Why IVIVC Has More Than One Level
A dissolution profile describes drug release from a dosage form under specified laboratory conditions. A pharmacokinetic profile describes drug concentrations in the body over time. An in vitro–in vivo correlation (IVIVC) is a predictive relationship between an in vitro measure of drug release and an in vivo measure related to drug input or exposure. Its value depends not only on whether the two measures move together, but on how much of the in vivo behavior the relationship represents.
The levels distinguish different kinds of correspondence. Level A relates profiles point by point and can, when adequately developed and evaluated, support prediction of the full in vivo profile. Level B compares summary moments and compresses each profile into a small number of statistics. Level C relates one dissolution measure to one pharmacokinetic parameter. These are different evidential claims, not simply three labels for correlations of increasing numerical strength.
The level describes the structure of the relationship. It does not, by itself, establish that the relationship predicts well. That requires suitable data, a scientifically appropriate in vitro test, and evaluation of prediction performance. In all three levels, a correlation observed among formulations is most interpretable when the formulations differ in release behavior while other influential product attributes and study conditions are sufficiently controlled.
Level A: Point-to-Point Profile Relationship
A Level A relationship links the amount or fraction dissolved in vitro to the amount or fraction entering the body over time. The in vivo quantity is commonly a fraction absorbed or an estimate of drug input derived from pharmacokinetic data. The paired quantities should correspond across time, possibly after an appropriate alignment of their time scales. A relationship can be linear or more complex; the essential feature is point-to-point mapping rather than a comparison of only one summary value.
The data therefore need to represent both in vitro dissolution profiles and in vivo input profiles for relevant formulations. In vivo input is not usually observed directly: it must be estimated using an appropriate pharmacokinetic analysis. The validity of that estimate is part of the evidence. Further, a useful relationship requires that the in vitro test meaningfully distinguish release behavior and that the formulations span a relevant range of release rates. If all formulations dissolve almost identically, the data provide little information about how the relationship behaves across release rates.
A well-supported Level A relationship has the broadest potential predictive meaning of the three levels: it can connect an entire in vitro dissolution profile to the corresponding in vivo input profile and, with an appropriate pharmacokinetic model, support prediction of the concentration-time profile. That potential is not automatic. The relation must be supported over the range where it will be used, and its predictions must be checked. A visually close match in a small development data set is not, on its own, evidence of reliable prediction for other formulations.
Worked Example: Reading a Hypothetical Level A Pairing
Suppose a hypothetical modified-release product has the following aligned measurements. The in vivo values are estimates of cumulative fraction absorbed, and all percentages refer to the administered dose.
| Time (h) | In vitro dissolved (%) | In vivo absorbed estimate (%) | Absolute difference (percentage points) |
|---|---|---|---|
| 0.5 | 10 | 8 | 2 |
| 1 | 30 | 28 | 2 |
| 2 | 60 | 58 | 2 |
| 4 | 85 | 83 | 2 |
| 6 | 95 | 94 | 1 |
At each listed time, the absolute difference is calculated as \(|\text{dissolved}-\text{absorbed estimate}|\). For example, at 2 h it is \(|60-58|=2\) percentage points. The mean absolute difference across these five listed pairs is
The close paired values illustrate the form of a point-to-point relationship. They do not establish a validated Level A IVIVC: these five observations are from one illustrative profile, and there is no evidence here about variability, other release rates, or prediction for an untested formulation. The example is useful for identifying the data structure, not for claiming predictive performance.
Level B: Relationships Between Moments
A statistical moment summarizes a profile. For dissolution, one useful summary is the mean dissolution time (MDT), the average time associated with release of the dose. For interval data, it can be approximated by assigning each increment of dissolved drug to the midpoint of the interval in which it dissolved, then averaging those times with the increments as weights. A Level B correlation relates such a summary to an appropriate in vivo moment, such as mean absorption time or a suitable mean residence time measure.
This approach needs dissolution profiles and in vivo data from which the relevant moment can be estimated. It uses more profile information than a single-time-point comparison, but it reduces each profile to a summary. Consequently, a relationship between moments does not identify the full shape of the in vivo profile. Different release patterns can have the same MDT, and a matching average time does not guarantee matching concentrations at particular times.
