The training set of a universal near infrared (NIR) model for quantitative analysis of a drug should cover as many samples of this drug in the market as possible. Inevitably the model may fail for new products that ha...The training set of a universal near infrared (NIR) model for quantitative analysis of a drug should cover as many samples of this drug in the market as possible. Inevitably the model may fail for new products that have different excipients and production processes. In such circumstances the model should be updated. We here propose a new strategy to iteratively update a universal NIR quantitative model for azithromycin. We prove that universal quantitative models generated from this new strategy are comparably effective for azithromycin injection powders and azithromycin tablets, compared to the strategy using hierarchical clustering method which we reported previously. Furthermore, we establish the correlation coefficient r between a new sample and the training set samples can be used to decide whether or not the model should be updated.展开更多
文摘The training set of a universal near infrared (NIR) model for quantitative analysis of a drug should cover as many samples of this drug in the market as possible. Inevitably the model may fail for new products that have different excipients and production processes. In such circumstances the model should be updated. We here propose a new strategy to iteratively update a universal NIR quantitative model for azithromycin. We prove that universal quantitative models generated from this new strategy are comparably effective for azithromycin injection powders and azithromycin tablets, compared to the strategy using hierarchical clustering method which we reported previously. Furthermore, we establish the correlation coefficient r between a new sample and the training set samples can be used to decide whether or not the model should be updated.
