Effect of experimental design on the prediction performance of calibration models based on near-infrared spectroscopy for pharmaceutical applications

Development of iterative optimization technology: Selecting pure component spectra using a small-scale feed frame simulator

The application of spectroscopic process analytical technology (PAT) for in-line data collection offers advantages to modern pharmaceutical manufacturing. Partial least squares (PLS) models are the preferred approach for predicting API potency from PAT data, particularly near-infrared (NIR) spectra. However, the calibration burden of PLS models is sometimes considered prohibitive. Pure component approaches, such as iterative optimization technology (IOT), have a reduced calibration burden for PAT applications. The IOT algorithm is dependent on several assumptions, including the harmonization of spectral collection conditions for pure component and mixture spectra. Collecting pure components under identical conditions to mixture spectra does not guarantee accurate predictions, and not all pure components are suitable for individual processing. This IOT assumption must be addressed to facilitate IOT application in PAT systems. In this work, IOT predicted API potency from in-line NIR spectra using combinations of stagnant and dynamic pure component spectra. A small number of mixture samples called a development set guided the selection of representative pure component spectral sets. Several model performance metrics from the development set predictions identified optimal pure component spectral sets for prediction of test sets. The combination of IOT and a development set generated accurate API potency predictions and potentiates the application of IOT in challenging pharmaceutical manufacturing settings. The IOT assumption of similar collection conditions should not be regarded as an assumption, but rather a consideration that the pure component spectral collection conditions should be representative of the mixture spectra to ensure appropriate predictions.

Leveraging visible and near-infrared spectroelectrochemistry to calibrate a robust model for Vanadium(IV/V) in varying nitric acid and temperature levels

Spectroelectrochemistry and optimal design of experiments can be used to rapidly build accurate models for species quantification and enable a greater level of process awareness. Optical spectroscopy can provide vital elemental and molecular information, but several hurdles must be overcome before it can become a widely adopted analytical method for remote analysis in the nuclear field. Analytes with varying oxidation state, acid concentration, and fluctuating temperature must be efficiently accounted for to minimize time and resources in restrictive hot cell environments. The classic one-factor-at-a-time approach is not suitable for frequent calibration/maintenance operations in this setting. Therefore, a novel alternative was developed to characterize a system containing vanadium(IV/V) (0.01-0.1 M), nitric acid (0.1-4 M), and varying temperatures (20-45 °C). Spectroelectrochemistry methods were used to acquire a sample set selected by optimal design of experiments. This new approach allows for the accurate analysis of vanadium and HNO₃ concentration by leveraging UV-Vis-NIR absorption spectroscopy with robust and accurate chemometric models. The top model's root mean squared error of prediction percent values were 3.47%, 4.06%, 3.40%, and 10.9% for V(IV), V(V), HNO₃, and temperature, respectively. These models, efficiently developed using the designed approach, exhibited strong predictive accuracy for vanadium and acid with varying oxidation states and temperature using only spectrophotometry, which advances current technology for real-world hot cell applications. Additionally, Nernstian analysis of the V(IV/V) standard potential was performed using traditional absorbance methods and multivariate curve resolution (MCR). The successful tests demonstrated that MCR Nernst tests may be valuable in highly convoluted spectral systems to better understand the redox processes' behavior.

Partial Least Squares, Experimental Design, and Near-Infrared Spectrophotometry for the Remote Quantification of Nitric Acid Concentration and Temperature

Near-infrared spectrophotometry and partial least squares regression (PLSR) were evaluated to create a pleasantly simple yet effective approach for measuring HNO₃ concentration with varying temperature levels. A training set, which covered HNO₃ concentrations (0.1-8 M) and temperature (10-40 °C), was selected using a D-optimal design to minimize the number of samples required in the calibration set for PLSR analysis. The top D-optimal-selected PLSR models had root mean squared error of prediction values of 1.4% for HNO₃ and 4.0% for temperature. The PLSR models built from spectra collected on static samples were validated against flow tests including HNO₃ concentration and temperature gradients to test abnormal conditions (e.g., bubbles) and the model performance between sample points in the factor space. Based on cross-validation and prediction modeling statistics, the designed near-infrared absorption approach can provide remote, quantitative analysis of HNO₃ concentration and temperature for production-oriented applications in facilities where laser safety challenges would inhibit the implementation of other optical techniques (e.g., Raman spectroscopy) and in which space, time, and/or resources are constrained. The experimental design approach effectively minimized the number of samples in the training set and maintained or improved PLSR model performance, which makes the described chemometric approach more amenable to nuclear field applications.

