Why Dose-Related Exposure Needs Careful Interpretation
Exposure often rises more than proportionally with dose. That pattern does not, by itself, show that bioavailability increases with dose. Saturable elimination can produce the same broad pattern: as concentration rises, elimination becomes less efficient in relation to concentration, so exposure can increase faster than dose even if the fraction absorbed is unchanged. The earlier material on saturable elimination establishes this competing explanation.
The question here is narrower: after a model accounts for saturable elimination, does a systematic dose-related discrepancy remain that is better explained by dose-dependent bioavailability? This is a structural model question. It requires comparing concentration-time profiles across dose levels, not simply comparing dose-normalized AUC values or inspecting one profile in isolation.
For oral data, one practical comparison is between a model with a common \(F\) at all doses and a model in which the systemic input depends on dose. In a simple one-compartment model with first-order absorption, the oral depot amount \(A_g\) and central concentration \(C\) can be represented as
Here \(k_a\) is the absorption rate constant, \(V\) is the apparent distribution volume, and \(V_{\max}\) and \(K_m\) describe saturable elimination. The constant-bioavailability model sets \(F(D)=F\); the alternative allows \(F(D)\) to vary across dose. In an analysis, both models should use the same elimination structure and other relevant assumptions so that improvement is not mistakenly attributed to bioavailability when it comes from changing another part of the model.
A Short Derivation: Why Saturable Elimination Can Mimic a Dose Effect
A rapid-input approximation makes the competing mechanism transparent. Suppose a systemic amount \(FD\) enters a one-compartment model quickly relative to elimination, giving an initial concentration \(C_0=FD/V\). Assume Michaelis–Menten elimination, for which
For the declining concentration, rearrange this equation to obtain \(dt=-\frac{V}{V_{\max}}\frac{K_m+C}{C}\,dC\). Integrating concentration over the complete profile gives
The quadratic term means AUC can increase more than proportionally with dose even though \(F\) is constant and \(C_0\) is proportional to dose. This derivation is an illustration, not a substitute for an oral absorption model: real oral input is spread over time, and the full concentration profile should be fitted when absorption is not rapid relative to elimination.
Worked Example: Recognizing Saturation Before Blaming Bioavailability
Consider hypothetical rapid-input data with \(V=10\) L, \(V_{\max}=20\) mg/h, \(K_m=2\) mg/L, and constant \(F=0.8\). At a 100 mg dose, \(C_0=0.8(100)/10=8\) mg/L. The derived equation predicts
At a 200 mg dose, the same constant \(F\) gives \(C_0=0.8(200)/10=16\) mg/L and
The dose doubles, but the predicted AUC rises from 24 to 80 mg·h/L, a 3.33-fold increase. Thus, superproportional exposure is entirely compatible with constant bioavailability in this example. A diagnosis of dose-dependent bioavailability would require evidence that the saturable-elimination model still systematically misses the observed profiles.
Finding a Residual Dose Effect After Fitting Elimination
Fit the constant-\(F\) model to concentration-time data from multiple dose levels, with adequate observations to inform absorption and the elimination curve. Then examine predictions and residuals by dose. The relevant signal is not merely that high-dose concentrations are larger: the model expects that. It is a persistent, organized discrepancy between observations and predictions at one or more dose levels, after the nonlinear elimination model has been fitted.
For example, if the model consistently overpredicts concentrations at the highest dose across much of the profile, the high-dose systemic input may be lower than the constant-\(F\) model implies. Consistent underprediction may indicate higher input. But the same pattern can arise from other misspecification, including an incorrect absorption model, time-varying processes, or inadequate characterization of elimination. A dose effect on bioavailability is a candidate explanation to assess, not a conclusion from a residual pattern alone.
A useful alternative parameterization allows relative bioavailability to change smoothly with dose while keeping predictions within a plausible range:
Here \(D_{\mathrm{ref}}\) is a reference dose, \(F_{\mathrm{ref}}\) is bioavailability at that dose, and \(\beta_F\) describes the direction and strength of the dose association on the logit scale. A negative \(\beta_F\) indicates decreasing \(F\) as dose rises; a positive value indicates increasing \(F\). This form is bounded between zero and one, but the particular functional form should not be assumed to describe behavior outside the studied dose range.
Worked Example: Estimating Relative Bioavailability From Model-Based Exposure
Suppose a hypothetical rapid-input analysis uses \(V=10\) L, \(V_{\max}=20\) mg/h, and \(K_m=2\) mg/L. Treat these parameters as known for this illustration. The fitted AUC at 50 mg is 8 mg·h/L, at 100 mg is 24 mg·h/L, and at 200 mg is 48 mg·h/L. The AUC equation simplifies to \(AUC=C_0+C_0^2/4\). At 50 mg, solving \(C_0+C_0^2/4=8\) gives \(C_0=4\) mg/L, since \(4+16/4=8\). Therefore \(F=VC_0/D=10(4)/50=0.8\).
At 100 mg, \(C_0=8\) mg/L satisfies \(8+64/4=24\), giving \(F=10(8)/100=0.8\). At 200 mg, \(C_0=12\) mg/L satisfies \(12+144/4=48\), giving \(F=10(12)/200=0.6\). Under the stated assumptions, the fitted pattern is consistent with unchanged bioavailability from 50 to 100 mg and lower bioavailability at 200 mg.
This calculation illustrates how a saturable-elimination relationship can be used to distinguish an exposure pattern from a change in systemic input. In practice, estimates should come from fitting the concentration-time data, not from treating a single AUC as a direct measure of \(F\). Uncertainty in the elimination parameters must also be carried into uncertainty about bioavailability.
