Documente Academic
Documente Profesional
Documente Cultură
ABSTRACT: The goal when developing an in vitro–in vivo correlation (IVIVC) model is
the ability to accurately predict the in vivo plasma concentration profile of a drug
formulation using only its in vitro dissolution data. The prediction accuracy of any model
depends on the reliability of the method used to develop it. Some statistical concerns
regarding methods based on deconvolution have been highlighted and a convolution
based technique has been proposed as an alternative. This comparison shows, by means
of a simulation study, that the modelling approach which uses convolution produces far
more accurate results, accurately predicting the observed plasma concentration–time
profile and, therefore, comfortably meeting the FDA validation criteria. The fact that the
model developed using the deconvolution based technique fails to describe the simulated
data and thus fails the FDA validation test when it ought to pass should be of great
concern to those currently implementing this method. ß 2007 Wiley-Liss, Inc. and the
American Pharmacists Association J Pharm Sci 97:3422–3432, 2008
Keywords: in vitro–in vivo correlations (IVIVC); mathematical model; deconvolution;
convolution
DOI 10.1002/jps JOURNAL OF PHARMACEUTICAL SCIENCES, VOL. 97, NO. 8, AUGUST 2008
3424 GAYNOR, DUNNE, AND DAVIS
parameters obtained from the reference data to tion has a negative impact on the model’s
produce plasma concentration profiles. predictions.
Once these fractions are plotted against one
another, an appropriate line or curve is fit to the
data. There is an abundance of instances docu-
mented in the literature where straight-line
Some Statistical Concerns
models are fit to data where a curve would be
There are many theoretical arguments against more appropriate.9–11 This can lead to inaccurate
using deconvolution based methods which are (either over or under) predictions. It is apparent
described in detail elsewhere.2 Descriptions of that care must be taken when choosing the model
those weaknesses which are believed to have the which describes the relationship between the
most significant effect on the model’s ability to in vitro and in vivo fractions dissolved.
predict with accuracy are given below, and are The process of plotting in vivo fractions dissolved
also summarised in Table 1. at a particular time against their in vitro counter-
Very often, the observed data are averaged parts is based on the assumption that their
across subjects or dosage units at each time point relationship is time invariant. This procedure
before analysis. This leads to a significant loss of forces the assumption that the link between
information particularly if the between-subject in vitro and in vivo dissolution does not change
(or dosage unit) variation is high. There may also over time; this is a somewhat ingenuous approach.
be a marked difference between the curve which is In effect, the medium in which the dosage unit
being analysed (made up of data averaged at each dissolves in vitro does not change as time passes,
time point) and the original individual profiles. but it may significantly alter in vivo as the dosage
Deconvolution can, of course, be carried out at the unit’s location in the gastrointestinal tract, and
individual level6–8 but this is not as frequently consequently the pH, changes. Of course this is
seen as analysis based on averages. In order to not necessarily the case, there are many solid
ensure comparable conditions, the methods of systems that are inert and whose dissolution is not
analysis being compared in this study were both influenced by their location in the gastrointestinal
carried out at the individual subject level. tract. It may, however, be unwise to assume that
When the in vivo dissolution rate has been this is always true and that the resulting IVIVC
converted to give the fraction dissolved in vivo at a relationship is time-invariant.
given time, these fractions are plotted against Generally, the method of ordinary least squares
the in vitro fractions dissolved at the same time. is used to fit the chosen model but two of its
Clearly only in vitro and in vivo data collected at main assumptions are violated in this case. The
common times can be used. This leads to the first is that the independent variable (usually
discarding of data where the observation times in vitro fraction dissolved) is measured without
are not coincident. The resulting loss of informa- error, which, in terms of this application, is an
JOURNAL OF PHARMACEUTICAL SCIENCES, VOL. 97, NO. 8, AUGUST 2008 DOI 10.1002/jps
PREDICTION ACCURACY OF TWO IVIVC MODELLING TECHNIQUES 3425
unattainable expectation and is known as the function of time and incorporates random effects
Errors in Variables issue.12 Secondly, it is which can account for some of the correlation
required that all observations be uncorrelated between repeated measurements on the same
or independent while it is clear that measure- subject or dosage unit. Observations in vitro and
ments made on the same subject or dosage unit in vivo do not have to be made at corresponding
cannot be uncorrelated. Violating these two times, eliminating the loss of information caused
underlying assumptions of the least squares by discarding some of the data. The use of this
method results in a model which can produce method removes the need for a deconvolution step
inaccurate predictions. and predicts plasma concentrations rather than
A fundamental shortcoming of this entire group their deconvolved equivalents.
of methods is that the process of deconvolution
is itself unstable13 and may be thought of as a
‘roughening’ of the data. It, therefore, by nature, METHOD—SIMULATION STUDY
introduces errors and reduces the accuracy with
which the model can predict the plasma concen- A simulation study to compare these two appro-
tration profiles. aches to establishing an IVIVC was undertaken.