Pursuit of the Ultimate Regression Model for Samarium(III), Europium(III), and LiCl Using Laser-Induced Fluorescence, Design of Experiments, and a Genetic Algorithm for Feature Selection

Laser-induced fluorescence spectroscopy, Raman scattering, and partial least squares regression models were optimized for the quantification of samarium (0-150 μg mL^-1), europium (0-75 μg mL^-1), and lithium chloride (0.1-12 M) with a transformational preprocessing strategy. Selecting combinations of preprocessing methods to optimize the prediction performance of regression models is frequently a major bottleneck for chemometric analysis. Here, we propose an optimization tool using an innovative combination of optimal experimental designs for selecting preprocessing transformation and a genetic algorithm (GA) for feature selection. A D-optimal design containing 26 samples (i.e., combinations of preprocessing strategies) and a user-defined design (576 samples) did not statistically lower the root mean square error of the prediction (RMSEP). The greatest improvement in prediction performance was achieved when a GA was used for feature selection. This feature selection greatly lowered RMSEP statistics by an average of 53%, resulting in the top models with percent RMSEP values of 0.91, 3.5, and 2.1% for Sm(III), Eu(III), and LiCl, respectively. These results indicate that preprocessing corrections (e.g., scatter, scaling, noise, and baseline) alone cannot realize the optimal regression model; feature selection is a more crucial aspect to consider. This unique approach provides a powerful tool for approaching the true optimum prediction performance and can be applied to numerous fields of spectroscopy and chemometrics to rapidly construct models.

Comparison Between Pure Component Modeling Approaches for Monitoring Pharmaceutical Powder Blends with Near-Infrared Spectroscopy in Continuous Manufacturing Schemes

Near-infrared (NIR) spectroscopy has become an important process analytical technology (PAT) for monitoring and implementing control in continuous manufacturing (CM) schemes. However, NIR requires complex multivariate models to properly extract the relevant information and the traditional model of choice, partial least squares, can be unfavorable on account of its high material and time investments for generating calibrations. To account for this, pure component-based approaches have been gaining attention due to their higher flexibility and ease of development. In the present study, the application of two pure component approaches, classical least squares (CLS) models and iterative optimization technology (IOT) algorithms, to pharmaceutical powder blends in a continuous feed frame was considered. The approaches were compared from both a model performance and practical implementation perspective. IOT were found to demonstrate superior performance in predicting drug content compared to CLS. The practical implementation of each modelling approach was also given consideration.

Design of Experiments, Chemometrics, and Raman Spectroscopy for the Quantification of Hydroxylammonium, Nitrate, and Nitric Acid

Selecting optimal combinations of preprocessing methods is a major holdup for chemometric analysis. The analyst decides which method(s) to apply to the data, frequently by highly subjective or inefficient means, such as user experience or trial and error. Here, we present a user-friendly method using optimal experimental designs for selecting preprocessing transformations. We applied this strategy to optimize partial least square regression (PLSR) analysis of Stokes Raman spectra to quantify hydroxylammonium (0-0.5 M), nitric acid (0-1 M), and total nitrate (0-1.5 M) concentrations. The best PLSR model chosen by a determinant (D)-optimal design comprising 26 samples (i.e., combinations of preprocessing methods) was compared with PLSR models built with no preprocessing, a user-selected preprocessing method (i.e., trial and error), and a user-defined design strategy (576 samples). The D-optimal selection strategy improved PLSR prediction performance by more than 50% compared with the raw data and reduced the number of combinations by more than 95.5%.

Development of calibration-free/minimal calibration wavelength selection for iterative optimization technology algorithms toward process analytical technology application

As continuous manufacturing (CM) processes are developed, process analytical technology (PAT) via NIR spectroscopy has become an integral tool in process monitoring. NIR spectroscopy requires the deployment of complex multivariate models to extract the relevant information. The model of choice for the pharmaceutical industry is Partial Least Squares (PLS). However, the development of PLS can be burdensome due to the time and resource intensive requirements of calibration. To overcome this challenge, calibration-free/minimal calibration approaches have become of increasing interest. Iterative optimization technology (IOT) algorithms are a favorable calibration-free/minimal calibration approach with only the requirement of pure component spectra for successful active pharmaceutical ingredient (API) quantification. IOT algorithms were utilized to monitor potency trends (qualitative) and API content (quantitative) in a CM system and compared to a traditional PLS model. To overcome the reduced prediction performance of IOT during non-steady state conditions, a novel wavelength method based on variable importance in projection scores was employed. Overall, the success and value of IOT algorithms for application in CM settings was demonstrated.

