Research On Causal Inference Based On Cox Proportional Hazard Model

To explore causation is the ultimate goal of many fields of research.Causal inference plays an important role in the development of statistics.Due to the influence of realistic factors,causal inference mainly studies causation based on observational data.Potential outcome model is one of the main models to study causal inference.The average treatment effect can be obtained by comparing and analyzing the difference between potential outcomes.Cox proportional hazard model is one of the most important models in survival analysis,which is of great significance to many research fields.In this thesis,the inverse probability weighted method is used to adjust the covariates in the observational data based on Cox proportional hazard model,and the average treatment effect is calculated by the adjusted data to study the causation between treatment and survival time.In observational data,the imbalance of covariates between the treatment group and the control group would bias the research results.In this thesis,under the counterfactual framework,logistic regression is used to estimate the stabilized weights.The original data is weighted in the case of censoring,so as to balance the influence of covariates on the estimation of the average treatment effect.Cox proportional hazard model is used to calculate the average treatment effect based on risk difference and restricted mean time.The Cox model with propensity score is used for statistical inference,and the hazard ratio between the two groups is calculated to analyze the causation between the treatment and the survival outcomes.R is used to simulate and generate datasets with different distributions of survival time.Adjust the datasets,and contrast the mean difference and distribution of covariate between two groups before and after adjustment.Compare the change of the average treatment effect,and analyze whether different censoring proportions has influence on matching results.The results show that the distribution of the adjusted variables is more balanced than before between the treatment group and the control group,and the average treatment effect also better reflects the influence of treatment variables on survival outcomes.Finally,an example is analyzed.The balance between the two groups is quantified by absolute standardized mean difference,and study causation between treatment and survival outcomes before and after the dataset adjustment.