Abstract:To address the limitations of existing causal heterogeneity effect estimation methods, which primarily focus on a binary treatment and fail to accommodate multiple treatment scenarios, this paper proposes an innovative two-stage multiple treatment-causal forest (MT-CF) method. This method achieves precise estimation of causal heterogeneity effects in multiple treatment settings through a two-stage analytical process. In the first stage, traditional causal forest methods are used to estimate pairwise treatment effect differences. In the second stage, these pairwise differences are extended to the multiple treatment context, enabling the estimation of the difference between each treatment and the average effect across all treatments. Two sets of simulation experiments were designed to verify the performance of the MT-CF method in different scenarios. In addition to the simulation experiments, the MT-CF method was applied to the analysis of the breast cancer data in the database of the surveillance, epidemiology, and end results (SEER) to evaluate the individualized treatment effects of postoperative adjuvant therapies. The results indicate that the MT-CF method performed well in both simulation experiments, with multiple evaluation metrics outperforming other benchmark methods. In evaluating the individualized treatment effects of postoperative adjuvant therapies, the MT-CF method can also estimate the relative effects of various treatment options. This provides a scientific basis for making individualized treatment decisions. Furthermore, the results validate the MT-CF methods practicality and reliability when applied to real-world medical data. The proposed method not only provides an effective tool for estimating multiple treatment causal effect, but also provides a new perspective and idea for solving complex causal inference problems.