What It Means for an IVIVC to Transport
An in vitro-in vivo correlation (IVIVC) is useful beyond the formulation that produced it only if it predicts in vivo behavior for formulations it did not use to establish the relationship. That question is narrower than asking whether two formulations have similar dissolution profiles, and different from asking whether an IVIVC can be made to fit a new formulation after its parameters are changed. Transportability asks whether the existing relationship, applied as specified, continues to predict the target formulation’s in vivo absorption or concentration profile.
The target formulation might differ in composition, excipient grade or amount, manufacturing process, or another feature that could affect release. Before testing transport, define which changes the intended claim covers. A model that transports between batches made by one process has not automatically been shown to transport to a formulation with a different release mechanism or to substantially different absorption conditions.
Let \(D_j(t)\) denote the fraction dissolved in vitro by time \(t\) for formulation \(j\), and let \(A_j(t)\) denote the fraction absorbed in vivo by time \(t\), defined on a consistent dose basis. Write the established Level A relationship as \(A_j(t)=g(D_j(t);\theta)\), where \(g\) is the specified mapping and \(\theta\) its fixed parameters. For a target formulation, transport testing applies this mapping to its dissolution profile and compares the result with in vivo observations for that target.
A transportability conclusion depends on more than a high correlation between two sets of points. The comparison must be prospective in logic: the target’s in vivo observations must not be used to choose a new time scale, lag time, slope, or transformation and then be presented as validation of the original model. Any allowed adjustments, such as a prespecified dose normalization, must be stated before examining the target results.
Conditions for a Meaningful Test
First specify the target domain and the prediction being tested. A prediction of the absorption profile is not identical to a prediction of the plasma concentration profile. When predicting concentrations, the disposition component used to translate absorption into concentrations must also be appropriate for the target. The comparison should use the same units, dose basis, sampling-time definition, and handling of any lag or time scaling as the established model.
The test is interpretable when the following conditions are reasonably supported:
- The dissolution method and sampling schedule are applied consistently, and the target profile is measured across the time range relevant to absorption.
- The target’s in vivo data are sufficiently informative to evaluate the prediction, and the absorption estimates or concentration measurements are obtained and processed consistently with the reference analysis.
- The drug’s disposition is stable across the comparison, or the concentration prediction accounts for any relevant difference using a prespecified approach.
- Dose, administration conditions, and population are comparable or explicitly accommodated. Changes in food, physiology, or other absorption conditions can alter in vivo behavior even when the in vitro profile is unchanged.
- The target lies within the scope of the claim. A result for one composition or process change does not establish transport across untested changes.
- Acceptance criteria and prediction summaries are defined before the target in vivo results are examined.
These conditions do not guarantee transport. They make a failure interpretable. If, for example, the dissolution method does not reflect the target’s release behavior, a discrepancy may point to a mismatch between the in vitro input and the in vivo process rather than simply to random prediction error. Likewise, if the target changes absorption conditions, a dissolution-based relationship may not capture the new in vivo behavior.
From Dissolution Prediction to an In Vivo Comparison
For an absorption-profile assessment, apply the frozen mapping at the target’s observed dissolution values to obtain predicted absorption fractions. Compare these with target absorption estimates at corresponding times. For a concentration-profile assessment, the absorption prediction must be translated through disposition. Under a linear, time-invariant disposition model, concentrations from successive absorbed amounts add by superposition.
Here \(h(t-u)\) is the concentration response at time \(t\) to a unit amount absorbed at time \(u\), expressed in units consistent with the chosen dose basis. The integral sums the contributions of absorption occurring from time zero through \(t\). This expression requires a suitable disposition function and linear superposition; it is not justified if the disposition response changes with concentration or with the target formulation.
