Robust calibrations on reduced sample sets for API content prediction in tablets: definition of a cost-effective NIR...

Area of Science:

• Analytical Chemistry
• Chemometrics
• Pharmaceutical Analysis

Background:

• Near-infrared (NIR) spectroscopy is valuable for active pharmaceutical ingredient (API) prediction in tablets.
• Traditional NIR model development can be costly due to spectral variations from non-target components.
• Robust and cost-effective modeling strategies are needed for routine API analysis.

Purpose of the Study:

• To develop cost-effective modeling strategies for routine API prediction in tablets using NIR spectroscopy.
• To improve the robustness and reduce the calibration requirements of NIR models.
• To investigate the impact of spectral clutter on model performance.

Main Methods:

• Utilized prior spectral information (intra- and inter-batch variation, pure component spectra) to define spectral clutter.
• Applied artificial data augmentation with spectral shapes from the clutter.
• Implemented net analyte pre-processing (NAP) before partial least squares (PLS) regression.
• Reduced calibration set variability without compromising prediction performance.

Main Results:

• Statistically significant improvement in model performance, with a 34-40% reduction in root mean square error of prediction (RMSEP).
• Reduced requirement for model latent variables.
• Demonstrated that exhaustive calibration is unnecessary.
• Highlighted the critical role of clutter completeness for model robustness.

Conclusions:

• The proposed strategy of data augmentation with spectral clutter and NAP enhances NIR model performance for API prediction.
• This approach reduces costs by minimizing the need for extensive calibration data.
• Model robustness is dependent on the completeness of the defined spectral clutter, especially for variations not co-linear with the target property.