Application Of Cox Hazard Proportional Model Based On Extreme Learning Machine In Survival Analysis Of Patients With Chronic Heart Failure

Objective:The traditional heart failure survival model is constructed using the Cox hazard ratio regression algorithm,but it receives many restrictions,such as the hazard ratio assumption and variable selection bias.Aiming at the problem of survival analysis and prediction of death in patients with chronic heart failure,this paper constructs a Cox survival analysis model based on extreme learning machine,realizes the modeling of EHRs data of inpatients with high censoring proportions and complex variable relationships,and improves the predictive ability of survival models.It provides a theoretical basis for clinicians to evaluate the prognosis of chronic heart failure patients and personalized treatment,and assists in the formulation of intervention plans for high-risk patients.Methods:Inpatients from Shanxi Medical University First Affiliated Hospital and Shanxi Provincial Cardiovascular Hospital from January 1,2014 to April 15,2019 were diagnosed with chronic heart failure and met the exclusion criteria and informed consent of this study 5279 cases of medical records.Univariate Cox analysis was performed on the variables consistent with this study,and then Lasso Cox regression and RSF were used to select the variables with the best performance as input variables.ELM Cox included all variables,and whether death was used as the outcome variable.Survival model prediction accuracy.In addition,the simulation data was used to model the performance of Lasso Cox,RSF,and ELM Cox prediction models at three different censoring ratios of 25%,50%,and 75%,respectively,before modeling the real data.Results:1.Simulation results show that when the data is censored at a rate of 25%,the performance of RSF and ELM Cox is almost the same,the C-index is above 0.75,and the performance of LASSO Cox is slightly worse.The three types of IBS are below 0.1,and the overall performance of the three algorithms is relatively stable.At 50% of the censored data,LASSO Cox and RSF perform worse than ELM Cox.The latters C-index and IBS indicators are better than the other two models.The performance is excellent;when 75% of the censored data is used,the performance of the three models is reduced,the C-index is below 0.6,and the IBS is above 0.15.In general,as the proportion of censored data increases,the model prediction performance of the three algorithms will gradually decrease.Among them,ELM Cox performs best among the three.2.Lasso Cox and RSF used single-factor Cox analysis to select meaningful 54 variable pairs for variable screening.The variables selected by Lasso Cox were: N-terminal forebrain natriuretic peptide,free triiodothyronine,New York classification,free thyroxine,age,red blood cell distribution width,albumin,renal insufficiency,body mass index,old myocardium There are 17 variables such as infarction,diastolic blood pressure,cardio tonics,diabetes,direct bilirubin,beta blockers,absolute neutrophil counts,valvar disease,etc.RSF finally selected N-terminal forebrain natriuretic peptide,New York grade,Eight variables including free thyroxine,red blood cell distribution width,albumin,age,neutrophil ratio,and body mass index were used for later modeling.3.Evaluation of model predictive ability: The lowest C-index of the traditional Cox risk ratio model is 0.644,and the highest IBS is 0.221,which is the worst compared to the other three models.The Cox model after Lasso penalty on the original data set performs better than the traditional model,and both indexes are better than the Cox model.The highest Cindex of the ELM Cox model is 0.77,which indicates that the model constructed by this algorithm has the highest accuracy rate,and the IBS model with only 0.185 has the most stable overall performance.The conclusion of this real data study is basically consistent with the conclusion of the previous simulation study.It can be considered that applying the Cox risk proportion model of extreme learning machine to the survival analysis of patients with chronic heart failure can have a higher prediction effect.Conclusion:In this study,we used a new clinical predictive modeling algorithm,the ELM Cox model,to establish a survival prediction model for chronic heart failure.Compared with traditional Cox regression,Lasso Cox regression,and random survival forest models,ELM Cox The higher prediction accuracy indicates that there are more complex interactions between the response variables of the predictors of chronic heart failure.Although some studies have shown that RSF has great advantages in submitting Cox data such as highdimensional,non-linear,and does not meet the analysis proportion assumptions,it is still not possible to establish high accuracy for high-dimensional,complex,and high-censored chronic heart failure HER data.ELM Cox has a great advantage in this regard.In this study,the use of extreme learning machines and the analysis and prediction of chronic heart failure survival were performed to evaluate the death of patients with heart failure,suggesting that high-risk groups of patients with chronic heart failure have an unfavorable prognosis,develop targeted treatment measures for patients,and provide theories in accordance with.