A Functionalized Method for Assessing the Risk of Adverse Events in Patients with Pulmonary Embolism

Keyu Sun^1,2Long Yu^1,3*Zichen Xie²Haisu Lu²Shengzhang Wang^1*

• ¹Fudan University, Shanghai, China
• ²Fudan University Minhang Hospital, Shanghai, China
• ³Zhejiang University, Hangzhou, China

Background: Pulmonary embolism (PE) is a common and potentially fatal condition in emergency medicine, and accurate early risk stratification is critical for guiding clinical management. Existing prognostic models, such as the Pulmonary Embolism Severity Index (PESI) and PERFORM, rely on discrete scoring systems and subjective variables, which may introduce bias and limit predictive precision. Methods: In this study, we propose CON-PERFORM, a novel functional prognostic model that incorporates three readily available objective parameters: age, heart rate, and arterial partial pressure of oxygen (PaO2). Weighting coefficients were determined through functional optimization rather than empirical cutoffs, yielding a continuous risk score designed to improve accuracy and threshold stability. A retrospective cohort of 559 objectively confirmed PE patients (373 in the training set, 186 in the validation set) was used to construct and evaluate the model, and an independent prospective cohort of 128 patients was collected for further validation. Results: Compared with the PERFORM method, CON-PERFORM demonstrated superior diagnostic performance, with higher area under the curve (AUC), accuracy, and specificity, while maintaining comparable sensitivity. In the training cohort, CON-PERFORM achieved an AUC of 0.841 versus 0.793 for PERFORM, with consistent improvements observed in both the validation and prospective cohorts. Moreover, CON-PERFORM effectively stratified patients into high-and low-risk groups, which displayed distinct survival and discharge patterns over 30 days. Conclusion: In conclusion, CON-PERFORM provides a simple, objective, and robust tool for individualized risk assessment in PE. Its improved diagnostic performance and stability across cohorts highlight its potential for clinical application and resource optimization in emergency care.