Worked Example: Calculating a Mean Dissolution Time
Consider hypothetical cumulative dissolution measurements from one formulation: 25% dissolved by 1 h, 60% by 2 h, 85% by 4 h, and 100% by 6 h. The fractions released in the successive intervals are 0.25, 0.35, 0.25, and 0.15. Their interval midpoints are 0.5, 1.5, 3, and 5 h.
Using the midpoint-weighted calculation,
The fractions sum to 1.00, so the weights account for the full dose. This result says the weighted average release time is approximately 2.15 h under this interval approximation. It does not say that 50% of the dose dissolves at 2.15 h, nor does it specify the complete dissolution curve. To make a Level B comparison, an appropriate in vivo moment would also be estimated for the same formulation and compared across formulations; this single MDT alone is not an IVIVC.
The limitation follows directly from what a mean retains. Suppose, ideally, all dose is released at 2 h. Its mean release time is 2 h. Alternatively, half the dose could be released at 1 h and half at 3 h:
Both patterns have the same mean, although their release profiles differ. Thus, a moment relationship can support comparisons of average timing or formulation ranking, but cannot by itself reconstruct a unique time course.
Level C: A Single Dissolution Measure and a PK Parameter
A Level C correlation pairs one selected dissolution measure—such as percentage dissolved at 2 h or time to a specified dissolution fraction—with one pharmacokinetic parameter. Common examples of the latter include \(C_{\max}\) and AUC. Its data requirements are less extensive than those of a profile-level relationship: each formulation needs the selected in vitro measure and the corresponding in vivo parameter. The selection should be scientifically justified rather than chosen only because it happens to produce a strong association in the available data.
Because the relationship uses one dissolution point and one pharmacokinetic summary, it does not describe the full in vivo time course. A strong association with AUC, for example, does not show that the concentration profile has the right shape or that peak concentration is predicted. A single Level C relationship can be useful for early formulation screening or ranking when the development question concerns that specific parameter. It generally offers more limited predictive scope than a supported Level A relationship.
Worked Example: Using a Single Level C Relationship
Suppose four hypothetical formulations give the following paired values. The dissolution measure is percentage dissolved at 2 h; AUC is in \(\text{mg}\cdot\text{h}/\text{L}\).
| Formulation | Dissolved at 2 h (%) | AUC (\(\text{mg}\cdot\text{h}/\text{L}\)) |
|---|---|---|
| A | 30 | 75 |
| B | 50 | 90 |
| C | 70 | 105 |
| D | 90 | 120 |
These hypothetical points lie on the line \(\mathrm{AUC}=52.5+0.75x\), where \(x\) is percentage dissolved at 2 h. For example, the change from formulation A to B is \(15\ \text{mg}\cdot\text{h}/\text{L}\) in AUC for a 20-percentage-point increase in dissolution, giving \(15/20=0.75\) AUC units per percentage point. For a hypothetical formulation with 60% dissolved at 2 h, the line gives
This is an illustrative interpolation within the observed dissolution range of 30% to 90%. The exact alignment of four points is deliberately simple and does not provide evidence of uncertainty or predictive performance. Also, this calculation predicts only AUC under the proposed relationship; it supplies no prediction for \(C_{\max}\), absorption timing, or the complete concentration-time profile.
A multiple Level C approach relates multiple dissolution time points to one or more pharmacokinetic parameters. It can contain more information than a single-point relationship and may be useful for formulation ranking. However, it remains distinct from Level A: multiple selected points do not automatically constitute a point-to-point mapping of the complete in vivo input profile.
