Use of diffuse reflectance Fourier transform near-infrared spectroscopy to confirm blend uniformity

Significance of the topic

The uniformity of powder blends is a critical control point in the manufacture of solid dosage pharmaceutical forms. Reliable, rapid, and non‑destructive monitoring of blend homogeneity during production reduces the risk of out‑of‑specification batches, shortens production cycle times, and supports regulatory expectations for blending validation and in‑process control. Near‑infrared (NIR) spectroscopy, particularly diffuse reflectance Fourier transform NIR (FT‑NIR), provides fast, reagent‑free, at‑line measurements well suited to these needs because it interrogates bulk sample volumes and yields immediate feedback to operators and quality systems.

Study objectives and overview

This application study evaluated diffuse reflectance FT‑NIR as a tool to confirm blend uniformity for two proprietary products. Objectives were to: develop quantitative calibrations from laboratory‑prepared blends; test those calibrations on real production samples; assess precision (instrument and sample) and predictive performance; and evaluate the method's suitability for at‑line process monitoring. Two product formulations were studied: Product #1 (binary blend) and Product #2 (ternary blend with a minor ~1% component). Laboratory mixtures spanning approximately 35–65% ranges of major components were used to build models which were then applied to production lots, including an independent set received four weeks later to probe intermediate precision and calibration stability.

Methodology

Sample preparation and measurement

• Calibration mixes were prepared by weighing raw materials and combining them in 2‑dram glass vials; composition ranges for major components were approximately 40–60% (some calibration points extended to 100% single components for linearity checks).
• Mixes were mechanically shaken briefly and hand‑tumbled; each calibration sample was measured in triplicate. Production samples were measured directly in vials without further preparation and measured at different physical locations within a lot (top, middle, bottom) to assess within‑vessel variability.

Spectroscopic acquisition and pre‑processing

• Spectral range: 10000–4000 cm−1; resolution: 8 cm−1; 32 co‑averaged scans (~24 s collection).
• Background: internal gold flag used for stability and to avoid spectral features in the NIR region.
• Preprocessing: mean‑centering and conversion to second‑derivative spectra using a Norris derivative (9‑point segment, no gap) to reduce multiplicative scatter and baseline effects.
• Chemometrics: Stepwise Multiple Linear Regression (SMLR) and Partial Least Squares (PLS) regression were evaluated; models were chosen based on formulation complexity and predictive performance.

Used instrumentation

• Thermo Scientific Antaris FT‑NIR analyzer with integrating sphere (method development instrument).
• Software: RESULT for data collection; ValPro for system qualification (NIST‑traceable standards and polystyrene for band position); TQ Analyst for chemometric model development.
• Note: authors commented that an improved Antaris II model is available providing enhanced speed and performance.

Results and discussion

Product #1 (binary system)

• SMLR using single spectral datapoints (one region at 5438 cm−1) produced adequate quantitative models because of the simple two‑component matrix.
• Model statistics: correlation coefficient r = 0.9918; RMSEC = 1.24 (approximately one standard deviation across the calibration range).
• Predictions for production samples averaged ≈49.7% component #1 and 50.3% component #2 (close to the theoretical 50:50 target). Extreme single‑component samples (100% of one component) were predicted accurately, supporting model linearity across a wide compositional range.
• Precision: instrument precision (six replicates without moving vial) yielded RSD ≈0.67% for one production sample (component #2). Sample precision (12 replicate measurements with shaking between runs) showed RSD ≈6.9% and a measurement range of 48.7–56.6% for component #2, indicating significantly higher sample heterogeneity than instrument variation.
• Practical implication: measured within‑vial variability suggests the need to consider sampling strategy and to use RSD or similar metrics as an in‑process end‑point criterion for blend uniformity.

Product #2 (ternary system)

• A PLS approach was required due to increased formulation complexity; calibrations used the 5200–6500 cm−1 region. Component #3 was essentially constant (~1%) so no separate calibration was developed for it.
• Model complexity: each PLS model used four latent factors (high relative to sample count but validated by predictive performance on extremes).
• Model statistics: for component #1, r = 0.9990 and RMSEC = 0.434; for component #2, r = 0.9993 and RMSEC = 0.377.
• Precision: instrument repeatability was excellent (six replicates produced SDs of 0.089% for component #1 and 0.072% for component #2), again showing sample variability dominated overall precision limits.
• Inter‑day/intermediate precision: production samples received four weeks later (measured in two portions) were predicted consistently; the two main components’ predictions summed to ~98.8–99.0%, consistent with composition expectations and indicating robustness of independent calibrations.

Overall interpretation

• FT‑NIR spectroscopy delivered accurate, precise predictions for both formulations when appropriate chemometric approaches were used (SMLR for simple binary mix, PLS for more complex ternary mix).
• Instrument contribution to variability was negligible relative to sample heterogeneity, highlighting the importance of sampling protocol and measurement replication to capture true blend variability.
• Successful prediction of extreme compositions and independent production lots supports calibration linearity and method ruggedness for these products.

Benefits and practical applications of the method

• Non‑destructive, reagent‑free measurement enabling at‑line process monitoring and rapid decision making.
• Representative bulk measurement via diffuse reflectance/integrating sphere, improving the chance of detecting inhomogeneity compared to surface‑only probes.
• Fast analysis times (~24 s per spectrum) suitable for routine process checks and end‑point determination during blending.
• Flexible chemometric approaches support both simple and more complex formulations; calibrations can be constructed from laboratory blends when allowed by formulation characteristics.

Future trends and potential applications

• Integration of FT‑NIR with process analytical technology (PAT) frameworks for automated blend end‑point detection using RSD or similar statistical control limits.
• Expansion to in‑line probe designs or multi‑point sampling to further reduce sampling variability and better represent large‑volume mixers and continuous processes.
• Use of robust spectral pre‑processing and advanced multivariate methods (e.g., variable selection, machine learning regressors) to improve performance for complex excipient matrices and low‑level actives.
• Instrument evolution (e.g., faster Antaris II and successors) will shorten acquisition times, enhance signal‑to‑noise, and enable denser process feedback loops.

Conclusion

Diffuse reflectance FT‑NIR implemented with integrating sphere sampling proved to be an effective at‑line tool to confirm blend uniformity for the two studied products. When matched with appropriate preprocessing and chemometric modeling (SMLR for simple binary mixtures, PLS for more complex blends), the technique produced accurate, linear, and precise predictions. Instrument repeatability was excellent; observed variability was dominated by sample heterogeneity, underscoring the need for representative sampling plans. The approach supports rapid in‑process checks and can be incorporated into PAT strategies to improve blending control and production efficiency.

References

• Porter J. Use of diffuse reflectance Fourier transform near‑infrared spectroscopy to confirm blend uniformity. Thermo Fisher Scientific Application Note (Antaris FT‑NIR). ©2022 Thermo Fisher Scientific Inc.