EXAMPLES OF IN VITRO-IN VIVO RELATIONSHIPS WITH A DIVERSE RANGE OF QUALITY

Russell 1. Rackley

Biopharmaceutics Purepac Pharmaceutical Co., a subsidiary of Faulding, Inc. 200 Elmora Avenue Elizabeth, New Jersey 07207

1

The purpose of this chapter is to present an introduction of in vivo - in vitro relation­ships (lVIVRs) and illustrate a number of examples which IVIVRs may be used to evalu­ate oral, modified-release formulations. Examples given are meant to portray applications to extended-release formulations, which may be defined as a dosage form allowing for a reduction in dosing frequency as compared to that drug presented as a conventional dos­age form (1). "Controlled-release" throughout this chapter refers to an extended-release dosage form. In this chapter, in accordance with the meeting, IVIVR is used interchange­ably with IVIVC (in vitro - in vivo correlation).

BACKGROUND

In vitro-in vivo correlation (lVIVC) has been defined by the United States Pharma­copeia (USP) Subcommittee on Biopharmaceutics as: "the establishment of a relationship between a biological property produced by a dosage form, and a physicochemical charac­teristic of the same dosage form" (2). A Food and Drug Administration (FDA) interpreta­tion of in vitro-in vivo correlation has been cited (3) as: "To show a relationship between two parameters. Typically a relationship is sought between in vitro dissolution rate and "in vivo" input rate. This initial relationship may be expanded to critical formulation parame­ters and "in vivo" input rate." As suggested by the either of the cited definitions ofIVIVC, physicochemical properties of a dosage form other than dissolution should not be over­looked as an in vitro measurement. However, with respect to quality control testing, more weight tends to be placed on the cumulative dissolution of a dosage form over time as an in vitro indicator of in vivo performance. In vivo performance may typically be assessed in man by rate and extent of absorption of an oral dosage form. For controlled-release dosage forms, it is especially desirable to determine the cumulative absorption-time profile. AI-

In Vitro-in Vivo Correlations, edited by Young et al. Plenum Press, New York, 1997

2 R. J. Rackley

though, other endpoints of in vivo performance of a dosage form might be investigated, such as measurement of drug efficacy. The ultimate goal of an IVIVC should be to estab­lish a meaningful relationship between in vivo behavior of a dosage form and in vitro per­formance of the same dosage form, which would allow in vitro data to be used as a surrogate for in vivo behavior.

Skelly and Shiu (4) have inferred that dissolution testing evolved as a tool for bio­pharmaceutical investigation of in vitro-in vivo correlation throughout the 1950's and 1960's. The United States Pharmacopeial Convention has greatly influenced the stand­ardization and general acceptance of dissolution since this time (5). As a result of the in­troduction of different generic forms of digoxin around the early 1970's, the use of dissolution testing began to be widely recognized throughout the industry as a quality con­trol test (4). This was in fact due to the in vitro-in vivo correlations that were demonstrated between different digoxin products (6-8). Although the evolution of in vitro-in vivo corre­lation may be rooted in conventional immediate-release dosage forms, the concepts are applicable toward the development and support of controlled-release dosage forms.

RATIONALE OF IVIVC

For controlled-release products, an in vitro-in vivo correlation should be based on an intra-product comparison. That is, only variations of the drug-release rate for a specific formulation controlled-release mechanism may be considered in the correlation of a given product. Because different controlled-release products generally employ different control­led-release mechanisms, it may not be possible to make a comparison between different controlled-release products. It must be emphasized that in vitro-in vivo correlations for controlled-release formulations should be considered to be product-specific.

In vitro-in vivo correlation for controlled-release dosage forms would be of benefit if utilized in one or more of the following ways:

• surrogate to bioequivalency studies which might typically be required with scale­up or minor post-approval changes (SUPAC), where minor post-approval changes may include site of manufacture, formulation, or strength;

• support and/or validate the use of dissolution testing and specifications as a qual­ity control tool for process control, dissolution specifications in quality control ranges may be shown to be relevant to in vivo data;

• predict in vivo performance of a formulation based on in vitro dissolution data, termed "biorelevant dissolution" (9), which may be used in the justification of dissolution specifications and may aid in the design of formulation release-time profiles resulting in optimal plasma concentration-time profiles;

• identify appropriate dissolution conditions for a formulation which result in data relevant to in vivo performance.

