Performance of risk assessment models for venous thromboembolism in critically ill patients receiving pharmacologic thromboprophylaxis: a post hoc analysis of the PREVENT trial
Hasan M Al-Dorzi,Hatim Arishi,Fahad M Al-Hameed,Karen Ea Burns,Sangeeta Mehta,Jesna Jose,Sami J Alsolamy,Sheryl Ann I Abdukahil,Lara Y Afesh,Mohammed S Alshahrani,Yasser Mandourah,Ghaleb A Almekhlafi,Mohammed Almaani,Ali Al Bshabshe,Simon Finfer,Zia Arshad,Imran Khalid,Yatin Mehta,Atul Gaur,Hassan Hawa,Hergen Buscher,Hani Lababidi,Abdulsalam Al Aithan,Abdulaziz Al-Dawood,Yaseen M Arabi,Saudi Critical Care Trials Group
DOI: https://doi.org/10.1016/j.chest.2024.07.182
IF: 9.6
2024-09-02
Chest
Abstract:Background: The diagnostic performance of the available risk assessment models for venous thromboembolism in critically ill patients receiving pharmacologic thromboprophylaxis is unclear. Research question: For critically ill patients receiving pharmacologic thromboprophylaxis, do risk assessment models predict who would develop venous thromboembolism or who could benefit from adjunctive pneumatic compression for thromboprophylaxis? Study design and methods: In this post hoc analysis of the PREVENT trial, we evaluated different risk assessment models for venous thromboembolism (ICU-VTE, Kucher, Intermountain, Caprini, Padua, and IMPROVE models). We constructed receiving operator characteristic curves and calculated the sensitivity, specificity, positive and negative predictive values, and positive and negative likelihood ratios. Additionally, we conducted subgroup analyses evaluating the effect of adjunctive pneumatic compression versus none on the study primary outcome. Results: Among 2003 patients receiving pharmacologic thromboprophylaxis, 198 (9.9%) developed venous thromboembolism. With multivariable logistic regression analysis, the independent predictors of venous thromboembolism were APACHE II score, prior immobilization, femoral central venous catheter, and invasive mechanical ventilation. All risk assessment models had areas under the curve <0.60 except for the Caprini model (0.64, 95% confidence interval 0.60, 0.68). The Caprini, Padua and Intermountain models had high sensitivity (>85%) but low specificity (<20%) for predicting venous thromboembolism, whereas ICU-VTE, Kucher, and IMPROVE models had low sensitivities (<15%), but high specificities (>85%). The positive predictive value was low (<20%) for all studied cutoff scores, whereas the negative predictive value was mostly >90%. Using the risk assessment models to stratify patients into high- versus low-risk subgroups, the effect of adjunctive pneumatic compression versus pharmacologic prophylaxis alone was not different across the subgroups (p for interaction >0.05). Interpretation: The risk assessment models for venous thromboembolism performed poorly in critically ill patients receiving pharmacologic thromboprophylaxis. None of the models identified a subgroup of patients who might benefit from adjunctive pneumatic compression.