A composite risk assessment model for venous thromboembolism.

Area of Science:

• Medical Informatics
• Clinical Risk Management
• Epidemiology

Background:

• Venous thromboembolism (VTE) poses a significant risk for hospitalization-related morbidity and mortality.
• Accurate VTE risk stratification is crucial for prevention, with federal mandates for universal risk assessment.
• Existing models like Caprini and Padua show limited predictive accuracy for general hospital admissions.

Purpose of the Study:

• To evaluate if combining risk factors from 23 available VTE risk-assessment models (RAMs) enhances VTE risk prediction.
• To assess the predictive performance of a novel composite RAM incorporating numerous risk factors.

Main Methods:

• Analysis of over 1.18 million Veterans Affairs hospital admissions (2016-2021).
• Logistic regression used to predict VTE within 90 days, utilizing risk factors from 23 VTE RAMs.
• Model performance evaluated using AUC, sensitivity, specificity, PPV, and NPV at optimized thresholds.

Main Results:

• A composite RAM with 102 risk factors demonstrated improved VTE prediction (AUC 0.74) compared to Caprini factors (AUC 0.63).
• The composite model showed modest gains in sensitivity and specificity, with high NPV but low PPV.
• Maximizing PPV in the composite model yielded only marginal improvements, limited by the trade-off with NPV and sensitivity.

Conclusions:

• A composite model integrating 102 risk factors significantly improved VTE prediction in a large national cohort.
• Despite improvements, neither the composite nor existing models possess sufficient sensitivity or PPV for reliable VTE prediction.
• Current VTE risk-assessment tools are insufficient for widespread implementation in general hospital populations.