Chemometrics and Experimental Design for the Quantification of Nitrate Salts in Nitric Acid: Near-Infrared Spectroscopy Absorption Analysis

Implementing remote, real-time spectroscopic monitoring of radiochemical processing streams in hot cell environments requires efficiency and simplicity. The success of optical spectroscopy for the quantification of species in chemical systems highly depends on representative training sets and suitable validation sets. Selecting a training set (i.e., calibration standards) to build multivariate regression models is both time- and resource-consuming using standard one-factor-at-a-time approaches. This study describes the use of experimental design to generate spectral training sets and a validation set for the quantification of sodium nitrate (0-1 M) and nitric acid (0.1-10 M) using the near-infrared water band centered at 1440 nm. Partial least squares regression models were built from training sets generated by both D- and I-optimal experimental designs and a one-factor-at-a-time approach. The prediction performance of each model was evaluated by comparing the bias and standard error of prediction for statistical significance. D- and I-optimal designs reduced the number of samples required to build regression models compared with one-factor-at-a-time while also improving performance. Models must be confirmed against a validation sample set when minimizing the number of samples in the training set. The D-optimal design performed the best when considering both performance and efficiency by improving predictive capability and reducing number of samples in the training set by 64% compared with the one-factor-at-a-time approach. The experimental design approach objectively selects calibration and validation spectral data sets based on statistical criterion to optimize performance and minimize resources.

Investigation of the effect of clinically relevant interferents on glucose monitoring using near-infrared spectroscopy

Near infrared spectroscopy (NIR) is a promising technique for continuous blood glucose monitoring for diabetic patients. Four interferents, at physiological concentrations, were introduced to study how the glucose predictions varied with a standard multivariate calibration model. Lactate and ethanol were found to interfere strongly with the glucose predictions unless they were included in the calibration models. Lactate was mistaken for glucose and gave erroneously high glucose predictions, with a dose response of 0.46 mM/mM. The presence of ethanol resulted in too low glucose predictions, with a dose response of -0.43 mM/mM. Acetaminophen, a known interferent in the glucose monitoring devices used for diabetes management today, was not found to be an interferent in NIR spectroscopy, nor was caffeine. Thus, interferents that may appear in high concentrations, such as ethanol and lactate, must be included in the calibration or model building of future NIR-based glucose measurement devices for diabetes monitoring.

Monitoring the Caustic Dissolution of Aluminum Alloy in a Radiochemical Hot Cell Using Raman Spectroscopy

Chemical processing of highly radioactive materials commonly takes place in heavily shielded hot cells. The remote, real-time monitoring of chemical processing streams via optical spectroscopic techniques in hot cells may be particularly useful. Here, we describe the implementation of Raman spectroscopy and chemometric analysis to monitor the dissolution of aluminum-clad targets containing irradiated aluminum-neptunium oxide cermet pellets in caustic solutions in a hot cell environment. Partial least squares regression analysis was used to generate calibration models to quantify the concentration of dissolved aluminum, nitrate, and hydroxide in solutions within the radiochemical hot cell. This work explored a systematic approach to optimize a matrix of calibration standards using a D-optimal experimental design. The Design of Experiments-based regression model, in comparison to more traditional analytical approaches, was found to be the more practical method for building calibration models, with fewer samples, to obtain informative analytical data from Raman spectra.

Development of an In-Line Near-Infrared Method for Blend Content Uniformity Assessment in a Tablet Feed Frame

Process analytical technology (PAT) has shown great potential for in-line tableting process monitoring. The study focuses on the development and validation of an in-line near-infrared (NIR) spectroscopic method for the determination of content uniformity of blends in a tablet feed frame. An in-line NIR method was developed after careful evaluation of the impact of potential experimental factors on the robustness and model accuracy and precision. The NIR method was validated according to the principles outlined in International Conference on Harmonization-Q2 for validation of analytical procedures and was demonstrated to be suitable for monitoring blend content for the formulation under evaluation. Reliable measurements of blend homogeneity rely on representative sampling. To reach the appropriate scale of scrutiny for a unit dose, the study assessed factors that influence the effective sample size measured by NIR. Spectral averaging, integration time, and feed frame paddle wheel speed were found to influence the effective sample size measured by the NIR probe. The effective sampling size was also estimated by comparing the distribution of predicted values with the reference values. The development of a robust, in-line PAT method was facilitated by thorough understanding of the sensitivity of PAT sensors to factors affecting pharmaceutical processes and products.

