Estimated Impact of Model-Guided Venous Thromboembolism Prophylaxis versus Physician Practice

Abstract

Background: The American Society of Hematology (ASH) recommends assessing venous thromboembolism (VTE) and major bleeding risk to optimize pharmacological VTE prophylaxis for medical inpatients. However, the clinical utility of model-guided approaches remains unknown.

Methods: Our objective was to estimate differences in VTE and major bleeding event rates and efficiency with prophylaxis guided by risk models versus prophylaxis based on physician judgment. Patients were adults admitted to one of 10 Cleveland Clinic hospitals between December 2017 and January 2020. We compared physician practice with hypothetical prophylaxis recommended by model-based prophylaxis strategies, including ASH-recommended risk scores (Padua and IMPROVE) and locally derived Cleveland Clinic risk prediction models. For each strategy we quantified the prophylaxis rate, VTE and major bleeding rates, and the incremental number-needed-to-treat (NNT) to prevent one event (VTE or bleeding).

Results: Physicians prescribed prophylaxis to 62% of patients whereas model-based strategies recommended prophylaxis for 17-87%. Model-guided prophylaxis produced more VTEs and fewer major bleeds than physicians, but total events varied among strategies. Overall, per 1,000 patients, model-based strategies produced 14.0-16.1 events compared with 14.3 for physicians. The Padua/IMPROVE models recommended prophylaxis for the fewest patients but caused the most total events. The most efficient model-based strategy recommended prophylaxis to 28% of patients with an incremental NNT (relative to no prophylaxis) of 80. Compared to physicians, it reduced prophylaxis by 55% and total events by 0.14%.

Conclusions: Physicians often prescribed inappropriate prophylaxis, highlighting the need for decision support. A model-based strategy maximized efficiency, reducing both events and prophylaxis relative to physicians.

Keywords: clinical decision support systems; clinical utility; hemorrhage; heparin; risk assessment; venous thromboembolism.

Figure 1.

Observed prophylaxis rates stratified by VTE risk (left) and major bleeding risk (right). Risk probabilities were calculated using the Cleveland Clinic VTE model (CCVM) and Cleveland Clinic Bleeding model (CCBM). The value beneath each bar is the number of patients within that risk group. The percentage at the top of each bar is the observed rate of events within that risk group, demonstrating good concordance between actual and predicted risk for both models. Event rates are not adjusted for prophylaxis, which reduces risk of VTE and increases risk of major bleeding. As a result, predicted rates of VTE appear somewhat overestimated and risks of bleeding appear underestimated. The vertical red lines indicate the high-risk thresholds for the CCVM (1.0%) and CCBM (0.78%).

Figure 2.

Prophylaxis and event rates for physicians (blue) versus four model-guided prophylaxis strategies: Cleveland Clinic (CC) models with prophylaxis for all patients regardless of VTE risk, except those with high major bleeding risk (pink); CC models with VTE and major bleeding equally balanced for high-risk patients (green); baseline CC models, with prophylaxis for high VTE risk and low major bleeding risk patients (purple); and guideline-recommended models, Padua and IMPROVE (orange). Bar heights are lower-bound event estimates based on high efficacy and high harm of prophylaxis; error bars are upper-bound estimates based on low efficacy and low harm.

Figure 3

Efficiency curve showing prophylaxis rates versus total event rates for each prophylaxis strategy. Event rates shown are lower bound estimates from sensitivity analyses. Incremental NNTs (boxed values) are shown only for strategies along the efficiency frontier, i.e., the set of points with dominant efficiency.