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

Direct transfer of a quantitative NIR model between Antaris FT-NIR instruments — concise expert summary

Significance of the topic

Near-infrared (NIR) spectroscopy is widely used in pharmaceutical development and GMP environments because it delivers rapid, non-destructive, high-throughput analysis suitable for routine QC and PAT. Quantitative NIR models commonly rely on complex multivariate calibrations and historically require time-consuming re-calibration or instrument-specific correction when moved between analysers. Demonstrating robust direct model transfer reduces downtime, cost and the need for additional calibration standards, enabling faster deployment of validated methods across sites and instrument generations.

Objectives and study overview

The study aimed to test whether a quantitative partial least squares (PLS) model for quantifying a minor polymorphic form (Form B) within a capsule formulation could be transferred directly from an older Antaris I MDS FT-NIR instrument to a new Antaris II MDS FT-NIR instrument without correction algorithms or transfer standards. The transfer was evaluated using independent validation samples and statistical tests to assess bias, prediction error and model applicability on the receiving instrument.

Method background and study design

The analytical challenge was quantifying a low-level polymorph (Form B) of an API present at approximately 10% w/w in a capsule matrix with strong excipient spectral contributions. Fifteen batches of capsules were prepared with varying levels of Form B and split into calibration, test and an independent validation set (different API/excipient/capsule lots) to assess model performance under realistic variation. Because capsules were large and non-circular, three reflectance measurements per capsule (different orientations) were acquired and averaged to represent each sample.

Methodology and model development

- Spectral acquisition: Reflectance spectra acquired with integrating-sphere accessory; each capsule represented by the mean of three orientations.
- Spectral region: Model built in the discriminating region 5800–6252 cm-1 where the polymorphs differ most.
- Pre-processing: second derivative, standard normal variate (SNV) and mean-centering to remove baseline/physical effects and normalize spectra.
- Chemometrics: PLS regression developed in TQ Analyst to predict % Form B.
- Calibration/validation metrics: standard error of calibration (SEC) = 1.11% (calibration set); independent validation produced SEP = 1.47%.

Instrumentation used

- Donor instrument: Thermo Scientific Antaris I MDS FT-NIR analyser with integrating sphere accessory (instrument >10 years old).
- Receiving instrument: Thermo Scientific Antaris II MDS FT-NIR analyser (newer system with slightly different electronics).

Outlier control and model domain definition

An auxiliary discriminant classification model (Mahalanobis distance) was implemented to screen unknown spectra and ensure they fell within the calibration space. Acceptance criterion for a pass was Mahalanobis distance < 1.8, chosen from calibration/validation distance distributions. This limited predictions to samples spectrally similar to the standards used for calibration.

Transfer procedure and validation

The calibrated spectra and TQ Analyst methods (classification and PLS) were copied from the Antaris I to the Antaris II instrument. The independent validation set (same capsules measured on both instruments) was used to compare predictions directly. Six capsules per validation batch were measured on both instruments with paired comparisons of results for the exact same capsules.

Main results and statistical assessment

- Agreement: Root-mean-square error of prediction (RMSEP) between donor and receiving instrument results = 0.92%, which is smaller than the method SEP of 1.47%.
- Mean difference and bias: Paired t-test showed no statistically significant difference in means at 95% confidence. When including Form B level as a factor, the 95% confidence interval for the difference in instrument means was −0.29 to +0.76% Form B, meeting AstraZeneca acceptance criteria (difference ≤ 1% and CI includes zero).
- Model applicability: All receiving-instrument spectra for validation samples had Mahalanobis distances < 1.8, so they were within the model space.
- PCA projection: Scores for donor and receiving instrument data overlapped completely in the principal-component space (PC1 driven by Form B level, PC2 by sampling variability), indicating no observable instrument-induced offset.

Discussion and interpretation

The results demonstrate that, for this case, a complex multivariate quantitative model can be transferred directly between two Antaris FT-NIR systems of different ages and electronics without corrective algorithms or transfer standards. Key contributors to successful direct transfer were strong spectral discrimination between polymorphs, consistent sampling strategy (three-orientation averaging), targeted spectral region selection, robust pre-processing, and strict outlier screening to confine predictions to samples similar to calibration standards. The instrument-to-instrument variability was smaller than the method’s inherent prediction error and within pre-defined acceptance limits.

Practical benefits and applications

- Time and cost savings by avoiding reconstruction of large calibration libraries or preparation of transfer standards.
- Simplified GMP method transfer and faster deployment of validated assays across sites and instrument generations.
- Applicability to QC tasks where spectral differences are strong and sampling/spectral protocols are tightly controlled (e.g., polymorph detection, content uniformity, rapid release testing).

Limitations and considerations

- Direct transferability depends on instrument optical stability, comparable sampling accessories/configurations, and robust pre-processing; not all models/instrument pairs will behave similarly.
- The calibration was created using artificially spiked samples; native production variability could introduce spectral differences not covered by the model, reinforcing the need for strict outlier checks and periodic monitoring.

Future trends and opportunities

- Continued improvements in instrument reproducibility and manufacturing consistency will increase the feasibility of direct transfers.
- Development of standardized instrument qualification and spectral reference procedures to support robust model portability across sites and vendors.
- Hybrid approaches that combine direct transfer with minimal standardization samples or instrument standardization spectra when full direct transfer is marginal.
- Cloud-based model management and remote validation could streamline multi-site deployment and model lifecycle management aligned with regulatory expectations and PAT frameworks.

Conclusions

This case study demonstrates that direct transfer of a quantitative NIR model between an Antaris I and an Antaris II FT-NIR analyser is achievable without correction algorithms when: strong analyte-specific spectral features exist, sampling and pre-processing are rigorously controlled, and model domain screening is enforced. Successful transfer reduced method migration time and resource requirements while preserving validated performance within defined acceptance criteria.

References

• European Medicines Agency. Guideline on the use of Near Infrared Spectroscopy by the pharmaceutical industry and the data requirements for new submissions and variations. 2014.
• Ph. Eur. Monograph 2.2.40. Near-Infrared Spectroscopy.
• United States Pharmacopeia. USP <1119> Near-Infrared Spectroscopy.