchanged to another model.