Differential Carryover Between Formulations

Why Formulation-Specific Residual Exposure Matters

A crossover comparison is intended to contrast each participant’s response to the test formulation, \(T\), with that participant’s response to the reference formulation, \(R\). The interpretation becomes less straightforward if the formulation used in one period leaves residual drug, metabolite, or pharmacodynamic activity that affects the next period. In particular, if the two formulations leave different residual effects, the measured contrast can combine the direct treatment difference with an influence from the preceding treatment.

Differential carryover is not simply a difference in concentrations measured before the second period. A concentration is evidence about residual exposure; carryover is the resulting influence of a previous treatment on the outcome measured in a later period. How strongly a residual concentration affects the outcome depends on the endpoint and the relevant pharmacokinetic or pharmacodynamic relationship.

Definition: Let \(\lambda_T\) be the additive effect on the measured outcome of residual influence from \(T\) when it is given in the preceding period, and let \(\lambda_R\) be the corresponding effect from \(R\). The differential carryover effect is \(\Delta_\lambda=\lambda_T-\lambda_R\). A nonzero \(\Delta_\lambda\) means the two preceding formulations have unequal effects on the later-period outcome scale.

The values of \(\lambda_T\) and \(\lambda_R\) are effects on the chosen analysis scale, not concentrations. For example, if the endpoint is log-transformed AUC, the carryover terms are shifts on that log scale. Their sizes and even their practical relevance can differ for AUC, \(C_{\max}\), a pharmacodynamic endpoint, or another outcome.

When Carryover May Differ Between Formulations

Formulations containing the same active ingredient can produce different concentration-time profiles. Differences in release rate, absorption duration, or the persistence of a formulation-associated depot can leave different concentrations at the start of the subsequent period. A long terminal phase, including one influenced by prolonged absorption, may make such residual exposure more consequential than a simple half-life calculation suggests. A metabolite can also matter if it persists and contributes to the measured endpoint.

Unequal predose concentrations are a reason to investigate differential carryover, but do not by themselves establish it. The measured concentration may not be pharmacologically active at the observed level; a short-lived concentration difference may have little effect on the endpoint; or a persistent pharmacodynamic response may remain even when measured concentrations are low. Conversely, if the relevant species is not measured, a low predose concentration of the parent drug does not rule out residual influence.

The important question is therefore two-part: could the formulations leave different residual exposure, and could that difference change the later-period outcome? The first part is informed by concentration-time data and pharmacokinetic knowledge. The second depends on the endpoint, the exposure-response relationship, and any persistent pharmacodynamic effects.

How Differential Carryover Enters the Treatment Contrast

Consider a two-period, two-sequence design with equal numbers assigned to \(TR\) and \(RT\). Let \(\tau_T\) and \(\tau_R\) be the direct effects of the current treatments on the outcome scale, and define the direct treatment contrast as \(\delta=\tau_T-\tau_R\). Let \(\pi_1\) and \(\pi_2\) denote period effects. Use an additive first-order carryover model: a preceding treatment can affect the second-period outcome by its \(\lambda\) term, and no such term contributes in period 1.

This model assumes that the carryover contribution is additive on the selected outcome scale, is determined by the preceding formulation, and does not depend on the current formulation. It also assumes a stable direct treatment contrast across periods. These are working assumptions, not guarantees about the biology. If carryover depends on the current treatment or effects vary substantially by period, the simple calculation below need not describe the distortion adequately.

Let \(D_{TR}\) be the within-participant outcome under \(T\) in period 1 minus the outcome under \(R\) in period 2, for participants assigned to \(TR\). The participant-specific contribution that is common to both periods cancels in the difference. The expected contrast is

E(D_{TR})=\delta+(\pi_1-\pi_2)-\lambda_T.

The minus sign on \(\lambda_T\) appears because residual influence from \(T\) increases the period-2 outcome under \(R\), which is subtracted. For participants assigned to \(RT\), define \(D_{RT}\) as their period-2 outcome under \(T\) minus their period-1 outcome under \(R\). Then

E(D_{RT})=\delta+(\pi_2-\pi_1)+\lambda_R.

