Individual Calibration Models Research Articles

Towards point-of-care manufacturing and analysis of immediate-release 3D printed hydrocortisone tablets for the treatment of congenital adrenal hyperplasia

Hydrocortisone (HC) is the preferred drug in children with congenital adrenal hyperplasia due to its lower potency as well as fewer reports of side effects. Fused deposition modelling (FDM) 3D printing holds the potential to produce low-cost personalised doses for children at the point of care. However, the compatibility of the thermal process to produce immediate-release bespoke tablets for this thermally labile active is yet to be established. This work aims to develop immediate-release HC tablets using FDM 3D printing and assess drug contents as a critical quality attribute (CQA) using a compact, low-cost near-infrared (NIR) spectroscopy as a process analytical technology (PAT). The FDM 3D printing temperature (140 °C) and drug concentration in the filament (10%-15% w/w) were critical parameters to meet the compendial criteria for drug contents and impurities. Using a compact low-cost NIR spectral device over a wavelength of 900–1700 nm, the drug contents of 3D printed tablets were assessed. Partial least squares (PLS) regression was used to develop individual calibration models to detect HC content in 3D printed tablets of lower drug contents, small caplet design, and relatively complex formula. The models demonstrated the ability to predict HC concentrations over a wide concentration range (0–15% w/w), which was confirmed by HPLC as a reference method. Ultimately, the capability of the NIR model had preceding dose verification performance on HC tablets, with linearity (R2 = 0.981) and accuracy (RMSECV = 0.46%). In the future, the integration of 3DP technology with non-destructive PAT techniques will accelerate the adoption of on-demand, individualised dosing in a clinical setting.

Understanding the effect of temperature and relative humidity on sensor sensitivities in field environments and improving the calibration models of multiple electrochemical carbon monoxide (CO) sensors in a tropical environment

The data quality of low-cost air quality monitoring sensor systems in field environments needs to be improved through calibrations to compare with regulatory data. This research study focused on improving linear regression calibration models of multiple Alphasense CO-B4 sensors compared to the factory calibration, evaluating the sensor sensitivity and calibration model variations in different temperature and relative humidity (RH) ranges, evaluating the unit-to-unit variability of sensor performances and the use of averaged calibration models instead of individual calibration models. In this study, changes were observed in field sensor sensitivities and manufacturer-given sensor sensitivities, and the inclusion of field sensor sensitivities improved the calibration model performances significantly. Sensor calibration models performed best in the temperature range of 20–24 0C and the RH range of 75–90%, with averaged r and RMSE values around 0.98 and 50 ppb. When compared with raw sensor data (RMSE = 172.1 ppb), the averaged RMSE values of the calibrated concentration values were 97.4 ppb or below for different calibration models. Additionally, results clearly illustrated that the use of averaged calibration model was a reliable alternative for individual calibrations. Furthermore, it was observed that the field calibrations significantly reduce the effect of temperature and RH on sensor outputs and improve the factory calibrations.

Use of mid-infrared spectroscopy to predict the content of bioactive compounds of a new non-dairy beverage fermented with water kefir

The present work aimed at development and characterization of a functional beverage and the construction of calibration models to predict the content of phenolic compounds, antioxidants, and anthocyanins in the fermented product. Mid-infrared spectra of the samples were obtained for different fermentation times, and PLS regression was applied to elaborate predictive models. The beverages were prepared and characterized by traditional methods, and, later, individual calibration models were built for each parameter analyzed. Through PCA it was possible to obtain a certain separation degree between the samples submitted to different fermentation times. The values of the coefficients of determination (R2) gathered in the validation stage of the models ranged from 0.74 to 0.89 and pointed to a good correlation between the data experimentally collected and the values predicted by the models. The results showed that the best prediction was obtained for estimating the content of total anthocyanins, which presented RMSEP, R2 and RPD values of 0.18, 0.89, and 3.11, respectively. This study demonstrated the applicability of MIR spectroscopy with chemometrics to predict some bioactive compounds of interest during kefir fermentation.

