Nondestructive prediction of the drug content of an aspirin suppository by near-infrared spectroscopy

• Published in: Drug development and industrial pharmacy 2010-07, Vol.36 (7), p.839-844
• Main Authors: Otsuka, Eri, Abe, Hiroyuki, Aburada, Masaki, Otsuka, Makoto
• Format: Article
• Language: English
• Subjects: Anti-Inflammatory Agents, Non-Steroidal - administration & dosage; Anti-Inflammatory Agents, Non-Steroidal - analysis; Aspirin - administration & dosage; Aspirin - analysis; Drug Carriers; Drug Compounding; Drug content; Drug Design; Molecular Weight; near-infrared spectrometry; nondestructive prediction; partial least-squares regression analysis; Regression Analysis; Spectroscopy, Near-Infrared; Suppositories - chemistry; suppository
• ISSN: 0363-9045; 1520-5762
• DOI: 10.3109/03639040903532053

Background: A suppository dosage form has a rapid effect on therapeutics, because it dissolves in the rectum, is absorbed in the bloodstream, and passes the hepatic metabolism. However, the dosage form is unstable, because a suppository is made in a semisolid form, and so it is not easy to mix the bulk drug powder in the base. Aim: This article describes a nondestructive method of determining the drug content of suppositories using near-infrared spectrometry (NIR) combined with chemometrics. Method: Suppositories (aspirin content: 1.8, 2.7, 4.5, 7.3, and 9.1%, w w) were produced by mixing an aspirin bulk powder with hard fat at 50 °C and pouring the melt mixture into a plastic mold (2.25 mL). NIR spectra of 12 calibration and 12 validation sample sets were recorded 5 times. A total of 60 spectral data were used as a calibration set to establish a calibration model to predict drug content with a partial least-squares (PLS) regression analysis. NIR data of the suppository samples were divided into two wave number ranges, 4000-12500 cm−1 (LR), and 5900-6300 cm−1 (SR). Calibration models for the aspirin content of the suppositories were calculated based on LR and SR ranges of second-derivative NIR spectra using PLS. Results: The models for LR and SR consisted of five and one principal components (PC), respectively. The plots of predicted values against actual values gave a straight line with regression coefficient constants of 0.9531 and 0.9749, respectively. The mean bias and mean accuracy of the calibration models were calculated based on the SR of variation data sets, and were lower than those of LR, respectively. Conclusion: Limiting the wave number of spectral data sets is useful to help understand the calibration model because of noise cancellation and to measure objective functions.