The pharmaceutical industry faces considerable challenges due to an increasing complexity of products, greater competition in the generic drug arena, and higher regulatory demands to ensure safe and effective pharmaceutical products. Process analytical technology (PAT) tools can provide knowledge of the pharmaceutical drug, both during development and throughout the production. The usage of PAT tools is encouraged by the regulatory authorities, and therefore the interest in new and improved PAT tools is increasing. The main purpose of introducing Quality by Design (QbD) and PAT in pharmaceutical production is to introduce innovation in the full scale production and to continuously increase the knowledge of the pharmaceutical processes, products, and how they relate to each other, so it is possible to produce pharmaceutical products at a high and consistent quality. This gives benefits to the consumer, the regulatory authorities, and the pharmaceutical manufacturers. The PAT tools include a wide range of disciplines, including process analyzers (e.g. spectroscopic sensors), multivariate statistical analysis (e.g. chemometric data treatment), process control, continuous improvement, and knowledge management. The real advantage of PAT tools lay in the combination of the different disciplines. This could be spectroscopy-based process analyzer that monitors the pharmaceutical product including latent factors that could influence on the final product quality. Chemometric data treatment can then be used to extract the relevant information from the data, and process control can ensure low variation in the final product quality. This can help to optimize the production, allow real-time product release, and potentially replace the expensive, destructive, and time consuming laboratory testing that is currently the standard practice in traditional pharmaceutical industry. In this thesis, three spectroscopic PAT tools are investigated, near-infrared-, terahertz-, and fluorescence-spectroscopy. These techniques have been evaluated with chemometrics and theory of sampling. The first study focused on the critical but rather overlooked sampling uncertainty that exist in all analytical measurements. The sampling error was studied using an example involving near infrared transmission (NIT) spectroscopy to study content of uniformity of five batches of escitalopram tablets, produced at different active pharmaceutical ingredients (API) concentrations and at different hardness (compression force) levels. The aim of this study was to investigate the influence of sample orientation, powder segregation, and compression force on the NIT spectra. Therefore, samples were taken at different production time points, at different compression forces, and measured with the spectrometer in different orientations. The study showed that a minimum of 18 tablets from each level of API concentrations (90 spectra in total) were required to establish a robust NIT calibration. Further, it was shown that the spectra from tablets with the scored line facing upwards gave better calibrations than those tablets in which the scored line were facing downwards. However, the largest fraction of the variation between spectra was found when moving the tablet in the tablet sampler. The NIT data was also used to show variable reduction in the API calibration development, which eliminate dspectral interfer-ence from moisture uptake by the tablets during storage. It was possible to improve the prediction error of the quantitative API model considerately by spectral variable selection. The second study investigated the potential use of terahertz time domain spectroscopy (THz-TDS) to quantify crystallinity in binary mixtures of amorphous and crystalline lactose, and this technique was compared to near infrared (NIR) spectroscopy. THz-TDS gave higher cross validated errors than NIR spectroscopy for both full concentration and low concentration range models. The third study investigated the use of fluorescence spectroscopy to simultaneously predict two APIs in a tablet formulation: flupentixol (FLU) in low dosage (0.208-0.625 % w/w free base) and melitracen (MEL) (4.17-12.5 % w/w free base). Despite internal quenching between
the ingredients and the two APIs, it was possible to establish calibrations using partial least squares regression (PLS) on unfolded fluorescence landscapes with relative errors of 9.1 % for
FLU and 4.1 % for MEL, respectively. Both fluorescence spectroscopy and terahertz time domain spectroscopy are new tools in pharmaceutical applications and the possibilities and limitations, in relation to the abundant NIR spectroscopy, is discussed. PAT tools can together with Quality by Design contribute to important product and process knowledge that is valuable for optimizing production methods and in the development of new pharmaceuticals.