Acomparative Study Between Parameters And COX Regression Model Used In Survival Analysis In Patients With ALL

In the follow-up study on cancer and chronic diseases, survival analysis has been concernedby many clinical researchers. In order to make a better selection and evaluation of analysismodel for clinical follow-up data, this study used numerous prognosis factors to build significantmodels, made an impact analysis and assessment of data quality in survival analysis, comparedparameter and semi-parameter models of survival data in patients with ALL.This study involved filtering prognostic factors, analysising and diagnosising regressionmodel of survival analysis, and comparing parameter and semi-parameter models of survivaldata in patients with ALL; All those steps were implemented in SAS software. On the parametricregression, in this study several common models were fitted and made comparative analysis:exponential regression model, Weibull regression model, generalized Gamma regression modeland standard Gamma regression model. The goodness of fit test was used in the parameter model,with graphic method and likelihood ratio goodness of fit test. In graphic method, Cox-Snellresiduals were used, which were equivalent to the so-called standard residual test based onlog-linear model expressions. This method is theoretically appealing and easy to use, but thedrawback is not sensitive to differences between the model fit. Using likelihood ratio statistic forgoodness of fit test make up the shortage of front diagnostics law. The results of fitting andtesting showed that the standard Gamma model was the optimal regression model for thesurvival data in patients with ALL; which model made interpretation of results more reasonablefor clinical practice, and could provide main evidence for analysis and evaluation of prognosis.The results also indicated that better fitting parameter model is clearly able to take advantage ofmore information in order to produce a better parameter estimation performance.This study conducted formal hypothesis testing with time dependent covariates in CoxProportional Hazards Assumption, and showed the indexes of the impact analysis: ProposedSchoenfeld residuals, weighted score residuals, martingale residuals, the remaining residuals,likelihood distance and maximum impact curvature. For follow-up data, comparing models anddiagnosising influential case were carried out.This study showed that the influential case affected partial likelihood estimate of regressioncoefficients in the Cox model, which could lead that the results of model fitting were not steady,and Sometimes individual strong influence would change the statistical significance of the regression coefficients. The influential case may be derived from the data, such as the extremepoints of the sample and covariate imbalance. Or model aspects, such as loss of the proposedmodel, contrary to the model assumes the like. The analysising survival data in patients withALL could examine the reasonableness of the Cox proportional hazards model assumes anddetect outliers in the data. This analysis also could find that some variables in this case did notmeet the proportional hazards model assumes, and there were two strong influential cases. Dueto the wide scope of the Cox model, In practical applications, analysts often ignore itsapplication conditions and model diagnostics, therefore directly affect the stability of the model.so, the analysis on the combination of the survival time distribution, the proportional hazardsmodel assumes, and detecting influential cases, should be the primary content of Coxproportional hazards regression model.Parametric regression model can provide more information by used the experience orhistorical data in survival analysis, When it meets the application conditions and assumedrequirements, the parameters model not only have a high precision estimates but also used moreinformation than half parametric model if sample size is limited. Through clinical experiencecomparative analysis in diagramming method and likelihood ratio goodness of fit test.This studyfurther confirmed the instances that Standard Gamma model has better results in fitting the ALLrelapse data. Prognostic factors included in the model were more in line with clinical practice,and have a more reasonable interpretation of results.Use time dependent covariates in Cox proportional hazards assumption as formal testingMeanwhile do affect diagnosis should be the first step in Cox regression analysis. This not onlyallows partial likelihood estimation of regression coefficient in Cox results more credible, Andstill can corrected the results of regression coefficient estimates issues which due to the stronginfluence of individual points. So the interference of strong impact point in proportional hazardsassumption can be combined consider in analysis of the data.