Risk stratification utilizing sequential organ failure assessment (SOFA) score, antithrombin activity, and demographic data in sepsis-associated disseminated intravascular coagulation (DIC)

Figure 1. Patient selection. The data from the patients with sepsis-associated DIC with antithrombin activity of 70% or less and treated with antithrombin concentrate were utilized. Regarding antithrombin activity, sequential organ failure assessment (SOFA) score, and Japanese Association for Acute Medicine (JAAM) disseminated intravascuclar coagulation (DIC) score, when the data are absent, these data are categorized as 'missing data'.

Figure 2. Cox multivariate regression analysis for survival factors. Cox multivariate regression analysis identified male sex, age 75 years or older, baseline sequential organ failure score (SOFA) score ≥ 8, and the presence of pneumonia or skin/soft tissue infection as significant factors associated with worsened survival. On the other hand, baseline antithrombin activity > 40% was found to be associated with better survival.

Figure 3. Kaplan–Meier curves for the cumulative survival of the patients with sepsis-associated DIC. Patients were divided into high (≥ 12) and low (< 12) sequential organ failure assessment (SOFA) score groups, high (> 47%) and low (≤ 47%) antithrombin activity groups, presence or absence of pneumonia or soft tissue infection, and male sex and older than 75 years old or no. Significant differences were seen between the groups.

Figure 4. Receiver operating characteristic curves of the SOFA score and antithrombin activities for 28-day survival. Receiver operating characteristic (ROC) curves of the baseline sequential organ failure assessment (SOFA) score and baseline antithrombin activity for 28-day survival are depicted. The respective areas under the ROC curves (AUCs) were 0.700 for the SOFA score and 0.614 for the antithrombin activity.

Figure 5. Decision tree model for risk stratification. A decision tree model was used to categorize patients to either 28-day mortality of high (> 40%), medium (20–40%), and low (< 20%) range according to the sequential organ failure assessment (SOFA) score, baseline antithrombin activity, infectious site, sex, and age. Two categories (13.9% of the total cases) were not appropriately classified.

Figure 6. Evaluation of a predictive model using decision tree risk factors. A predictive model was constructed using logistic regression, and its performance was evaluated using Receiver operating characteristic (ROC) analysis. Four decision tree variables were used in the prediction model, and the objective variable was 28-day mortality. The predictive score was constructed as 3.350 * SOFA cutoff + 2.067 * antithrombin activity (AT) cutoff + 2.452 * infection site + 1.403 * male and ≥ 75 years old. The results of ROC analysis showed high predictive ability with an area under the curve (AUC) of 0.723.