Study and ethics

This was a prospective, controlled, single-center study of patients with sepsis (the sepsis group) and without sepsis (the control group). The study was registered at the Chinese Clinical Trials Registry (ChiCTR2200060418) on June 1, 2022, and approved by the Ethics Review Committee (Ethics No. B2022-241(2)). All participants provided signed informed consent. The procedures followed were in accordance with the of the responsible committee on human experimentation and with the ethical guidelines of the 2003 Helsinki Declaration.

Design, setting, and participants

From July 2022 to October 2022, septic patients were enrolled in the sepsis group. Inclusion criteria were (1) volunteered to participate in the trial and signed informed consent; (2) planned to undergo emergency surgery; (3) aged ≥ 18 and ≤ 80 years; (4) had American Society of Anesthesiologists (ASA) physical statuses I–IV; and (5) met the Sepsis 3.0 diagnostic criteria. Exclusion criteria were (1) comorbidity with tumor disease; (2) pre-existing cardiac insufficiency with left ventricular ejection fraction (LVEF) < 40% before the current infection; (3) pre-existing severe hepatic insufficiency (prothrombin ratio < 15%) before this infection; (4) pre-existing severe renal insufficiency (estimated glomerular filtration rate [eGFR] < 30 ml/min/1.73 m2) before this infection; and (5) pre-existing mental illness, refusal to participate or failure to sign the informed consent. Withdrawal criteria for the sepsis group were infection found to be caused by a tumor during the operation.

Patients undergoing elective surgery during the same period were included in the control group. Inclusion criteria were (1) volunteered to participate in the trial and signed informed consent; (2) ASA I–II; (3) aged ≥ 18 years and ≤ 80 years; and (4) planned to undergo surgery unrelated to infection or tumor removal. Exclusion criteria were (1) ASA ≥ III; (2) pre-existing cardiac insufficiency; (3) pre-existing severe hepatic insufficiency; (4) pre-existing severe renal insufficiency; and (5) pre-existing mental illness, refusal to participate, or failure to sign the informed consent. Withdrawal criteria for the control group were infection or tumors found in the lesion during the operation.

Blood sampling and mass spectrometry

Venous blood samples were taken before surgery on the day of surgery (D0) for both groups and on postoperative day 7 (D7) in the sepsis group. Serum samples were collected and stored at − 80 °C. Five randomized serum samples per group were analyzed via mass spectrometry (timsTOF Pro, Bruker, Germany), and mass spectrometry databases were established for both groups to screen out the significantly altered proteins in the septic patients. RAW files were identified, quantified, and analyzed using Maxquant1.4.1.2 (Max Planck Institute of Biochemistry, Martinsried, Munich, Germany). The ion peak intensity reflected the relative protein abundances and the quantitative ratio of the proteins was normalized against the median ratio value of the internal standard sample.

Gene ontology (GO) functional analysis and the protein­-protein interaction (PPI) network

We used the “cluster Profiler” package in R for the GO annotation of the differentially expressed proteins (DEPs). The GO database was used to analyze the cellular components, molecular function, and biological processes of the proteins (p value cutoff 0.05, q value cutoff 0.05). The PPI network of these DEPs was constructed using the Search Tool for the Retrieval of Interacting Genes (STRING) database (https://www.string-db.org/).

DEP validation

ELISA was used to validate the candidate biomarkers obtained from the GO analysis using ELISA kits (Shanghai Xinle Biotechnology Co., Ltd., China) per the manufacturer’s instructions.

Follow-up

Patients’ baseline clinical characteristics were recorded, including age, sex, height, weight, history of previous diseases, history of special drug use, and results of major laboratory tests such as routine blood counts (white blood cell count, neutrophil ratio hemoglobin count, platelet count), infection indicators (procalcitonin, C-reactive protein, plasma lactate, interleukin-1β, interleukin-2 receptor, interleukin-6, interleukin-8, interleukin-10, tumor necrosis factor-α), liver and kidney function parameters (bilirubin, albumin, creatinine), myocardial dysfunction markers (cardiac Troponin T, myoglobin, creatine kinase, N-terminal brain natriuretic peptide precursor), and serum electrolytes (serum sodium, potassium, calcium, magnesium). The follow-up time points were D1 (1 day after surgery), D7, at discharge, and 1 month after surgery. The last visit at 1 month was done via telephone; the other visits were on-site. Clinical data, including vital signs; laboratory tests; SOFA score; organ dysfunction; intensive care unit (ICU) admission, length of hospital stay, and discharge outcomes, were collected at each on-site follow-up visit. Multiple organ dysfunction score (Marshall et al. 1995) was used to assess organ dysfunction events. Oxygenation index, serum creatinine, serum bilirubin, platelet count, Glasgow coma scale and pressure-adjusted heart rate were selected as variables to reflect an important system functional status. The prognostic variables were septic shock incidence, organ dysfunction incidence, proportion of ICU admissions, length of hospital stay, hospital mortality, and 1-month mortality.

Statistical analysis

Data analyses were performed in R, project 3.5.3, for Windows and IBM SPSS Statistics, version 22.0. Normality and homogeneity of variance were tested via the Shapiro-Wilk and Levene’s tests. Categorical variables are reported as frequencies and percentages. Continuous variables with a normal distribution are presented as the means ± standard deviation; all others are described as medians and interquartile range (IQR). Baseline characteristics between the groups were compared using Fisher’s precision probability test or Pearson’s chi-square test for categorical variables, using Student’s t-test for normally distributed continuous variables or Wilcoxon rank-sum test for non-normally distributed continuous variables. Spearman’s correlation test was conducted to measure the statistical relationship between preoperative blood tests and selected serum proteins. LR analysis was used to estimate the relative risk of selected serum protein levels and clinical outcomes of septic patients. The incidence was entered as a dependent variable, Y, in the LR model and was coded as 0 for absent (did not occur) or 1 for present (occurred). We applied and compared two regression methods: stepwise logistic regression (SL) and logistic least absolute shrinkage and selection operator (LASSO) regression. To obtain the logistic LASSO estimator, we used the glmnet package in R. For the stepwise selection, we used the Akaike information criterion to select the covariates. For the logistic LASSO regression, we used cross-validation to select λ. We calculated the area under the receiver operating characteristic curve (AUROC) for the data to measure the predictive performances of the fitted models. P < 0.05 was considered statistically significant.

Calibration curves were plotted to assess the calibration of the models. A significant test statistic implies that the models do not calibrate perfectly. To quantify the discrimination performance of the model, Harrell’s C-index was measured. The models were subjected to bootstrapping validation to calculate a relatively corrected C-index.