The Caprini score for venous thromboembolism risk assessment: A scoping review of applications, validation, and...

The aim of this study was to characterize the contemporary academic landscape surrounding the Caprini venous thromboembolism (VTE) risk assessment model (RAM), including its application patterns, reported performance, evolving scholarly perspectives, and dominant refinement strategies.

A scoping review was conducted following the Preferred Reporting Items for Systematic reviews and Meta-Analyses extension for Scoping Reviews framework. PubMed, Web of Science, Embase, and the Cochrane Library were systematically searched for studies published between January 1, 2021, and December 31, 2025, using terms including "Caprini score," "venous thromboembolism," and "risk assessment." Eligible studies reported on the validation, application, or modification of the Caprini RAM in adult patients. Data on study characteristics, predictive performance metrics (eg, area under the curve [AUC]), author conclusions, and model refinement details were extracted.

From 754 identified records, 275 studies met the inclusion criteria. Analysis of 275 studies with codable author perspectives revealed a distribution of supportive (113 studies; 41.1%), neutral (96 studies; 34.9%), and critical (66 studies; 24.0%) stances. Fifty-two studies providing paired AUC data reported a mean baseline Caprini score AUC of 0.702 (standard deviation [SD], 0.104), compared with a mean AUC of 0.832 (SD, 0.080) for refined or new models. The mean AUC improvement was 0.130 (SD, 0.117), with a median improvement of 0.105 (interquartile range, 0.106); improvements were observed in 98.1% of these comparisons. Critical perspectives frequently cited poor accuracy in specific populations (eg, medical inpatients) and operational complexity. In response, 111 studies proposed refinements categorized into four domains: integration of biomarkers (33 studies), development of specialty-specific/simplified models (45 studies), application of artificial intelligence/machine learning (23 studies), and optimization of assessment processes (30 studies).

This review quantifies a fragmented scholarly discourse on the Caprini RAM, reflecting its validated utility in certain contexts alongside recognized limitations driving extensive model refinement. The significant performance gains reported for refined models, particularly in specialties where the original score underperforms, highlight the ongoing evolution of VTE risk assessment toward more precise and context-adapted tools.