At a selected time \(t_i\), an absorption-scale residual can be written \(e_i=A_{\mathrm{obs}}(t_i)-A_{\mathrm{pred}}(t_i)\). Positive values mean the observed fraction absorbed exceeds the prediction at that time; negative values mean it is below. Residual patterns across time are often more informative than a single summary. A persistent negative discrepancy near the end of the profile, for example, may suggest a difference in extent or incomplete absorption, while a discrepancy concentrated early may suggest a timing mismatch. Neither pattern alone proves a mechanism.
Prediction limits should be justified for the intended decision and specified in advance. A mean absolute error, maximum absolute error, or comparison of selected profile features can summarize the differences, but no one numerical cutoff is universal. The uncertainty in the estimated IVIVC, dissolution measurements, disposition, and target in vivo data should also be considered; otherwise, a small apparent discrepancy can be overinterpreted, or an imprecise prediction can appear reassuring.
Worked Transportability Assessments
Worked Example: A Target Profile Within Prespecified Limits
Hypothetical data. Suppose a reference IVIVC has the fixed relationship \(A=D\), and the planned target assessment uses absorption fractions at four matching times. Before examining the target in vivo results, the team specifies that each absolute difference must be no greater than 0.10 fraction units and that the mean absolute difference must be no greater than 0.06. These are illustrative study criteria, not universal standards.
| Time (h) | Target \(D\) | Predicted \(A\) | Observed \(A\) | Observed minus predicted |
|---|---|---|---|---|
| 1 | 0.20 | 0.20 | 0.22 | 0.02 |
| 2 | 0.45 | 0.45 | 0.42 | -0.03 |
| 4 | 0.72 | 0.72 | 0.75 | 0.03 |
| 8 | 0.90 | 0.90 | 0.88 | -0.02 |
Because \(A_{\mathrm{pred}}=D\), the predicted values are 0.20, 0.45, 0.72, and 0.90. The absolute differences are 0.02, 0.03, 0.03, and 0.02 fraction units. Their mean is
The largest absolute difference is 0.03, below 0.10, and the mean is 0.025, below 0.06. Thus these hypothetical point estimates meet both prespecified criteria. This supports transport for the defined target and assessment, subject to the precision of the observations and the validity of the comparison conditions. It does not establish that the relationship will transport to other formulations.
Worked Example: A Persistent Difference Suggesting an Excipient Effect
Hypothetical data. The same reference model predicts \(A=D\). A target formulation contains a changed excipient level. Its in vitro dissolution fractions at 1, 2, 4, and 8 hours are 0.20, 0.45, 0.72, and 0.90; observed in vivo absorption fractions are 0.18, 0.40, 0.65, and 0.82. Use the illustrative limits from the preceding example.
The predictions equal the dissolution values. The observed-minus-predicted differences are \(0.18-0.20=-0.02\), \(0.40-0.45=-0.05\), \(0.65-0.72=-0.07\), and \(0.82-0.90=-0.08\). Their absolute values sum to \(0.02+0.05+0.07+0.08=0.22\), so the mean absolute difference is \(0.22/4=0.055\). The largest absolute difference is 0.08.
The maximum difference meets the 0.10 limit, but the mean difference of 0.055 is below the 0.06 limit—so this dataset actually meets both specified criteria. It nevertheless shows a consistent negative pattern, with observed absorption increasingly below prediction. The pattern deserves investigation: it could reflect a systematic target-specific effect that the summary criteria do not rule out. Possible explanations include an excipient effect on absorption or a mismatch between in vitro dissolution and in vivo release. These values alone cannot distinguish those explanations. The correct conclusion is that the prespecified numerical criteria are met, but the directional pattern is a diagnostic signal to assess before making a broader transportability claim.
Worked Example: A Release-Profile Mismatch That Fails the Criteria
Hypothetical data. Suppose the fixed reference relationship is again \(A=D\). For a target manufactured by a changed process, the dissolution fractions at 1, 2, 4, and 8 hours are 0.10, 0.30, 0.60, and 0.85. The corresponding observed absorption fractions are 0.02, 0.15, 0.45, and 0.78.
