Prediction of Symptomatic Venous Thromboembolism in Critically Ill Patients: The ICU-Venous Thromboembolism Score*

Abstract

Objectives: To identify risk factors and develop a prediction score for in-hospital symptomatic venous thromboembolism in critically ill patients. Design: Retrospective cohort study. Setting: Henry Ford Health System, a five-hospital system including 18 ICUs. Patients: We obtained data from the electronic medical record of all adult patients admitted to any ICU (total 264 beds) between January 2015 and March 2018. Interventions: None. Measurements and Main Results: Symptomatic venous thromboembolism was defined as deep vein thrombosis, pulmonary embolism, or both, diagnosed greater than 24 hours after ICU admission and confirmed by ultrasound, CT, or nuclear medicine imaging. A prediction score (the ICU-Venous Thromboembolism score) was derived from independent risk factors identified using multivariable logistic regression. Of 37,050 patients who met the eligibility criteria, 529 patients (1.4%) developed symptomatic venous thromboembolism. The ICU-Venous Thromboembolism score consists of six independent predictors: central venous catheterization (5 points), immobilization greater than or equal to 4 days (4 points), prior history of venous thromboembolism (4 points), mechanical ventilation (2 points), lowest hemoglobin during hospitalization greater than or equal to 9 g/dL (2 points), and platelet count at admission greater than 250,000/μL (1 point). Patients with a score of 0–8 (76% of the sample) had a low (0.3%) risk of venous thromboembolism; those with a score of 9–14 (22%) had an intermediate (3.6%) risk of venous thromboembolism (hazard ratio, 6.7; 95% CI, 5.3–8.4); and those with a score of 15–18 (2%) had a high (17.7%) risk of venous thromboembolism (hazard ratio, 28.1; 95% CI, 21.7–36.5). The overall C-statistic of the model was 0.87 (95% CI, 0.85–0.88). Conclusions: Clinically diagnosed symptomatic venous thromboembolism occurred in 1.4% of this large population of ICU patients with high adherence to chemoprophylaxis. Central venous catheterization and immobilization are potentially modifiable risk factors for venous thromboembolism. The ICU-Venous Thromboembolism score can identify patients at increased risk for venous thromboembolism.

Incidence, prognostic factors, and outcomes of venous thromboembolism in critically ill patients: data from two prospective cohort studies

“…Occurrence of PE-LDVT and not NLDVT was associated with increased 90-day mortality. This is in line with recent cohort studies in which critically ill patients who developed PE-LDVT (or a comparable composite) had higher mortality rates, while NLDVT was not associated with increased mortality [3,15,16]. Our study also replicates results from earlier studies that used surveillance ultrasound to detect VTE, confirming the value of previous VTE and renal failure as a prognostic factors for PE-LDVT in critically ill patients [1,13,17].…”

“…NLDVT occurred in 2.2% of the patients in the PROTECT randomized trial [3]. Several prognostic factors for the development of VTE have been identified in critically ill patients, including central venous catheters, higher illness severity scores, renal failure, a history of VTE, higher body mass index, and vasopressor therapy [1,13,16]. The latter three were also recognized to be associated with failure of thrombosis prophylaxis in a post hoc analysis of the PROTECT trial [17].…”

Genetic factors increase the identification efficiency of predictive models for dyslipidaemia: a prospective cohort study

“…Previously, validated disease prediction models for CVD and diabetes were developed based on the Framingham study [10,11]. Moreover, some researchers have also focused on effective risk models for other diseases to help diagnosis and prevention [12][13][14][15]. However, few studies involved the prediction model for dyslipidaemia [16][17][18][19], and most of them were limited to specific populations such as children and adolescents to a certain extent.…”

Genetic factors increase the identification efficiency of predictive models for dyslipidaemia: a prospective cohort study

“…Previously, validated disease prediction models for CVD and diabetes were developed based on the Framingham study [10,11]. Moreover, some researchers have also focused on effective risk models for other diseases to help diagnosis and prevention [12][13][14][15]. However, few studies involved the prediction model for dyslipidaemia [16][17][18][19], and most of them were limited to speci c populations such as children and adolescents to a certain extent.…”