Abstract
BACKGROUND: Sepsis is characterized by organ dysfunction as a response to infection and is one of the leading causes of mortality and loss of health. The heterogeneous nature of sepsis, along with ethnic differences in susceptibility, challenges a thorough understanding of its etiology. This study aimed to propose prediction models by leveraging genetic-risk scores and clinical variables that can assist in risk stratification of patients. METHODS: A total of 1,403 patients from Taiwan, diagnosed with sepsis, were utilized. Genome-wide survival analysis was conducted, with death within 28 days from sepsis onset, as the primary event to report significantly associated SNPs. A polygenic risk score (PRS-sepsis) was constructed via clumping and thresholding method which was added to clinical-only models to generate better performing prognostic models for identifying high-risk patients. Kaplan-Meier analysis was conducted using PRS-sepsis. RESULTS: A total of five single-nucleotide-polymorphisms (SNPs) reached genome-wide significance (p < 5e-8), and 86 SNPs reached suggestive significance (p < 1e-5). The prognostic model using PRS-sepsis showed significantly improved performance with c-index [confidence interval (CI)] of 0.79 [0.62-0.96] and area under receiver operating characteristic curve (AUROC) [CI] of 0.78 [0.75-0.80], in comparison to clinical-only prognostic models (c-index [CI] = 0.63 [0.45- 0.81], AUROC [CI] = 0.61 [0.58-0.64]). The ethnic specificity was established for our proposed models by comparing it with models generated using significant SNPs from prior European studies (c-index [CI] = 0.63 [0.42-0.85], AUROC [CI] = 0.60 [0.58-0.63]). Kaplan-Meier plots showed that patient groups with higher PRSs have inferior survival probability compared to those with lower PRSs. CONCLUSIONS: This study proposed genetic-risk models specific for Taiwanese populations that outperformed clinical-only models. Also it established a strong racial-effect on the underlying genetics of sepsis-related mortality. The model can potentially be used in real clinical setting for deciding precise treatment courses for patients at high-risk thereby reducing the possibility of worse outcomes.