Identifiability Of Bioavailability And Clearance

Why Identifiability Matters

A pharmacokinetic model can describe observed concentrations well while leaving some of its parameters impossible to estimate separately. This distinction matters when a parameter has a direct interpretation: a fitted value for absolute bioavailability, for example, is not informative if the same concentration curve is equally compatible with a different value of bioavailability and a different clearance. Before interpreting estimates, ask whether the available data can distinguish the parameters at all.

For oral data, a key distinction is between absolute bioavailability—the fraction of an administered dose reaching systemic circulation—and relative bioavailability—the exposure from one formulation relative to another. Oral concentration-time data can often support a relative formulation contrast even when they cannot identify absolute bioavailability and clearance separately.

Definition: A parameter is structurally identifiable under a specified model and design if, in ideal noise-free data, distinct values of that parameter cannot produce exactly the same observations. It is practically identifiable if the available finite, noisy data estimate it with useful precision. Structural non-identifiability cannot be repaired simply by collecting more observations of the same type; practical weakness may improve with more informative or more precise data.

An Oral Model and Its Identifiable Combinations

Consider a single oral dose \(D\) in a one-compartment model with first-order absorption and elimination. Let \(F\) be absolute bioavailability, \(V\) the apparent volume of distribution, \(CL\) systemic clearance, and \(k=CL/V\) the elimination rate constant. Let \(k_a\) be the first-order absorption rate constant. Assume linear, time-invariant pharmacokinetics; complete concentration sampling in principle; no lag time; and a single dose entering an absorption compartment. For \(k_a\ne k\), the predicted concentration at time \(t\ge 0\) is

C(t)=\frac{F D k_a}{V(k_a-k)} \left(e^{-kt}-e^{-k_a t}\right), \qquad k=\frac{CL}{V}.

The data may contain information about the shape of the curve, including \(k\) and \(k_a\), and about its scale. But the scale involves \(F/V\), while the elimination rate involves \(CL/V\). Equivalently, the total area under the concentration-time curve (AUC) is

\begin{aligned} AUC_{0-\infty} &=\int_0^\infty C(t)\,dt\\ &=\frac{F D k_a}{V(k_a-k)} \left(\frac{1}{k}-\frac{1}{k_a}\right)\\ &=\frac{F D}{V k} =\frac{F D}{CL}. \end{aligned}

Thus, even if the oral profile estimates \(k\) and the scale precisely, oral data alone do not generally reveal \(F\), \(CL\), and \(V\) individually. They inform combinations of these quantities. In particular, AUC identifies \(F/CL\) for a known dose, not \(F\) and \(CL\) separately.

Key result: In this oral model, multiplying \(F\), \(CL\), and \(V\) by the same positive constant leaves both \(F/V\) and \(CL/V\) unchanged. It therefore leaves the entire predicted concentration curve unchanged, provided the transformed \(F\) remains physically admissible. The data cannot choose among parameter values related by this transformation.

Why the Oral Data Cannot Separate the Parameters

The non-identifiability follows directly from an invariance in the model. For a positive constant \(c\), define \(F'=cF\), \(CL'=cCL\), and \(V'=cV\). Then

\frac{CL'}{V'}=\frac{cCL}{cV}=\frac{CL}{V}=k, \qquad \frac{F'}{V'}=\frac{cF}{cV}=\frac{F}{V}.

The absorption rate \(k_a\) is unchanged as well. Every term in the oral concentration equation is therefore unchanged. This is a structural property of the model, not a consequence of an unlucky sample or a particular estimation algorithm. The physical restriction \(0<F\le 1\) may limit the range of \(c\), but usually does not select one unique value.

Worked Example: Two Parameter Sets, One Oral Curve

Suppose a hypothetical study gives an oral dose of \(100\) mg. Consider these two parameter sets, with \(k_a=1.20\ \mathrm{h}^{-1}\):

Set A: \(F=0.500\), \(CL=5.00\ \mathrm{L/h}\), and \(V=50.0\ \mathrm{L}\).
Set B: \(F=0.750\), \(CL=7.50\ \mathrm{L/h}\), and \(V=75.0\ \mathrm{L}\).

For both sets, \(k=CL/V=0.100\ \mathrm{h}^{-1}\) and \(F/V=0.0100\ \mathrm{L}^{-1}\). At \(t=2.00\) h, Set A predicts

\begin{aligned} C(2) &=\frac{(0.500)(100\ \mathrm{mg})(1.20\ \mathrm{h}^{-1})} {(50.0\ \mathrm{L})(1.20-0.100)\ \mathrm{h}^{-1}} \left(e^{-(0.100)(2)}-e^{-(1.20)(2)}\right)\\ &=(1.0909\ \mathrm{mg/L})(0.7280)\\ &=0.7942\ \mathrm{mg/L}. \end{aligned}

Set B gives the same value because its \(F/V\) and \(k\) are identical; indeed, the whole predicted curve is identical. Both bioavailability values are between zero and one, so the physical bound does not distinguish the sets. A precise oral profile cannot establish which set is the true one under this model.

