Validation And Maintenance Of Universal Near-infrared Models

For the purpose of quick inspection of medicines in countryside and fighting against counterfeit drugs,the National Institute for the Control of Pharmaceutical and Biological Products(NICPBP) have designed mobile vehicles(mobile labs) which was equipped with many instruments including near-infrared(NIR) spectrometers and put them into use in China in the past few years.In NIR fast drug prescreening system, universal identification and quantification models are combined to screen counterfeit medicines,while the former models identify and discriminate suspicious medicines rapidly and the latter models quantify the contents of API(Active Principal Ingredient) or moisture in the formulations to determine whether the contents meet the requirements.NIR models are the essential part of NIR fast drug prescreening system and vital for mobile labs to exert their practical actions.In this dissertation,the current universal identification models were validated comprehensively and the performances of them were evaluated on the whole.The results indicate that they have favorable identification performances,and thus confirm the development of universal identification models is appropriate and feasible.In the meantime,it is also successful to construct universal quantification models for concrete formulations in previous work,however,are they special examples or of universal sense? In this paper,nine types of different beta-lactam antibiotics powder injections were chosen to validate the feasibility of universal quantification models specifying moisture content as the interested component, simulating the process of building universal calibration models for the samples with same INN(International Nonproprietary Name) from diverse formulations and sources.It has been proved that the universal identification and quantification models are effective though different manufactures employed different process,excipients and packages.However,when new enterprises emerge or new formulations are employed in pharmaceutical industry,original universal identification and quantification models might become invalid for the new products and may yield wrong discriminate results or unacceptable prediction errors due to the occurrence of new products.In this dissertation, the principle and methods to maintain and update universal identification models have been discussed and investigated.The normality plot of spectral distances indicating the internal differences of the spectra of the same medicine preparations is the basic principle to judge whether the spectra grouping is rational,and the strategies to update original universal identification models using the spectra of new products were proposed under three different circumstances.As to the limit of universal quantification models,two solutions were proposed to solve the problem of model invalidation.One is the classical model updating method by which information of new products are added to original models so as to include more variations and expand the applicability.Another solution is calibration transfer from known samples to new products by PDS algorithm to extend the performance of universal models.