Sepsis-associated DIC is a serious life-threatening complication with a reported mortality rate that ranges from 30–60% , due to inflammation, coagulopathy, that cause tissue injury and multiorgan failure [13–14]. Anticoagulant therapy in septic DIC patients may have important effects in inhibiting thromboinflammation to improve potential outcomes [15–20]. In septic patients who develop DIC, a timely and accurate evaluation of the severity and prediction of outcomes is important for clinicians.
In the current study, we evaluated the predictive performance for 28-day mortality using a new scoring system, the SOFAComb, calculated by SOFAΔ + absolute SOFA score, and validated its potential applicability using a sepsis-associated DIC patient database. We previously reported the superiority of SOFAΔ compared to DIC scores , and our goal was to improve the performance of SOFAΔ in the present study. The performance of SOFAΔ has been repeatedly validated for predicting the prognosis in sepsis patients, but the results are inconsistent.
Minne et. al. reported the predictive performance of SOFAΔ for the mortality, but the AUCs varies widely from 0.510 to 0.828 in their systemic review . This inconsistency may be due to the consideration that SOFAΔ represents only the changes of the SOFA score and ignores the absolute SOFA score. Indeed, among the patients with the same SOFAΔ, the mortality should be different depending on the absolute SOFA score. For instance, the significance of a -2 point in SOFAΔ should be different between from two to 0 and from 24 to 22 point. However, de Grooth et. al.  validated the treatment effects, and mortality evaluated by SOFAΔ and reported SOFAΔ reflected the efficacy more accurately than the absolute SOFA score on a fixed day after randomization. They also reported the association between SOFAΔ and mortality did not change even after the adjustment by SOFA score on admission. Karakike et al.  reported SOFA score changes evaluated by a percentage of the initial score on day 7 or later was a better predictor of mortality, and the 25% decrease of initial SOFA was the best cut-off value. In our study, SOFAΔ on day 7 did not exhibit a better predictive value over the SOFA score on day 7, but the SOFAComb on day 2 and later had a better predictivity than baseline SOFA, and the performance increased over time. Based on the timing of evaluation, SOFAΔ cannot detect the status change or the treatment effect at an early timing since there should be a time lag until the SOFA score improves. The absolute SOFA score included in SOFAComb may help to reduce this drawback. In fact, SOFAComb demonstrated a better performance than SOFAΔ at early timing in the present study since SOFAComb reflects both the time-trend of disease status and real-time severity. The early detection of the status change or treatment effect is particularly helpful for clinicians to reconsider their therapeutic strategy.
For designing SOFAComb, we did not use a multivariable logistic regression model because its predictive value did not improve the performance, and the day 7 AUC was 0.866, and identical to the value calculated by our proposed method. As a result, we selected a simple method using absolute SOFA score plus SOFAΔ to calculate SOFAComb.
A logistic regression curve revealed the relationship between the scores and the estimated mortality as shown in Fig. 3. Both curves showed similar sigmoid shape, with standard error, implying that the performance of both SOFAΔ and SOFAComb was able to predict 28-day mortality over the entire range of the scores. However, the R2 calculated from the logistic regression curve analysis for SOFAComb on day 7 was 44.0% and higher than that of SOFAΔ (32.0%), which are consistent with the superior performance of SOFAComb shown by the higher accuracy, sensitivity, and specificity.
The present study has some limitations. First, the performance of SOFAComb was developed using the data from a post-marketing survey, and the timing of the evaluation was pre-specified on day 1 (before the treatment), 2, 4 (after the treatment), and 7. In addition, all patients received antithrombin supplementation. It is uncertain whether the results obtained in this study are generalizable and need additional validation. Second, in the post-marketing database used, missing data was present, and patients were excluded from the analysis. Because this data is a retrospective analysis, the results need to be confirmed in a prospective study.