Polygenic Risk Score for Early Prediction of Sepsis Risk in the Polytrauma Screening Cohort

Abstract

Background: Increasing genetic variants associated with sepsis have been identified by candidate-gene and genome-wide association studies, but single variants conferred minimal alterations in risk prediction. Our aim is to evaluate whether a weighted genetic risk score (wGRS) that aggregates information from multiple variants could improve risk discrimination of traumatic sepsis.

Methods: Sixty-four genetic variants potential relating to sepsis were genotyped in Chinese trauma cohort. Genetic variants with mean decrease accuracy (MDA) > 1.0 by random forest algorithms were selected to construct the multilocus wGRS. The area under the curve (AUC) and net reclassification improvement (NRI) were adopted to evaluate the discriminatory and reclassification ability of weighted genetic risk score (wGRS).

Results: Seventeen variants were extracted to construct the wGRS in 883 trauma patients. The wGRS was significantly associated with sepsis after trauma (OR = 2.19, 95% CI = 1.53-3.15, P = 2.01 × 10^-5) after being adjusted by age, sex, and ISS. Patients with higher wGRS have an increasing incidence of traumatic sepsis (P _trend = 6.81 × 10^-8), higher SOFA (P _trend = 5.00 × 10^-3), and APACHE II score (P _trend = 1.00 × 10^-3). The AUC of the risk prediction model incorporating wGRS into the clinical variables was 0.768 (95% CI = 0.739-0.796), with an increase of 3.40% (P = 8.00 × 10^-4) vs. clinical factor-only model. Furthermore, the NRI increased 25.18% (95% CI = 17.84-32.51%) (P = 6.00 × 10^-5).

Conclusion: Our finding indicated that genetic variants could enhance the predictive power of the risk model for sepsis and highlighted the application among trauma patients, suggesting that the sepsis risk assessment model will be a promising screening and prediction tool for the high-risk population.

Keywords: genetic variants; prediction; sepsis; trauma; weighted genetic risk score.

FIGURE 1

Random forest model including 17 variants previously associated with sepsis. The first 64 variables with the highest mean decrease accuracy are plotted. Seventeen sepsis-associated variants were shown to induce a positive change in mean decrease accuracy. These variants were thus considered to have a relevant influence on the model and were chosen for inclusion in the wGRS.

FIGURE 2

Distributions of the wGRS among sepsis cases and controls. (A) The percentage of wGRS of 17 variants displaying a significant difference among cases and controls. (B) The distributions of wGRS of 17 variants among cases and controls.

FIGURE 3

Model comparisons and clinical usefulness of the nomogram. (A) The nomogram incorporating ISS and wGRS was constructed for the prediction of sepsis after trauma. (B) ROC curves of three models for sepsis risk. The area under the ROC curves (AUCs) are based on logistic regression models of only clinical risk factor (ISS), only genetic factor (wGRS), and both clinical risk factor and genetic factor (ISS + wGRS). (C) DCA for the nomogram. The net benefit was plotted vs. the threshold probability. The red line represents the nomogram. The gray and black lines represent the hypothesis that all patients and no patients had sepsis, respectively.