Near-Infrared Spectroscopy for Predicting Dry Granulated Ribbon Density: Effect of Modeling Approach, Chemical Composition, and Spectroscopic Signature On Calibration Model's Performance

The powder properties of pharmaceutical materials play a critical role in determining the manufacturability and performance of solid dosage forms.  For a product manufactured by dry granulation, a commonly utilized unit operation in pharmaceutical industry; the properties of the resultant granules can have a direct impact on the downstream operations (milling, tabletting) as well as on product performance (disintegration, dissolution and friability).   Dry granulation is often accomplished by roller compaction process; wherein the density of the output ribbons is often a critical quality attribute (CQA) and can impact the particle size of milled granules, blend flowability and tabletability.  Defining and maintaining optimal ribbon density is therefore one of the key objectives during drug product process development and scale-up.

Our research objective was to develop an objective NIR based spectroscopic method for on/at-line determination of ribbon density.  The raw NIR spectrum of a sample displays information pertaining to its chemical composition, as well as the three-dimensional physical structure.  Thus, it was pertinent to investigate the impact of chemical variability on the spectral representation of a physical attribute, i.e. density.  Additionally, with respect to the NIR calibration model, the choice of modeling algorithm can influence model's specificity to the property of interest, and its accuracy.  Thus, the specific aims were to evaluate the impact of (i) chemical composition on spectroscopic signatures, and (ii) mathematical algorithms on the NIR model performance.

The calibration model was developed utilizing standardized compacts of excipient based formulations comprising of diluents/binders (microcrystalline cellulose + lactose), disintegrant (0-3%), glidant (0-1.5%) and lubricant (0-1.5%).  The model was then tested on roller compacted ribbons and compacts containing an active pharmaceutical ingredient (API) at various loads (0-20 % w/w).  The standard compacts were prepared using a compaction emulator (Stylcam^TM) and characterized using AOTF-NIR spectrometer.  Key formulation and process variables evaluated were: grade of excipients, ratio of diluents, levels of lubricant and glidant, and compression pressure.  The densities of standard compacts were also determined using volumetric techniques (GeoPyc, AcuPyc and dimensional measurements).  Two modeling approaches were evaluated used for density calibration: 1) a univariate model between density and spectral slope, and 2) multiple linear regression (MLR) approach which modeled density as a function of spectral slope augmented with formulation information.

Near-infrared (NIR) spectroscopy is a powerful technique for characterization of the physical attributes of pharmaceutical samples, in addition to its utility in chemical analysis.  The physical integrity of a sample affects its absorption of NIR radiations.  An increase in sample density produces a lesser proportion of air-material interfaces, in comparison to material-material interfaces.  This suggests a lower probability of NIR photons encountering a change in refractive index, which causes radiation scattering.  This results in a net increase in absorption, depicted as a positive baseline shift in a sample's NIR spectrum.  Consequently, the slope of the baseline referred to as “spectral slope” can be used to model for sample density via an univariate modeling approach.  NIR analysis of the compact and ribbon density samples revealed that the main effect of chemical design variability on spectroscopic signature of density was a decrease in spectral slope values.  An increase in API concentration in the ribbons and compacts produced smaller slopes even at higher compression forces.  The reasons were attributed to the appearance of new shapes/chemical peaks in the raw spectrum.  Additionally, due to the poor deformation behavior of the API, an increase in drug loads produced compacts of lower densities with smaller spectral slopes.

As a next step, the NIR data in conjunction with the formulation and process related variables was used to build a model based on a MLR approach.  The data analysis indicated that the MLR model (error <1%) statistically outperformed the spectral slope based univariate model.  Further, the model also enabled prediction of ribbon density in presence of API, with >95% accuracy for 1-10% drug load. Addition of formulation information in the MLR model provided a better description of ribbon density; which led to an improved model performance compared to spectral slope approach.

The use of NIR for ribbon density measurement has several benefits, e.g. non-destructive, rapid, improved statistics, potential for at/in-line monitoring, over traditional volumetric methods.  The NIR method showed capability for prediction of ribbon density at pilot-scale, using calibrations developed on compacts created at small scale.  The study also demonstrated that a calibration model based on standard compacts can effectively be utilized to predict density of ribbons samples obtained from product manufacturing operations.  The MLR model approach has a wide-ranging applicability during development of low dose formulations, as the model is largely unaffected by the variations in API or excipients encountered during formulation robustness and process development stages.