Venous thromboembolism risk models may be ready for a makeover

Lal et al challenge the accuracy of the most common Risk Assessment Models (RAMs) for quantification of venous thromboembolic (VTE) risk and for clinical guidance in thromboprophylaxis. During a 5-year period (2016-2021), >1 million hospital admissions at Veterans Administration Medical Centers were reviewed to assess the ability of two RAMs to predict VTE within 90 days of admission.1Hayssen H. Sahoo S. Nguyen P. et al.Ability of Caprini and Padua risk-assessment models to predict venous thromboembolism in a Nationwide VA Study.J Vasc Surg Venous Lymphat Disord. 2024; 12101693Google Scholar The Caprini RAM, first published in 1991 and designed for general surgery patients, has been validated in >250,000 patients worldwide.2Caprini J.A. Arcelus J.I. Hasty J.H. Tamhane A.C. Fabrega F. Clinical assessment of venous thromboembolic risk in surgical patients.Semin Thromb Hemost. 1991; 17: 304-312Google Scholar,3Cronin M. Dengler N. Krauss E.S. et al.Completion of the updated caprini risk assessment model (2013 version).Clin Appl Thromb. 2019; 25: 1-10Google Scholar The Padua RAM, first published in 2010, was designed for acutely ill general medical patients.4Barbar S. Noventa F. Rossetto V. et al.A risk assessment model for the identification of hospitalized medical patients at risk for venous thromboembolism: the Padua Prediction Score.J Thromb Haemost. 2010; 8: 2450-2457Google Scholar Although both models achieved appropriate calibration (higher scores associated with higher risk), their predictive accuracy was questioned in this study. Prediction at alternate (30- and 60-day) time points, type of outcome (composite vs VTE alone), and type of prophylaxis status did not differ significantly between RAMs or improve predictive ability. The area under the curve for predicting VTE at 0 to 90 days of the Caprini and Padua models were 0.56 and 0.59. An area under the curve of <0.6 suggests a predictive capability slightly more accurate than a coin flip (0.7-0.8 acceptable, 0.8-0.9 excellent, >0.9 outstanding).5Mandrekar J.N. Receiver operating characteristic curve in diagnostic test assessment.J Thorac Oncol. 2010; 5: 1315-1316Google Scholar The Padua model was significantly (P < .001) better than the Caprini model, although the authors are keen to acknowledge that this was of no clinical importance, even when adjusting their criteria for statistical significance to a P value of <.01.6Cummings P. Rivara F.P. Reporting statistical information in medical journal articles.Arch Pediatr Adolesc Med. 2003; 157: 321-324Google Scholar Should the criterion be even more stringent? Computerized clinical decision support systems are associated with a significant increase in use of VTE prophylaxis (odds ratio, 2.35; 95% confidence interval, 1.78-3.1; P < .001) and, more important, a significant decrease in hospital acquired VTE events (0.33%-0.15%; P < .01).7Borab Z.M. Lanni M.A. Tecce M.G. Pannucci C.J. Fischer J.P. Use of computerized clinical decision support systems to prevent venous thromboembolism in surgical patients: a Systematic review and meta-analysis.JAMA Surg. 2017; 152: 638-645Google Scholar,8Titi M.A. Alotair H.A. Fayed A. et al.Effects of computerised clinical decision support on adherence to VTE prophylaxis clinical practice guidelines among hospitalised patients.Int J Qual Health Care. 2021; 33: mzab034Google Scholar The authors suggest that the two RAMs may be better suited for niche patient populations. Should all admitted patients be assessed using the same tools, or should surgical, medical, or critically ill patients each have their own RAMs? Clearly, a profusion of differing risk profiles will only silo the confusion. Can artificial intelligence (AI) build new models or improve upon current risk models developed before AI was widely available? The VA has used AI for other risk assessments for >10 years.9Atkins D. Makridis C.A. Alterovitz G. Ramoni R. Clancy C. Developing and implementing predictive models in a learning healthcare system: traditional and artificial intelligence approaches in the veterans Health administration.Annu Rev Biomed Data Sci. 2022; 5: 393-413Google Scholar The authors have suggested several modifications to the RAMs evaluated, which may improve their predictive capabilities. Despite the Veterans Health Administration demographic (mostly males), its large integrated network may still be an appropriate location to pilot an improved RAM before validating it in a more general population. Readers will note the repetitive disclaimers in their manuscript indicating the absence of a clinical difference when a statistical difference was reported from this large database. This raises a question for journal editors and readers interpreting the findings from large database analyses such as this. It may be time for a makeover of venous risk assessments and statistical reporting standards for database studies. The opinions or views expressed in this commentary are those of the authors and do not necessarily reflect the opinions or recommendations of the Journal of Vascular Surgery: Venous and Lymphatic Disorders or the Society for Vascular Surgery. None. Ability of Caprini and Padua risk-assessment models to predict venous thromboembolism in a nationwide Veterans Affairs studyJournal of Vascular Surgery: Venous and Lymphatic DisordersVol. 12Issue 2PreviewVenous thromboembolism (VTE) is a preventable complication of hospitalization. Risk-stratification is the cornerstone of prevention. The Caprini and Padua are two of the most commonly used risk-assessment models (RAMs) to quantify VTE risk. Both models perform well in select, high-risk cohorts. Although VTE RAMs were designed for use in all hospital admissions, they are mostly tested in select, high-risk cohorts. We aim to evaluate the two RAMs in a large, unselected cohort of patients. Full-Text PDF Open Access