Here residual influence from \(R\) contributes to the period-2 outcome under \(T\), which is not subtracted. When sequence sizes are equal, averaging the two sequence-specific contrasts cancels the period difference. The resulting treatment contrast is

\begin{aligned} E\left(\frac{D_{TR}+D_{RT}}{2}\right) &=\frac{\delta+(\pi_1-\pi_2)-\lambda_T +\delta+(\pi_2-\pi_1)+\lambda_R}{2}\\ &=\delta-\frac{\lambda_T-\lambda_R}{2}\\ &=\delta-\frac{\Delta_\lambda}{2}. \end{aligned}

Key result: Under the stated additive model and equal sequence balance, the expected treatment contrast is the direct contrast minus half the differential carryover effect. If \(\lambda_T=\lambda_R\), the common carryover contribution cancels in this balanced average. If \(\lambda_T\ne\lambda_R\), the contrast is shifted by \(-\Delta_\lambda/2\).

This result explains why “some carryover” and “differential carryover” are not interchangeable claims. Equal carryover can affect both second-period outcomes while cancelling from this particular balanced contrast. Unequal carryover can shift the contrast. The cancellation does not show that carryover is absent, and it depends on the model and balance conditions. With unequal sequence sizes, simply pooling the two sequence contrasts need not cancel the period effect in the same way.

Worked Example: Greater Residual Effect After the Test Formulation

Suppose a hypothetical balanced study has a direct contrast \(\delta=4.0\) outcome units, a period difference \(\pi_2-\pi_1=0.6\) units, and carryover effects \(\lambda_T=1.2\) and \(\lambda_R=0.2\) units. Apply the sequence-specific formulas.

For the \(TR\) sequence,

E(D_{TR})=4.0+(0-0.6)-1.2=2.2\text{ units}.

For the \(RT\) sequence,

E(D_{RT})=4.0+(0.6-0)+0.2=4.8\text{ units}.

The balanced average is \((2.2+4.8)/2=3.5\) units. Equivalently, \(\Delta_\lambda=1.2-0.2=1.0\) unit, so the expected contrast is \(4.0-1.0/2=3.5\) units. The estimate is shifted downward because the preceding \(T\) has the larger positive carryover effect. All values here are hypothetical and are expressed on the same outcome scale.

Worked Example: Equal Carryover Can Still Affect Both Sequences

In a second hypothetical study, let \(\delta=2.0\) outcome units, \(\pi_2-\pi_1=0.4\) units, and \(\lambda_T=\lambda_R=1.5\) units. The sequence-specific contrasts are

E(D_{TR})=2.0-0.4-1.5=0.1\text{ units}, \qquad E(D_{RT})=2.0+0.4+1.5=3.9\text{ units}.

Their balanced average is \((0.1+3.9)/2=2.0\) units, equal to the direct contrast because \(\Delta_\lambda=0\). The large difference between the two sequence-specific contrasts does not show that the balanced average is biased by differential carryover: in this model the carryover effects are equal, and the period effects also cancel in the average. It does show why interpreting one sequence alone would be misleading.

Using Predose Information Carefully

Predose concentrations before period 2 can help assess whether residual exposure differs by the preceding formulation. The comparison should be made with the design in mind: participants who received \(T\) in period 1 provide information about residual exposure after \(T\), and those who received \(R\) provide information after \(R\). Concentrations should be interpreted in relation to the assay, sampling time, relevant analytes, and endpoint.

A concentration difference does not directly supply \(\Delta_\lambda\). To translate concentrations into endpoint effects would require a justified exposure-outcome relationship, including its scale and relevant timing. For example, a linear relationship may be a useful illustration but can be inappropriate if the response is nonlinear, delayed, saturated, or driven by a different analyte. Predose data are therefore diagnostic evidence, not a direct correction factor.

Worked Example: From a Predose Difference to an Illustrative Bias

Suppose, hypothetically, the mean predose concentration attributable to the preceding formulation is \(0.6\) ng/mL after \(T\) and \(0.2\) ng/mL after \(R\). For illustration only, assume the later-period endpoint shifts linearly by \(3.0\) outcome units per ng/mL of residual concentration, with no other carryover mechanism.

