Direct Transfer of a Quantitative Model between Antaris FT-NIR Instruments

Significance of the topic

Near-infrared (NIR) spectroscopy is widely used in pharmaceutical development and manufacturing because it is rapid, non-destructive, amenable to high throughput and suitable for GMP environments. Quantitative NIR methods often require complex multivariate calibration models and extensive spectral libraries that can be difficult and costly to transfer between instruments. Demonstrating reliable direct transfer of a quantitative NIR model between instruments—without correction algorithms or transfer standards—reduces validation burden, shortens deployment time and lowers operational expense for critical quality assays such as polymorph quantification in finished dosage forms.

Objectives and overview of the study

• Develop and validate a reflectance NIR method to detect and quantify a minor, unwanted polymorphic form (Form B) present at low levels in a capsule formulation containing the API at ~10% w/w.
• Demonstrate whether the validated quantitative model built on an older Antaris I MDS FT-NIR instrument could be directly transferred to a newer Antaris II MDS FT-NIR instrument without corrective algorithms or additional calibration standards.
• Statistically evaluate equivalence of results between donor and receiving instruments using independent validation samples.

Methodology

• Sample preparation and design
  • Fifteen separate batches of capsules were manufactured, each spiked with different known levels of Form B. These were split into calibration, model test and an independently prepared validation set (different lots of API, excipients and capsule shells).
  • Because capsules were large and non-circular, each capsule was measured in three orientations; the three reflectance spectra per capsule were averaged to produce a single representative spectrum.
• Spectral acquisition and preprocessing
  • Reflectance spectra were acquired with an integrating sphere accessory.
  • Preprocessing included second derivative, standard normal variate (SNV) and mean centering to reduce baseline and scatter effects and to normalize the data.
  • The model spectral range was restricted to 5800–6252 cm⁻¹, the region showing the strongest discrimination between polymorphs.
• Calibration and validation
  • A partial least squares (PLS) regression model was developed using TQ Analyst software to predict % Form B.
  • Model metrics on the original (donor) instrument: standard error of calibration (SEC) = 1.11%; independent validation produced a standard error of prediction (SEP) = 1.47%.
  • An outlier-screening classification model using Mahalanobis distance was implemented; acceptance required Mahalanobis distance < 1.8.
• Transfer and comparative testing
  • Spectral data and TQ Analyst methods (quantification and outlier detection) were moved to a new Antaris II MDS analyzer without applying instrument-to-instrument correction algorithms.
  • Validation samples (six capsules per validation batch) were measured on both donor and receiving instruments. Each capsule was measured on both instruments to allow paired comparisons.

Instrumentation used

• Thermo Scientific Antaris I MDS FT-NIR analyzer (donor)
• Thermo Scientific Antaris II MDS FT-NIR analyzer (receiving)
• Integrating sphere reflectance accessory
• TQ Analyst software for PLS calibration, classification (Mahalanobis distance) and preprocessing

Main results and discussion

• Calibration and validation performance
  • SEC on the donor instrument was 1.11% Form B; SEP on independent validation samples was 1.47% Form B.
• Direct transfer evaluation
  • Paired analysis of the same validation capsules on donor and receiving instruments yielded a root-mean-square error of prediction between instruments (RMSEP) = 0.92% Form B, which is substantially below the method SEP (1.47%).
  • Paired t-test showed no statistically significant difference between instrument means at the 95% confidence level; the confidence interval for the mean difference was –0.29% to +0.76% Form B, meeting AstraZeneca acceptance criteria (difference < 1% and CI including zero).
  • All validation spectra measured on the receiving instrument passed the Mahalanobis distance outlier criterion (<1.8), indicating the transferred spectra fell within the original model space.
  • PCA (Hotelling T2, PC1 vs PC2) projection showed complete overlap of donor and receiving instrument scores; PC1 correlated with Form B level and PC2 with sampling variability, and no systematic bias or offset between instruments was observed.
• Implications
  • The small inter-instrument differences observed were within expected prediction error and did not indicate a need for instrument-specific correction or re-calibration in this case.
  • Successful direct transfer occurred despite substantial differences in instrument age and some electronics between Antaris I (>10 years old) and Antaris II.

Benefits and practical applications

• Direct model transfer reduces the need to rebuild large calibration libraries on new instruments, saving time and consumables required to prepare multiple calibration/validation standards.
• Simplifies GMP implementation of NIR assays by minimizing validation scope associated with instrument replacement.
• Applicable where instruments have comparable optical performance and where strong, localized spectral discrimination exists for the analyte of interest (here, polymorph-specific bands in 5800–6252 cm⁻¹).

Future trends and potential applications

• Modern FT-NIR instrument standardization, improved detector stability and tighter manufacturing tolerances will increase the feasibility of direct model transfer across broader instrument fleets.
• Standardized preprocessing pipelines and robust outlier/transfer diagnostics (Mahalanobis distance, PCA projections, paired statistical tests) will become routine steps in method transfer protocols.
• Automation of transfer checks and centralized model repositories could enable remote deployment of validated models across multiple production sites, supporting PAT and real-time quality control.
• Research into instrument-independent spectral representations and advanced domain adaptation methods may further reduce the need for any transfer standards or recalibration.

Conclusion

This case study demonstrates that, under favorable conditions (strong local spectral markers, controlled sampling, robust preprocessing and comparable instrument performance), a quantitative PLS NIR model for low-level polymorph quantification can be directly transferred from an older Antaris I analyzer to a newer Antaris II analyzer without corrective algorithms. Validation using independently prepared samples, paired statistics, Mahalanobis outlier screening and PCA-based diagnostics confirmed equivalence of results and supported adoption of the transferred method, yielding practical savings in time and resources for GMP deployment.

References

1. European Medicines Agency Guideline on the use of Near Infrared Spectroscopy (NIRS) by the pharmaceutical industry and the data requirements for new submissions and variations.
2. Ph Eur monograph 2.2.40; USP <1119> Near-Infrared Spectroscopy.
3. Brookes S. Direct Transfer of a Quantitative Model between Antaris FT-NIR Instruments. Application Note AN52624. Thermo Fisher Scientific; 2014.