Development of dynamic prediction models for venous thromboembolism recurrence and bleeding risks to guide personalized anticoagulant therapy.

BACKGROUND

Oral anticoagulants are commonly used for venous thromboembolism (VTE), but the optimal treatment duration remains uncertain, highlighting the need for personalized strategies based on individual risk.

OBJECTIVES

We developed predictive models to assess VTE recurrence and major bleeding risks during and after anticoagulant treatment, incorporating time-varying data.

METHODS

This study included 14,653 patients with VTE who received oral anticoagulants between July 2018 and December 2020 using a claims database. Two cohorts were constructed: an on-treatment cohort, segmented into 90-day intervals for dynamic updates, and an off-treatment cohort. VTE and major bleeding were identified using diagnosis codes, hospitalization records, and procedure data. Three predictive models were developed: models 1 and 2 predicted 90-day risks of VTE recurrence and major bleeding during treatment, and model 3 predicted VTE recurrence after discontinuation. Time-varying effects were identified through interaction-term testing. Models were developed using least absolute shrinkage and selection operator regression, with optimism-corrected areas under the receiver operating characteristic curves (AUROCs) assessed via bootstrapping.

RESULTS

The incidence proportions were 2.7% for on-treatment VTE recurrence, 1.3% for major bleeding, and 0.8% for VTE recurrence within 90 days after discontinuation. The final models showed optimism-corrected AUROCs of 0.715, 0.828, and 0.679 for models 1, 2, and 3, respectively. Calibration was suboptimal for models 2 and 3.

CONCLUSION

These models may provide supportive information for individualized treatment decisions to dynamically assess VTE recurrence and bleeding risks, but their clinical value must be confirmed through external validation and recalibrated in larger cohorts before clinical use.