Background
Patients with unprovoked venous thromboembolism (VTE) have a high recurrence risk and are candidates for extended anticoagulation. However, many patients stay recurrence free and are unnecessarily exposed to anticoagulants. The Vienna Prediction Model has been developed to discriminate patients with unprovoked VTE with a low recurrence risk from those with a high risk based on the patient’s sex, the location of VTE, and D-Dimer, but allows risk assessment only at one single time point (3 weeks after anticoagulation).
Aim
To update and expand the model based on a larger number of events and a longer observation time, in order to assess the recurrence risk also from time points later than three weeks after anticoagulation on.
Methods
We analysed the data set of the Austrian Study on Recurrent Venous Thromboembolism, a prospective cohort study in patients of legal age with a first VTE who had received anticoagulants for 3 to 18 months. Patients with VTE provoked by surgery, trauma, pregnancy, or female hormone intake; with a natural inhibitor deficiency, the lupus anticoagulant, or cancer were excluded. The study end point was recurrent symptomatic deep vein thrombosis (DVT) and/or pulmonary embolism (PE). We integrated D-Dimer levels measured at several time points after anticoagulation with the patient’s sex and location of VTE. We generated nomograms to calculate individual risk scores and cumulative recurrence rates from 3 weeks, 3, 9, and 15 months on after discontinuation of anticoagulation using a dynamic landmark competing risks regression approach. The ethics committee approved the study and all patients gave written informed consent.
Results
134 of 553 patients had recurrence during a mean follow-up of 6 years. D-Dimer levels varied between patients, but did not substantially - albeit statistical significantly (p<0.001) - increase over time. The updated model has two improvements: we accounted for the competing risk of death or informative drop out by competing risks regression, and we considered various time points of prediction rather than predicting just once after anticoagulation. Subdistribution hazard ratios (95% CI) dynamically changed from 3 weeks to 3, 9, and 15 months from 2.4 (1.6-3.8), 2.3 (1.5-3.5), 2.0 (1.3-3.0) to 1.7 (1.1-2.7) in men vs. women, from 1.8 (1.0-3.4), 1.7 (0.9-3.1), 1.5 (0.8-2.8) to 1.4 (0.8-2.7) in patients with proximal DVT or PE compared to distal DVT, and from 1.3 (1.1-1.6), 1.3 (1.1-1.5), 1.2 (1.0-1.4) to 1.1 (0.9-1.4) per doubling D-Dimer levels. We created nomograms based on subdistribution hazard ratios from the multivariable dynamic model to predict the recurrence risk from 3 weeks, 3, 9 or 15 months on after anticoagulation. A web-based calculator allows risk assessment from random time points on between 3 weeks and 15 months.
Conclusions
The updated Vienna Prediction Model integrates patient’s sex, location of first VTE and serial D-Dimer measurements and allows prediction of recurrent VTE at a random time point after discontinuation of oral anticoagulation.
Disclosures:
No relevant conflicts of interest to declare.
Quantifying time-varying cause-specific hazard and subdistribution hazard ratios with competing risks data