There are a number of ways of evaluating Four batches of a drug, each with a different
accuracy of prediction, two of the most commonly release rate, were simulated. The first three
used measures are bias and efficiency. A biased batches will be referred to as the fast, medium
estimate will tend to consistently over or under and slow formulations while the fourth was
predict the quantity of interest, for example, AUC included for the purposes of external validation.
or Cmax. A low bias is therefore usually preferable. The data were simulated such that a marked
An inefficient estimate will produce predictions relationship between in vitro and in vivo dissolu-
that are widely scattered around the true value. tion would exist. The three kinds of data: in vitro,
The combination of the issues mentioned above in vivo ER and reference were produced for 12 ER
is believed to have a substantial impact on the dosage units and 12 subjects with 22 observations
accuracy with which the deconvolution method made from time of administration (0 h) up to 34 h
can predict the plasma concentration curve, and it in order to simulate a study with a crossover
is expected that this method will produce biased design.
and inefficient predictions. The reference data were simulated according to
a one compartment pharmacokinetic model with
first order absorption as follows:
CONVOLUTION BASED APPROACH
l3k
cdk ðtÞ ¼ ðel2k t el3k t Þ (2)
Vk ðl3k l2k Þ
An alternative modelling approach which incor-
porates the convolution integral (Eq. 1) has been where Vk, l2k and l3k represent the kth subject’s
developed. This method, like the deconvolution volume of distribution and rate constants of drug
based method described above, assumes the elimination and absorption, respectively. The
linearity and time invariance of the system being plasma drug concentration for the kth subject at
studied.4 The version of this convolution based time t following administration of a unit reference
method proposed by O’Hara et al.3 first fits a dose is given by
standard compartmental model to the unit
Y3k ðtÞ ¼ cdk ðtÞ þ "3k ðtÞ (3)
impulse response data in order to estimate each
subject’s pharmacokinetic parameters separately. where e3k is approximately normally distributed
The in vivo plasma concentrations resulting from ðNð0; s 22 ÞÞ with s 22 accounting for in vivo measure-
administration of the ER dosage unit to each ment error.
subject, and the in vitro fractions dissolved for The observed in vitro fraction of drug dissolved
each dosage unit are modelled directly (i.e. from the ith dosage unit at time t is described as
without averaging) in a one-step process. The follows:
resulting IVIVC model does not suffer from the
Yi1 ðtÞ ¼ f1 Fi1 ðtÞ þ "i1 ðtÞ; "i1 ðtÞ Nð0; s 1 2 Þ (4)
same theoretical flaws as the deconvolution based
method previously described. It allows the rela- where Fi1(t) is the true fraction dissolved at time t,
tionship between the fraction dissolved in vitro s 21 represents the in vitro measurement error and
and that dissolved in vivo to be described as a the parameter f1 accounts for any difference
DOI 10.1002/jps JOURNAL OF PHARMACEUTICAL SCIENCES, VOL. 97, NO. 8, AUGUST 2008
3426 GAYNOR, DUNNE, AND DAVIS
between actual dose and label claim. Variation in The in vivo plasma concentration measured
Fi1(t) between dosage units was modelled as: from the kth subject at time t following admin-
istration of the ith dosage unit is described as
logitðFi1 ðtÞÞ ¼ logitðF1 ðtÞÞ þ ui ; ui Nð0; v1 2 Þ (5) Z
0
Yi2k ðtÞ ¼ Dose f2 cdk ðt tÞFi2k ðtÞdt þ ei2k ðtÞ;
The term ui is a random effect which allows ei2k ðtÞ Nð0; s 2 2 Þ ð7Þ
for differences between dosage units while also
accounting for the correlation between repeated where Fi2k0 (t) is the in vivo dissolution rate at time
observations made on the same dosage unit. The t, f2 allows for a difference in bioavailability
inter dosage unit variation is given by v21 . F1(t) is between the reference dose and the ER dose and
the fraction of the typical dosage unit (where ui is s 22 represents in vivo measurement error as
zero) dissolved at time t and is modelled according before.
to the Weibull function as follows: The error terms ei1, ei2k and e3k are all modelled
according to truncated normal distributions to
F1 ðtÞ ¼ 1 expðl1 ta Þ (6)
ensure that the values of Yi1, Yi2k and Y3k remain
positive.
