A Modified Causal Interaction Tree Model Based On Multivariate Treatment Variables

Heterogeneity exists widely in various fields of scientific research.Researchers hope to identify subgroups under heterogeneity.Based on this purpose,this paper builds a new causal interaction tree model in the presence of interactions between multiple treatment variables.The split statistic in the tree model is built on the basis of standardized difference of conditional average treatment effect between groups to measure the heterogeneity of interaction between treatment variables,so that the model can be applied to more complex situations.In the model,three estimation methods of causal effect are used,which are Inverse Probability Weight method based on generalized propensity score,g-formula estimation method based on generalized linear model and Doubly Robust estimation method.In the simulation study,the performance of the model for subgroup identification is evaluated by constructing simulated data with different characteristics.The results show that,among the three methods,the new causal interaction tree model using g-formula estimator has the highest subgroup identification accuracy and the lowest mean square error.Although the overall performance of the new causal interaction tree using Doubly Robust estimator is not optimal,it does show doubly robustness.The paper applies the new causal interaction tree model with g-formula estimator to the study of children with asthma aged 3-10 years in Denmark,and conducts subgroup identification based on the number of recurrence of children with asthma,providing a basis for putting forward more targeted medical prevention and treatment suggestions.According to the analysis,in terms of disease prevention and treatment,researchers are supposed to pay more attention to whom with poor family economic conditions and whose mothers have low education levels among male asthmatic children with low birth weight.