The Use Of Random Survival Forest In The Prognostic Analysis Of Sakubatril Valsartan In The Treatment Of Heart Failure

Objective:Considering the limitations of the risk proportion assumption and variable screening bias in the Cox return model imputation,this paper establishes a stochastic survival forest prediction model for the risk of heart failure readmission for heart failure patients treated with sakubatril valsartan,realizes the treatment of complex relationships of variables,improves the prediction accuracy of the survival model,and provides a guiding basis for chronic disease management and clinical decision-making.Methods:Clinical electronic medical records of 300 cases of heart failure treated with sakubatril valsartan at the First Hospital of Shanxi Medical University between October2018 and April 2022 were collected and retrospectively analyzed.Clinical data such as gender,age,blood pressure,comorbidities and laboratory tests at the time of discharge were retrieved,and the patients’ blood pressure,N-terminal brain natriuretic peptide precursor(NT-pro BNP)and left ventricular ejection fraction(LVEF)indexes and heart failure readmission within 1 year were recorded 6 months after discharge.The Lasso Cox proportional risk model and random survival forest model for heart failure readmission within 1 year were developed,and the predictive accuracy of the two models was compared using consistency indices,time ROC curves,time AUC curves,calibration curves,Brier scores,decision analysis curves,NRI and IDI.The study followed the Declaration of Helsinki,medical record information was reviewed and approved by the Ethics Committee of the First Hospital of Shanxi Medical University {[2020] Lun Audit No.(K071)},and patient informed consent was waived.Results:1.Patients with heart failure treated with sakubatril valsartan had increased LVEF levels and statistically significant reductions in NT-pro BNP,SBP and DBP at 6 months of discharge compared to discharge(P < 0.05),indicating improved cardiac function with ARNI treatment,as well as a hypotensive effect.2.The Lasso Cox model and the RSF model were used to model the screening of all29 independent variables.14 variables were finally screened by the Lasso algorithm(age,heart rate,weight,ARNI daily dose,hospital stay,NYHA classification,LVEF,Hb,TC,HDL-C,TG,LDL-C,Na,Cl)for Cox multifactor analysis,and the results showed that LVEF,Hb,TC,HDL-C,TG and Cl were influential factors for heart failure readmission(P < 0.05).After the RSF model was adjusted for parameters(nodesize=7,mtry=8,ntree=500),10 variables were screened using the variable importance method combined with the minimum depth method(TC,hospital stay,Na,HDL-C,LDL-C,Hb,K,TG,WBC and LVEF)for post-modelling.3.Model prediction accuracy evaluation: C-index,time ROC,time AUC,calibration curve,decision analysis curve,net reclassification index and comprehensive discrimination improvement index were used to evaluate the prediction ability of Lasso Cox and RSF models.The consistency index of RSF models(training set 0.906,test set0.886)was higher than the C-index of Lasso Cox model(training set 0.745,test set0.695);The time ROC curve is plotted based on the risk score calculated by the model and the area under the time ROC curve is also larger for the RSF model corresponding to three months,six months and one year AUC;The calibration curve shows that the predicted value of RSF model is closer to the actual value,and Brier score is smaller;The decision analysis curve is above the None and All lines,with both NRI and IDI greater than 0,it suggests that the RSF model has clinical utility and is more suitable than the Lasso Cox model for predicting the prognosis of this group of patients.4.RSF model output: measures are provided for patients to develop treatment plans based on the values of the corresponding predictor variables at the lowest number of outcome events.Cutoff values were calculated based on the patient’s risk score,divided into high and low risk groups,and Kaplan-Meier curves showed that high risk groups were associated with poor prognosis(P < 0.001).Conclusions:In clinical real-world studies,HF patients treated with ARNI showed significant decreases in systolic and diastolic blood pressure,as well as NT Pro BNP,and significant increases in LVEF levels after 6 months of discharge.This also confirms the effectiveness of sakubatril valsartan in improving cardiac function.After using multiple indicators to evaluate the predictive ability of the model,the RSF model still outperforms the Cox regression model with Lasso penalty on variables,even if the Cox regression is a Cox proportional risk model that satisfies the PH assumption.The RSF model can take corresponding treatment measures for patients’ condition based on the relationship between predicted variables and outcome events;Provide early intervention measures for high-risk populations based on risk level grouping.In this study,a randomised survival forest model was developed to assess unplanned heart failure readmission within 1 year of discharge in ARNI-treated heart failure patients,to screen and identify high-risk groups and provide a theoretical basis for healthcare,disease prediction and management of chronic disease.