Global calibration models for temperature-modulated metal oxide gas sensors: A strategy to reduce calibration costs

Tolerances in the fabrication of metal oxide (MOX) gas sensors lead to inter-device variability in baseline and sensitivity, even for sensors of the same fabrication batch. This has traditionally forced the use of individual calibration models (ICMs) built specifically for each sensor unit, which requires an expensive and time-consuming calibration process and hinders sensor replacement. We propose Global calibration models (GCMs) built using the responses of multiple sensor units, and then applied to a new sensor unit that is not part of the calibration set. GCM have been already successfully applied to transfer calibration models between sensor arrays (electronic noses) for classification tasks. In this work, we investigate the use of such models for regression purposes in temperature-modulated sensors, aiming at the quantification of low concentrations of carbon monoxide (CO) in the presence of variable humidity levels (20–80% r.h. at 26 ± 1 °C). Using a laboratory dataset containing data from 6 replicas of the FIS SB-500–12 model, we evaluate the performance of global models built with data from 1 to 4 sensors when applied to unseen sensor units. Results show that the performance of global models improves with an increasing number of sensors in the calibration set, approaching the performance of individual calibration models (1.38 ± 0.15 ppm for GCM; 1.05 ± 0.24 ppm for ICM), and surpassing their performance only if few calibration conditions per sensor are available (2.09 ± 0.10 ppm for GCM;; 2.76 ± 0.22 ppm for ICM, if only 5 samples per sensor are used).

Rapid and nondestructive evaluation of soluble solids content (SSC) and firmness in apple using Vis–NIR spatially resolved spectroscopy

Abstract Visible–near infrared (Vis–NIR) spectroscopy is a rapid and nondestructive method used to characterize organic compounds in postharvest fruit and vegetable assessment. However, developing robust calibration models is a challenge as conventional spectrometers collect only the cumulative effects of light absorption and scattering. In this study, a multifiber-based Vis–NIR spatially resolved spectra measurement system was designed for simultaneous evaluation of soluble solid content (SSC) and firmness in apple. Thirty silica fibers separated into five groups at 1, 2, 3, 4, and 5 mm away from the light illumination point and connected to a cost-effective Vis–NIR hyperspectral imaging camera were used to acquire spectral data with an improved signal-to-noise ratio (S/N) by a two-step signal averaging process (i.e., 30 camera pixels per fiber and six optical fibers per group). Reflectance ratio spectra were then calculated by dividing the diffusely reflected light intensity detected at distance d +△ by that detected at distance d to realize a light reference-free approach. Finally, the useful explanatory variables were selected by competitive adaptive reweighted sampling (CARS) to construct individual calibration models for various regions. The coefficients of determination ( R c a l 2 ) and the root mean square errors (RMSEcal) of the best-performing calibration models were approximately 0.97 and 0.20 % for SSC and 0.96 and 0.37 N for firmness, respectively. Furthermore, the predicted results were 0.92 and 0.35 % for SSC and 0.87 and 0.71 N for firmness. Our method offers low-cost and portable detection of SSC and firmness for postharvest fruit evaluation.

Multi-unit calibration rejects inherent device variability of chemical sensor arrays

Inherent sensor variability limits mass-production applications for metal oxide (MOX) gas sensor arrays because calibration for replicas of a sensor array needs to be performed individually. Recently, calibration transfer strategies have been proposed to alleviate calibration costs of new replicas, but they still require the acquisition of transfer samples. In this work, we present calibration models that can be extended to uncalibrated replicas of sensor arrays without acquiring new samples, i.e., general or global calibration models. The developed methodology consists in including multiple replicas of a sensor array in the calibration process such that sensor variability is rejected by the general model. Our approach was tested using replicas of a MOX sensor array in the classification task of six gases and synthetic air, presented at different background humidity and concentration levels. Results showed that direct transfer of individual calibration models provides poor classification accuracy. However, we also found that general calibration models kept predictive performance when were applied directly to new copies of the sensor array. Moreover, we explored, through feature selection, whether particular combinations of sensors and operating temperatures can provide predictive performances equivalent to the calibration model with the complete array, favoring thereby the existence of more robust calibration models.