Choosing a Level for the Development Question
The appropriate level depends on the decision to be supported. If the question is whether a formulation’s full release behavior corresponds to in vivo input across time, Level A is the relevant target. If the question concerns average release and absorption timing, a Level B comparison may be informative. If the immediate task is to screen or rank formulations against a particular exposure parameter, Level C may be adequate for that limited purpose.
| Level | Paired quantities | Typical development use | Main limitation |
|---|---|---|---|
| A | In vitro dissolution profile and in vivo input profile | Assess profile-level correspondence and potential prediction of an in vivo profile | Requires credible input estimates and evidence across relevant release behavior |
| B | In vitro and in vivo statistical moments | Compare average timing or rank formulations by summary behavior | Different profiles can share the same moment |
| C | One dissolution measure and one PK parameter | Screen or rank formulations for a specified parameter | Does not characterize the complete PK profile |
The key distinction is between association and predictive scope. Each level can reveal a relationship in development data, but the level determines what that relationship directly represents. Prediction beyond the observed formulations requires additional evidence; a label such as “Level A” or a high correlation coefficient cannot substitute for that evaluation.
Assumptions, Diagnostics, and Common Pitfalls
Interpretation at any level depends on whether the measurements are suitable for the intended comparison. The dissolution method should be sufficiently reproducible and sensitive to meaningful release differences. The in vivo parameter or profile should be estimated comparably across formulations, and the study should not confound release-rate changes with unrelated changes that strongly affect absorption or disposition. The formulation set should cover the range where conclusions will be applied. These are scientific conditions to assess, not guarantees that can be inferred from a correlation statistic alone.
Warning signs include dissolution profiles that are nearly indistinguishable, an apparent relationship driven by one formulation, inconsistent rankings across repeated data, or a relationship that bends or changes direction over the observed range when a simple linear form is assumed. For Level A, poor correspondence at particular times can be obscured by an overall summary. For Level B and C, residuals and plots can expose systematic patterns, but even a good-looking fit does not remove the information lost by reducing profiles to moments or individual values.
- Pitfall: treating a Level C association as a full-profile prediction. A learner may reason, “AUC tracks dissolution at 2 h, so the complete concentration-time curve is predictable.” The conclusion does not follow: AUC is one summary and the dissolution value is one point. The correction is to limit the claim to the specific parameter and range evaluated.
- Pitfall: assuming a higher level guarantees validity. Calling a relationship Level A does not show that it predicts accurately. The correction is to separate the type of relationship from evidence of predictive performance and to evaluate the intended use.
- Pitfall: interpreting a Level B match as matching profiles. Equal mean times can arise from different release patterns. The correction is to state that the moment summaries correspond, not that the underlying curves are identical.
These levels are complementary rather than interchangeable. A simple Level C analysis can answer a narrow early-development question efficiently; a moment comparison can add information about average timing; and a Level A approach can address profile-level correspondence when its demanding data and analysis requirements are met. None should be selected merely because it is easiest to calculate: the claim should match the evidence and the decision at hand.
Check Your Understanding
Answer each question, then compare your reasoning with the solutions below.
- A team relates percentage dissolved at 3 h to \(C_{\max}\) across formulations. What level is this, and what does it not establish?
- Why can a Level B relationship fail to determine the shape of an in vivo concentration profile?
- For a hypothetical dissolution profile, 40% dissolves in the first hour and the remaining 60% dissolves from 1 to 3 h. Using interval midpoints, calculate the approximate MDT.
- A formulation has 65% dissolved at 2 h. A proposed Level C equation is \(\widehat{\mathrm{AUC}}=40+0.8x\), where \(x\) is percent dissolved at 2 h and AUC is in \(\text{mg}\cdot\text{h}/\text{L}\). Calculate the prediction and name one conclusion it does not support.
- What additional feature distinguishes a multiple Level C analysis from a Level A relationship?
Solutions. (1) This is Level C: one dissolution measure is related to one PK parameter. It does not establish prediction of the full concentration-time profile or other parameters. (2) A moment compresses a profile into a summary; distinct profiles can have the same mean, so the moment does not uniquely specify the curve. (3) The interval increments are 0.40 and 0.60, with midpoints 0.5 h and 2 h. Thus, \(\mathrm{MDT}\approx(0.5)(0.40)+(2)(0.60)=0.20+1.20=1.40\) h. (4) \(\widehat{\mathrm{AUC}}=40+0.8(65)=92\ \text{mg}\cdot\text{h}/\text{L}\). This does not establish a prediction for \(C_{\max}\) or the full profile. (5) Multiple Level C uses several selected dissolution points, but it is not automatically a point-to-point mapping across the complete in vivo input profile, which is the defining feature of Level A.