(424f) PAT Based Batch Statistical Process Control for Pharmaceutical SEMI-SOLID and Liquid Manufacturing Processes

N. Bostijn^1,*, C. Vervaet², J.P. Remon², T. De Beer¹

¹Laboratory of Pharmaceutical Process Analytical Technology, Ghent University, Ottergemsesteenweg 460, 9000 Ghent, Belgium

²Laboratory of Pharmaceutical Technology, Ghent University, Ottergemsesteenweg 460, 9000 Ghent, Belgium

Introduction

Pharmaceutical semi-solids and liquids are produced in a batch-wise manner with off-line time-consuming and less efficient quality control (1). The aim of this study is to simulate the batch process on a laboratory scale of five commercial available semi-solid and liquid formulations and to control the critical quality attributes of the process by monitoring the chord length distribution (CLD) of the particles, viscosity, temperature and active pharmaceutical ingredient (API) concentration/homogeneity in-line. For each formulation, reference batches were used to develop a reference batch partial least squares (PLS) regression model. From this model, score control charts were created describing the average batch trajectory. These control charts were used to monitor test batches and to detect any deviations from the expected batch trajectory. This approach allows early fault detection and real-time process adjustments, hence avoiding batch loss.

MATERIALS AND METHODS

Materials

For this study, five commercial available pharmaceutical formulations were selected. Two liquids (shampoo and suspension) and three semi-solids (ointment, cream and gel).

Methods

All experiments were performed using a customized IKA LR2000 modular mini plant reactor system (IKA, Staufen, Germany). The mixing vessel was equipped with a double wall for controlling the temperature of the process. A peristaltic pump (Watson-Marlow, Falmouth, Cornwall, UK) was used for pumping water from a water bath (Type 1032, GFL, Burgwedel, Germany) to the mixing vessel. In-line Raman spectra were recorded using a Raman Rxn1 spectrometer (Kaiser Optical Systems, Ann Arbor, MI, USA), equipped with an air-cooled CCD detector and a fiber-optic immersion probe. The laser wavelength was the 785 nm line from an Invictus NIR diode laser. Every 30 s a spectrum was acquired in the 0â??1900 cm^-1 region with an exposure time of 15 s and a resolution of 4 cm^-1.

There were also Raman spectra recorded using a Raman Rxn2 spectrometer (Kaiser Optical Systems, Ann Arbor, MI, USA), equipped with a fiber-optic PhAT probe. The laser wavelength was 785 nm. Spectra were collected every 30 s with an exposure time of 15 s and a resolution of 5 cm^-1. The spectral range of the system was from 150â??1890 cm^-1. Viscosity was measured using an in-line viscometer (Sofraser, Villemandeur, France), equipped with an internal temperature probe. The viscometer had a range from 0.1â??1000 mPa.s. Each second the viscosity and temperature were recorded. A focused beam reflectance measurement (FBRM) probe (model C35, Mettler-Toledo, Columbus (Ohio), USA) was used for monitoring the chord length distribution of particles in the system. The probe window was cleaned every 10 s by a pressurized air activated scraper and each 2 s the averaged chord length distribution was saved. The batch model was developed and evaluated using the SIMCA software (Version 14, Umetrics, Umeå, Sweden).

RESULTS AND DISCUSSION

To investigate whether the reference batch models were able to distinguish good from bad batches, process and formulation deviations were induced in the test batches. Only the results of the ointment will be discussed in this abstract.

For test batch 1 the API was added in three different steps. For all the reference batches, API was added in one step after ± 1200 s. In this test batch, 50 % (w/w) of the API was added after 1200 s, 45 % (w/w) after 2520 s and 5 % (w/w) after 3780 s. This induced error resulted in a decrease of the PLS component 1 (PLSC 1) scores in the middle part of the batch evolution scores plot. To monitor and evaluate the overall performance of new batches, a Hotellingâ??s T2 plot can be calculated. When detecting deviations in the Hotellingâ??s T2 plot, it is possible to identify the variables responsible for the deviations via contribution plots. The contribution plots at 1800 s (50 % API) and 3030 s (95 % API), showed that the deviation was determined by a lower viscosity and API concentration. After addition of all the API, the viscosity and API concentration recovered to the average values of the reference model.

In test batch 2 the set point of the water bath was set 10 °C higher compared to the reference batches and corrected after 2580 s. The PLSC 2 scores fell out of the control limits but evolved towards the average batch trajectory at the end of the process when the water bath temperature reached it correct value. At 1800 s the temperature was higher and the viscosity lower compared to the average observations, as seen from the contribution plot.

CONCLUSION

The batch model, developed from four reference batches, was able to detect both the process and formulation induced deviations in the test batches by monitoring the temperature, viscosity, CLD and API concentration in-line.

CHALLENGES AND FUTURE WORK

A next step is to investigate the possibility to produce the same formulations with a continuous production system and to monitor the process using the same and/or new PAT tools.

REFERENCES

1. De Beer T et al. Near infrared and Raman spectroscopy for the in-process monitoring of pharmaceutical production processes. Int. J. of Pharm. 2011; 417: 32â??47.