Logistic regression analysis for identifying the factors affecting development of non-invasive blood glucose calibration model by near-infrared spectroscopy

This study was conducted to develop a statistical method for investigating the factors that hamper the development of individual non-invasive blood glucose calibration model using near-infrared (NIR) spectroscopy. First, the individual calibration models for the patients with diabetes mellitus were developed by applying partial least squares regression to relate data of NIR spectra measured at the hand with blood glucose concentrations. Second, items obtained via medical examinations were analyzed using logistic regression in order to specify the factors that hamper the development of successful models. Consequently, the individual calibration models for approximately 40% of patients were successfully developed with a mean standard error of cross-validation (SECV) of 25.0mg/dL. For more than 60% of these patients, over 80% of the validation samples fell within zone A of the Clarke error grid representing clinically acceptable accuracy. According to the results of logistic regression analyses, body mass index (BMI) which may affect the variation in physical measurement conditions was considered to be an effective factor for developing successful calibration models besides the change in blood constituents. A statistical approach using logistic regression analysis represents a novel efficient method for investigating the factors contributing to develop successful individual calibration models.

Development of a comprehensive near infrared spectroscopy calibration model for rapid measurements of moisture content in multiple pharmaceutical products

Near infrared (NIR) spectroscopy has been widely used for the determination of water content in a wide variety of samples. With few exceptions, all methods employ a calibration model developed and applicable for a single product. The current study describes a NIR method using a single, comprehensive calibration model to predict the water content in tablets containing different active pharmaceutical ingredients (API). The calibration model was developed for water content range of 2–13% (w/w) using tablets containing three different APIs and different formulation compositions. To develop a robust comprehensive model, individual calibration models were sequentially developed starting from a simple model for one product to including tablets from all three projects in the final model using partial least square analysis method. Data pretreatments and spectral region selections were performed during the method development to optimize the number of factors and the correlation coefficients for cross-validation and prediction by the comprehensive model. The model reliably predicted the water content in tablet samples of these three products, and can be updated for water measurements of new drug products by adding to the model two samples of the new product for calibration purpose.

Robustness Considerations and Effects of Moisture Variations on near Infrared Method Performance for Solid Dosage Form Assay

Quantitative analysis by near infrared (NIR) spectroscopy involves the establishment of the relationship between spectra, related to both physical and chemical information of a sample, and the corresponding parameter(s) of interest. To make a model useful and robust, other sources of variability, not directly related to the element(s) to predict should be included in the calibration set. One of the potential sources of variability is moisture. Raw materials may have different moisture levels as a function of the manufacturing lots, the geographic situation of a plant, storage conditions or the season. In a traditional calibration effort, tablets are often made at the same time and no robustness to moisture is built into the model. The present article investigates how moisture variations affect the predictive ability of a NIR calibration model for active ingredient in solid oral dosage forms. Examples of variable selection and orthogonalisation techniques are presented as an alternative to including the moisture variability in the calibration data. Tablets composed of acetaminophen, lactose, microcrystalline cellulose, hypromellose and magnesium stearate were manufactured using laboratory scale equipment. A full-factorial design was used to vary acetaminophen (five levels) and excipient ratios (three levels) to generate tablets for calibration and test. Tablets were placed in humidity chambers over saturated salt solutions and equilibrated to 11%, 32%, 52% and 75% relative humidity, respectively. Calibration and test tablets were scanned at each moisture level. Following spectral collection, the acetaminophen content was determined by HPLC. From each sample set representing tablets equilibrated at a single relative humidity, individual calibration models for acetaminophen were constructed. Test samples, stored at the alternate relative humidity conditions, were predicted. When the moisture level was different between calibration and test sets, the prediction error increased, indicating a degradation of the model performance when moisture variance was unaccounted for. Models developed using selected variable, orthogonalisation and global approaches gave significantly lower prediction errors for the test set than the individual models applied to all samples. These findings demonstrated the importance of accounting for expected sources of variance, such as moisture in order to achieve robust calibrations.

Selectivity for glucose, glucose-6-phosphate, and pyruvate in ternary mixtures from the multivariate analysis of near-infrared spectra

Near-infrared spectroscopy offers the potential for direct in situ analysis in complex biological systems. Chemical selectivity is a critical issue for such measurements given the extent of spectral overlap of overtone and combination spectra. In this work, the chemical basis of selectivity is investigated for a set of multivariate calibration models designed to quantify glucose, glucose-6-phosphate, and pyruvate independently in ternary mixtures. Near-infrared spectra are collected over the combination region (4,000-5,000 cm(-1)) for a set of 60 standard solutions maintained at 37 degrees C. These standard solutions are composed of randomized concentrations (0.5-30 mM) of glucose, glucose-6-phosphate, and pyruvate. Individual calibration models are constructed for each solute by using the partial least-squares (PLS) algorithm with optimized spectral range and number of latent variables. The resulting standard errors are 0.90, 0.72, and 0.32 mM for glucose, glucose-6-phosphate, and pyruvate, respectively. A pure component selectivity analysis (PCSA) demonstrates selectivity for each solute in these ternary samples. The concentration of each solute is also predicted for each sample by using a set of net analyte signal (NAS) calibration models. A comparison of the PLS and NAS calibration vectors demonstrates the chemical basis of selectivity for these multivariate methods. Selectivity of each PLS and NAS calibration model originates from the unique spectral features associated with the targeted analyte. Overall, selectivity is demonstrated for each solute with an order of sensitivity of pyruvate > glucose-6-phosphate > glucose.

