A Risk Assessment Model for Predicting Perioperative Venous Thromboembolism in Patients receiving surgery under anesthesia care.

Abstract

Background: Perioperative venous thromboembolism (VTE), including pulmonary embolism (PE) and deep vein thrombosis (DVT), contributes significantly to morbidity, mortality, and healthcare costs of care. A reliable risk assessment model is essential for identifying patients at risk for perioperative VTE. This study aimed to develop and validate a model to predict VTE aligned with the Agency for Healthcare Research and Quality's (AHRQ) Patient Safety Indicator 12 (PSI-12), which tracks VTE occurrences from hospital admission through discharge. This approach may improve early identification and targeted prevention.

Methods: We retrospectively analyzed hospital registry data from surgical patients at two tertiary care hospitals in the US: Montefiore Medical Center (MMC) in the Bronx, NY, and Beth Israel Deaconess Medical Center (BIDMC) in Boston, MA. Data from MMC between 2016 and 2021 were used for prediction model creation, while data from 2021 to 2023 served for internal temporal validation. We classified perioperative VTE if patients carried a new ICD code for DVT or PE, and a VTE-related imaging order was documented. Stepwise backward logistic regression and bootstrap resampling were employed for model development. Model performance was evaluated using the receiver operating characteristic (ROC) curves, and Brier score.

Results: Among 319,134 surgical patients included in the study, 2,647 (0.8%) were diagnosed with perioperative VTE following hospital admission. The model exhibited robust discriminatory performance across all cohorts, with areas under the receiver operating characteristic curve (AUC) of 0.87 (95%-confidence-interval [95%CI], 0.86-0.89) in the development cohort, 0.84 (95%CI, 0.81-0.87) in the internal temporal validation cohort, and 0.76 (95%CI, 0.75-0.77) in the external validation cohort. By contrast, the Caprini Score and Roger's risk assessment model exhibit significantly lower predictive accuracies of 0.66 and 0.51 respectively. Additionally, the prediction score exhibited strong performance in predicting VTE both in patients before surgery (AUC=0.91; 95%CI, 0.89-0.93) and in those after surgery (AUC=0.84; 95%CI, 0.82-0.86).

Conclusions: We developed a clinically intuitive risk assessment model that predicts perioperative VTE across diverse surgical populations, based on the AHRQ's definition. This model demonstrates superior performance compared to existing instruments, offering the potential for improved VTE prevention during hospitalization.