Representing Formulation Effects In Pharmacokinetic Models

Why Formulation Effects Need Explicit Parameters

A pharmacokinetic model describes how a dose enters the body and how drug already in the body is distributed and eliminated. A formulation may change the input process without changing disposition: for example, it may alter the fraction reaching systemic circulation, the absorption rate, or the delay before absorption begins. A useful formulation comparison therefore makes explicit which parameters are allowed to differ and which are assumed to be shared.

This distinction is a modeling assumption, not a consequence of assigning different formulation labels. If clearance or volume is allowed to vary freely by formulation, the model can attribute a profile difference to disposition rather than absorption. Conversely, forcing a genuinely formulation-dependent input process to be identical can leave systematic structure in the residuals.

The parameterization below is for a hypothetical, linear, one-compartment model with first-order elimination and oral input. It assumes the compared treatments have appropriately comparable doses, the disposition is linear over the observed range, and the structural model is adequate for the concentration-time data. For this example, clearance \(CL\) and volume \(V\) are shared, so \(k_e=CL/V\) is also shared. The model does not establish that these assumptions are true; they need to be supported by study design, pharmacological knowledge, and model diagnostics.

A Reference Formulation and Parameter Contrasts

Choose one formulation as the reference, denoted \(R\), and another as the test, denoted \(T\). A direct parameterization assigns each formulation its own value, such as \(F_R\) and \(F_T\). A contrast parameterization instead defines a reference value and a difference or ratio. For positive parameters such as relative availability and absorption rate, ratios are often convenient because they preserve positivity and describe multiplicative changes.

The reference value sets the scale; the contrast describes the formulation change. If availability is normalized by setting \(F_R=1\), then \(F_T\) is directly the test-to-reference ratio. This normalization does not mean the reference has 100% absolute bioavailability. With oral concentration data alone, absolute bioavailability is generally not separated from disposition without additional information; the earlier material on identifiability of bioavailability and clearance addresses that limitation.

A log contrast is useful because it converts a ratio into an additive parameter and ensures that the implied ratio is positive. A zero contrast means equal values; a positive contrast means the test value exceeds the reference value; and a negative contrast means it is lower. For lag time, which is measured in time units and may be zero, it is often clearer to use a direct difference, \(\Delta_{t_{\mathrm{lag}}}=t_{\mathrm{lag},T}-t_{\mathrm{lag},R}\), rather than a log ratio.

Mapping Formulation Effects to an Oral Input Model

For a single oral dose \(D\), first-order absorption with a lag time can be represented by the concentration prediction

Here \(j\) indexes formulation, \(F_j\) is relative availability, \(k_{a,j}\) is the absorption rate constant in inverse time, and \(t_{\mathrm{lag},j}\) is the lag time. In the formulation-effect model, \(CL\) and \(V\) are common across \(j\), while \(F_j\), \(k_{a,j}\), and \(t_{\mathrm{lag},j}\) may differ. The displayed expression assumes \(k_{a,j}\ne k_e\); when they are equal, the limiting form is used rather than evaluating a zero denominator.

This equation separates three distinct input features. \(F_j\) scales the available amount, \(k_{a,j}\) controls the rate of first-order absorption after the lag, and \(t_{\mathrm{lag},j}\) shifts the onset of absorption. These parameters can produce related-looking changes in a concentration-time curve, so allowing all of them to vary does not guarantee that each will be estimated precisely. A model should include only formulation differences that are scientifically plausible and supported by the sampling information.