The predictions are 0.10, 0.30, 0.60, and 0.85. The observed-minus-predicted differences are \(-0.08\), \(-0.15\), \(-0.15\), and \(-0.07\). Therefore,
The maximum absolute difference is 0.15 fraction units, and the mean absolute difference is 0.1125. Both exceed the illustrative limits of 0.10 and 0.06. The frozen model therefore fails this target-profile assessment. The discrepancy is largest at 2 and 4 hours, rather than increasing steadily to the final observation. That pattern is compatible with a difference in release timing or mechanism, but it does not identify the cause. The next step is to check dissolution conditions, sampling and time alignment, manufacturing changes, and the target in vivo data—not to alter the mapping until the discrepancies disappear and then call the altered model a successful test of the original.
Diagnosis, Alternatives, and Limitations
When predictions are poor, first check the comparison itself. Confirm that the frozen model was applied as specified; that dissolution values and in vivo observations refer to corresponding times and dose bases; and that the target profile has not been shifted, truncated, or otherwise processed differently. Examine both the dissolution and in vivo profiles, not only a summary error. A changed release mechanism, an excipient with an in vivo effect not represented by the dissolution test, or changed absorption conditions can all produce a real transportability failure. They are hypotheses to investigate, not conclusions supplied by the residual pattern alone.
Three broad approaches answer different questions:
| Approach | What it answers | Main limitation |
|---|---|---|
| Apply the frozen IVIVC to the target | Whether the existing relationship predicts this target without outcome-driven change | A failure identifies a problem for the tested domain but may not identify its mechanism |
| Refit or recalibrate using target data | Whether a revised relationship can describe the combined data | Fit after using target outcomes does not independently demonstrate transport of the original model |
| Develop a formulation-specific or mechanistic model | Whether additional formulation or process features can explain behavior | Requires suitable data and assumptions; greater complexity does not guarantee reliable prediction |
Refitting may be appropriate when the purpose is to develop a revised model. It should be distinguished from testing transport, and the revised model requires its own assessment on data not used to fit it. Likewise, a formulation-specific model may be scientifically useful without implying that the original IVIVC was transportable.
A transportability claim remains limited by the tested formulations, processes, dissolution conditions, in vivo conditions, and prediction range. Passing one set of criteria does not prove that the underlying release and absorption mechanisms are identical. Conversely, failure does not establish that an IVIVC is useless; it may remain predictive within a narrower domain. The intended use determines whether the observed accuracy and uncertainty are adequate.
Common Pitfall
A related error is to adjust the target’s time scale or model parameters using its in vivo outcomes, then report the improved fit as evidence that the original model transported. That procedure evaluates a revised relationship, not the unchanged model. Keep the initial transport test separate from any subsequent model redevelopment.
Check Your Understanding
Answer each question using the definitions and criteria in this tutorial.
- A target formulation’s dissolution profile falls within the reference formulation’s observed range. Does that alone establish transportability? Why or why not?
- A frozen model predicts an observed absorption fraction of 0.68, while the target estimate is 0.61. Calculate the observed-minus-predicted residual and interpret its sign.
- Why does changing the IVIVC parameters after examining target in vivo data prevent the result from being a test of the original model?
- A target’s residuals are consistently negative at late times. Give two possible explanations and one reason the pattern does not prove either explanation.
Solutions. (1) No. In vitro similarity alone does not establish in vivo predictive accuracy; apply the fixed model and compare its prediction with target in vivo observations under suitable conditions. (2) \(0.61-0.68=-0.07\) fraction units. The observed estimate is 0.07 below prediction at that time. (3) The target outcomes have then been used to change the model being assessed, so the resulting fit describes a revised model rather than independently testing the frozen relationship. (4) Possibilities include a difference in absorption extent or a mismatch in release behavior captured by the dissolution method. The residual pattern is not mechanistic proof; other model, measurement, or comparison issues may also explain it.