Inline Monitoring of a Hot Melt Extrusion Process by Near Infrared Spectroscopy

Significance of the Topic

Hot melt extrusion (HME) is an increasingly important continuous manufacturing technology in the pharmaceutical industry for producing solid dosage forms with improved bioavailability and controlled solid-state properties. Real-time, in-line monitoring of critical quality attributes such as drug load homogeneity is a central tenet of Process Analytical Technology (PAT) and Quality by Design (QbD). Near-infrared (NIR) spectroscopy, particularly Fourier-transform NIR (FT-NIR), offers rapid, non-destructive, in-line measurement capability and can directly monitor product attributes at the extruder die, enabling more reliable determination of steady-state operation than indirect process signals (e.g., torque).

Objectives and Study Overview

This study demonstrates development and validation of an in-line FT-NIR method for monitoring theophylline drug load in polyethylene oxide (PEO) extrudates processed by twin-screw hot melt extrusion. Main goals were to define practical pre-requisites for successful quantitative models, evaluate reflection-mode measurement for opaque melts, build a calibration model across 0–20% drug loads, and compare FT-NIR predictions against conventional process indicators such as extruder torque during start-up and steady-state operation.

Methodology

• Formulations: PEO (Sentry WSR N10) matrix with theophylline anhydrous at 0, 5, 10, 15, and 20% w/w. Lactose was added in some experiments to increase reflectivity of the extrudate surface and improve NIR signal quality.
• Sample preparation: Pre-blends were manually mixed in polyethylene bags for ~3 minutes.
• Extrusion conditions: Single-screw feeder at 500 g/h fed a Pharma 16 HME twin-screw extruder operating at 100 rpm with barrel temperatures up to 120 °C; screw configuration included two kneading sections for melting and mixing.
• Spectroscopy: FT-NIR spectra were collected in reflection mode at a probe positioned near the die using a fiber-optic reflection probe mounted in a ½”-20 UNF Dynisco port. Spectra were acquired continuously during runs; for calibration spectra were recorded for ~30 minutes per concentration.
• Data analysis: Spectral preprocessing (second derivatives illustrated for visual separation), principal component analysis (PCA) for exploratory discrimination, and partial least squares (PLS) regression for quantitative calibration. Model quality assessed by PRESS plots, correlation coefficient (R), RMSEC, and RMSECV where applicable.

Used Instrumentation

• Antaris MX FT-NIR spectrometer (Thermo Fisher Scientific) with a reflection fiber-optic probe.
• Pharma 16 HME twin-screw extruder (Thermo Fisher Scientific) with a Dynisco ½”-20 UNF port for probe mounting.
• Single-screw feeder FW 18 (Brabender Technologies) delivering 500 g/h.
• Materials: PEO Sentry WSR N10 (Dow Wolff Cellulosics), theophylline anhydrous (BASF), lactose (Meggle).

Main Results and Discussion

• Measurement mode selection: Because melt opacity increases with drug load and amorphous-state formation was not achieved in these blends, reflection-mode acquisition at the extruder die was chosen as the practical approach for in-line monitoring.
• PCA findings: When lactose was included to enhance surface reflectivity, PCA showed clear clustering by drug concentration. Principal component 1 aligned with drug content, enabling discrimination among concentrations. Principal component 2 indicated that cluster dispersion decreased with higher drug load, consistent with improved spectral consistency as reflectivity increased.
• PLS calibration: A robust quantitative calibration for theophylline content (with lactose-containing batches) achieved a correlation coefficient of 0.9938 and RMSEC = 0.837. PRESS analysis indicated only two PLS factors were needed, suggesting the spectral variance relevant to the target analyte was captured with minimal model complexity.
• In-line monitoring and process dynamics: The PLS model implemented in automated RESULT software monitored a production run targeted at 15% theophylline. FT-NIR predictions of theophylline content indicated that chemical homogeneity at the die was achieved later in the start-up sequence than suggested by extrusion torque alone. In other words, torque reached a nominal steady value earlier, but the FT-NIR showed the product composition remained transient for a longer period — underscoring the advantage of direct, in-line chemical measurement.
• Performance for target concentration: The Antaris MX system demonstrated the ability to predict a 15% theophylline content within the calibration model’s error bounds; individual predicted values may vary (an example figure in the study showed a calculated value of ~13.4% for a test run), emphasizing the importance of robust calibration and drift control.

Benefits and Practical Applications of the Method

• Real-time assurance of critical quality attributes (drug load homogeneity) at-line or in-line without delay, enabling faster decisions about start-up and material acceptance.
• Direct measurement at the product surface provides more relevant information about content uniformity than indirect process variables like torque or screw speed.
• Minimal spectroscopy model complexity (two PLS factors) reduces overfitting risk and simplifies routine implementation and maintenance.
• Ability to detect when true steady-state product quality is reached, enabling more consistent product release criteria and reduced waste from transient start-up periods.

Future Trends and Potential Applications

• Extending FT-NIR calibration models to cover broader formulation variability and process conditions (temperature, screw configuration, residence time distribution) to improve robustness and transferability across equipment.
• Systematic study of matrix and excipient choices (e.g., reflective fillers vs. glassy amorphous dispersions) to optimize spectral quality and enable transmission-mode options when feasible.
• Integration with multivariate process control and advanced chemometric strategies (transfer learning, adaptive calibration, model updating) for reliable long-term in-line monitoring in commercial production.
• Combined multi-sensor PAT strategies (NIR + Raman + process sensors) to provide orthogonal confirmation of physical and chemical product attributes and to support real-time release testing.

Conclusion

FT-NIR spectroscopy in reflection mode can be successfully implemented as an in-line PAT tool for monitoring drug load in hot melt extrusion. Key practical prerequisites include selecting the appropriate measurement geometry (reflection vs. transmission) based on extrudate optical properties and, where necessary, adding reflective excipients to improve signal quality. Robust chemometric models with few latent variables (two PLS factors in this study) provided accurate quantitation (R = 0.9938, RMSEC = 0.837). Importantly, direct spectroscopic monitoring identified that chemical steady state at the die can lag behind apparent mechanical steady state indicated by torque, highlighting the value of in-line NIR for process control and quality assurance. Further work should expand model scope to accommodate varying process and formulation conditions, enabling broader industrial adoption.

Reference

1. Tumuluri SV, Prodduturi S, Crowley MM, Stodghill SP, McGinity JW, Repka MA, Avery BA. Drug Dev Ind Pharm. 2004 May;30(5):505-11.
2. Rohe T, Becker W, Kölle S, Eisenreich N, Eyerer P. Talanta. 1999 Sep 13;50(2):283-90.
3. Gendrin C, Roggo Y, Spiegel C, Collet C. Eur J Pharm Biopharm. 2008 Mar;68(3):828-37. Epub 2007 Aug 10.