CONSIDERATIONS

A number of rational considerations should be taken into account before attempting an in vitro-in vivo correlation for solid oral dosage forms (3,10-12). Factors affecting the success of an in vitro-in vivo correlation may include one or more of the following:

Examples of in Vit1'trin Vivo Relationships 3

• if a drug has a fairly narrow therapeutic window, an in vitro-in vivo correlation may still not be acceptable as a surrogate for bioequivalency testing (13)

• the pharmacodynamic properties (therapeutic or adverse) of the drug have been evaluated and there is minimal lag (or hysteresis) for effect vs. plasma concentra­tion (12)

• degree of linearity in pharmacokinetics of the drug may limit degree of correla­tion, and variation in the rate and extent of absorption or disposition may limit correlation

• the physicochemical nature of the drug is not foreseen as a limiting factor poten­tially leading to absorption variability in vivo

• the physical absorption of drug should not be the rate-limiting step in the absorp­tion process (i.e. is not permeation rate-limited)

• release from the dosage form is the rate limiting step in the overall absorption process (i.e. is dissolution rate-limited)

• ideally, the formulation is relatively insensitive to the range of variation expected within the in vivo environment (effect of pH, surfactants, agitation, etc.)

• dose (mg) and solubility (mg/mL) will determine the volume of dissolution media necessary to evaluate the formulation in vitro, sink conditions may be difficult to maintain for large volumes and may indicate problems in vitro and in vivo; how­ever, in these cases the USP Apparatus III and USP Apparatus IV may be investi­gated (12).

Logically, a higher degree of correlation may be expected with controlled-release formulations, since release from the dosage form tends to become the rate-limiting step in absorption, overcoming permeation rate limitations. Also, the time-frame allowed to char­acterize the profile of dissolution or absorption is much longer than with immediate-re­lease dosage forms, which permits a greater degree of accuracy and precision in characterizing the dissolution and absorption profile.

It has been recommended that validation of a Level A correlation for a controlled-re­lease formulation would be accomplished by preparing one or more batches of drug for­mulation which release at different rate(s) by varying critical formulation components that are likely to vary during normal manufacturing (1). A small pharmacokinetic study (N=6) may then be conducted to determine whether the Level A correlation is still supported by the new batches.

It should be realized that success with one drug formulation may not be extrapolated to all drugs or necessarily to different formulations of the same drug. All formulation de­velopment efforts for new drugs should be treated on a case-by-case basis.

DECONVOLUTION AND CONVOLUTION

Scheme I summarizes the relationships of deconvolution or convolution to in vitro­in vivo correlation or biorelevant dissolution. Based on a knowledge of the pharmacoki­netic system for a drug, the plasma concentration-time profile resulting from administration of an oral dosage form may be taken apart, or deconvoluted, to give an ab­sorption-time profile for the oral dosage form in vivo. Also, based on the assumption that release of drug from the controlled-release formulation is the rate-limiting factor in the ab­sorption process, then the absorption-time profile resulting from deconvolution may be considered to be indicative of in vivo dissolution. It is necessary to determine the absorp-

4

Observed Cp-tlme -. IVIVC profile

Simulated .- Cp-tlme

profile

~ Deconvol.

~

t .- PK -+ Convol.

system t Absorp. Dlssln.

:~:f~e -+ IVIVC .- :~:f~e

R. J. Rackley

Scheme 1. Relative approaches to IVIVC may be based on deconvolution or convolution methodologies.

tion-time profile in the assessment of Level A Correlations. The more common examples of deconvolution are the Wagner-Nelson equation (14) for one-compartment model drugs and the Loo-Riegelman equation (15) for two-compartment model drugs. Numerical de­convolution may have an advantage in that in vivo dissolution may be determined, using a program such as PCDCON (16). In general, the use of deconvolution methods (14---17) re­quires a fairly good understanding of a drug's pharmacokinetics and pharmacokinetic principles. These methods theoretically imply investigation of the drug administered intra­venously; but, in reality the use of oral solution or an immediate-release form of the drug might be an acceptable alternative.

