Enhanced prediction of thrombotic events in hospitalized COVID-19 patients with soluble thrombomodulin

Abstract

We aimed to determine the predictive accuracy of elevated soluble thrombomodulin (sTM) and angiopoietin-2 (Ang2) for thrombotic events (TE) in hospitalized COVID-19 patients. We conducted a nested case-control study within a cohort of people admitted to hospital with COVID-19 from March 2020 to August 2022. The cases (people with TE within 28 days after hospital admission) were matched by propensity score to comparable patients without TE. We determined plasma levels of sTM and Ang2 in all available frozen samples, prioritizing the earliest post-admission samples, using an automated immunoassay technique. Among 2,524 hospitalized COVID-19 patients (43% females; median age 67 years), 73 had TE (incidence 1.15 events per 1000 patient-days of follow-up). Frozen plasma samples were available for 43 cases and 176 controls. Elevated plasma concentration of sTM was significantly associated with TE (2.8 [1.8, 4] vs. 1.52 [1.1, 2.65] ng/mL; p = 0.001) and mortality (median [Q1, Q3], 3.32 [2.16, 4.65] vs. 1.58 [1.11, 2.73] ng/mL; p = 0.001), while D-dimer showed a specific association with TE (2.3 [0.8, 7.4] vs. 0.75 [0.4, 1.6] mcg/mL; p = 0.001). In contrast, Ang2 was not associated with any of these events. The association with thrombotic events remained in adjusted models (HR [95%CI] per unit increase, 1.24 [1.04-1.47] for sTM; 1.07 [1.03-1.10] for D-dimer). The adjusted regression model that included both biomarkers, sTM and D-dimer, improved (AUC 73%, sensitivity 77% and specificity 65% for TE diagnosis; p = 0.007) the predictive capacity of the same model without sTM. In conclusion, determination of soluble thrombomodulin along with D-dimer enhances thrombotic risk assessment in hospitalized COVID-19 patients.

Fig 1. Boxplots illustrating soluble thrombomodulin (Panel a) and D-dimer (Panel b) plasma concentration in relation to 28-day mortality (right) and thromboembolicevents (left).

All patients were accounted for within the 28-day follow-up period, with no losses to follow-up before the censoring date.

Fig 2. Multivariate Cox proportional hazards regression analysis of baseline predictors for 28-day thrombotic events in the propensity score cohort.

HR, Hazard Ratio; CI, confidence interval; RT-PCR, reverse transcriptase polymerase chain reaction. The figure presents adjusted HRs with 95% CIs for baseline predictors included in the multivariate model: soluble thrombomodulin (sTM), D-dimer, interleukin-6, and RT-PCR cycle threshold. The cohort was also matched by propensity score for age, sex, Charlson comorbidity index, and WHO COVID-19 severity score.

Fig 3. Comparison of ROC curves for the prediction of thrombotic events using D-dimer, soluble trombomodulin (sTM), and their combination.

Evaluation of the predictive performance of D-dimer alone, sTM alone, and the combination of D-dimer with sTM in hospitalized COVID-19 patients. ROC, receiver operating characteristic; AUC, area under the ROC curve; s-TM, soluble thrombomodulin. In the legend, the first number represents sensitivity (%) and the second specificity (%). The cutoff value for the regression models in all three cases was 0.2. The p-value for the comparisons between the combined model (D-dimer +  sTM) and the models with D-dimer alone or sTM alone are 0.007 and 0.001, respectively.