Circulating metabolic biomarkers predict incident sepsis: a large-scale population study in the UK Biobank

Abstract

Background: Currently, there is an absence of large-scale research focusing on the metabolome profiles of individuals prior to the development of sepsis. This study aimed to evaluate the associations of circulating Nuclear Magnetic Resonance (NMR) metabolic biomarkers with the risk of incident sepsis and the predictive ability of these metabolites for sepsis.

Methods: The analysis utilized plasma metabolomic data measuring through NMR from the UK Biobank, which involved baseline plasma samples of 106,533 participants. The multivariable-adjusted Cox proportional hazard models were used to assess the associations of each circulating NMR metabolite biomarker with risk of incident sepsis. The full cohort was randomly assigned to a training set (n = 53,267) and a test set (n = 53,266) to develop and validate the sepsis risk prediction model. In training set, the least absolute shrinkage and selection operator (LASSO) and stepwise Cox regression analyses were used to develop the prediction model. In test set, the predictive ability of conventional risk factors-based and combined metabolic biomarkers prediction model was assessed by Harrell's C-index. The incremental predictive power of the metabolic biomarkers was evaluated with continuous net reclassification improvement (NRI) and integrated discrimination improvement (IDI).

Results: A total of 90 circulating metabolic biomarkers were significantly associated with risk of incident sepsis (all FDR adjusted P value < 0.05). Of these, triglycerides related lipid sub-classes, glycolysis, ketone bodies, and inflammation related metabolite biomarkers, creatinine, and phenylalanine were positively associated with risk of incident sepsis, while most of other lipid sub-classes, albumin, histidine, fatty acid and cholines related metabolic biomarkers were negatively associated with risk of sepsis. The Harrell's C-index of the conventional prediction model was 0.733 (95% CI: 0.722, 0.745) for incident sepsis; after adding the circulating NMR metabolic biomarkers to the conventional prediction model, the Harrell's C-index increased to 0.741 (95% CI: 0.730, 0.753) for incident sepsis. In addition, the continuous NRI and IDI were 0.022 (95% CI: 0.015, 0.043, P < 0.05) and 0.009 (95% CI: 0.006, 0.014, P < 0.05).

Conclusion: This study identified multiple plasma metabolic biomarkers were associated with risk of incident sepsis. The addition of these metabolic biomarkers to the conventional risk factors-based model significantly improved the prediction precision.

Keywords: Metabolic biomarkers; Sepsis; UK biobank.

Fig. 1

A flowchart of analysis of plasma metabolic landscape and associations of circulating metabolic biomarkers with risk of incident sepsis. Each NMR circulating metabolic biomarkers concentrations were natural log-transformed and standardized by z-score. Model was adjusted for age, sex, ethnicity, qualifications, socio-economic status, body mass index, smoking status, alcohol drinking, C-reactive protein level, and medical history of diabetes, hypertension, hyperlipidemia, myocardial infarction, stroke, renal failure, liver disease, and cancer. ^*P < 0.05 for Benjamini–Hochberg adjusted false discovery rate. Apo, apolipoprotein; DHA, docosahexaenoic acid; FA, fatty acids; acid; HDL, high‑density lipoproteins; IDL, intermediate‑density lipoproteins; L, large; LA, linoleic acid; LDL, low‑density lipoproteins; LP, lipoprotein; M, medium; MUFA, monounsaturated fatty acids; PUFA, polyunsaturated fatty acids; S, small; SFA, saturated fatty acids; VLDL, very low‑density lipoproteins; XL, very large; XS, very small; XXL, extremely large

Fig. 2

Associations of circulating metabolic biomarkers with risk of incident sepsis by age (< 60 vs. ≥60 years) and sex (male vs. female). A age; B sex. Each NMR metabolite concentrations were natural log-transformed and standardized by z-score. Models were adjusted for age, sex, ethnicity, qualifications, socio-economic status, body mass index, smoking status, alcohol drinking, C-reactive protein level, and medical history of diabetes, hypertension, hyperlipidemia, myocardial infarction, stroke, renal failure, liver disease, and cancer. *P for interaction < 0.05 for of subgroup by age and sex. Apo, apolipoprotein; DHA, docosahexaenoic acid; FA, fatty acids; acid; HDL, high‑density lipoproteins; IDL, intermediate‑density lipoproteins; L, large; LA, linoleic acid; LDL, low‑density lipoproteins; LP, lipoprotein; M, medium; MUFA, monounsaturated fatty acids; PUFA, polyunsaturated fatty acids; S, small; SFA, saturated fatty acids; VLDL, very low‑density lipoproteins; XL, very large; XS, very small; XXL, extremely large

Fig. 3

Receiver operating characteristic curves for conventional prediction model and combined model