Risk Measurement of TAVR Surgical Complications Based on Unbalanced Multilabel Classification Approaches

Transcatheter aortic valve replacement (TAVR) is a high-risk cardiovascular interventional procedure with a high incidence of postoperative complications, urgently requiring more refined risk identification and mitigation strategies. The main challenges in assessing the risk of TAVR complications lie in the scarcity of real-world data and the co-occurrence of multiple complications. This study developed an adjustment evaluation model that adapts randomised clinical trial (RCT) evidence to real-world data (RWD) and adopted multi-label classification methods that incorporate a LocalGLMnet-like regularization term, enabling data-adaptive parameter shrinkage for more accurate estimation. In the empirical analysis, with real surgical data from a hospital in the United States, a combination of multi-label random sampling and representative multi-label classification algorithms was used to fit the data. The model was compared across multiple evaluation metrics, including Hamming loss, ranking loss, and micro-AUC, to ensure robust results. The model used in this paper bridges the gap between medical risk prediction and insurance actuarial science, provides a practical data modelling foundation and algorithmic support for the future development of post-operative complication insurance products that precisely align with clinical risk.