Comparison of several chemometric methods of libraries and classifiers for the analysis of expired drugs based on Raman spectra

•Both libraries and classifiers were used to detect the expired drugs effectively.•Cosine and correlation were used and evaluated by ROC and four evaluations.•Three supervised classifiers were used and compared subsequently.•First derivative was the best preprocessing method for libraries.•Max–min normalization was optimal preprocessing method for classifiers.

Some expired drugs are difficult to detect by conventional means. If they are repackaged and sold back into market, they will constitute a new public health challenge. For the detection of repackaged expired drugs within specification, paracetamol tablet from a manufacturer was used as a model drug in this study for comparison of Raman spectra-based library verification and classification methods. Raman spectra of different batches of paracetamol tablets were collected and a library including standard spectra of unexpired batches of tablets was established. The Raman spectrum of each sample was identified by cosine and correlation with the standard spectrum. The average HQI of the suspicious samples and the standard spectrum were calculated. The optimum threshold values were 0.997 and 0.998 respectively as a result of ROC and four evaluations, for which the accuracy was up to 97%. Three supervised classifiers, PLS-DA, SVM and k-NN, were chosen to establish two-class classification models and compared subsequently. They were used to establish a classification of expired batches and an unexpired batch, and predict the suspect samples. The average accuracy was 90.12%, 96.80% and 89.37% respectively. Different pre-processing techniques were tried to find that first derivative was optimal for methods of libraries and max–min normalization was optimal for that of classifiers. The results obtained from these studies indicated both libraries and classifier methods could detect the expired drugs effectively, and they should be used complementarily in the fast-screening.

Graphical abstractFigure optionsDownload full-size imageDownload as PowerPoint slide