Establishment and Verification of the Prediction Model for the Inspection Results of the Category Ⅲ Medical Device Business License
Objective To establish and verify the prediction model for the inspection results of the category Ⅲ medical device business license,and to provide a reference for scientific and intelligent supervision.Methods The application materials for category Ⅲ medical device business license submitted by enterprises in Guangzhou from June 2018 to December 2022 were retrospectively selected.The inclusion and exclusion criteria were determined,the relevant information and examination results from the application materials in the internal system were extracted,the univariate analysis,forward-backward stepwise regression analysis and multivariate Logistic regression analysis were conducted on the extracted variables,and variables significantly related to the examination results were screened.On this basis,a predictive model was constructed and evaluated by the receiver operating characteristic curve(ROC),calibration curve,Hosmer-Lemeshow goodness of fit test,and decision curve analysis(DCA).Results A total of 182 enterprises were included,and six predictive variables were selected through univariate analysis,including the number of professional and technical personnel(factor A),the number of quality management personnel(factor B),the education level of quality director(factor C),the area of business site(factor D),the area of warehouse(factor E),and the volume of cold storage(factor F).The forward-backward stepwise regression results showed that the minimum Akaike Information Criterion(AIC)value could be obtained as 216,the education level of the quality director and the volume of cold storage,at which point the model combined excellent predictive performance and low complexity.Three influencing factors(A,C,and F)were selected as the final predictive variables,and a multivariate Logistic regression analysis was performed to construct a prediction model by drawing a nomogram.The area under ROC curve(AUC)of the prediction model was 0.73 with good model discrimination.The result of the Hosmer-Lemeshow goodness of fit test showed a good model calibration(P = 0.387).DCA results showed that when the cost-benefit ratio was above 0.5,the net benefit of inspection based on the prediction results of the model was significantly higher than inspection of all enterprises and no inspection at all.Conclusion Factors A,C,and F are the main influencing factors on whether an enterprise's on-site inspection is passed or not.The established nomogram of the prediction model has a certain value for the inspection results of the category Ⅲ medical device business license.
medical device enterprisesbusiness license inspectionprediction modelnomogram
国家科技期刊平台
NSTL
万方数据
