在医学领域，新药物、治疗方法以及其他医疗手段在投入临床使用前，通常需要进行大量对比实验以验证其有效性。因此，随机对照试验在医学研究中具有重要地位。然而，实际操作中常受到各种现实因素和伦理道德问题的干扰，影响试验的进行和结果精度。现实世界中存在大量非随机对照数据，因此对非随机对照试验的研究具有重要意义。本研究利用倾向性评分等方法消除组间混杂因素的影响，构建伪随机对照试验，提高实验结果的准确性。首先，使用R语言通过蒙特卡洛方法进行多种数据情形下的数值模拟实验，验证了不同机器学习方法结合不同策略的效果。模拟结果显示，在简单混杂因素和较少样本量时，逻辑回归构建倾向性评分进行匹配、加权或分层处理能获得较好的混杂均衡效果；随着样本量增加，GBM方法的效果也得到提升。然而，当混杂因素较复杂时，使用GBM方法计算倾向性评分进行匹配处理的效果相对较好。最后，我们进行了新型瓣膜成型环器材的疗效评估研究，采用前瞻性、多中心的研究设计，以确保数据的全面性和代表性。我们构建了未治愈率的风险比值和治愈率的差值指标衡量结果，结合模拟部分的结论，采用逻辑回归方式构建倾向性评分，并结合匹配、加权、匹配后再加权和分层等方法消除组间混杂因素的干扰，对非随机对照试验数据进行了事后随机化。各种方法得到的结果均显示出新型瓣膜成型环器材的治愈率差值小于非劣效下界-10%。因此，我们得出新型瓣膜成型环器材非劣效于已上市器材的结论，为新器材后续研究和上市使用提供了数据支持。

 In the field of medicine, before new drugs, treatment methods, and other medical interventions are introduced into clinical practice, extensive comparative experiments are typically conducted to verify their effectiveness. Consequently, randomized controlled trials hold a prominent position in medical research. However, practical implementation is often influenced by various real-world factors and ethical considerations, which can affect the conduct and precision of trials. A large amount of non-randomized control data exists in the real world, highlighting the significance of studying non-randomized controlled trials. This study employs methods such as propensity score matching to eliminate the influence of confounding variables and construct pseudo-randomized controlled trials, thereby enhancing the accuracy of experimental results. Initially, numerical simulation experiments under various data scenarios were conducted using Monte Carlo methods based on the R language to validate the effectiveness of different machine learning methods combined with different strategies. Simulation results indicate that, under simple confounding factors and small sample sizes, using logistic regression to construct propensity scores for matching, weighting, or stratification can achieve good balance in confounding effects; as sample sizes increase, the effectiveness of the GBM method also improves. However, when confounding factors are more complex, using the GBM method to calculate propensity scores for matching processing yields relatively better results. Finally, a prospective, multicenter study was conducted to evaluate the efficacy of a novel valve molding ring device, ensuring the comprehensiveness and representativeness of the data. We constructed outcome indicators for the risk ratio of non-cure rate and the difference in cure rate, and combined with the conclusions from the simulation section, employed logistic regression to construct propensity scores, and utilized methods such as matching, weighting, matching followed by weighting, and stratification to eliminate intergroup confounding factors, conducting post-randomization on non-randomized controlled trial data. The results obtained from various methods all indicate that the difference in cure rates of the novel valve molding ring device is less than the non-inferiority margin of -10%. Thus, we conclude that the novel valve molding ring device is non-inferior to the devices already on the market, providing data support for further research and subsequent market use of the new device.