Identifiability, Diagnostics, and Model Comparison
Oral concentration data often do not identify absolute bioavailability separately from other scale parameters. For example, the concentration scale can depend on \(F/V\); changing \(F\) and \(V\) together can preserve similar predictions. Without an intravenous reference, a justified external estimate, or other information that separates these quantities, the data may support a relative change in bioavailability across doses without establishing absolute \(F\). Saturable elimination parameters can also be difficult to estimate if the observed concentrations do not span a range informative about the curvature.
Assess whether the model is adequate before interpreting a dose effect. Check that dose levels have enough concentration-time information to characterize absorption and the elimination phase, and that the concentrations cover a useful range relative to \(K_m\). If all observations are far below or far above that range, \(K_m\) and \(V_{\max}\) may be weakly distinguished. Compare observed and predicted profiles by dose, inspect residuals for systematic dose ordering, and consider whether a few influential profiles drive the apparent improvement. The later material on concentration-dependent residuals addresses a different diagnostic question: whether the residual error structure itself is unsuitable.
Compare the constant-\(F\) and dose-dependent-\(F\) models using an approach appropriate to the model and estimation method, such as likelihood-based comparison for nested models when its conditions hold, supported by diagnostic plots and parameter uncertainty. A more flexible model will often fit the data at least as well; fit improvement alone does not establish that the dose effect is reliable or scientifically meaningful. Check whether the dose-effect estimate is precise, whether it is supported across the profile rather than by a narrow time region, and whether conclusions are robust to reasonable alternative absorption or elimination specifications.
| Model or explanation | What it allows | Interpretive caution |
|---|---|---|
| Constant bioavailability with saturable elimination | Nonproportional exposure due to nonlinear elimination | Does not imply exposure is proportional to dose |
| Dose-dependent bioavailability with saturable elimination | Systemic input fraction changes with dose as well as nonlinear elimination | May be weakly identified or confounded with scale and other model features |
| Separate empirical fit at each dose | Different profile shapes or scales by dose without a shared dose-effect structure | Can describe the data but may not support interpolation or a mechanistic explanation |
The separate-fit approach can be a useful descriptive check, but it does not directly estimate a coherent dose relationship. Conversely, a single smooth \(F(D)\) function is economical, yet can hide abrupt changes or misfit if the functional form is wrong. The appropriate choice depends on the dose range, design, and strength of the data.
Worked Assessment of a Dose-Dependent Model
Worked Example: Interpreting a Bounded Dose Relationship
Suppose a hypothetical analysis has independently anchored absolute \(F\) at a reference dose of 100 mg and estimates \(F(100)=0.8\) and \(F(200)=0.6\). For the logit model, \(\operatorname{logit}(0.8)=\log(0.8/0.2)=1.386\) and \(\operatorname{logit}(0.6)=\log(0.6/0.4)=0.4055\). Since \(\log(200/100)=\log 2=0.6931\),
The estimated association is negative: in this model, the log-odds of bioavailability fall as dose rises. At 400 mg, the fitted logit is \(0.4055+(-1.415)\log 2=-0.5753\), corresponding to \(F(400)=1/(1+\exp(0.5753))\approx0.360\). This is a model-based extrapolation, not a directly observed result; without data near 400 mg, it should not be treated as established behavior. If the analysis had only oral data without an absolute bioavailability anchor, the estimates might instead be interpreted as relative dose effects.
Common Pitfall and Practical Limits
Wrong reasoning: “Dose-normalized AUC is higher at the larger dose, so bioavailability must increase with dose.” This ignores saturable elimination, which can raise dose-normalized exposure even when \(F\) is constant. Correction: first fit and assess a suitable elimination model, then ask whether a dose-related input effect remains in the concentration-time data.
The inference is limited by the dose range, sampling schedule, and ability to estimate competing model components. Sparse sampling may leave absorption and elimination poorly separated. A dose-dependent \(F\) model may also absorb a different form of structural error, so its coefficient is not automatically a causal explanation for the observed pattern. Strong conclusions require an adequate design, plausible model assumptions, uncertainty assessment, and restraint about extrapolation.
Use the dose levels together and allow saturable elimination where supported by the data.
Look for persistent, organized discrepancies rather than interpreting nonproportional exposure alone.
Allow \(F(D)\) to vary and assess fit, uncertainty, identifiability, and sensitivity to model assumptions.
Check Your Understanding
Answer each question, then compare your reasoning with the solutions.
- With constant bioavailability and saturable elimination, can dose-normalized AUC rise with dose? Why?
- In the rapid-input approximation, let \(V=10\) L, \(V_{\max}=20\) mg/h, \(K_m=2\) mg/L, and \(C_0=4\) mg/L. Calculate AUC.
- What does a consistent overprediction of high-dose concentrations by a constant-\(F\) model suggest, and what does it not prove?
- Why might oral data support a relative dose effect on bioavailability without identifying absolute \(F\)?
Solutions. (1) Yes. The derived expression is \(AUC=(V/V_{\max})(K_mC_0+C_0^2/2)\); the quadratic term means AUC can rise more than proportionally as \(C_0\) increases, even with constant \(F\). (2) \(AUC=(10/20)(2(4)+4^2/2)=0.5(8+8)=8\) mg·h/L. (3) It suggests that systemic input at high dose may be lower than predicted, so dose-dependent bioavailability is one candidate explanation. It does not prove that explanation; other structural errors or model assumptions may be responsible. (4) Oral concentration scale can depend on combinations such as \(F/V\), so different absolute \(F\) and \(V\) values may give similar predictions. An intravenous reference or other identifying information may be needed to separate them.