Pharmacokinetic modeling for a drug may be necessary to use some deconvolution techniques correctly. Plasma concentration-time data may be analyzed with standard pro­grams such as NONLIN (18), SAS (19), or NONMEM (20) in the determination of an ap­propriate model and parameter estimates for the model. Here it is important to keep in mind the drug-related considerations listed previously for potential evaluation of an in vi­tro-in vivo correlation.

A knowledge of the pharmacokinetic system of a drug may be combined with, or convoluted with, the dissolution-time profile of an oral dosage form to simulate a plasma concentration-time profile for administration of the controlled-release dosage form (Scheme I). Leeson et al. (21) used this process as a method to evaluate prototype control­led-release formulations in vitro without performing more costly and time-consuming in vivo testing, and the concept has been termed "biorelevant dissolution." It is essential that sensitivity of the oral controlled-release dosage form be investigated by testing over a di­verse range of in vitro testing conditions in order to simulate the in vivo environment (see Dissolution Specifications). As with deconvolution, convolution may only be used when a valid pharmacokinetic system for the drug has been established. Simulations resulting from the combination of dissolution data and the pharmacokinetic system are only relative to the in vivo behavior of the reference form of the drug on which the pharmacokinetic system is based (21).

A relatively simple method of convolution may be based on the numerical integra­tion of a dissolution profile, transformed to rate, and input into the absorption compart­ment of an appropriately defined pharmacokinetic model. The simulation software known as STELLA II (22), provides a quick and efficient means of convoluting data, making model modifications, or testing different dosing regimens; however, meaningful parameter estimates must be provided to the model and both drug- and formulation- related consid-

Examples of in Vitro-in Vivo Relationships

100~----------------~

~80 C 60

~ 40

8? 20

o 4 8 12 16 20 24

Time (hr)

Figure 1. Qualitative, visualization of in vitro - in vivo relationship.

5

erations previously mentioned must be taken into account in order to generate meaningful simulations of in vivo performance.

LEVEL AIN VITRO-IN VIVO CORRELATION METHODS

• Qualitative:

The first test of a Level A correlation is to visualize the dissolution and absorption profiles together to assess the degree of superimposition. If the profiles do not appear to superimpose each other with respect to rate and extent of drug release, then pursuit of a more quantitative correlation may not be warranted. If, however, the profiles to appear to superimpose each other in this manner, possibly with a lag-time for absorption, then a more quantitative correlation may be explored.

This formulation represents a developmental formulation of highly soluble drug salt in an oral osmotic controlled-release system with an outer immediate-release component. Visualization of Figure 1 demonstrates a qualitative in vitro-in vivo correlation in which it may be concluded that there is a fairly good Level A correlation.

• Quantitative:

Linear Relationship of Absorption as a Function of Dissolution. Using x-y data pairs representing estimates of dissolution and absorption at common time points, linear regres­sion is performed. By far, this has been one of the more popular methods used initially to investigate a quantitative relationship between dissolution and absorption data. Superim­posable data should have a one-to-one relation; therefore, linear regression of absorption

-j---+----,.~I Lj y = -8.6 + 1.07*x

-- R2 =0.970

100 ~ 0 80 :;:::; c.

60 ~

0 !II

40 .D < :R. 20 0

0 0 20 40 60 80 100

"10 Dissolution

Figure 2. Quantitative, linear regression of in vitro - in vivo relationship.

6 R. J. Rackley

versus dissolution at common time points should demonstrate a linear relationship with a slope of one, an intercept of zero, and a coefficient of determination (r-squared) of one. It has been suggested that a y-intercept less than zero might be explained by a lag-time in absorption, whereas a positive y-intercept may require additional evaluation (3).

Quantitatively, the slope of this apparent relationship is close to unity; the intercept is slightly negative, indicating an overall lag of absorption relative to dissolution; and a fairly high coefficient of determination (r-squared) is observed (Figure 2). This further supports the notion of a Level A correlation for the data presented in Figure I, although there is systematic variation about the regression.

Hwang et al. (23) have mathematically demonstrated that the lag of absorption-time data in relation to dissolution-time data would be eliminated if in vivo dissolution-time data were estimated using numerical deconvolution methods. In cases where drug-release is constant (zero-order), then a simple time-shift in the absorption-time profile, equivalent to the reciprocal of the first-order absorption rate constant, may be used to superimpose dissolution-time data and establish a Level A correlation (23). When drug release is a slow first-order process, the same approximation appears to apply. In order to accommodate the

a 100

80

c: 0 60 .~

:; III

~ 40 lf1.

