Development and validation of a novel risk assessment model to estimate the probability of pulmonary embolism in postoperative patients.

Pulmonary embolism (PE) is a leading cause of mortality in postoperative patients. Numerous PE prevention clinical practice guidelines are available but not consistently implemented. This study aimed to develop and validate a novel risk assessment model to assess the risk of PE in postoperative patients. Patients who underwent Grade IV surgery between September 2012 and January 2020 (n = 26,536) at the Affiliated Dongyang Hospital of Wenzhou Medical University were enrolled in our study. PE was confirmed by an identified filling defect in the pulmonary artery system in CT pulmonary angiography. The PE incidence was evaluated before discharge. All preoperative data containing clinical and laboratory variables were extracted for each participant. A novel risk assessment model (RAM) for PE was developed with multivariate regression analysis. The discrimination ability of the RAM was evaluated by the area under the receiver operating characteristic curve, and model calibration was assessed by the Hosmer-Lemeshow statistic. We included 53 clinical and laboratory variables in this study. Among them, 296 postoperative patients developed PE before discharge, and the incidence rate was 1.04%. The distribution of variables between the training group and the validation group was balanced. After using multivariate stepwise regression, only variable age (OR 1.070 [1.054-1.087], P < 0.001), drinking (OR 0.477 [0.304-0.749], P = 0.001), malignant tumor (OR 2.552 [1.745-3.731], P < 0.001), anticoagulant (OR 3.719 [2.281-6.062], P < 0.001), lymphocyte percentage (OR 2.773 [2.342-3.285], P < 0.001), neutrophil percentage (OR 10.703 [8.337-13.739], P < 0.001), red blood cell (OR 1.872 [1.384-2.532], P < 0.001), total bilirubin (OR 1.038 [1.012-1.064], P < 0.001), direct bilirubin (OR 0.850 [0.779-0.928], P < 0.001), prothrombin time (OR 0.768 [0.636-0.926], P < 0.001) and fibrinogen (OR 0.772 [0.651-0.915], P < 0.001) were selected and significantly associated with PE. The final model included four variables: neutrophil percentage, age, malignant tumor and lymphocyte percentage. The AUC of the model was 0.949 (95% CI 0.932-0.966). The risk prediction model still showed good calibration, with reasonable agreement between the observed and predicted PE outcomes in the validation set (AUC 0.958). The information on sensitivity, specificity and predictive values according to cutoff points of the score in the training set suggested a threshold of 0.012 as the optimal cutoff value to define high-risk individuals. We developed a new approach to select hazard factors for PE in postoperative patients. This tool provided a consistent, accurate, and effective method for risk assessment. This finding may help decision-makers weigh the risk of PE and appropriately select PE prevention strategies.

Clinical symptoms and related risk factors in pulmonary embolism patients and cluster analysis based on these symptoms.

Pulmonary embolism (PE) remains largely underdiagnosed due to nonspecific symptoms. This study aims to evaluate typical symptoms of PE patients, their related predictors, and to differentiate typical clusters of patients and principal components of PE symptoms. Clinical data from a total of 551 PE patients between January 2012 and April 2016 were retrospectively reviewed. PE was diagnosed according to the European Society of Cardiology Guidelines. Logistic regression models, system clustering method, and principal component analysis were used to identify potential risk factors, different clusters of the patients, and principal components of PE symptoms. The most common symptoms of PE were dyspnea, cough, and tachypnea in more than 60% of patients. Some combined chronic conditions, laboratory and clinical indicators were found to be related to these clinical symptoms. Our study also suggested that PE is associated with a broad list of symptoms and some PE patients might share similar symptoms, and some PE symptoms were usually cooccurrence. Based on ten symptoms generated from our sample, we classified the patients into five clusters which represent five groups of PE patients during clinical practice, and identified four principal components of PE symptoms. These findings will improve our understanding of clinical symptoms and their potential combinations which are helpful for clinical diagnosis of PE.