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 pharmaceutical dosage forms. Reliable, rapid, and non‑destructive verification of blend homogeneity reduces batch failures, improves process control, and accelerates decision making on blending end‑points. Diffuse reflectance Fourier transform near‑infrared (FT‑NIR) spectroscopy provides a practical at‑line technique for routine blend assessment because it requires no chemical reagents or sample preparation and interrogates a representative sample volume quickly.

Objectives and study overview

This application study evaluated the applicability of diffuse reflectance FT‑NIR (Thermo Scientific Antaris analyzer) to confirm blend uniformity for two proprietary product formulations. Calibration models were built from laboratory prepared blends spanning the expected composition ranges and then used to predict the composition of real production samples (including repeat lots measured after four weeks) to test prediction accuracy, precision and intermediate stability.

Methodology

Key experimental elements:

• Calibration samples: Laboratory blends prepared gravimetrically in 2‑dram vials to cover ~35–65% ranges for the main components. For Product 1 (two components) six calibration mixtures were used; for Product 2 (three components) six mixtures were prepared with component #3 ≈1% and essentially constant.
• Sampling/measurement: Samples were measured directly through the bottom of 2‑dram glass vials placed on an integrating sphere (diffuse reflectance). Each calibration sample was analyzed in triplicate. Production samples were measured from multiple positions within blending vessels (top, middle, bottom) to probe spatial homogeneity.
• Spectrometer settings: Spectral region 10000–4000 cm−1, resolution 8 cm−1, 32 co‑averaged scans (~24 s collection). Background was collected on an internal gold flag; instrument qualification used ValPro with NIST‑traceable standards.
• Chemometrics: Spectra were mean‑centered and converted to second‑derivative form using a Norris derivative (9‑point segment, no gap). Calibrations were built using Stepwise Multiple Linear Regression (SMLR) and Partial Least Squares (PLS) as appropriate.

Used instrumentation

• Thermo Scientific Antaris FT‑NIR analyzer with integrating sphere module (study instrument).
• Thermo Scientific RESULT software for data collection and TQ Analyst software for chemometrics.
• ValPro system qualification software with internal NIST‑traceable standards and polystyrene band check; internal gold flag used for background.

Main results and discussion

Product 1 (two‑component system):

• SMLR models using a single selected spectral point (5438 cm−1 region) yielded good performance: correlation coefficient 0.9918, RMSEC 1.24% (approx. one standard deviation across the calibration range).
• Predictions for production samples returned averages near the expected 50/50 composition (example prediction average 49.7% vs theoretical 50%).
• Instrument repeatability (six replicates without moving vial) gave RSD 0.67% for component content; sample precision (12 replicate measurements with shaking between readings) showed RSD 6.9% indicating substantial sample heterogeneity within the measured vial (range ~48.7–56.6%). This indicates actual blend spatial variability dominates measurement noise.

Product 2 (more complex, three‑component system):

• PLS models were required for adequate calibration; four PLS factors were used for each major component (component #3 was essentially constant and not modeled).
• Model metrics: component #1 correlation 0.9990 (RMSEC 0.434%); component #2 correlation 0.9993 (RMSEC 0.377%).
• Instrument precision was excellent (<0.1% RSD for replicate determinations), again showing that sample heterogeneity is the larger source of variance.
• Predictions for production samples taken at different locations (top, middle, bottom) consistently summed to ~99% for components #1 and #2, consistent with component #3 being ~1%, supporting calibration linearity and quantitative accuracy.

General observations:

• Laboratory‑prepared calibration blends successfully predicted real production samples for both formulations, demonstrating that representative lab calibrations can be used for in‑process monitoring for these products.
• Non‑destructive, through‑vial measurement using an integrating sphere provided representative interrogation, enabling at‑line use and rapid feedback.
• Disparities between instrument precision and sample precision emphasize the importance of sampling strategy (multiple locations, mixing) when assessing blend uniformity.

Practical benefits and applications

• Rapid at‑line confirmation of blend end‑point to support real‑time process decisions and reduce over‑ or under‑mixing risks.
• Non‑destructive analysis without reagents or preparation reduces operational complexity and waste.
• High instrument precision makes FT‑NIR suitable for monitoring small changes and for establishing RSD‑based criteria for blend acceptance.
• Ability to measure multiple production lots and different sampling positions enables quantification of lot‑to‑lot and spatial variability, informing process control and sampling plans.

Future trends and potential applications

• Integration with process analytical technology (PAT) frameworks and automated sampling systems to provide closed‑loop control of blending operations.
• Use of improved FT‑NIR hardware (e.g., Antaris II or newer) and faster acquisition schemes to shorten analysis time and increase throughput.
• Advanced chemometrics and machine learning approaches to improve robustness when product matrices are more complex or when minor components influence spectra.
• Development of standardized RSD or other statistical end‑point criteria to allow regulatory‑acceptable in‑process blend release based on FT‑NIR measurements.

Conclusions

Diffuse reflectance FT‑NIR with integrating‑sphere sampling is a practical and effective technique for at‑line blend uniformity confirmation in pharmaceutical powder processing. The study demonstrated that laboratory‑based calibrations can accurately predict production sample composition for the formulations tested, and that instrument repeatability is excellent. Observed variability was primarily driven by sample inhomogeneity rather than spectrometer noise, highlighting sampling procedure and measurement location as key considerations. Overall, FT‑NIR enables fast, non‑destructive, and quantitative monitoring that can support blend validation and routine process control.