Demographics and baseline qSOFA and SIRS scores in survivals and nonsurvivals
Of 370 suspected SFTS patients who were admitted to our hospital from May 2013 to July 2017, 321 were confirmed by detected positive SFTSV and were included in this study. Mean age of all included patients was 63.8±11.2 years. Eighty seven patients (27.1%) died during hospitalization who were older than those who survived (70.6±9.2 vs 61.3±10.8 years, p<0.0001). Percentage of male patients in nonsurvivors was greater than in survivors (p<0.05). Proportions of patients with qSOFA and SIRS score >2 were dramatically increased in nonsurvivors compared with survivors (p<0.0001). Hospital stays and biochemical parameters had significant differences between the two groups. Data are shown in Table 1 and Figure 1.
Predictive values of independent risk factors, baseline qSOFA and SIRS scores and established risk score models for in-hospital mortality of SFTS patients
Predictive values of independent risk factors and risk models for all patients
Multivariate logistic regression analysis showed that age, AST, qSOFA and SIRS scores were the independent risk factors for in-hospital mortality of all SFTS patients. AUCs (95% CI) , cutoff values, SEN, SPE, PPV and NPV of these factors for the prediction of in-hospital mortality are included in Table 2. Based on values and regression coefficient of these risk factors, a risk score model was constructed as M1=0.102×age+0.002×AST+1.296×qSOFA score+0.486×SIRS score. AUC (95% CI) of M1 was 0.919 (0.883-0.946) with odd ratio (OR) (95% CI)=2.95 (2.308-3.771) at the cutoff value of 9.22 (Table2, Figure 2A). Kaplan-Meier survival analysis showed a strong difference between high-risk and low-risk groups (log-rank test, c2 = 1551.1, p<0.0001) (Figure 3A).
Considering relative small regression coefficient of AST and SIRS scores, we modified the models into simpler ones as:
M2 (model 2) = 0.102×age+1.296×qSOFA+0.486×SIRS
M3 (model 3) = 0.102×age+0.002×AST+1.296×qSOFA
M4 (model 4) = 0.102×age+1.296×qSOFA
These three models had comparable predictive power and was a relative less power than M1 ( Table 2, Figure 2).
Survival analysis demonstrated a strong statistical significance between high and low-risk groups of M1 and M4 (selection based on the highest AUC and OR values) (p<0.0001) (Figure 3A, B).
Predictive values of independent risk factors and risk models for patients with age ≥60 years
The four models also have high predictive values for in-hospital mortality in those patients at age ≥60 years (Table 3). Logistic regression analysis indicated that age, plasma lactate, serum AST and hs-CRP levels, and qSOFA and SIRS scores were independent risk factors for in-hospital mortality in this group of patients . Based on these factors, we established another risk score model:
This model has the highest predictive value among these models for in-hospital mortality in this group of patients (Table 3, Figure 2B).
Survival analysis manifested a profound statistical significance between high and low-risk groups of M4 and M5 (selection based on the highest AUC and OR values) p<0.0001) (Figure 3C, D).
Predictive values of independent risk factors and risk models for patients in ICU
The former four models also have high predictive values for in-hospital mortality in those patients enrolled in ICU at admission or transferred to (Table 4, Figure 2C).
Univariate logistic regression showed that age and qSOFA score were the independent risk factors for mortality of this subgroup of patients. Multivariate logistic regression analysis demonstrated that qSOFA score was the only independent risk factor for mortality of these patients. Based on these two parameters we built another model that has the same index of M4 with different regression coefficient:
M6 (model 6)= 0.071×age+1.877×qSOFA
This model has the highest predictive value among these models for in-hospital mortality in this subgroup of patients (Table 4, Figure 2C).
Survival analysis showed a significant statistical difference between high and low-risk groups of M4 and M6 (selection based on the highest AUC and OR values) p<0.0001) (Figure 3E, F).
Of note, no multi-colinearity between SOFA and SIRS were detected in these three models.