Application Of Patient-reported Outcomes In The Study Of Prognosis Of Heart Failure

Objective:The high hospitalization rate,high mortality rate and poor prognosis of heart failure have become public health issues of global concern,so it is extremely important to build a disease prognosis prediction model for patients with heart failure.Patient-reported outcomes(PRO)data is different from the objective medical record data,it reports the patient's condition from the patient's point of view,quantifies all aspects of the patient's physical status,and can fully express the overall health of the patient.On the basis of the above,the prognosis model of patients with heart failure is constructed based on random forest(RF),support vector machine(SVM)and BP neural network(BPNN).The optimal model is used to compare the prediction performance of the model before and after the inclusion of PRO data,and to explore the application value of PRO in the prognosis of heart failureMethods:1.Based on the PRO data of patients with heart failure diagnosed in the first Hospital of Shanxi Medical University and Shanxi Cardiovascular Hospital from May 2017 to November 2019,796 valid data were obtained after data preprocessing.The statistically significant independent variables were selected by univariate analysis.With the selected variables as input variables and whether major adverse cardiovascular events including cardiogenic death and heart failure re-hospitalization occurred within one year after discharge as outcome variables,RF,SVM and BPNN classification prediction models were constructed.Their classification performance was evaluated and compared by ROC2.The optimal model was selected by comparison.The prognostic model of heart failure was constructed before and after the inclusion of PRO data.The NRI,the IDI and the AUC were used to compare the added value of PRO data in the prognostic model of heart failureResults:1.14 variables related to the prognosis of heart failure were selected by univariate analysis.they are age,sex,BMI,diastolic blood pressure,NYHA,smoking history,drinking history,hypertension,valvular disease,renal insufficiency,patient reported outcome scores in four areas with physiology,psychology,society and treatment.As input variables,they are introduced into RF,SVM and BPNN to construct the risk prediction model of major adverse cardiovascular events in patients with heart failure2.Evaluation and comparison of model prediction performance:the evaluation indexes of RF classification prediction model are:AUC value 0.806,sensitivity 67.9%,specificity 93.4%,accuracy 85.3%;SVM classification prediction model evaluation indicators are:AUC value 0.783,sensitivity 65.5%,specificity 89.7%,accuracy 83.2%;The evaluation indexes of BPNN classification prediction model are:AUC value 0.752,sensitivity 61.9%,specificity 87.3%,accuracy 79.2%.In this study,the RF model performs best,the SVM takes the second place,and the BPNN performs generally3.The sensitivity of the RF model not included in PRO data is 26.2%,the specificity is 90.6%,and the AUC value is 0.584.After including PRO data,the AUC value of the model is 0.806,which increases by 0.222.Compared with the prediction model that is not included in PRO data,the NRI value is 0.444,the IDI value is 0.105,and the bilateral P<0.05,the degree of improvement is statistically significant.In the importance score of RF model variables,the scores of physiological,psychological,social and therapeutic dimensions of PRO data were 16.491,13.797,5.925 and 5.531 respectively,which were higher than other variables,suggesting that these variables may have clinical significanceConclusion:1.Based on RF,SVM and BPNN,the prediction performance of RF model is higher than that of the other two models,indicating that RF model has high application value in the study of prognosis of heart failure2.Based on the comparison of the prediction performance of the model before and after the inclusion of PRO data based on the RF model,it can be seen that the prediction ability of the model is significantly improved after the inclusion of PRO data,and the four areas of PRO data score are higher in the importance score of RF model variables,indicating that PRO data play an important role in the study of the prognosis of heart failure.