Why Extrapolate a Formulation Contrast Across Doses?
A model fitted at one or more studied doses can be used to predict how two formulations compare at another dose. Such predictions can help characterize a dose range or identify where a formulation contrast might change. But a model does not make an unstudied dose equivalent to a studied one: the prediction depends on whether the pharmacokinetic relationships learned from the data remain credible at the target dose.
The central question is whether dose changes merely scale exposure or also change absorption or disposition. Under dose-proportional pharmacokinetics, a dose increase scales concentrations and exposure measures in a predictable way, and a formulation ratio can remain constant. If bioavailability or absorption rate depends on dose, however, the predicted contrast may itself change with dose.
Model, Contrasts, and Assumptions
Consider an oral one-compartment model with first-order absorption and elimination. For formulation \(j\), at dose \(D\), let \(F_j(D)\) be bioavailability, \(k_{a,j}(D)\) the absorption rate constant, \(V\) the apparent volume of distribution, and \(k_e=CL/V\) the elimination rate constant, where \(CL\) is clearance. With no lag time, the model-predicted concentration at time \(t\) is
This expression assumes linear disposition over the relevant range, complete absorption as represented by \(F_j(D)\), and first-order absorption and elimination. It also assumes the model structure is an adequate description of the concentration-time profiles. If \(k_a=k_e\), the expression is evaluated by its limit rather than by dividing by zero.
For this model, integrating the concentration curve to infinity gives \(AUC_j(D)=F_j(D)D/CL\), provided absorption is complete in the model and \(k_e=CL/V\). The peak time and peak concentration are
Using these results to extrapolate requires more than fitting the model at the observed doses. One must assess whether the dose-response assumptions, formulation-specific absorption behavior, and disposition parameters can be transported to the target dose. In particular, the model should not be assumed to have dose-independent \(F\) or \(k_a\) merely because such a restriction makes extrapolation convenient.
When Dose Scaling Preserves the Contrast
A simple case occurs when, over the observed and target dose range, each formulation has dose-independent bioavailability and absorption rate, and disposition is linear. Thus \(F_j(D)=F_j\), \(k_{a,j}(D)=k_{a,j}\), and \(CL\), \(V\), and \(k_e\) do not change with dose. Substituting a dose \(qD\), for a positive multiplier \(q\), into the concentration equation then multiplies every concentration by \(q\). Integration and maximization preserve that scaling: AUC and \(C_{\max}\) at \(qD\) are \(q\) times their values at \(D\), while \(T_{\max}\) is unchanged. Therefore, for either AUC or \(C_{\max}\), the test-to-reference ratio is unchanged across doses under these assumptions.
For AUC, the reason is especially direct. If both formulations share the same systemic clearance,
The dose and clearance cancel. This does not imply that every formulation contrast is constant in every pharmacokinetic model. For \(C_{\max}\), the ratio is constant under dose scaling only if dose changes do not alter the relevant absorption and disposition parameters. If \(k_a\) differs between formulations, their \(C_{\max}\) ratio can differ from their AUC ratio even when both ratios remain constant across dose.
Worked Example: Scaling an AUC Contrast Under Dose Proportionality
Suppose a hypothetical fitted model has \(F_T=0.72\) and \(F_R=0.80\), with shared \(CL=10\) L/h. Assume these parameters are constant from 100 mg through the target dose of 250 mg. Predict the AUC values and their ratio at both doses.
At 100 mg, \(AUC_T=0.72(100)/10=7.20\) mg·h/L and \(AUC_R=0.80(100)/10=8.00\) mg·h/L. The test-to-reference ratio is \(7.20/8.00=0.900\).
At 250 mg, \(AUC_T=0.72(250)/10=18.0\) mg·h/L and \(AUC_R=0.80(250)/10=20.0\) mg·h/L. The ratio is \(18.0/20.0=0.900\). The prediction follows from the assumed constant bioavailabilities and shared linear clearance. It is not evidence that the ratio will remain 0.900 if those assumptions fail at 250 mg.
Absorption Can Change a Peak Contrast
AUC primarily reflects the extent of systemic availability in this model, whereas \(C_{\max}\) also depends on the rate of absorption. A faster absorption rate can produce a higher and earlier peak without changing AUC, if bioavailability and clearance are unchanged. Consequently, a model-based comparison should calculate each formulation's peak at its own predicted \(T_{\max}\); comparing the concentrations at one shared time does not generally compare their \(C_{\max}\) values.
Worked Example: Predicting Cmax With Different Absorption Rates
Consider hypothetical parameter estimates at 100 mg: \(k_e=0.10\) h\(^{-1}\), \(k_{a,T}=0.61255\) h\(^{-1}\), \(k_{a,R}=0.80\) h\(^{-1}\), and dose-scale terms \(F_TD/V=0.900\) mg/L and \(F_RD/V=1.00\) mg/L. Assume all these parameters remain constant per unit dose through 200 mg.
The test peak time is \(T_{\max,T}=\ln(0.61255/0.10)/(0.61255-0.10) =\ln(6.1255)/0.51255\approx3.53616\) h, or 3.536 h to four significant figures. The reference peak time is \(T_{\max,R}=\ln(0.80/0.10)/(0.80-0.10) =\ln(8)/0.70\approx2.97063\) h.
At the peak, the model gives \(C_{\max}=(FD/V)e^{-k_eT_{\max}}\). This follows because \(e^{-k_aT_{\max}}=(k_e/k_a)e^{-k_eT_{\max}}\), so the absorption term in the concentration equation simplifies to \(e^{-k_eT_{\max}}\). Thus the test prediction is \(0.900e^{-0.10(3.53616)}\approx0.632\) mg/L. The reference prediction is \(1.00e^{-0.10(2.97063)}\approx0.743\) mg/L. Their \(C_{\max}\) ratio is \(0.632/0.743\approx0.851\), while the AUC ratio is \(0.900\). The different ratios reflect the different absorption rates.
