Abstract
Background:
Predicting the risk of recurrent venous thromboembolism (rVTE) during
and after anticoagulation is complex. Current models do not adapt to changing patient
conditions, fail to account for the direct impact of anticoagulation in real-world
settings, and are often derived without concurrent data on major bleeding.
Objectives:
To develop a dynamic, broad, treatment-adjusted risk score for rVTE.
Methods:
We analyzed data from a retrospective cohort of patients from the UK
Clinical Practice Research Datalink (2001-2020) with a first venous thromboembolism
(VTE) who received 3 months of anticoagulation. rVTE risk was modeled using baseline
and time-varying covariates (post-VTE). Hazard ratios with 95% CIs were estimated; a
Fine–Gray model identified predictors. A scoring scheme was developed from
subdistribution hazard ratios, with discrimination (C-statistic) estimated via crossvalidation.
Results:
A total of 51 465 patients with VTE were included; 4041 cases of rVTE
occurred over 200 698 person-years. rVTE incidence was 2.01 per 100 person-years.
The model included 19 predictors, of which 13 were associated with increased rVTE
risk and 6 with decreased rVTE risk. Risk correlated well with outcomes; calibration
showed good agreement between predicted and observed risks (calibration slope, 0.99
[95% CI, 0.94-1.04]). rVTE risk decreased over 10 years, but within each year, higher
scores were associated with higher risk (0.92%-21.23% in the first year [–5 to +8
points]). The C-statistic (0.65 [95% CI, 0.64-0.66]) was stable over 10 years.
Conclusion:
Our dynamic risk score effectively identifies the risk of rVTE after 3
months of treatment and supports ongoing management over 10 years, aiding clinical
decision-making regarding treatment duration and strategy adjustments as patient
characteristics and comorbidities evolve.