Abstract
OBJECTIVE: To develop and validate a predictive model for the early identification of concomitant pulmonary embolism (PE) in patients with deep vein thrombosis (DVT) upon hospital admission. METHODS: We retrospectively collected data from a cohort of patients diagnosed with DVT, including baseline demographics, clinical characteristics, laboratory parameters, and imaging-based measurements of compression of the iliac vein to develop a predictive model. The least absolute shrinkage and selection operator regression, widely used in clinical decision-making algorithms for its ability to perform variable selection and regularization simultaneously, was used for variables selection. A multivariate logistic regression was then conducted to construct a predictive model. The model's discriminatory ability was assessed using the area under the curve. Calibration analysis and decision curve analysis were performed. RESULTS: Patients were randomly divided into a development dataset (69.8% [143 with PE and 130 without PE]) and a validation dataset (30.2% [63 with PE and 55 without PE]) for model construction and internal validation. Seven predictors, including female gender, hypertension, cardiovascular disease, fracture, age, D-dimer, and compression of the iliac vein percentage were identified by least absolute shrinkage and selection operator regression and finally incorporated into the nomogram. The model achieved an area under the curve of 0.727 (95% confidence interval, 0.667-0.787) in the training set, and 0.707 (95% confidence interval, 0.611-0.803) in the validation set. The model was well-calibrated, and decision curve analysis demonstrated a net benefit for predicting PE at threshold probabilities ranged between 18% and 80%. CONCLUSIONS: A novel predictive model with strong calibration and discriminative power was developed for assessing concomitant PE risk in patients with DVT. This model may facilitate early estimating of PE probability before obtaining definitive CT angiography results and support timely management processes.