| Objective:Kidney transplant recipients need to take immunosuppressive medicine(IM)for life,and low drug adherence is a direct cause of long-term survival of the patient and the graft.The development of medication non-adherence predictive model for kidney transplantation patients is the entry point and breakthrough to form the scientific management and intervention threshold of medication non-adherence.The objective of this study including:1.To clarify the status of IM non-adherence and explore the theoretical research on IM non-adherence of kidney transplantation patients by constructing the theoretical framework for early warning of IM adherence of kidney transplantation,and explore the risk factors of early warning of IM non-adherence on this basis.2.To compare the consequence in difference between adherent patient and non-adherent patient and construct a warning scheme for IM non-adherence of kidney transplantation.3.To construct,evaluate and verify the early warning model of IM non-adherence of kidney transplantation patients based on machine learning and theoretical framework and warning situation division scheme.Methods:1.The status of IM non-adherence in renal transplant recipients and the construction of the theoretical framework for prediction of IM non-adherence of them: This was a cross-sectional study and a total of1,357 renal transplant recipients from transplant follow-up outpatient clinic were enrolled within one year from August 2019 to July 2020.Study was used to continue to explore the characteristics of IM non-adherence in kidney transplantation patients,and theoretical studies on medication non-adherence early warning were carried out based on health belief model(HBM)and planned behavior theory(TPB)to form a theoretical framework for IM non-adherence early warning.2.Construction of alertness partition for prediction of IM non-adherence: Qualitative interview combined with Delphi method was used to construct the early-warning warning classification of IM non-adherence of kidney transplantation patients,and the warning classification standard was obtained.3.Prediction model of IM non-adherence for renal transplant recipients based on machine learning trchnology: Based on Big Data from multiple centers,the theoretical framework of mining the IM non-adherence risk factors after renal transplantation and machine learning methods.The synthetic minority over sampling technique(SMOTE)method was adopted to solve the problem of unbalanced classification of patient data samples,K-fold cross validation was adopted to find the optimalhyperparameters.Five IM adherence risk machine learning models and five IM adherence risk grade machine learning models were built.Through the comparison among models and external validation,an optimal prediction model was obtained.Results:1.The status of IM medication non-adherence among kidney transplant patients: 33.53% of kidney transplant patients reported IM medication non-adherence in the past four weeks,among which,31.10%of patients’ non-adherence behavior was not taking medication at the prescribed time as prescribed by the doctor.Regression analysis showed marital status,religious family income,preoperative drinking history,and post-transplant time,HBM perceived disease susceptibility,HBM perceived severe disease,HBM perceived behavioral benefit,HBM perceived prevention disorders,TPB attitude to taking medication,TPB perceived behavioral control,TPB intention to take medication,TPB past taking medication behavior were all factors that independently predicted IM medication adherence in kidney transplantation recipients2.The theoretical framework for prediction of IM non-adherence of kidney transplant recipients : the modified HBM/TPB integration theoretical model had fitting index GFI=0.995,CFI=0.992,RMSEA=0.033,CMIN/DF=2.469,which meet the requirements of model parameters.3.The alertness partition for prediction of IM non-adherence: a warning situation division scheme for IM non-adherence,including 4first-level items,7 second-level items and 7 third-level items,and the warning situation division standard was formed.4.Prediction model of IM non-adherence for renal transplant recipients based on machine learning trchnology: through internal and external validation,the Support Vector Machine(SVM)model was the best among all IM adherence risk warning models,with ROC value of0.750 for internal validation and 0.668 for external validation.SHAP method was used to evaluate the importance of its predictors.The SHAP values of patients’ age,marital status,HBM perception impairment,post-transplant drug box use and SPSS family support were 0.07,0.06,0.04,0.03 and 0.03,respectively.The Random Forest(RF)model was the best among all the IM adherence risk grade models.The overall prediction accuracy of internal verification was 0.682,and the overall prediction accuracy of external verification was 0.633.Conclusion:1.IM non-adherence among kidney transplantation patients in China was common.2.The constructed TPB/HBM integrated theoretical model framework can improve the explanatory power of IM non-adherence behavior among kidney transplantation patients,and provide theoretical basis for related early warning modeling research.3.The constructed early-warning classification scheme for IM non-adherence of kidney transplant patients enabled the establishment of an early-warning predictive model,which had important clinical implication..4.The establishment of the early-warning predictive model for IM non-adherence in kidney transplantation provided the possibility of rapid identification of high-risk patients with medication non-nonadherence in early clinical stage and carrying out interventions for IM non-adherence.Figures: 30;Tables: 50;Conference: 177... |
