Development and validation of a clinical prediction model for post thrombotic syndrome following anticoagulant therapy for acute deep venous thrombosis.

OBJECTIVES

To identify independent prediction factors for post thrombotic syndrome (PTS) following acute deep vein thrombosis (DVT) and develop a clinical prediction model assessing the risk of PTS in individual patient.

METHODS

We prospectively recruited consecutive adult patients with acute DVT who were managed at Shanxi Bethune Hospital, China between June 2014 and December 2016. Investigator assessed PTS using the Villalta scale at 1, 6, 12, 18 and 24 months following diagnosis of DVT. Variable selection was performed by applying the least absolute shrinkage and selection operator (LASSO) with 10-fold cross-validation. Based on these data, we established a clinical prediction model for the development of PTS following DVT. The Bootstrap method was used for internal validation. During the process of model development, we re-collected the information of DVT patients from 2016 to 2017 for a temporal validation. The performance of the prediction model included discrimination and calibration, and clinical utility of prediction model was also evaluated using a decision curve analysis.

RESULTS

A total of 808 consecutive patients with acute DVT were enrolled in the training and validation datasets, of which 540 patients were included in the training dataset for the development of prediction model and the other 268 patients were in the other dataset for temporal validation. Seventy-six patients in training dataset developed PTS. The prediction factors associated with PTS were ilio-femoral DVT (OR = 4.835, 95% CI: 2.471-9.463), active cancer (OR = 3.006, 95% CI: 1.404-6.435), history of chronic venous insufficiency (OR = 7.464, 95% CI: 3.568-15.616), previous venous thromboembolism (OR = 6.326, 95% CI: 2.872-13.932), and chronic kidney disease (OR = 9.916, 95% CI: 2.238-43.937), duration of compression therapy <6 months (OR = 2.894, 95% CI: 1.595-5.251). The c index of the prediction model was 0.825 (0.774-0.877), and the c index of internal validation and temporal verification were 0.816 and 0.773 (95% CI: 0.699-0.848), indicated that the prediction model had a good discrimination in predicting PTS risk following DVT. All the calibration curve showed the model had a good calibration. The decision curve analysis showed a better net benefit of prediction model predicting PTS risk within threshold probability ranged from 0% to 72% and 86% to 98% in training dataset, and 0% to 58% in the validation datasets.

CONCLUSION

Our prediction model can accurately estimate the likelihood of PTS risk and identify high-risk patients who may develop PTS following DVT based on individual characteristics, but further external validation is still required.