Abstract 6207: Derivation of a Novel Prediction Algorithm for Idiopathic Venous Thromboembolism in Women
Background: Established risk factors for idiopathic venous thromboembolism (VTE) include age, prothrombotic mutations, and obesity. A risk prediction algorithm might allow clinicians to focus preventive efforts.
Methods: We derived a prediction model for idiopathic VTE using 22 possible predictor variables and their interactions in 19,458 initially healthy Caucasian women who were followed for incident idiopathic VTE. Minimization of the Bayes Information Criterion (BIC) was used to help identify the best, most parsimonious model, and estimates of discrimination (C-index) and calibration (Hosmer-Lemeshow (H-L) statistic) comparing observed and predicted risk were computed and compared with a model including only age and weight.
Results: The mean (SD) age was 54.1 (7.1) years, the mean (SD) weight 69.9 (14.1) kilograms, and 2.8% and 5.3% of participants carried the prothrombin and factor V Leiden mutations, respectively. After a median of 12.4 years of follow up, we observed 130 idiopathic cases of VTE. The best fitting model included log(age), log(weight), log(apolipoprotein A1), and the presence of the prothrombin or factor V Leiden mutations (C-index 0.704; H-L statistic 5.9). When compared to model including only log(age) and log(weight) (C-index 0.636; H-L statistic 3.9), the full model achieved higher discrimination at similar calibration and led to a net reclassification of 34.5 percent (6705) of participants, all correctly.
Conculsion: In our population of middle-aged, initially healthy Caucasian women, a model that includes age, weight, apolipoprotein A1, and the prothrombin and factor V Leiden mutations predicts incident idiopathic VTE more accurately than one including age and weight alone.