where the rate of dissolution of the drug is
The relationship between in vitro and in vivo
determined by l1 and a. The use of the logit
dissolution is defined as follows:
transformation logit(x) ¼ log(x/(1x)) in equation
five ensures that the simulated fraction, even with logitðFi2k ðtÞÞ ¼ logitðF1 ðtÞÞ þ u1 þ u2 t þ ui þ sik ;
the addition of a random effect (ui), remains
within the interval [0,1]. with ui Nð0; v1 2 Þ; and sik Nð0; v2 2 Þ ð8Þ
JOURNAL OF PHARMACEUTICAL SCIENCES, VOL. 97, NO. 8, AUGUST 2008 DOI 10.1002/jps
PREDICTION ACCURACY OF TWO IVIVC MODELLING TECHNIQUES 3427
DOI 10.1002/jps JOURNAL OF PHARMACEUTICAL SCIENCES, VOL. 97, NO. 8, AUGUST 2008
3428 GAYNOR, DUNNE, AND DAVIS
Figure 2. In vivo reference data and fitted curves for each subject.
JOURNAL OF PHARMACEUTICAL SCIENCES, VOL. 97, NO. 8, AUGUST 2008 DOI 10.1002/jps
PREDICTION ACCURACY OF TWO IVIVC MODELLING TECHNIQUES 3429
DOI 10.1002/jps JOURNAL OF PHARMACEUTICAL SCIENCES, VOL. 97, NO. 8, AUGUST 2008
3430 GAYNOR, DUNNE, AND DAVIS
which lie quite close to the fitted curve in Figure 4, can see the resulting underprediction of in vivo
resulting in good predictions in Figure 5b. Most of plasma concentration values in Figure 5c at early
the points in Figure 4 which belong to the slowest time points and, similarly, concentrations at later
formulation are above the fitted curve and one times are overpredicted. This lack of accuracy in
JOURNAL OF PHARMACEUTICAL SCIENCES, VOL. 97, NO. 8, AUGUST 2008 DOI 10.1002/jps
PREDICTION ACCURACY OF TWO IVIVC MODELLING TECHNIQUES 3431
Table 3b. Percentage Prediction Errors (Used for based method may also have an impact on the
External Validation) results. This is a question which has previously
been studied, albeit not in this context. Zhang
Method of et al.17 point out that ‘conditioning on some of the
Batch Analysis %PE AUC %PE Cmax
data or on prior fits to them may yield inaccurate
External Deconvolution 0.2730 11.0301 standard errors as there is not a proper account-
Convolution 0.0025 0.0232 ing for the uncertainty in conditioning quantities’.
In general, it is agreed that simultaneous analysis
of data is, statistically, preferable to first estimat-
prediction leads to a very high %PE value for ing parameters, which are then fixed and used in
the Cmax of this formulation, resulting in the the second stage of the analysis.17,18
failure to satisfy the internal validation criteria. The findings of this study confirm the expecta-
The formulation used for external validation tion that the convolution based method would
has a release rate which lies between the medium outperform its counterpart. The concerns about
and slow formulations as seen in Figure 1a, the deconvolution method are supported by the
predictions made using deconvolution for this results and its lack of accuracy in prediction is a
formulation are not sufficiently accurate to meet result of the combined influence of its theoretical
the FDA external validation criteria. flaws. While these may have seemed like purely
The fact that the convolution method is, in statistical considerations, they have a very strik-
effect, fitting the correct model leads to results ing practical effect in terms of quality of predic-
that are, perhaps, over optimistic. The degree to tions and, therefore, ability to meet the FDA
which the predictions would deteriorate when validation criteria. These results substantiate the
using real data would depend on the extent to statistical theory, illustrating and quantifying the
which the model used is mis-specified. superiority of the convolution based approach and
The deconvolution method not only fails to providing support for its use in preference to the
produce a model which meets the FDA validation deconvolution-based group of methods.
criteria but, as can be seen in Figure 5 (particu-
larly in panel c), the predicted plasma concentra-
tion curve is entirely the wrong shape. This is
potentially very misleading if the IVIVC model is REFERENCES
to be used during product development.