20

------r-----~F:~=r~~~:I:--' _ .. ( ... _ .••..•.........• c.· ..... ··.·· ...... i ..................... i

I : : :

I ! i i

···················r-···_· .. ··········r····.·.·.·· .. ·.· ................ -................ :! .......................... ·· .. ·· .. ·.·.·.·.· ...... :1 ........ ::· .. ··.· .. ·.·.·.·· .. ··.· .. ·.·· .. ·· .. ·.·.·.·.· ............ :1.:

.................. .1. ............ _ .... .1. : i 1

y. ml + m2·ma

.. Vaua ml -38.495 m2 1.3449

Chisq 20.35 ., R2 0.99562

0

a 20 40 60 80 100

% Dissolution

b 100 : y_ ml .m2·mO

au. 80 m1 ·5.2238

g~ '.;::s=.

m2 ·0.98497 ChisQ 24.509

E!-'" 0'1 60 R' 0.99565

.8':"

~~ 40 -1: o(/)

IZ 20

0

0 20 40 60 80 100

% Dissolution

Figure 3. a - Linear regression of absorption vs. dissolution, with a large negative y-intercept. This type of regres­sion may suggest a lag in estimated absorption. relative to dissolution. b - Linear regression for cubic spline fits of absorption (-1.35 hr) vs. dissolution. Shifting the absorption data by l/ka and correcting for lag in estimated ab­sorption demonstrates an improved linear relationship with dissolution data.

Examples of in Vitro-in Vivo Relationships 7

time shift, interpolation of data may be used if there is a high degree of confidence in the profile or process (Le. with dissolution in a controlled system).

For this example, a different set of data initially indicated that the dissolution and absorption profiles were quite similar, with the exception that the absorption appeared to lag behind the dissolution profile approximately 1 to 2 hr. Linear regression of absorption versus dissolution at common time points indicates a fairly large negative y-intercept and a slope greater than one, although there appears to be a strong linear relationship (Figure 3a). Based on the concepts presented by Hwang et al. (23), an investigation of a time shift was conducted in an effort to improve the correlation. The first-order absorption rate con­stant for the drug in this formulation was previously estimated as 0.741 hr- I , based on ad­ministration of a rapidly releasing oral dosage formulation. Therefore, an appropriate time shift of absorption data should be 1/0.714 or 1.35 hr. The absorption time profile was then shifted by a time interval or -1.35 hr to account for lag of absorption estimates to in vivo dissolution. Linear regression was again performed using absorption data shifted by -1.35 hr and the slope was nearly one with a y-intercept close to zero, and a strong linear rela­tionship remained (Figure 3b). The linear regression was weighted to 1/y1\2 to account for the higher distribution of data near 100% and mO is a variable representing the abscissa (x-axis).

NON-LINEAR RELATIONSHIP OF ABSORPTION AS A FUNCTION OF DISSOLUTION

There are a variety of possibilities in which non-linear functions may be investigated and the utility of this approach appears to be somewhat empirical. When linear regression does not indicate a good correlation, a non-linear function may indicate an excellent rela­tionship. Higher order polynomial equations may be used in many cases to describe data with curvature. Unlike a linear relationship for superimposable data, the parameter esti­mates for higher order polynomial equations are more difficult to interpret. Nevertheless, this approach might be considered as a valid method of expressing a relationship between dissolution and absorption data.

When a third order polynomial is explored for the data in Figure 1, a very good rela­tionship may be defined between the absorption and dissolution data (Figure 4); however, the coefficients should be reproducible for a Level A correlation.

Other methods suggested for investigation of nonlinear relationships include the use of Emax equation and its variations (24), the Weibull equation (25), and exponential or Gompertz equations (13).

100 c

80 0 ;:: Q.

60 ... 0 1/1

40 .a -< tft. 20

0 0 20 40 60 80 100

% Dissolution

Figure 4. Quantitative, non~linear in vitro - in vivo relationship.