A useful reparameterization makes the limitation explicit. Oral data are often described using apparent parameters such as \(CL/F\) and \(V/F\). Their ratio still gives \(k\), since \((CL/F)/(V/F)=CL/V\). These apparent parameters can be useful for predicting concentrations after oral dosing, but they are not absolute clearance and volume unless \(F\) is known.

In real data, an additional issue is practical identifiability. Sparse sampling, substantial measurement error, or weak information about the terminal portion of the curve may make even the identifiable combinations imprecise. That does not remove the structural invariance: better sampling can sharpen the estimate of an identifiable combination, but cannot make the oral-only data distinguish parameter sets that predict exactly the same curve.

Relative Bioavailability From a Formulation Contrast

A formulation comparison can cancel clearance when the same subject receives both formulations and their clearance is unchanged between periods. Under linear pharmacokinetics, the dose-normalized AUC for formulation \(j\) in subject \(i\) is \(F_{ij}/CL_i\). For test (\(T\)) and reference (\(R\)) formulations in that subject, the ratio is therefore

\frac{AUC_{T,i}/D_{T,i}}{AUC_{R,i}/D_{R,i}} = \frac{F_{T,i}/CL_i}{F_{R,i}/CL_i} = \frac{F_{T,i}}{F_{R,i}}.

This estimates relative bioavailability, not either formulation’s absolute \(F\). The cancellation relies on a meaningful within-subject contrast: clearance must be sufficiently stable across periods, and the exposure measures must be comparable. If doses differ, dose normalization is appropriate only when exposure is dose-proportional over the doses used.

Worked Example: Dose-Normalized Relative Bioavailability

In a hypothetical crossover comparison, one participant receives \(100\) mg of test formulation and \(200\) mg of reference formulation in separate periods. The respective measured AUC values are \(18.0\ \mathrm{mg\,h/L}\) and \(32.0\ \mathrm{mg\,h/L}\). Assume the participant’s clearance is unchanged between periods and pharmacokinetics are linear over this dose range.

The dose-normalized exposures are

\frac{AUC_T}{D_T} =\frac{18.0\ \mathrm{mg\,h/L}}{100\ \mathrm{mg}} =0.180\ \mathrm{h/L}, \qquad \frac{AUC_R}{D_R} =\frac{32.0\ \mathrm{mg\,h/L}}{200\ \mathrm{mg}} =0.160\ \mathrm{h/L}.

The relative bioavailability estimate is

\frac{0.180\ \mathrm{h/L}}{0.160\ \mathrm{h/L}}=1.125.

In this hypothetical comparison, the test formulation’s dose-normalized exposure is estimated to be \(1.125\) times the reference formulation’s, or \(12.5\%\) higher. This is a relative estimate; it does not imply that either formulation has absolute bioavailability of \(1.125\). The calculation also does not by itself establish bioequivalence.

A randomized crossover design is especially useful because each participant can serve as their own contrast, reducing the influence of stable between-subject clearance differences. Randomization helps protect the comparison from systematic sequence effects, and an adequate washout is needed to avoid residual drug from one period affecting the next. A parallel design can also estimate a formulation contrast under appropriate randomization and assumptions, but it does not obtain the same direct within-subject cancellation and is more exposed to between-subject differences.

Worked Example: Why a Within-Subject Contrast Helps

Suppose three hypothetical participants have the following AUC values after equal doses, in \(\mathrm{mg\,h/L}\). Assume each participant’s clearance is stable across their two periods and the doses are in the linear range.

Participant	Reference AUC	Test AUC	Test/reference ratio
1	10.0	12.0	1.20
2	5.00	6.00	1.20
3	20.0	24.0	1.20

The raw AUC values vary fourfold between participants, consistent with substantial differences in their exposure scale. Yet each within-participant ratio is \(1.20\): \(12.0/10.0=6.00/5.00=24.0/20.0=1.20\). The common clearance factor within each participant’s pair cancels. This simple pattern illustrates why a paired formulation contrast can be informative even when the absolute clearance and bioavailability for each participant are not separately known. With real data, period variation, measurement error, and participant-to-participant differences in the formulation contrast must also be considered.

When Absolute Bioavailability Can Be Estimated

Absolute bioavailability requires an exposure reference that fixes the systemic input scale. An intravenous dose provides such a reference because its systemic availability is defined as one. Under linear pharmacokinetics and the same clearance for oral and intravenous administration, intravenous AUC gives \(CL=D_{\mathrm{IV}}/AUC_{\mathrm{IV}}\). Oral AUC can then be compared with the exposure expected from that clearance to estimate \(F\):

CL=\frac{D_{\mathrm{IV}}}{AUC_{\mathrm{IV}}}, \qquad F=\frac{AUC_{\mathrm{oral}}/D_{\mathrm{oral}}} {AUC_{\mathrm{IV}}/D_{\mathrm{IV}}}.

