Safety and efficacy evaluation of laparoscopic surgical robots based on multicenter randomized controlled trials and propensity score matching

With the rapid development of laparoscopic surgical robots, minimally invasive surgical techniques, the laparoscopy surgical robot has been used extensively and has great potential for improving the precision and outcomes of surgery, but empirical validation of their safety and effectiveness remains limited to multicenter RCTs. In this study, we propose a methodological evaluation framework that combines multicenter RCTs and an enhanced PSM approach. Stemming from the classical logistic regression PSM model, we show that a variational autoencoder (VAE) can map high-dimensional patient covariables to a latent space for accurate matching, while an attention-based gradient boosting matcher captures weighted correction for heterogeneous features of different centers. Additionally, integrating the Bayesian hierarchical structure with the survival analysis module allows for the joint evaluation of intraoperative complications and postoperative recovery as multidimensional indicators, improving the robustness and external generalizability of causal estimates. The experimental results show that the proposed model achieved a standardized mean difference of less than 0.1 on 94% of covariates, significantly improving matching balance.