The deconvolution method and software used in 1. Food and Drug Administration. 1997. Guidance for
this study (CoDe15) has been chosen from a large industry: Extended release oral dosage forms:
number of such techniques. The CoDe method was Development, evaluation, and application of in
selected on the basis of its good performance in vitro/in vivo correlations.
testing such as that conducted by Madden et al.5 2. Dunne A, Gaynor C, Davis J. 2005. Deconvolution
based approach for level A in vivo-in vitro correla-
There are also many variations possible at the
tion modelling: Statistical considerations. Clin Res
implementation stage of both deconvolution and
Regul Aff 22:1–14.
the convolution-based methods so a number of 3. O’Hara T, Hayes S, Davis J, Devane J, Smart T,
further decisions were necessary. The choice of Dunne A. 2001. In vivo–in vitro correlation (IVIVC)
whether or not to average the data before analysis modeling incorporating a convolution step. J Phar-
is believed to be a crucial one. Current advice on macokinet Pharmacodyn 28:277–298.
best practise16 suggests that deconvolution should 4. Dunne A. 2007. Approaches to developing in vitro-
take place on an individual subject (or dosage in vivo correlation models. In: Chilukuri DM,
unit) level. Consequently, all of the analysis Sunkara G, Young D, editors. Pharmaceutical pro-
conducted as part of this study was carried out duct development in vitro-in vivo correlation.
entirely at the individual subject level for both the pp 47–70.
5. Madden FN, Godfrey KR, Chappell MJ, Hovorka R,
deconvolution and convolution-based methods.
Bates RA. 1996. A comparison of six deconvolution
Averaging of the raw data is believed to be so
techniques. J Pharmacokinet Biopharm 24:283–
fundamental an issue that it warrants further 299.
exploration and, as such, this point will form the 6. Abuzarur-Aloul R, Gjellan K, Sjölund M, Löfqvist
basis of a separate study. The use of a one-stage M, Graffner C. 1997. Critical dissolution tests
rather than a two-stage approach to fitting the of oral systems based on statistically designed
in vitro and in vivo ER data in the convolution experiments. I. Screening of critical fluids and
DOI 10.1002/jps JOURNAL OF PHARMACEUTICAL SCIENCES, VOL. 97, NO. 8, AUGUST 2008
3432 GAYNOR, DUNNE, AND DAVIS
in vitro/in vivo modelling of extended release coated 12. Snedecor GW, Cochran WG. 1980. Statistical meth-
spheres. Drug Dev Ind Pharm 23:749–760. ods, 7th edition. Ames, Iowa: Iowa State University
7. Abuzarur-Aloul R, Gjellan K, Sjolund M, Graffner Press.
C. 1998. Critical dissolution tests of oral systems 13. Wing GM. 1991. A primer on integral equations of
based on statistically designed experiments. II. In the first kind—The problem of deconvolution and
vitro optimization of screened variables on ER- unfolding. Philadelphia: Society for Industrial and
coated spheres for the establishment of an in Applied Mathematics.
vitro/in vivo correlation. Drug Dev Ind Pharm 14. Beal SL, Sheiner LB. 1992. NONMEM user’s
24:203–212. guides, NONMEM Project Group. San Francisco:
8. Abuzarur-Aloul R, Gjellan K, Sjolund M, Graffner University of California.
C. 1998. Critical dissolution tests of oral systems 15. Hovorka R, Chappell MJ, Godfrey KR, Madden FN,
based on statistically designed experiments. III. In Rouse MK, Soons PA. 1998. CODE: A deconvolution
vitro/in vivo correlation for multiple-unit capsules program implementing a regularization method of
of paracetamol based on PLS modeling. Drug Dev deconvolution constrained to non-negative values.
Ind Pharm 24:371–383. Description and pilot evaluation. Biopharm Drug
9. Liu Y, Schwartz JB, Schnaare RL, Sugita ET. 2003. Dispos 19:39–53.
A multi-mechanistic drug release approach in a 16. Farrell C, Hayes S. 2007. IVIVC for oral drug
bead dosage form and in vitro/in vivo correlations. delivery: Immediate and extended release dosage
Pharm Dev Technol 8:409–417. forms. In: Chilukuri DM, Sunkara G, Young D,
10. Cardot JM, Beyssac E. 1993. In vitro/in vivo corre- editors. Pharmaceutical product development in
lations: Scientific implications and standardisation. vitro-in vivo correlation. pp 47–70.
Eur J Drug Metab Pharmacokinet 18:113–120. 17. Zhang L, Beal S, Sheiner L. 2003. Simultaneous vs.
11. Kortejarvi H, Mikkola J, Bacjman M, Antila S, sequential analysis for population PK/PD data I:
Marvola M. 2002. Development of level A, B Best-case performance. Journal of Pharmacoki-
and C in vitro-in vivo correlations for modified- netics and Pharmacodynamics 30:387–403.
release levosimendan capsules. Int J Pharm 241: 18. Verbeke G, Molenberghs G. 2000. Linear mixed
87–95. models for longitudinal data. New York: Springer.
JOURNAL OF PHARMACEUTICAL SCIENCES, VOL. 97, NO. 8, AUGUST 2008 DOI 10.1002/jps