The comparison requires compatible measurements and conditions: clearance must be comparable across routes, the relevant dose-exposure relation must be linear, and AUC must represent the same elimination process. An intravenous reference is not the only possible source of the missing information: an independently known clearance or a justified external constraint can also break the oral-only ambiguity. Such external information must be treated as an assumption or data source, not as something learned from the oral curve alone.

Worked Example: Using an Intravenous Reference

In a hypothetical study, an intravenous dose of \(100\) mg produces an AUC of \(20.0\ \mathrm{mg\,h/L}\). An oral dose of \(100\) mg produces an AUC of \(10.0\ \mathrm{mg\,h/L}\). Assume linear pharmacokinetics and the same clearance by both routes.

First, estimate clearance from the intravenous administration:

CL=\frac{100\ \mathrm{mg}}{20.0\ \mathrm{mg\,h/L}} =5.00\ \mathrm{L/h}.

Then estimate absolute oral bioavailability:

F=\frac{10.0/100}{20.0/100}=0.500.

The hypothetical estimate is \(F=0.500\), or \(50.0\%\). Here the intravenous data supply the information that oral data alone lack. If clearance differed between the route-specific observations or exposure were not dose-proportional, this calculation would not have the stated interpretation.

Diagnostics, Alternatives, and Limitations

A model can converge numerically despite structural non-identifiability. The following patterns are reasons to investigate whether parameters are supported by the design:

Large changes in \(F\), \(CL\), and \(V\) across starting values with little or no change in predicted concentrations.
Very high parameter correlations, a nearly singular information matrix, or a long, flat likelihood profile along combinations of \(F\), \(CL\), and \(V\).
Uncertainty intervals that are extremely wide, or estimates that repeatedly reach imposed parameter bounds.
Stable estimates of \(CL/V\) or \(F/CL\) alongside unstable estimates of the individual parameters.

These signs do not automatically prove structural non-identifiability. Sparse or noisy data can produce similar symptoms through practical weakness, and a misspecified structural model can also distort estimates. Check whether the parameter transformation above leaves the model predictions unchanged under the fitted assumptions. If it does, no optimizer or additional oral samples from the same design can select a unique absolute \(F\) and \(CL\).

The main alternatives answer different questions. Oral-only modeling can estimate concentration-relevant apparent parameters or identifiable combinations. A within-subject formulation contrast can estimate relative bioavailability when clearance cancels. An intravenous reference or independently supported clearance can make absolute bioavailability estimable, subject to its assumptions. These approaches are not interchangeable: a relative estimate does not supply absolute bioavailability, and an apparent clearance is not absolute clearance.

The simple ratio argument has limits. If clearance changes between formulation periods, the cancellation is incomplete; if pharmacokinetics are nonlinear, dose normalization may fail; and if sampling does not adequately characterize exposure, AUC itself may be uncertain. Carryover, period effects, or formulation-dependent changes in systemic disposition can also undermine a direct interpretation as a bioavailability ratio. Identifiability is therefore a property of the model together with the design and assumptions, not of a parameter name alone.

Common pitfall: A fitted oral model reports \(F=0.60\) and \(CL=4.0\ \mathrm{L/h}\), so the analyst concludes that both quantities were estimated separately from the oral profile. A numerical output is not evidence of identifiability: another pair related by the model’s scaling invariance may give precisely the same predictions. Report only the combinations supported by the data, or state what external information or constraint supports separate estimates.

Check Your Understanding

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

In the oral model, what happens to the predicted curve if \(F\), \(CL\), and \(V\) are all multiplied by the same admissible positive constant?
A participant has dose-normalized AUC values of \(0.24\ \mathrm{h/L}\) for test and \(0.20\ \mathrm{h/L}\) for reference. What is the estimated relative bioavailability, under the required assumptions?
Why does precise oral sampling not, by itself, identify absolute bioavailability and clearance separately?
What additional information in the worked intravenous example makes an estimate of absolute bioavailability possible?

Solutions. (1) The curve is unchanged because \(CL/V\) and \(F/V\) remain constant. (2) The ratio is \(0.24/0.20=1.20\), so the test formulation has estimated dose-normalized exposure \(1.20\) times the reference, assuming clearance is stable and pharmacokinetics are linear. (3) The oral model has an exact parameter invariance: distinct \(F\), \(CL\), and \(V\) values can give identical ideal curves, regardless of sampling precision. (4) The intravenous AUC provides clearance through \(CL=D_{\mathrm{IV}}/AUC_{\mathrm{IV}}\), fixing the systemic exposure scale; comparing dose-normalized oral and intravenous AUC then supports estimation of \(F\), if clearance is comparable and pharmacokinetics are linear.