Clinical characteristics of trauma cohort
Demographic data of 1000 trauma patients was summarized in Table S2. Patients were mostly young (mean age: 42.89±12.56 years) and severely injured (mean ISS: 19.59±8.99). Incidence of sepsis was 26.20% (n=262). Pneumonia and primary bloodstream infection was approximately 49.24% of all the documented infections. Gram-negative infections accounted for approximately 83.20%, gram-positive infections for 6.87%, and mixed gram-negative/gram-positive infections for 2.29% of sepsis patients. The median time for sepsis occurrence in the whole trauma cohort was 7.02±6.95 days. The maximum of SOFA score and APACHE II score in hospital were 3.45±2.79 and 8.27±6.01, respectively. 21 (2.10%) trauma patients died during the hospital days.
Isolated variants have only a small impact on sepsis risk
In the present study, 64 variants were successfully genotyped by the SNPscan method in 1000 trauma patients. The overall calling rate was greater than 96%. All variants meet the criteria of MAF >0.01 and P(HWE) >0.01 (Table S3). Due to genotyping failure in some samples, 883 patients with complete genotyping data for all 64 variants were finally selected for further analysis. Firstly, we evaluated the association between 64 genetic variants and sepsis risk in additive genetic model using unadjusted logistic regression analysis (Table S4). The results indicated that four variants were significantly related to the sepsis risk at a nominal level: rs2297518, located in the NOS2 gene (OR=1.53, 95%CI=1.12-2.10, P=0.01); rs10865710, located in the PPARG gene (OR=1.32, 95%CI=1.06-1.63, P=0.01); rs740598, located in the HSPA12A gene (OR=1.25, 95%CI=1.01-1.53, P=0.04); and rs5743551, located in the TLR1 gene (OR=1.26, P=0.04). The associations of the four variants with sepsis were confirmed using logistic regression analysis, adjusting for age and sex. None of the other variants was associated with sepsis. These results indicted a relatively limited effect of single variants on sepsis in our trauma cohort.
A wGRS is significantly associated with traumatic sepsis
To evaluate the joint effect of these genetic variants on sepsis risk, a random forest algorithm was applied. As shown in Table 1 and Figure 1, 17 genetic variants induced a positive effect (MDA >1) by random forest algorithm (Table S4) were selected in the subsequent calculation of the wGRS. For all trauma patients, the wGRS distribution was ranging from 0.68 to 3.69. The incidence of sepsis increased significantly along with the increase of wGRS (Figure 2A) and cases had more risk alleles than controls (Figure 2B) using the wGRS of 17 variants (P=3.47×10-6). As shown in Table S5, unadjusted logistic regression analyses indicted the significant association between traumatic sepsis risk and wGRS (OR=2.42, 95%CI=1.73-3.39, P=3.03×10-7), which was also significantly associated with sepsis after adjusted by age, sex, and ISS through multivariable logistic regression analysis (OR=2.19, 95%CI=1.53-3.15, P=2.01×10-5).
We further classified all trauma patients (883 individuals) into four subgroups according to the WGRS quartiles: low risk group (wGRS<1.80), Medium risk group (wGRS=1.80-2.20), high risk group (wGRS=2.20-2.50), and extremely high risk group (wGRS≥2.50). The results demonstrated that 237 patients were classified into the low risk group with 42 (17.72%) sepsis cases and 205 patients were classified into the extremely high risk group with 83 (40.48%) sepsis cases. Compared with those individuals who had the lowest score (wGRS<1.80), the trauma patients with higher score had higher incidence of sepsis, with odds ratios of 1.47 (95%CI=0.93-2.30, P=0.10), 1.87 (95%CI=1.20-2.92, P=6.00×10-3), and 3.16 (95%CI=2.05-4.88, P=1.20×10-7), respectively (Ptrend=6.81×10-8) (Table 2). Furthermore, we compared the SOFA score and APACHEII score in different wGRS subgroups, which also demonstrated higher SOFA score (Ptrend=5.00×10-3) and APACHEII score (Ptrend=1.00×10-3) were observed in patients with higher score, respectively (Table 3).
Discriminative ability for traumatic sepsis
The wGRS and ISS were identified as independent risk predictors of sepsis in trauma patients using the multivariate logistic regression algorithm (Table S5). Furthermore, the VIF of the two candidate predictors was 1.012, indicating that there was no collinearity. Therefore, wGRS and ISS were used to construct the prediction model of traumatic sepsis. To validate whether the wGRS could enhance the predictive value, we conducted ROC to evaluate the predictive ability of three models: only ISS, only wGRS, and ISS plus wGRS. The AUCs of only wGRS and only ISS were 0.619 (95%CI=0.586-0.651) and 0.734 (95%CI=0.703-0.763), respectively. Our results demonstrated that when incorporating wGRS into the ISS, the AUC of the prediction model increased to 0.768 (95%CI=0.739-0.796), with an increase of 3.40% (P=8.00×10-4) (Figure 3B). To confirm the improvement, we considered NRI to estimate the reclassification of the prediction model when wGRS was included. Compared with the ISS, these reclassification rates gave an estimated NRI of 25.18% by including the wGRS into the ISS (95%CI=17.84-32.51%, P=6.00×10-5) (Table 4). Therefore, when wGRS was added to the clinical model, the ability of the prediction model improved significantly.
Clinical usefulness of the prediction model
For the clinical usefulness, a nomogram incorporating the two predictors was constructed based on the multivariate logistic regression model that showed good calibration and discrimination in the trauma cohort (Figure 3A, B). The AUC of nomogram was 0.768, which was superior to either the wGRS or the ISS alone. As presented in the Figure 3C, the DCA of nomogram model indicated when the threshold probability is between 0 and 0.56, the nomogram performed more net benefit than the "treat-all" or "treat-none" strategies, which indicated that the nomogram was clinically useful.