Why Bridge a Formulation to an Established Model?
A new oral formulation may have a different extent or rate of absorption while the active drug and its systemic disposition remain unchanged. A model for an established formulation can provide a useful reference: the new data are analyzed in relation to that reference, and the resulting model can be used to predict exposure under a proposed regimen. The bridge is only informative, however, if the shared assumptions about disposition and the formulation-specific absorption differences are made explicit.
The goal is not simply to compare two concentration curves. The analysis separates relative bioavailability, which describes the ratio of absorbed fractions, from absorption-rate parameters such as \(k_a\). It then uses the fitted model to predict quantities such as average steady-state concentration and peak concentration under the intended dose and interval.
A Shared-Disposition Model With Formulation-Specific Absorption
A useful starting model is one-compartment disposition with first-order absorption and elimination. For formulation \(j\), let \(F_j\) be bioavailability, \(D_j\) the dose, \(k_{a,j}\) the absorption-rate constant, \(V\) the apparent volume of distribution, and \(CL\) systemic clearance. Define \(k_e=CL/V\). After a single oral dose, with no lag time, the model concentration at time \(t>0\) is:
The single-dose concentration formula requires \(k_{a,j}\ne k_e\); the repeated-dose concentration and peak-time formulas require \(k_a\ne k_e\). Equal-rate cases are excluded here.
For a bridge, \(V\) and \(CL\) are typically shared between formulations, while \(F_j\) and \(k_{a,j}\) may differ. A convenient parameterization sets \(F_{\mathrm{new}}=F_{\mathrm{rel}}F_{\mathrm{ref}}\), or estimates the log ratio \(\log(F_{\mathrm{new}}/F_{\mathrm{ref}})\). Since oral data alone generally identify apparent parameters, fixing or anchoring disposition to an established model can support relative comparisons. It does not make absolute bioavailability identifiable without additional information, such as an intravenous reference or a justified known reference value.
The shared-disposition assumption means that the formulation changes input into the body, not systemic clearance or distribution. Other key assumptions are linear pharmacokinetics over the studied dose range, an adequate absorption model, comparable assay and sampling procedures, and no unmodeled time-varying or formulation-dependent changes in disposition. These assumptions should be judged against the data, not adopted merely because the model is convenient.
Why Dose-Normalized AUC Estimates Relative Bioavailability
The AUC result follows directly by integrating the single-dose concentration equation from zero to infinity. The integral of \(e^{-kt}\) over this range is \(1/k\), for \(k>0\). Therefore:
Thus, if clearance is shared, the dose-normalized AUC ratio equals the bioavailability ratio. The cancellation does not hold if clearance differs between formulations or if the kinetics are nonlinear. AUC can also be biased by inadequate terminal sampling or an unreliable estimate of the extrapolated terminal area.
Worked Example: Estimating Relative Bioavailability From AUC
Suppose a hypothetical comparison gives dose \(D_{\mathrm{ref}}=100\) mg and \(\mathrm{AUC}_{\mathrm{ref}}=6.00\) mg·h/L for the established formulation. For the new formulation, the dose is also 100 mg and \(\mathrm{AUC}_{\mathrm{new}}=5.40\) mg·h/L. Assume linear pharmacokinetics and the same clearance.
The dose-normalized AUCs are \(6.00/100=0.0600\) and \(5.40/100=0.0540\) (mg·h/L)/mg. The relative bioavailability estimate is:
Under the shared-clearance assumption, the new formulation provides an estimated 90.0% of the systemic exposure per milligram of the reference. This point estimate alone does not establish that the formulations are equivalent or that the difference is clinically acceptable; uncertainty and the relevant decision criteria still matter.
Testing Whether Absorption Rates Differ
A change in absorption rate is distinct from a change in extent. A formulation may have approximately the same AUC but a different \(k_a\), and hence a different concentration-time profile. To compare rates, define \(\delta=\log(k_{a,\mathrm{new}}/k_{a,\mathrm{ref}})\). The null hypothesis of equal absorption rates is \(H_0:\delta=0\); a two-sided alternative is \(H_a:\delta\ne0\).
If the fitted model estimates \(\hat{\delta}\) with standard error \(SE(\hat{\delta})\), the large-sample Wald statistic is \(z=\hat{\delta}/SE(\hat{\delta})\). Under \(H_0\), this statistic is approximately standard normal when the model is adequately specified and the estimator is sufficiently close to normal. The two-sided p-value is \(2P(Z\ge |z|)\) for \(Z\sim N(0,1)\). The standard error must account for covariance between the formulation-specific parameter estimates; treating two correlated estimates as independent can give the wrong uncertainty.
Worked Example: Wald Test for an Absorption-Rate Difference
In a hypothetical joint fit, the estimated absorption rates are \(k_{a,\mathrm{new}}=1.20\ \mathrm{h}^{-1}\) and \(k_{a,\mathrm{ref}}=0.800\ \mathrm{h}^{-1}\). The model estimates \(\hat{\delta}=\log(1.20/0.800)=\log(1.50)=0.4055\), with \(SE(\hat{\delta})=0.180\ \) on the log scale. Test at \(\alpha=0.05\).
The Wald statistic is:
The two-sided p-value is \(2P(Z\ge2.253)\approx0.02426\). Since \(0.02426<0.05\), the data provide evidence against equal absorption rates under this model. The estimated ratio is 1.50, so the fitted absorption-rate constant is higher for the new formulation. This is evidence of a rate difference, not evidence that the formulation causes a clinical outcome or that the formulations fail any prespecified equivalence criterion.