8 R. J. Rackley

~

:::!i! t... 80

'E 60 Q) () 40 .... Q)

20 D-

o 0 4 8 12 16 20 24

Time (hr) ~

AI" 0.028 ref ABC =0.057

Figure 5. Quantitative, area between the curves; Resigno Index (RI) was 0.028, based on unweighted data up to 24 hr. Relative area between the curves (rei ABC) was 0.057, which means the area between the dissolution and absorption curves represents a deviation of 6% relative to the area under the dissolution curve up to 24 hr, which serves as the reference.

RESCIGNO INDEX

It has been previously proposed by Dr. J. Powers (13) that an index reported by Re­scigno (26) be considered for comparing the similarity of dissolution- and absorption­time curves. The method was originally proposed for the assessment of concentration-time profiles for bioequivalency. However, cumulative absorption and dissolution tend to pla­teau at some asymptote; therefore, the time up to which absorption and dissolution data are collected should be specified. For the example given in Figure 1, index of Rescigno (26) was calculated for data up to 24 hr with an exponent of one without weighting and the resulting index was 0.028.

RELATIVE AREA BETWEEN THE CURVES

A variation of the Rescigno Index is to relate the difference of area between the cu­mulative dissolution and absorption curves to the area under the cumulative dissolution curve, as dissolution is assumed to be the independent variable. When the index is calcu­lated in this way, it directly gives the fractional difference of absorption to dissolution. Again, since cumulative data are analyzed, the final time point used in the evaluation should be specified. The relative area between the curves for the data in Figure 1 was cal­culated as 0.057, indicating that the area between the curves up to 24 hr was less than 6% of the area under the dissolution-time curve up to 24 hr (see Figure 5).

If these methods do not appear to demonstrate a relationship of absorption to dissolu­tion, then an alternate method of dissolution should be investigated, which generates a disso­lution-time profile resulting in a closer match of the absorption-time profile. In any case the goal of an in vitro test should be to generate results that are indicative of in vivo performance.

F2 CALCULATION

With the possible exception of the F2 calculation (27), quantitative evaluation of dissolution data, as indicator of bioequivalence, has rarely been used. However, just in the same manner that the Rescigno Index was applied to dissolution and absorption data, it is conceivable that the F2 calculation be applied in a similar manner. For comparisons of ab-

Examples of in Vitrtrin Vivo Relationships 9

sorption and dissolution data, it may be appropriate to correct for lag in the absorption profile.

Except for the concept of "mapping", the issue of variability in dissolution data and its relation to absorption variability has not been taken into account in the discussion of these methods. Thus, there appears to be a need for more statistically based guidelines in the establishment of equivalency of dissolution and absorption profiles. Future develop­ments of in vitro-in vivo correlation and biorelevant dissolution will probably include the use of nonlinear mixed effect modeling in the evaluation of dissolution and absorption data, for solid oral formulations.

IVIVR FOR FOOD EFFECTS

There has been a growing interest in the ability to demonstrate in vitro - in vivo cor­relations for extended-release formulations, which may be subject to food effects. A some­what simplistic approach has been to examine the absorption time profiles for fasting and fed conditions, relative to the dissolution profile. As seen in Figure 6, this approach may only demonstrate a correlation for one condition or the other. Figure 6a represents rela­tively good linear correlation for absorption vs. dissolution for a fasting study of a modi­fied-release formulation. However, for the fed study phase it is apparent that a linear correlation would not be appropriate (Figure 6b).

A modern approach is exemplified by that of McCall et al. (28), where these authors attempted to optimize an in vitro test which mimicked post-prandial in vivo performance (Figure 7a-7d). This type of approach may be useful for screening and evaluation of ex­tended-release formulations that may be sensitive to food effects.

DISSOLUTION SPECIFICATIONS

With controlled-release formulations, there is an inherent need to "profile" the re­lease over time. Dissolution specifications define the acceptable range of dissolution-time data and should be representative of the profile and variability associated with a control­led-release dosage form. The USP (29) offers a guide suggesting the time over which the dissolution profile be related to the labeled dosing interval. And, there should be a mini­mum of three dissolution time points for dissolution testing of controlled-release dosage forms (12, 13): the first time point should assess dose dumping, the second or more time points should "profile" the dissolution-time curve, and the last time point should provide information as to the recovery of drug in the dosage form.