Transit Absorption as an Alternative Input Representation

A single lag time represents a fixed delay followed by absorption. Some formulations instead produce a spread of delays before drug becomes available for first-order absorption. A transit model represents that spread with a sequence of \(m\) equal-rate transit stages. For the convention used here, the transit delay has a gamma density with shape \(m\) and rate \(k_{\mathrm{tr},j}\):

Here \(m\) is a positive integer, and \(\mathrm{MTT}_j\) is the mean transit delay under this gamma-delay convention. The transit delay is followed by first-order absorption at rate \(k_{a,j}\). One way to express the systemic input rate is

The integral combines the distribution of transit delays with the first-order absorption process. Transit effects can be formulation-specific through \(k_{\mathrm{tr},j}\) or, equivalently, through \(\mathrm{MTT}_j\); \(m\) is commonly held fixed across formulations unless there is a strong reason and sufficient information to estimate a difference. This representation is not interchangeable with a single fixed lag: it spreads input over time. Which representation is appropriate depends on the scientific model and the shape of the data.

Why Log Contrasts Preserve the Intended Interpretation

The log-ratio representation follows directly from the requirement that a positive parameter remain positive. For a baseline value \(\theta_R>0\), define a formulation contrast \(\delta=\log(\theta_T/\theta_R)\). Exponentiating both sides gives \(\exp(\delta)=\theta_T/\theta_R\), and multiplying by \(\theta_R\) yields \(\theta_T=\theta_R\exp(\delta)>0\). Thus the model can estimate an unconstrained real-valued contrast \(\delta\), while every finite estimate implies a positive test parameter.

The contrast is also interpretable on the original scale: \(\exp(\delta)\) is the test-to-reference ratio. For instance, \(\delta=0.20\) implies a ratio of \(\exp(0.20)\), approximately 1.221, or about 22.1% higher—not exactly 20% higher. This distinction matters when translating fitted log contrasts into formulation-specific values.

Assumptions, Diagnostics, and Failure Modes

Before interpreting formulation parameters, check the model’s structural and data assumptions. Shared disposition means that, conditional on the model, \(CL\) and \(V\) do not change by formulation. Linear disposition means the model’s dose-proportional behavior is credible over the studied range. The chosen input model must also represent the observed onset and shape of absorption sufficiently well. Sparse or poorly timed samples can make a lag, absorption rate, and transit delay difficult to distinguish, even if each parameter has a precise scientific definition.

• Inspect residuals by formulation and time. A repeated early-time pattern in one formulation can indicate an inadequate lag or absorption-input model. A systematic late-time difference can be inconsistent with the assumed shared disposition, though residual patterns alone do not identify the cause.
• Check parameter uncertainty and dependence. Large standard errors, strong parameter correlations, estimates near imposed bounds, or substantial changes under reasonable starting values can signal weak identification or model instability.
• Check the implied profile. A positive \(k_a\) or \(k_{\mathrm{tr}}\) is mathematically valid but may still imply an implausible profile. Plot model predictions against observations by formulation and examine whether the estimated changes match the parts of the curve they are intended to represent.
• Do not interpret a normalized \(F\) as absolute availability. Setting \(F_R=1\) defines a comparison scale; it does not establish that the reference dose is completely available systemically.

A key failure mode is allowing \(F_j\), \(k_{a,j}\), lag time, transit parameters, \(CL_j\), and \(V_j\) all to vary without enough information to distinguish their effects. Several combinations can produce similar predictions. The remedy is not to assume that the most flexible fit is automatically best: specify a scientifically defensible shared-versus-formulation-specific structure, and examine whether the data can support each contrast.

Comparing Parameterization Choices

Two common approaches are direct formulation-specific parameters and a shared reference plus contrasts. Direct values, such as \(F_R\) and \(F_T\), are easy to state but may be inconvenient for optimization or comparison. A baseline-plus-contrast form, such as \(F_T=F_R\exp(\delta_F)\), makes the comparison explicit and ensures positivity. When several formulations are present, each can be compared with a chosen reference, but the reference choice changes the parameter values and not the fitted formulation predictions, provided the model is reparameterized consistently.

For input shape, a lag-time model is parsimonious when a fixed onset delay is plausible; a transit model represents a distribution of delays and has additional structural choices, including the number of stages. Neither is universally preferable. The model should not be selected solely because one parameterization gives a more convenient numerical result. Compare scientific plausibility, fit diagnostics, parameter stability, and whether sampling is informative about the added complexity.