Integrating D-Dimer Thresholds into the Revised Caprini Risk Stratification to Predict Deep Vein Thrombosis Risk in Preoperative Knee Osteoarthritis Patients

Introduction: Preoperative patients with knee osteoarthritis have a significantly increased risk of venous thromboembolism (VTE). While the Caprini risk assessment model offers some clinical guidance in predicting deep vein thrombosis (DVT), it has a relatively low predictive accuracy. Enhancing the model by integrating biomarkers, such as D-dimers, can potentially improve its accuracy. In this study, we explored the effectiveness of combining the Caprini risk model with D-dimer levels for individualized DVT risk assessment in patients with knee osteoarthritis.

Materials And Methods: This retrospective cohort study included 1605 knee osteoarthritis patients scheduled for total knee arthroplasty from Peking Union Medical College Hospital, screened between January 2015 and December 2018. A revised Caprini risk stratification model was developed, and a predictive DVT model was developed based on this revised system. The sensitivity, specificity, and the area under the curve (AUC) were used to determine predictive effectiveness of the model.

Results: In the revised Caprini risk stratification, the incidence of DVT increased with higher risk levels: 2.52% in the low-risk group (scores 0-2), 2.88% in the moderate-risk group (score 3), 6.47% in the high-risk group (score 4), and 9.09% in the highest-risk group (score ≥ 5). The incidence of DVT was 3.869-fold higher in the highest-risk group and 2.676-fold higher in the high-risk group compared to the low-risk group (p = 0.013 and p = 0.014, respectively). Combining the revised Caprini risk stratification with D-dimer level demonstrated an improved AUC of 0.792, compared to D-dimer level alone (AUC 0.774) and the revised Caprini model alone (AUC 0.598). Furthermore, applying specific D-dimer thresholds across the four Caprini risk stratifications outperformed the combination of the revised Caprini model and D-dimer level in terms of AUC, specificity, and reduction in unnecessary ultrasonography. Using the Youden index, the AUC for the threshold-based method was slightly higher (0.775 vs 0.754, p = 0.310), with significantly better specificity (76.8% vs 63.6%, p < 0.001) and a greater reduction in ultrasound use (74.1% vs 61.4%). At a sensitivity of 85.5%, the differences were modest but still favored the threshold-based approach. At a sensitivity of 100%, the specificity (36.0% vs 24.7%, p < 0.001) and ultrasound reduction (34.8% vs 23.9%) were significantly better.

Conclusion: The revised Caprini risk stratification improves preoperative DVT prediction in patients with knee osteoarthritis. Incorporating specific D-dimer thresholds into the four-level Caprini risk model enhances specificity and reduces unnecessary ultrasonography, outperforming both the use of individual indicators and the combination of the revised Caprini model with D-dimer level.