Development Of A Data Mining-based Prediction Model For UTI In T2DM Patients

Objective:The increasing number of T2DM patients in China has become an important factor affecting the health of our population,and the risk of UTI in T2DM patients is higher than that of normal people,which is a great burden on patients and society.Therefore,in order to prevent and reduce the occurrence of UTIs in T2DM patients,this study is based on real-world medical data,by analyzing the factors influencing the occurrence of UTI in T2DM patients,and using a variety of machine learning algorithms to establish predictive models to provide reference for the formulation of individualized treatment plans for clinical T2DM patients,in order to reduce the incidence of UTIs in T2DM patients.Methods:Firstly,the published literature on UTI-related influencing factors in patients with T2DM was systematically searched and analyzed to fully understand the possible factors that have an impact on the establishment of predictive models.Secondly,the diagnosis and treatment data of T2DM patients hospitalized in Sichuan Provincial People’s Hospital from September 1,2018 to August 30,2021 were collected,and a real data information extraction table was formulated according to the results of literature research,and patients were screened and patient data were extracted by including exclusion criteria.Then statistical methods are used to describe the data statistically and discuss the distribution characteristics of variables in groups.Then,4 data filling methods,3 data sampling methods and 4 feature screening methods are used to process the data and filter important variables,and finally 18 machine learning algorithms are used to establish a UTI prediction model for T2DM patients,and the model is verified by the set of data,and finally the five models with the best performance are screened out according to the corresponding model evaluation indicators.Results:A total of 119 articles were included in the literature study,and the main influencing factors extracted included basic patient information,comorbidity information,combined medication information and related laboratory tests.Real-world data were screened by inclusion exclusion criteria to obtain a study dataset with a sample size of 1340.According to whether patients developed UTI after admission,they were divided into two groups,including 440 in the UTI group and 900 in the non-UTI group.After preprocessing the data,a total of 106 variables were obtained,including 46 continuous variables and 60 discrete variables.Through data filling,data sampling,and feature screening,a total of 864 prediction models were established by machine learning,and five models with the best prediction performance were selected through model evaluation indicators,the filling methods used were no filling and random forest filling,the sampling methods were no sampling and random upsampling,and the feature screening methods were no screening and Lasso screening,and the AUC values of the five best models were>0.9,indicating that the model had good predictive performance and had certain clinical use.Conclusion:In this study,the UTI prediction model of T2DM patients constructed by machine learning algorithm can effectively predict whether patients will develop UTI,which has certain clinical applicability,and can provide reference for the treatment plan and lifestyle of T2DM patients,and provide methodological reference for data mining research and prediction model construction of other chronic diseases.