A study of two risk assessment models (RAMs) for predicting the bleeding risk in patients considered for pharmacologic prophylaxis to prevent venous thromboembolism (VTE) has found that though an increasing risk score correlated with higher bleeding rates, both models had a low predictive ability for major bleeding post-admission.
This was the conclusion of Hilary Hayssen (University of Maryland School of Medicine, Baltimore, USA), who presented the findings of the analysis of more than 1.2 million patients undergoing both surgical and non-surgical interventions at this year’s Society for Vascular Surgery (SVS) Vascular Annual Meeting (VAM 2023; 14–17 June, National Harbor, USA).
Half of all VTE events are associated with hospitalisation, Hayssen told delegates, explaining that hospital-associated pulmonary embolism (PE) is a leading preventable cause of death. Pharmacologic prophylaxis reduces the incidence of PE but can cause bleeding, meaning that there is a need to balance the risk of VTE against the risk of bleeding when considering this avenue of treatment.
There are two current risk assessment models to evaluate bleeding risk in those being considered for pharmacologic prophylaxis, the IMPROVE and Consensus models. Hayssen and colleagues assessed the predictive ability of each model for bleeding within 30, 60 and 90 days post-admission, comparing the performance of the two models in patients admitted at all 1,298 Veterans Health Administration (VHA) facilities nationwide between January 2016 and December 2021.
In total, data were analysed from 1,228,448 patients, 26.5% (n=324,959) of whom underwent surgical procedures, and 73.5% (n=903,489) non-surgical interventions. To review the two scores, researchers calculated the IMPROVE and Consensus scores using medical record data, which enabled them to assess the predictive ability of the models for bleeding at 90 days in both surgical and non-surgical patients.
A total of 5.6% of patients had major bleeding, as defined by the International Society on Thrombosis and Hemostasis (ISTH), within 90 days post-admission, occurring in 5% of the surgical patients, and 5.8% of non-surgical patients. A total of 68,372 bleeding events occurred within 90 days of admission, and 29% of events occurred between 31 and 90 days, Hayssen reported.
In terms of the performance of the two models, Hayssen reported that higher scores were associated with higher bleeding rates. Results showed that the IMPROVE scores ranged from 0 to 22, while Consensus scores ranged from -5.60 to -1.21.
However, she reported that the ability of either RAM to predict 90-day bleeding, calculated by computing the areas under the respective receiver operating-characteristic curves (AUC) was “no better than a coin toss” (AUCs: IMPROVE 0.61, Consensus 0.59), a finding that was similarly low at 30 and 60 days post-admission. The predictive ability for either score was consistent across both surgical and non-surgical patients.
“In this validation study evaluating these two bleeding RAMs for patients being considered for pharmacoprophylaxis, we found that increasing scores were associated with increasing bleeding rates, but that the scores actually had low predictive ability for 90-day bleeding in a general hospitalised setting,” Hayssen offered in her concluding remarks.
“The bleeding RAMs possibly are not ready for general implementation in a hospital setting. More validation studies and more evaluation of the components of these models is needed to improve predictive ability to evaluate these models in conjunction with our more commonly used VTE risk assessment models.” In the discussion that followed the presentation, Hayssen was asked by session moderator Michael Dalsing (Indiana University School of Medicine, Indianapolis, USA) what factors could improve the two scores.
“It is a combination of adding risk factors that are not included,” she commented, noting that one of the risk models has 11 risk factors and the other has seven. It could be possible, she said, to include risk factors that are related but also possibly removing risk factors that are not related.
“When we looked at the prevalence of the factors in the bleeding group and the non-bleeding group, there were [factors] that were possibly, in isolation, considered more protective based upon their incidence in the two groups. So, while we haven’t evaluated which risk factors may be more important, that is definitely an aim of future research that we have.”