An investigation of the dependence of the formulation on pH and surfactants is rec­ommended in media of various compositions (12). Also, a dependence on dissolution equipment, and range of equipment settings, should also be considered in the investiga­tion. However, for controlled-release formulations that are sensitive to changes in the dis­solution environment, dissolution should be examined to determine the in vitro conditions which achieve an optimal IVIVC (1). Again, one or more batches of drug formulation which release at different rate(s} could be examined in an in vivo study to determine whether the correlation is supported. Once a formulation has been finalized for clinical use, dissolution data should be collected and tracked. The dissolution data should include batch-to-batch variation. Dissolution specifications have implications related to critical manufacturing parameters. Batches manufactured near the limit of a critical manufactur-

10

a 913

Ba

713 x

~ 613 L.J !XI 0: 0 to 513 !XI II:

to-Z

40 L.J U 0: L.J a...

30

213

10

0

b 90

B0

70 ~

x

~ 60 L.J !XI 0: 0 to 50 !XI II:

to-Z

40 L.J U 0: L.J a...

30

20

10

0

o

50 613

DISSOLUTION (%)

o

o

o o

30 40 50 60

DISSOLUTION (%)

.10p •• 1.178

loto,cop\--11.153

,2_ .9644

70 B0

o

110p.- 1.189

tn'.rc.p1.--2l.922

~2_ • B~93

B0

R. J. Rackley

o

90 100

90 100

Figure 6. a - Approximated linear relation of a modified release formulation administered under fasting condi­tions. b - Approximated linear relation of the same modified release formulation administered under fed conditions. Nonlinear nature relationship becomes more apparent under fed conditions.

Examples of in Vitro-in Vivo Relationships 11

120 100

a b • 100

80 'C .. ~ 80 0 .. .. 2i :;; 60

'C 60 .. of 0 ..

~ .. '" 40 c(

• .. 40 '" ?F-:! ~

20 20 •

Test: y -1.16>< + 0.17 rZ - 0.98 0 0

0 4 8 12 0 20 40 60 80 100

Time (hours) % Dissolved

C 100 d 100 --0--- % Dissolved (Type 111- OJI· 15 cpm)

% Absorbed 80 80 -

'C " ~ 0 .. 60 .. ~

'C 60

" € 0

• .. en of 40 0 .. '"

'" 40 c(

?F-•

:! ~ 20 20 -

• y = 0.9Sx - 0.67 r = 0.98 0 0

0 4 8 12 0 20 40 60 80 100

Time (hours) % Dissolved

Figure 7. a - Qualitative assessment of in vitro - in vivo relationship for TIMERx formulation administered under fasting conditions, for absorption compared to dissolution in different apparatus and media. b - Quantitative linear relationship of in vitro - in vivo relationship for TIMERx formulation administered under fasting conditions, for absorption vs. dissolution optimized for fasted conditions. c - Qualitative assessment of in vitro - in vivo relation­ship for TIMERx formulation administered under fed conditions, for absorption compared to dissolution optimized for fed conditions. d - Quantitative linear relationship of in vitro - in vivo relationship for TIMERx formulation ad­ministered under fed conditions, for absorption vs. dissolution optimized for fed conditions.

ing parameter must pass dissolution specifications. Extension of the range of a critical manufacturing parameter may be justified if the batch will pass dissolution specifications. The USP acceptance tables (29) appear to be based on an empirically developed "decision tree." These USP tables are used as a guide for setting specifications and possibly have roots in the evaluation of digoxin, mentioned previously.

Typically, the historically-based average plus or minus some measure of variation has been used to set dissolution specifications (Figure 8). Experience has demonstrated that the average ± 3.0 SD's may be used as a guide to setting dissolution specifications for some modified release formulations. A more meaningful method of developing dissolution specifi­cations would be through the use of in vitro-in vivo correlation or biorelevant dissolution based on a Level A correlation. Two ways have been suggested to go about this (1,12):

Deconvolution. Limits of variation observed in the actual plasma concentrations ob­served in clinical studies are assessed to determine absorption-time profiles (Scheme I, in

12

100% Dissolution

R. J. Rackley

Time

Figure 8. Example of dissolution-time data for a modified-release formulation with typical variation in dissolu­tion at set time points.

vitro-in vivo Correlation). For example, the 95% confidence intervals of observed plasma concentrations may be deconvoluted to provide absorption-time profiles that may be used as a guide to set dissolution specifications. This approach, however, has potential prob­lems associated with it because not all sources of variation in plasma concentration-time profiles are taken into account, and this approach tends to over-estimate the degree of variation that should be allowed in the dissolution specifications (Figure 9).