At 200 mg, dose proportionality doubles each predicted \(C_{\max}\): approximately 1.264 mg/L for test and 1.486 mg/L for reference. The ratio remains approximately 0.851. This conclusion depends on the stated dose-invariant absorption and disposition parameters.
When the Contrast Changes With Dose
If bioavailability changes with dose, AUC need not be proportional to dose. If absorption rate changes with dose, the \(C_{\max}\) ratio can also change, even when the AUC ratio does not. The model must represent such behavior using a specified dose relationship, and predictions are credible only to the extent that the data support that relationship over the target range.
Worked Example: Dose-Dependent Bioavailability Changes the AUC Ratio
Suppose, hypothetically, a fitted model describes bioavailability in two formulations as \(F_T(D)=0.72/(1+D/400)\) and \(F_R(D)=0.80/(1+D/200)\), with dose \(D\) measured in mg and shared \(CL=10\) L/h. Calculate the AUC ratio at 100 and 400 mg.
At 100 mg, \(F_T=0.72/1.25=0.576\) and \(F_R=0.80/1.50=0.5333\). The predicted AUCs are \(0.576(100)/10=5.76\) mg·h/L and \(0.5333(100)/10\approx5.333\) mg·h/L, giving a ratio of \(5.76/5.333\approx1.08\).
At 400 mg, \(F_T=0.72/2=0.360\) and \(F_R=0.80/3=0.2667\). The predicted AUCs are \(0.360(400)/10=14.4\) mg·h/L and \(0.2667(400)/10\approx10.67\) mg·h/L, giving a ratio of \(14.4/10.67\approx1.35\). The contrast changes because the assumed bioavailability functions change differently with dose. These calculations illustrate a consequence of the specified model; they do not establish that these functions describe an actual formulation.
Checking Whether an Extrapolation Is Defensible
Before interpreting a prediction, check whether the target dose is supported by the design and model. A target just beyond the observed range is still an extrapolation, but a prediction far beyond that range is usually more dependent on unverified assumptions. Inspect dose-normalized concentration profiles and exposure summaries: systematic dose trends in dose-normalized AUC or \(C_{\max}\) are warnings against simple proportional scaling. Also examine whether residual patterns or parameter estimates suggest the chosen model is inadequate. Structural fit and identifiability are important: distinct parameter combinations that fit the observed data similarly can produce very different predictions at an unstudied dose.
Uncertainty should include more than uncertainty in the mean contrast at the observed doses. A prediction at a new dose depends on estimated parameters and their covariance, and may also depend on uncertainty about the model structure. Propagating parameter uncertainty, for example by bootstrap resampling, can show how much the predicted contrast varies under repeated samples. Sensitivity analyses that vary plausible dose-dependence or absorption assumptions reveal whether the conclusion rests on a narrow modeling choice. A precise-looking point prediction is not persuasive if reasonable alternatives give substantially different results.
The main alternatives have different strengths and limitations:
| Approach | What it provides | Main limitation |
|---|---|---|
| Scale an observed contrast under dose proportionality | A simple prediction when exposure scales with dose and the formulation ratio is stable. | Can be misleading if bioavailability, absorption, or disposition changes with dose. |
| Use a dose-dependent pharmacokinetic model | A prediction that can represent specified changes in formulation parameters with dose. | Relies on the correctness and data support of the chosen dose-response relationships. |
| Study the target dose directly | A formulation comparison based on observations at the dose of interest. | Requires data at that dose and does not remove the need for sound analysis. |
A model-based extrapolation is useful for prediction and planning, but its scope is limited by the studied dose range, the adequacy of the structural model, and the transportability of its parameters. It should not be presented as a direct observation or as proof of bioequivalence. A prediction can inform whether further study is warranted; any decision about equivalence must use an appropriate decision framework and account for uncertainty.
Common Mistake
A common error is to calculate the test-to-reference AUC ratio at one dose and assume the same ratio at every dose because dose appears in both numerator and denominator. That cancellation is valid only when the relevant bioavailability and clearance assumptions hold across doses. If \(F_T(D)\) and \(F_R(D)\) change differently, then \(AUC_T(D)/AUC_R(D)=F_T(D)/F_R(D)\) can vary with \(D\). For \(C_{\max}\), dose-dependent absorption can change the contrast even when the AUC ratio is stable. State the assumptions first, calculate formulation-specific predictions under the model, and report the dose range to which the interpretation applies.
Check Your Understanding
Answer each question, then compare with the short solutions below.
- Under what assumptions does doubling dose double both AUC and \(C_{\max}\), while leaving \(T_{\max}\) unchanged?
- In the first worked example, why does the AUC ratio remain 0.900 at 250 mg?
- Can two formulations have the same AUC ratio but different \(C_{\max}\) ratios? Explain using the model.
- For the dose-dependent bioavailability example, what is the predicted AUC ratio at 400 mg, and why is it not 1.08?
- Name two checks that would help assess whether an unstudied-dose prediction is trustworthy.
Solutions. (1) Bioavailability, absorption rates, and linear disposition parameters must remain dose-invariant over the range, and the model structure must be adequate. (2) The shared dose and clearance cancel in the AUC ratio, and the assumed \(F_T\) and \(F_R\) are constant. (3) Yes. AUC depends on bioavailability and clearance here, while \(C_{\max}\) also depends on absorption rate. (4) It is approximately 1.35 because the formulation-specific bioavailabilities change differently with dose. (5) Examples include checking dose-normalized exposures for trends, assessing model fit and parameter identifiability, and examining prediction uncertainty or sensitivity to plausible alternative assumptions.