Abstract
In order to solve the problems of missing, discontinuous, and unstructured data in past cardiovascular interventional studies, this work constructed a database of specific cardiovascular interventional diseases. Within one year of its implementation in a top-three hospital in Zhejiang Province, the database collected a total of 728 cases of cardiovascular interventional patients, realizing the structuring of patient data and 360° whole-cycle management. With the support of a specific disease database, we proposed an improved LSTM-BLS model to predict the risk of death after cardiovascular interventions. Compared with the traditional long-short term memory (LSTM) model, the parallel learning structure-width learning system (BLS) was introduced to calculate the weights directly, which can solve the problems of overfitting and delay caused by the deep structure of LSTM, so as to improve the accuracy of prediction. The experimental results showed that the accuracy rate of the proposed model was 87.46%, the precision was 90.74%, and recall rate was 93.61%, which can objectively reflect the postoperative death risk of patients, and help doctors to make timely medical intervention for patients with high death risk. SUPPLEMENTARY INFORMATION: The online version contains supplementary material available at 10.1038/s41598-026-39788-7.