Rather, it would be more appropriate to deconvolute the variation associated with pharmacokinetics, which generally constitute a large portion of the overall variation ob­served in estimation of absorption.

Sources of Variation

Var(abs) Var(diss) Var(pk)

= F[Var(diss), Var(pk), Var(error)] = F[Var(intra-batch), Var(inter-batch), Var(assay), Var(error)] = F[Var(intra-subj), Var(inter- subj), Var(assay), Var(error)]

100% Absorbed Dose

Time

Figure 9. Example of absorption-time data with 95% confidence intervals as deconvoluted from mean plasma concentration-time data and respective 95% confidence intervals.

Examples of in Vitro-in Vivo Relationships

�0...----------

8

e 5

·~0~2~.~&~I~IO~I~2~1.~I8~I~I~~~2~2~2.~

Time (hrs)

13

Figure 10. Average, steady-state plasma concentrations (±SD) observed for patients administered a control-release formulation compared to simulations based on convolution of low, mid, and high dissolution specifications. This convolution approach justifies proposed dissolution specifications.

Here VarO represents a some function of variation for either absorption (abs), disso­lution (diss), or pharmacokinetics (pk). For each of these there may be further sources of variation, which are defined in the function ofF[].

Convolution. Limits of dissolution are used to simulate drug input to a pharmacoki­netic model associated with the active ingredient of the dosage form (Scheme I, Biorele­vant Dissolution). This is essentially the "biorelevant dissolution" approach, coined by Dr. L. J. Leeson (9). The resulting simulations of plasma concentrations are compared to vari­ation observed in the clinical studies.

Wagner-Nelson method (14) of deconvolution was used to estimate the absorption­time profile, which could be compared to cumulative dissolution-time data. Excellent lin­ear correlations of absorption versus dissolution at common time points were observed. The Rescigno index (calculated with an exponent of one and no weighting) was found to be 0.043 up to 24 hr, while the relative area between the curves was 0.072 up to 24 hr. Dissolution data from 5 clinical batches were pooled (individual tablets, n=60), and speci­fications were established based on the average ± 2.5 SD at 3, 6, 12, and 24 hr. The upper and lower limits of the dissolution specifications were then convoluted, based on an ap­propriate pharmacokinetic system for actual patients, and compared to variation observed in clinical, steady-state studies (Figure 10).

Although variation in resulting simulations are within the variation observed for the actual plasma concentration-time profiles, this approach still fails to take into account all pertinent sources of variation. That is, variation associated with the pharmacokinetic sys­tem have not been convoluted in with the simulations, giving plasma concentrations which are dependent on absorption.

A PROPOSED APPROACH TO VALIDATION OF DISSOLUTION SPECIFICATIONS

Newer methods for establishing dissolution specifications for controlled-release for­mulations have now been proposed in the recent guidance for development, evaluation and application of in vitro - in vivo correlations for extended release solid oral dosage

14 R. J. Rackley

forms (30). Certainly, any two lots of a product on the market should bioequivalent. How­ever, batches manufactured with averages centered at the limits of dissolution specifica­tions should not necessarily be required to be bioequivalent. A biostudy to investigate IVIVR over this in vitro range would provide some insight as to whether a correlation was valid over this range. Traditionally, dissolution specifications have been set to encompass the limits of variation associated with lot-to-lot manufacture. And, the specifications are set with sensitivity to detect production of units outside the normal lot-to-lot variation. However, a lot with average dissolution centered on the lower limit (or upper limit) of dis­solution specifications may very well not pass the dissolution specifications themselves and would theoretically not be on the market. For appropriate drug formulation candi­dates, side-batch formulations representing the limits of bioequivalence should be deter­mined. These formulation side-batches would then be used to determine the appropriate limits of dissolution. Comparison of side-batches representing dissolution specifications to a manufacturing target reference, or mid-point batch, may be a more reasonable approach, but it too may suffer from some of the same limitations just mentioned.

