Records of 2 325 patients with CHD were extracted from 20 hostpial information systems, who were divided into two groups, one group has 768 patients using Dengzhan Xixin, the other group has 1 557 patients without using Dengzhan Xixin. Using generalized boosted models (GBM) with propensity scores to balance confounding factors and using three Logistic regressions to analysis the cure rates of coronary heart disease. The results is that 72 knowing confounding factors between two groups, such as: age, admission condition, dying days, regression coefficients of three Logistic regressions were negative (P < 0.05), statistically, the result of using Dengzhan Xixin injection to cure coronary heart disease is significantly higher than do not using Dengzhan Xixin injection. Propensity score could be a good method to balance confounding factors in a retrospective data analysis. However it is not a prospective research, the information from this study should be carefully referred to.