Near Infra-Red spectroscopy for content uniformity of powder blends - Focus on calibration set development, orthogonality transfer and robustness testing

The aim of this work was to develop and validate a NIR method for the quantification of three active ingredients from powder blends. Calibration set formulations were selected based on a D-optimal experimental design with three factors (ibuprofen, paracetamol, caffeine) and five variation levels (80-90-100-110-120%). NIR spectra were recorded in transmittance mode using a rotating sample configuration. Prior to model development the effect of spectral pre-processing was assessed by evaluating its impact over the transfer of orthogonality from concentration space to spectral space. NIR method was validated on the full calibration range with external prediction sets, using the accuracy profile approach. Robustness testing results showed that the accuracy of predictions for the analyte found in lower concentrations (caffeine) was influenced by relative humidity, while paracetamol/ibuprofen predictions were robust to all factors. Redefinition of interfering factor variation level was beneficial to reduce the bias in caffeine content predictions. Also, alternative solutions are provided for ensuring robustness and successful routine use.

Diagonal designs for a multi-component calibration experiment

Modern spectroscopic and sensor technologies combined with multivariate modelling are increasingly used for the quantitative analysis of complex mixtures. Their performance depends directly on the data design chosen for model training and validation. A well-balanced calibration experiment with the fewest samples possible presents additional challenges when several mixture components (factors) need to be calibrated on the same dataset and subsequently quantified from the same multivariate measurement. This practically important problem stays poorly addressed by the theory of experimental design. This theoretical work systematically formulates the requirements to an optimal calibration/validation dataset and introduces a new family of calibration designs, where the samples are placed along the diagonals of an experimental space that is a hypercube. Such placement is appropriate due to reasonable assumptions about the linear nature of analytical response. Suggested filling schemes allow economical diagonal designs with intrinsic validation to be built for multiple factors presented in as many levels as the number of samples. The most important practical cases of two and three factors are considered in detail, and generalization to higher dimensions is outlined. Diagonal designs of any complexity can be generated using a simple geometrical scheme or with a supplied script.

Effect of Sampling Frequency for Real-Time Tablet Coating Monitoring Using Near Infrared Spectroscopy

While the sampling of pharmaceutical products typically follows well-defined protocols, the parameterization of spectroscopic methods and their associated sampling frequency is not standard. Whereas, for blending, the sampling frequency is limited by the nature of the process, in other processes, such as tablet film coating, practitioners must determine the best approach to collecting spectral data. The present article studied how sampling practices affected the interpretation of the results provided by a near-infrared spectroscopy method for the monitoring of tablet moisture and coating weight gain during a pan-coating experiment. Several coating runs were monitored with different sampling frequencies (with or without co-adds (also known as sub-samples)) and with spectral averaging corresponding to processing cycles (1 to 15 pan rotations). Beyond integrating the sensor into the equipment, the present work demonstrated that it is necessary to have a good sense of the underlying phenomena that have the potential to affect the quality of the signal. The effects of co-adds and averaging was significant with respect to the quality of the spectral data. However, the type of output obtained from a sampling method dictated the type of information that one can gain on the dynamics of a process. Thus, different sampling frequencies may be needed at different stages of process development.

Selection method of quasi-continuous wavelength combination with applications to the near-infrared spectroscopic analysis of soil organic matter

Equidistant combination multiple linear regression (EC-MLR) for the quasi-continuous wavelength selection of spectroscopic analysis was proposed and successfully applied to the reagent-free determination of soil organic matter with near-infrared spectroscopy. For comparison, the continuous-mode moving window partial least squares (MWPLS) and the discrete-mode successive projections algorithm (SPA) were improved by considering the stability and applied to the same analysis object as well. All methods exhibited good effect, but the modeling accuracy, stability, and validation effect of EC-MLR were better than that of the other two methods. Compared with MWPLS, the optimal EC-MLR model contained only 16 wavelengths, and method complexity was substantially reduced. Compared with SPA-MLR, the optimal EC-MLR model could easily undergo spectral preprocessing to improve predictive capability. Moreover, appropriate equidistant discrete wavelength combination with EC-MLR corresponded to the spectral absorption band with proper resolution and can effectively overcome co-linearity interruption for the MLR model. Thus, the EC-MLR method has great potential in practical application and instrument design.