CONCLUSION

Applicability of in vitro-in vivo correlation has currently evolved to include demon­stration of a relationship of dissolution release specifications, based on variation in disso­lution, to variation in plasma concentrations resulting from administration of a controlled-release formulation. Ideally, an in vitro-in vivo correlation should demonstrate a relationship between critical formulation parameters and dosage form performance, in vitro and in vivo. The prospect of using an in vitro-in vivo correlation as a surrogate for bioequivalency studies appears to have come of age, especially for situations involving scale-up or post-approval changes of a controlled-release dosage form. Inference of opti­mal biological performance should come from a Level A correlation. The most practical use of IVIVCs in support of dissolution specifications would be to demonstrate a relation­ship holds true over the limits ofbioequivalency. However, actual specifications should be expected to be extend above and below the specific range validated so that the dissolution specifications will be useful as a tool for monitoring process control.

REFERENCES

1. AAPS/USP/FDA Workshop on Scale-Up of Extended-Release Dosage Forms; Crystal Gateway Mariott Hotel, Arlington, VA; Sept. 8-10, 1992; and Skelly JP, Van Buskirk GA, Arbit HM, et al.; Pharm Res 10(12):1800-1805,1993.

2. Pharmacopeial Forum; July-August, 1988; pg. 4160. 3. Cardot JM and Beyssac E, Eur J Drug Metab Pharmacokin, 18(1):113-120, 1993. 4. Skelly JP and Shiu GF; Eur J Drug Metab Pharmacokin, 18(1):121-129, 1993. 5. Cohen JL, Hubert BB, Leeson LJ, et al.; Pharm Res, 7:983-987, 1990. 6. Lindenbaum J, Butler VP, Murphy JE et al.; Lancet, June 2(1):1215-1217, 1973. 7. Johnson BF, McCrerie J, Greer H, et al.; Lancet, June 30(1):1473-1475,1973. 8. Shaw TRD, Raymond K, Howard MR, et al.; Br Med J, 4:763-766, 1973. 9. Leeson LJ; LJL Associates, Inc., personal communication.

10. Siewert M; Eur J Drug Metab Pharmacokin, 18(1):7-18, 1993. 11. Oosterhuis Band Jonkman JHG; Eur J Drug Metab Pharmacokin, 18(1): 19-30, 1993. 12. Pharmacopeial Forum, 19(3):5366-5379; May-June, 1993. 13. Generic Drugs Advisory Committee, Open Session; Rockville, MD; Jan. 11-12, 1994.

Examples of in Vitro-in Vivo Relationships 15

14. Wagner lG and Nelson E; J Pharm Sci, 52(6):61()...{j1l, 1963. 15. Loo lCK and Riegelman S; J Pharm Sci, 57:918-928,1968. 16. Gillespie WR; PCDCON: Deconvolution for Pharmacokinetic Applications; The University of Texas at

Austin, Austin, TX. 17. Chan K, Langenbucher F, and Gibaldi M; J Pharm Sci, 76(6):446-450, 1987. 18. PCNONLIN; Statistical Consultants, Inc.; Lexington, KY. 19. SAS; SAS Institute, Inc.; Cary, NC. 20. NONMEM; NONMEM Project Group, University of California; San Francisco, CA. 21. Leeson LJ, Adair D, Clevenger J, et al.; J Pharmacokin Biopharm, 13(5):493-514, 1985. 22. STELLA II; High Performance Systems, Inc.; Hanover, NH. 23. Hwang SS, Bayne W, Theeuwes F; 1 Pharm Sci, 82(11): 1145-1150, 1993. 24. Holford NHG and Sheiner LB; Clin Pharmacokin, 6:429-453, 1981. 25. Langenbucher F; Pharm Ind, 38(5):472-477, 1976. 26. Rescigno A; Pharm Res, 9(7):925-928, 1992. 27. Moore JW and Flanner HH, Pham Tech, June 1996, pg 64-74. 28. McCall TW. Diehl D, Baumgarner C, et al.; II th Annual AAPS Meeting, October, 1996, Poster PDD 7279

[TiMERx is a trademark of TIMER x Technologies]. 29. USP XXII; Chapter <711>, pg 1578-1579; and Chapter <724>, pg 1580--1581. 30. Malinowski H, Marroum P, Uppoor YR, et al.; Guidance for Industry; Extended Release Solid Oral Dosage

Forms; Development, Evaluation, and Application of in vitro - in vivo Correlations; Center for Drug Evalu­ation and Research (CDER), July I, 1996.