Postoperative pulmonary embolism in a Chinese surgical cohort (2015-2023): Natural language processing of electronic medical records for surveillance.

Abstract

Postoperative pulmonary embolism remains a leading cause of morbidity, mortality, and health care costs, underscoring the need for improved surveillance and prevention. This study aimed to determine the incidence of postoperative pulmonary embolism following major surgeries in a Chinese population using a natural language processing algorithm and to compare its performance with that of the traditional International Classification of Diseases-based approach. A retrospective analysis was conducted on 56,709 adult patients who underwent major surgeries between 2015 and 2023 at a Chinese tertiary hospital. A natural language processing algorithm was developed to detect pulmonary embolism events within 30 days after surgery using clinical data from consecutive encounters, with manual chart reviews for validation. Annual incidence of pulmonary embolism and International Classification of Diseases-based reporting sensitivity were analyzed. The natural language processing algorithm achieved a recall of 100% and precision of 84.6%. A total of 239 pulmonary embolism events were confirmed within 30 days after surgery, yielding an overall incidence of 0.42%. The incidence of pulmonary embolism varied across surgeries, ranging from 0.09% in hip replacement to 1.18% in large intestine excision. Statistically significant increases in the incidence of pulmonary embolism were observed for pancreatic excision, total abdominal hysterectomy, and spinal fusion. During the index hospitalization, the International Classification of Diseases-based approach identified 44.4% to 92.9% of pulmonary embolism events annually. Compared with the International Classification of Diseases-based approach, the natural language processing algorithm demonstrates superior performance in postoperative pulmonary embolism surveillance, offering a scalable alternative to resource-intensive manual chart reviews or registry programs. Integrating natural language processing algorithms could help health care systems identify trends in postoperative complications, yielding reliable data to drive and measure improvements in surgical care quality.