Under that assumed relationship, the carryover contributions would be

\lambda_T=3.0\frac{\text{outcome units}}{\text{ng/mL}}\times0.6\text{ ng/mL}=1.8\text{ outcome units}, $$ $$ \lambda_R=3.0\frac{\text{outcome units}}{\text{ng/mL}}\times0.2\text{ ng/mL}=0.6\text{ outcome units}.

Thus \(\Delta_\lambda=1.8-0.6=1.2\) outcome units, and the balanced contrast would be shifted by \(-1.2/2=-0.6\) outcome units. If the direct contrast were \(5.0\) units, the expected balanced contrast under these assumptions would be \(5.0-0.6=4.4\) units. This arithmetic shows what an exposure-outcome model would imply; the predose difference alone does not justify the assumed slope or establish that the actual bias is \(0.6\) units.

Diagnostics, Alternatives, and Limitations

Useful checks include examining predose concentrations by preceding formulation, reviewing concentration-time profiles near the period boundary, and considering active metabolites or persistent pharmacodynamic effects. A pattern of higher residual exposure after one formulation makes differential carryover more plausible. Lack of a detected concentration difference is not conclusive if sampling is sparse, the relevant analyte is unmeasured, or the important effect is pharmacodynamic rather than concentration-mediated.

The treatment contrast also deserves attention by sequence. A strong difference between \(D_{TR}\) and \(D_{RT}\) can arise from period effects, differential carryover, or departures from the simple model; it is not, by itself, a diagnosis of any one cause. In a two-period design, direct treatment and carryover effects cannot generally be separated from the outcome data alone without additional assumptions. As established in the earlier material on carryover identifiability, the observed contrast can combine these effects rather than identify each independently.

The main alternatives address different parts of the problem:

Improve the washout. A longer interval may reduce residual exposure, but the appropriate interval depends on the relevant concentration and effect time courses. A half-life rule alone may not capture prolonged absorption, active metabolites, or persistent pharmacodynamic effects.
Use exposure information as a diagnostic. Predose and concentration-time measurements can support a mechanistic assessment, but they do not automatically correct the treatment contrast. Adjustment requires a defensible model and adequate measurements.
Choose a different design when carryover risk is material. A parallel comparison avoids a participant receiving both formulations in sequence, though it gives up the within-participant comparison and its associated advantages. Design choice should be made in light of the scientific question and expected residual effects.

A common mistake is to reason, “The two formulations have the same active ingredient, so any carryover must be equal and will cancel.” The identity of the active ingredient does not guarantee equal residual concentrations, metabolites, or endpoint effects. The correction is to assess both the residual exposure and its expected influence on the specific outcome. A second mistake is to treat different predose concentrations as proof that the treatment contrast is biased by a known amount. The correction is to distinguish a measured concentration difference from the unobserved or modeled carryover effect on the endpoint.

Key takeaway: In a balanced two-period crossover under an additive first-order model, differential carryover shifts the treatment contrast by \(-\Delta_\lambda/2\). Whether formulation-specific residual exposure creates a meaningful shift depends on how that exposure affects the measured endpoint, not on concentration differences alone.

Check Your Understanding

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

If \(\lambda_T=0.8\) and \(\lambda_R=0.8\) outcome units, what is the differential carryover bias in the balanced contrast under the stated model?
If \(\lambda_T=1.4\) and \(\lambda_R=0.4\) units, in which direction and by how much is the balanced contrast shifted?
Why does a predose concentration difference not by itself determine the carryover bias?
A sequence-specific contrast is \(1.0\) unit in \(TR\) and \(5.0\) units in \(RT\), with equal sequence sizes. What is their average? Can that difference alone identify differential carryover?

Solutions. (1) \(\Delta_\lambda=0.8-0.8=0\), so the shift is \(-\Delta_\lambda/2=0\) units; equal carryover can still affect both second-period outcomes. (2) \(\Delta_\lambda=1.4-0.4=1.0\) unit, so the contrast shifts by \(-1.0/2=-0.5\) unit, downward. (3) The concentration difference must be translated into an effect on the selected endpoint, which requires information or assumptions about the exposure-outcome relationship and other possible residual mechanisms. (4) The average is \((1.0+5.0)/2=3.0\) units. No: the sequence difference alone can also reflect period effects or departures from the additive model, so it does not identify differential carryover on its own.