Table 5.

Characteristics of retrospective studies on venous thromboembolism prediction models

Study	Country	Study type, time period	Total number of cases	Venous thromboembolism incidence rate	Prediction model	Performance
Kampouri et al.	Switzerland	Retrospective, February 28th to April 30th, 2020	491	9.3%	Wells score for PE ≥ 2 points and D-dimer value ≥ 3,000 ng/mL	PPV: 18.2% NPV: 98.5 Accuracy: 0.905
Dujardin et al.	Netherlands	Retrospective, March 13th to April 9th, 2020	127	41.7%	Binary linear regression model; D-dimer is > 9 μg/mL and C-reactive protein > 280 mg/mL	Predicted probability: 92%
Tsaplin et al.	Russia	Retrospective, April 30th to May 29th, 2020	168	6.5%	Modified Caprini score > 12; D-dimer > 3 upper limit of normal	Sensitivity: 73%; Specificity: 84%
Spyropoulos et al.	United States	Retrospective, March 1st, 2020 to April 27th, 2020	9407	2.9%	The International Medical Prevention Registry on Venous Thromboembolism and D‐Dimer (IMPROVE‐DD) risk assessment model	AUC: 70%; sensitivity: 97%; specificity: 22%
Freund et al.	France, Spain, Belgium, Italy, Chile, and Canada	Retrospective, February 1st to April 10th, 2020	974	15%	Revised Geneva score and D-dimer [D-dimer below the age-adjusted threshold (i.e., 500 µg/mL under 50 years and age × 10 over 50 years)]	AUC: 0.81

AUC area under the curve, NPV negative predictive value, PE pulmonary embolism, PPV positive predictive value