IMPROVING PHOSPHORUS SENSING BY ELIMINATING SOIL PARTICLE SIZE EFFECT IN SPECTRAL MEASUREMENT

This study investigated the effects of soil particle size on the reflectance spectra of sandy soils using ultraviolet,visible, and near-infrared spectroscopy in sensing phosphorus (P) concentration. Pure sandy soil was graded into threeparticle sizes. Sieve sizes were 125, 250, and 600 .m for fine, medium, and coarse, respectively. Phosphorus application ratesfor the soil samples were 0.0, 12.5, 62.5, 175.0, 375.0, 750.0, and 1000.0 mg kg-1. Concentrations of P in the soil sampleswere analyzed. The reflectance of the samples was measured between 225 and 2525 nm at 1 nm intervals. Overall, soils withcoarse particles absorbed light more than those with medium and fine particles. Detection analysis for soil particle sizes wasconducted using ratio and discriminant analysis methods. Prediction analyses for P concentration were performed usingmultiple linear regression (MLR; stepwise and maximum R2 methods) and linear partial least squares (PLS). Results showedthat detection of the particle size in a spectrum and then the prediction of P using individual calibration models for each soilparticle size produced lower prediction errors. For the maximum R2 MLR, stepwise MLR, and linear PLS analyses,respectively, the standard errors of prediction (SEPs) for determining P concentration without removing the particle size effectwere 105.8, 106.2, and 69.8 mg kg-1 and after removing the particle size effect were 52.8, 73.4, and 64.4 mg kg-1.

Genetic algorithms applied to the selection of factors in principal component regression

Using principal component regression (PCR) as a multivariate calibration tool, always brings up the question what subset of factors, i.e. principal components (PCs) gives the best calibration model. Normally factor selection is based on deterministic methods like top–down procedures, forward–backward-stepwise variable selection or correlated principal component regression (CPCR). In contrast to this, we applied a stochastic method, i.e. a genetic algorithm (GA) for factor selection in this paper. A new kind of fitness function was applied which combined the prediction error of the calibration and an independent validation set. The performance of eigenvalue and correlation ranking was compared. A general statistical criterion for judging the significance of differences between individual calibration models is introduced. In this context it could be shown that for the uncertainties of the standard deviations representing the prediction errors a very simple approximation formula holds which only includes the number of standards. For the current applications it is shown that the GA gives a result very close to CPCR-solutions.

Strategies for coupling digital filtering with partial least-squares regression: Application to the determination of glucose in plasma by Fourier-transform near-infrared spectroscopy

Protocols are established for coupling digital filtering techniques with partial least-squares (PLS) regression for use in constructing multivariate calibration models from Fourier transform near-infrared absorbance spectra. Calibration models are developed to predict glucose concentrations in bovine plasma samples. Employing a calibration data set of 300 spectra collected from 55 plasma samples and 3 plasma lots, individual calibration models are developed based on four spectral ranges selected from the region 5000-4000 cm-1. A separate test set of 69 spectra collected from 14 plasma samples is used to evaluate the computed models. Gaussian-shaped bandpass digital filters are implemented by use of Fourier filtering techniques and employed to preprocess spectra to remove variation due to the background absorbance of the plasma matrix. PLS regression is used with the filtered spectra to compute calibration models for glucose. The optimization of the filter bandpass parameters is explored through the use of response surface methods. Through these optimization studies, calibration models are developed that achieve standard errors of estimate and standard errors of prediction in the range 0.4-0.5 mM across the concentration range of 2.5-25.5 mM. It is determined that the use of digital filtering as a preprocessing step significantly improves the performance of the resulting calibration models, minimizes the importance of spectral range in the calibration model development, and reduces the required number of PLS factors in each model.