Performance and Head-to-Head Comparison of Three Clinical Models to Predict Occurrence of Postthrombotic Syndrome: A Validation Study

Abstract

Objective The SOX-PTS, Amin, and Méan models are three different clinical prediction scores stratifying the risk for postthrombotic syndrome (PTS) development in patients with acute deep vein thrombosis (DVT) of the lower limbs. Herein, we aimed to assess and compare these scores in the same cohort of patients. Methods We retrospectively applied the three scores in a cohort of 181 patients (196 limbs) who participated in the SAVER pilot trial for an acute DVT. Patients were stratified into PTS risk groups using positivity thresholds for high-risk patients as proposed in the derivation studies. All patients were assessed for PTS 6 months after index DVT using the Villalta scale. We calculated the predictive accuracy for PTS and area under receiver operating characteristic (AUROC) curve for each model. Results The Méan model was the most sensitive (sensitivity 87.7%; 95% confidence interval [CI]: 77.2–94.5) with the highest negative predictive value (87.5%; 95% CI: 76.8–94.4) for PTS. The SOX-PTS was the most specific score (specificity 97.5%; 95% CI: 92.7–99.5) with the highest positive predictive value (72.7%; 95% CI: 39.0–94.0). The SOX-PTS and Méan models performed well for PTS prediction (AUROC: 0.72; 95% CI: 0.65–0.80 and 0.74; 95% CI: 0.67–0.82), whereas the Amin model did not (AUROC: 0.58; 95% CI: 0.49–0.67). Conclusion Our data support that the SOX-PTS and Méan models have good accuracy to stratify the risk for PTS.