A 95% Wald interval for \(\delta\) is \(0.4055\pm1.96(0.180)=[0.0527,0.7583]\). Exponentiating its endpoints gives an interval of approximately \([1.054,2.135]\) for \(k_{a,\mathrm{new}}/k_{a,\mathrm{ref}}\). Under the model and its large-sample approximation, this interval is compatible with rate ratios from about 1.05 to 2.135.
Predicting Exposure Under the Intended Regimen
Once the formulation parameters have been estimated, predictions should use the proposed dose and dosing interval rather than extrapolating from a single-dose summary alone. For repeated oral doses with interval \(\tau\), the steady-state concentration at time \(t\) after a dose, for \(0\le t<\tau\), under the same first-order model is:
The average steady-state concentration over one interval is \(\bar{C}_{\mathrm{ss}}=FD/(CL\tau)\). The steady-state peak occurs at:
These formulas assume repeated doses at a constant interval, linear kinetics, and that the model's absorption and disposition parameters apply at steady state. They provide model-based predictions, not guarantees about an individual patient's concentrations.
Worked Example: Predicting Steady-State Exposure
Use hypothetical new-formulation estimates \(F=0.540\), \(k_a=1.20\ \mathrm{h}^{-1}\), \(CL=10.0\) L/h, and \(V=50.0\) L for a 100 mg dose every 12 h. Assume linear one-compartment disposition. Then \(k_e=CL/V=10.0/50.0=0.200\ \mathrm{h}^{-1}\).
Average steady-state concentration is:
The peak time is:
At this time, the bracket in the steady-state concentration equation is approximately \(e^{-0.200(1.696)}/(1-e^{-2.4})-e^{-1.20(1.696)}/(1-e^{-14.4})=0.7833-0.1306=0.6527\). The coefficient is \(0.540(100)(1.20)/[50.0(1.20-0.200)]=1.296\ \mathrm{mg/L}\), so \(C_{\max,\mathrm{ss}}\approx1.296(0.6527)=0.846\ \mathrm{mg/L}\). The estimate describes the model's predicted peak for this hypothetical regimen. It does not show that the regimen is safe or effective.
Diagnostics, Alternatives, and Limitations
A bridge is credible only if both the model and the shared-parameter assumptions are plausible. Examine observed and population-predicted concentration-time profiles by formulation, residuals against time and predicted concentration, and the distribution of individual predictions where individual-level data are available. Persistent early-time residuals can indicate an inadequate absorption model; systematic late-time residuals can indicate a wrong disposition or terminal-phase assumption. Check whether parameter estimates are stable, whether confidence intervals are excessively wide, and whether the data contain enough early samples to distinguish \(k_a\) from \(k_e\). When absorption is slower than elimination, flip-flop behavior can make the terminal slope reflect absorption rather than elimination, so the parameters may not be separately identified by oral data alone.
Compare the shared-disposition model with plausible alternatives, such as allowing formulation-specific disposition, adding a lag time, or using a more flexible absorption model, when the design and data support those comparisons. A materially improved fit after allowing different clearance is a warning that the simple relative-bioavailability interpretation may not hold. Such alternatives should be scientifically motivated; adding parameters just to improve fit can produce unstable estimates and misleading predictions.
| Approach | Useful when | Main limitation |
|---|---|---|
| Dose-normalized AUC ratio | Reliable AUCs are available and shared clearance is credible | Does not characterize absorption-rate differences or their time course |
| Separate model fits | Each formulation has rich data and the goal is descriptive | Differences in disposition estimates can obscure formulation comparisons |
| Joint model with shared disposition | A reference model can reasonably anchor systemic disposition | Predictions can be biased if the shared-disposition assumption is false |
A common mistake is to interpret \(F_{\mathrm{rel}}=0.90\) as proof that the new formulation is bioequivalent, or to interpret a significant \(k_a\) test as proof of clinically important impact. The correction is to distinguish the estimand and decision: relative bioavailability is an exposure ratio under stated assumptions; an absorption-rate test addresses a parameter difference; and equivalence requires a prespecified margin and an appropriate uncertainty-based analysis. The bridge also has limited value if the intended dose lies outside the range where linearity has been supported, if the established model is poorly estimated, or if the new formulation introduces unmodeled absorption behavior.
Check Your Understanding
Answer each question, then compare with the solutions below.
- A reference dose of 80 mg gives AUC \(4.80\) mg·h/L; a new dose of 120 mg gives AUC \(6.48\) mg·h/L. With shared linear clearance, calculate relative bioavailability.
- For a log absorption-rate ratio estimate of \(0.300\) with standard error \(0.150\), calculate the Wald statistic and state the two-sided conclusion at \(\alpha=0.05\).
- Why does equal dose-normalized AUC not imply equal absorption-rate constants?
- Name one diagnostic pattern that could challenge the shared-disposition model and one possible response.
Solutions. 1. \((6.48/120)/(4.80/80)=0.0540/0.0600=0.900\), so the estimated relative bioavailability is 0.900. 2. \(z=0.300/0.150=2.00\), giving two-sided \(p\approx0.0455\); reject \(H_0\) at 0.05, subject to the model and Wald approximation. 3. AUC depends on \(F/CL\), not \(k_a\), in this linear model; different absorption rates can therefore yield the same AUC but different concentration-time profiles. 4. For example, systematic late-time residuals by formulation challenge shared disposition; investigate model adequacy and scientifically assess a formulation-specific disposition alternative rather than assuming the simple bridge remains valid.