More than 100,000 patients are hospitalized in the United States each day. Half of all venous thromboembolism (VTE) events occur in hospitalized patients. Hospital-related pulmonary embolism was declared the leading preventable cause of in-hospital death by the US Surgeon General, Centers for Disease Control and Prevention, and Centers for Medicare and Medicaid Services. Risk-stratification and targeted-prophylaxis are keys to prevention. Pharmacoprophylaxis is highly effective, though with an increased risk of bleeding. Therefore, decisions for VTE prophylaxis must balance the predicted risk of VTE against the predicted risk of bleeding in individual patients. While there are several risk assessment models (RAMs) to predict the risk of VTE, there are only two RAMs proposed to predict bleeding from pharmacoprophylaxis and neither has been validated in a large, unselected cohort of general hospitalized patients. We analyzed all consecutive unique hospital admissions of surgical and nonsurgical patients at all 1298 VA facilities nationwide between January 2016 and December 2021. IMPROVE and Consensus RAM scores were generated using data from a repository of the unified national VA electronic medical record. We assessed the predictive ability of the RAMs for bleeding within 30, 60, and 90 days after admission and compared their performance in surgical and nonsurgical patients by computing the areas under their respective receiver operating-characteristic curves (AUC). We analyzed 1,228,448 patients; 26.5% (n = 324,959) surgical and 73.5% (n = 903,489) nonsurgical. 68,372 (5.6%) patients had major bleeding (as defined by the International Society on Thrombosis and Hemostasis) within 90 days postadmission (surgical 5.0%, nonsurgical 5.8%). IMPROVE scores ranged from 0 to 22 (median, 3.5; interquartile range, 2.5-5.0); Consensus scores ranged from −5.60 to −1.21 (medina, −4.93; interquartile range, −5.60 to −4.93). Higher scores were associated with higher bleeding rates, however, the ability of either RAM to predict 90-day bleeding was no better than a coin toss (AUCs: IMPROVE, 0.61: Consensus, 0.59), and was similarly low at 30 and 60 days postadmission (Table). Both RAMs performed better in surgical vs nonsurgical patients (AUCs: IMPROVE, 0.64 vs 0.61; P < .001; Consensus, 0.63 vs 0.58; P < .001) (Figure). In this first known validation study evaluating existing RAMs for bleeding from pharmacoprophylaxis in an unselected cohort, increasing scores correlated with bleeding rates; though both RAMs had low predictive ability for major bleeding postadmission. Both RAMs predicted bleeding better in surgical vs nonsurgical patients, but the differences were of limited clinical importance. Further studies are urgently required to improve existing bleeding RAMs to facilitate systematic decision-making for VTE prophylaxis in hospitalized patients.TableArea under receiver operating characteristic curves (AUC) for the IMPROVE and Consensus Bleeding risk assessment models (RAMs) at different follow-up intervalsTime from hospital admission, daysImprove bleeding RAMConsensus bleeding RAMAUC (95% confidence intervals)P valueAUC (95% confidence intervals)P value0-300.62 (0.62-0.62)<.0010.61 (0.60-0.61)<.0010-600.62 (0.61-0.62).0720.60 (0.60-0.60).0120-900.61 (0.61-0.62)1.00(ref)0.59 (0.59-0.60)1.00(ref) Open table in a new tab
Read full abstract