(99e) PAT for Pharmaceutical Extrusion Monitoring and Supervisory Control

Introduction

Hot Melt Extrusion (HME) is increasingly attracting
interest in pharmaceutical manufacturing, for two main reasons: (1) it allows
to produce controlled release formulations by embedding API crystals in
polymers, lipids or sugars (Kleinebudde, 2011) and (2) it significantly enhances the bioavailability
of APIs with low solubility. This can be achieved with a solid solution, in
which the API and the matrix are molecularly dispersed. Estimations predict
that up to 90% of future new chemical entities may be classified as poorly
soluble. Therefore, enhancing the bioavailability is a significant part for successful
drug development. Further dosage forms like transdermal patches and implants
among others can be produced by HME as well (Breitenbach, 2002).

HME also has the potential to implement continuous
production in a straightforward way. One way is to use a hot-strand cutter as a
downstream processing unit. The (hot) extrudate is cut directly at the die
face, thereby forming nearly spherical pellets. If necessary, pellets with
different APIs or release characteristics can be mixed and further processed by
a capsule filling machine or a tablet press. This production line is in
development at RCPE.

At present, many Process Analytical
Technology (PAT) tools are available for establishing physicochemical product
properties, such as the chemical composition. Multivariate data analysis (MVDA)
combined with spectroscopic techniques, including near infrared (NIR) and Raman,
allows quantitative and non-invasive process monitoring (Koller et al., 2011). Applying multivariate statistical process control (MSPC) reduces
real-time process data streams to a convenient control chart. Clearly, process
analysis calls for an IT infrastructure that can aggregate real-time process
data from multiple unit operations, raw material data, PAT data and equipment
status. SIPAT is a commercial PAT software solution that was created to meet
the above requirements.

In this study the PAT concept consisted of: (1)
In-line, real-time monitoring of extruder parameters and collection of NIR
spectra of the extrudate close to the die; (2) Analysis of the spectrum with
multivariate data analysis (MVDA) tools; (3) Integration and implementation of
this PAT concept in SIPAT.

In this presentation we intend to show the setup of
the HME system, Principal Component Analysis (PCA) to reduce the dimensionality
of the spectra and a Partial Least Squares (PLS) model to predict the response
of the extrusion process (the API content at the die) to step changes of the
feed rates.

Experimental Methods

Materials

Paracetamol, donated by G.L. Pharma GmbH, Lannach,
Austria, was used as an API (volume mean particle size 139.2 µm). The matrix
carrier system was calcium stearate (Werba-Chem GmbH, Vienna, Austria; mean
particle size 16.62 µm) (Roblegg et al., 2011). The paracetamol crystals were embedded in
CaSt, resulting in a solid dispersion.

Extruder setup

Extrusions were performed on a co-rotating twin-screw
extruder (ZSK 18, Coperion GmbH, Stuttgart, Germany) with 18mm screw diameter.
The barrel temperatures and the screw setup was chosen in such way that the
melt temperature at the die was 130°C. Two twin-screw gravimetric feeders
(KT20, K-Tron, Pitman, USA) were filled with a premix. Mixtures with a mass
fraction of paracetamol from 0% to 60% can be obtained by adjusting the feed
rates of the two feeders. In the experiments the throughput of the extruder was
0.6 kg/h. To monitor, control and store process parameters of the extruder, NIR
spectra and results of calculations (e.g. API content prediction) the software
SIPAT 3.1.1 (Siemens AG, Brussel, Belgium) was used.

NIR spectroscopy

NIR spectra were collected in transflexion mode with a
process spectrometer SentroPAT FO (Sentronic GmbH, Dresden, Germany). The
spectrometer has a spectral range of 1100 nm - 2200 nm and 2 nm resolution. A
fibre-optic Dynisco NIR probe was mounted close to the extrusion die, after the
screws, for in-line monitoring of the melt. For each measurement 120 spectra
were averaged, with an integration time of 0.014s per spectrum. However,
significant modifications of the die were necessary to obtain spectra which
reflect the current bulk composition of the melt.

Chemometric models

Around 100 spectra of the melt of each step of API
content were collected to develop a PCA and a PLS model. Principal component analysis applied to the observed data may
be used to classify
the main events affecting the process. The PLS model enables the direct prediction of the paracetamol content. The chemometric
model was built with the software Simca P+ 12.0 (Umetrics, Umeå, Sweden).
Simca-Q 12.0.1 (Umetrics, Umeå, Sweden) is used as an external calculation
engine in SIPAT, which has the advantage of directly integrate the models
developed in Simca P+ in SIPAT.

Results

Selecting a good NIR measurement position

Ensuring a good sample presentation to the Dynisco NIR
probe is crucial for correct API predictions. The main reason for improper
sample presentation is a too slow exchange of melt in the field of view of the
NIR probe. First, if the probe is not perfectly in line with the barrel
surface, melt will fill the dead volume once. The shear forces are not high
enough to exchange the material. This is likely for measurements above the
screws. A second possibility is a too large volume of melt directly before
exiting the die. Here a funnel flow might build up, which causes slow melt
exchange at the barrel walls, where the probe is situated. Although NIR
penetrates a transparent melt, in the mentioned cases the measurement could be
biased. Thus, a special probe set-up was designed and tested to represent the
actual melt composition.

Monitoring the process

The feeding dynamics are the most important factor for
content uniformity of the melt, especially for pharmaceutical powders, which
are often not free flowing. Despite the mixing elements of the screw, the
extruder has only a limited capability to compensate lower frequency feeding variations
by its backmixing ability.

Monitoring the process in the reduced PC space is a
straightforward way of implementing real-time process analysis. Adjusting the feed rates of the
feeders yielded different API concentrations in the extrudate that appeared in
the score plot as clusters of observations.

Moreover, the prediction of the API content by the PLS
model enables another way of monitoring the process in a convenient control
chart.

Using Simca P+ models or models defined in MATLAB, the
observations can be processed and visualized in SIPAT in real-time. The
integration of the PCA and PLS models facilitates different ways to detect in
real-time abnormalities in the process.

Conclusion

The presented hot melt extrusion system with an
integrated PAT concept is capable of monitoring the critical process
parameters. Special attention was paid to ensure a good NIR measurement
position. However, the MVDA models that are developed in this process can be
applied to only one formulation. In a typical campaign-driven manufacturing
environment (which even continuous manufacturing of the future will be)
formulations and/or products may change weekly or monthly. Automatic model
development can be performed based on the setup including the HME process,
SIPAT and Simca-Q. Using several models at the same time opens up the
opportunity to validate different PCA or PLS models in real-time.

References

[1] Breitenbach J (2002) Melt extrusion: from process to drug delivery
technology. Eur. J. Pharm. Biopharm. 54:107-117.

[2] Kleinebudde P (2011) Pharmaceutical
Product Design: Tailored Dissolution of Drugs by Different Extrusion
Techniques. Chem-Ing-Tech. 83:589-597.

[3] Roblegg E, Jäger E, Hodzic A, Koscher
G, Mohr S, Zimmer A and Khinast J (2011) Development of sustained-release
lipophilic calcium stearate pellets via hot melt extrusion. Eur. J. Pharm.
Biopharm. 79:635-645.

[4] Koller D M, Posch A, Hörl G, Voura C,
Radl S, Urbanetz N, et al. (2011) Continuous quantitative monitoring of powder
mixing dynamics by near-infrared spectroscopy. Powder Technology. 205:
87-96