This multicentric biomarker-blinded prospective observational clinical study was conducted in 14 ICUs in France, Switzerland, Italy, and the United Kingdom, see Additional file 1: Table 1. Approval was obtained from the ethics committees of all participating sites. The study was registered at ClinicalTrials.gov (no. NCT03474809).
All patients older than 18 years old admitted to the ICU and expected to stay at least 7 days and/or to be mechanically ventilated for at least 5 consecutive days were screened for study inclusion (Additional file 1:Table 2). Patients with a clinical diagnosis of sepsis or suspicion of sepsis at the admission were not included. Written informed consent to participation and any study-related assessment was obtained from all patients before their inclusion in the study. Written informed consent was obtained according to the specific requirements of each center’s ethics committee.
At least 40 patients who developed sepsis after study enrolment were planned to be included. Based on recently published data [17,18,19], the expected incidence of sepsis in the targeted ICU population was estimated conservatively to be 15%. This value corresponded to a sample of 267 patients, to which we added 10% to compensate for withdrawal and loss to follow-up.
Trial procedure and definitions
Patients were managed according to the centers’ standard clinical practices, including those for the diagnosis, assessment, and treatment of sepsis. Blood samples were collected daily for biomarker (PCT, CRP, and PSP) measurements. Patients were followed until death or discharge from the ICU or for 30 days, whichever occurred first. The Sepsis-3 criteria were used to define sepsis , and microbiological procedures were performed to diagnose infections responsible for sepsis according to local protocol.
Measurement of biomarkers
Daily PSP levels were determined at the end of the study period with the CE-marked IVD PSP capsule on the point-of-care abioSCOPE® device (Abionic SA; Additional file 2: Figure 1). The 5th and 95th percentiles of PSP values in healthy adults are 25.0 to 60.7 ng/ml (median: 41.7 ng/ml) (Abionic internal data). CRP and PCT measurements were performed by a central laboratory using a Tina-quant® C-Reactive Protein Gen.3 (Roche Diagnostics) and a Liaison® Brahms PCT® for PCT (DiaSorin), respectively. All instruments were used according to the manufacturers’ instructions. The upper limit of the normal range of the CRP assay is 5 mg/l, and of the PCT assay 0.1 ng/ml, as per the manufacturer’s information.
External adjudication committee
An independent endpoint adjudication committee (EAC) composed of three ICU experts was formed. Two non-chair EAC members retrospectively and independently reviewed case report forms for each patient and determined whether a septic event had occurred during the patient’s ICU stay (including on the day of inclusion), and on which day it started. Day 0 sepsis was determined using all information collected in the study’s eCRF and in accordance with the proposal published by Lambden et al. . Discordant determinations were arbitrated by the EAC chair. The EAC had access to all case report forms, including results of all microbiological investigations, CRP and PCT measurements taken in the ICUs as part of standard care, but was blinded to centrally measured PSP, PCT, and CRP levels and the investigators’ diagnoses of sepsis.
Data collected at the time of inclusion were: demographic characteristics, reason for ICU admission, medical history, and Charlson comorbidity index. Additional clinical data were collected on day 1 and daily thereafter when measurements were performed as part of standard care; these data included vital signs (temperature, heart rate, and blood pressure), Sequential Organ Failure Assessment (SOFA) score, PaO2/FiO2 ratio, hematological data, clinical chemistry parameters, need for organ support (ventilation and use of vasoactive drugs) and microbiological data. In addition to the collection of data on antibiotic treatments (drug name, dose, schedule, route, and indication), meticulous assessments for sepsis detection were carried out daily until ICU discharge, death, withdrawal, or day 30, whichever came first.
Discrete values are described as counts (percentages) and continuous variables are described as means with standard errors or medians with interquartile ranges (IQRs), as appropriate. The missing biomarker values were imputed by carrying the last observation forward, and multiple consecutive missing values were reported as missing. Based on EAC decisions, the patients were allocated to sepsis (a septic event occurred during the study period) or no-sepsis (no septic event during the study period) groups. In the sepsis group, the event day was set as the day on which the first septic events began, according to the EAC. In the no-sepsis group, the event day was set to day 7 of patients’ ICU stays, when more than 50% of adjudicated septic events occurred, or the days of discharge for patients with ICU stays of fewer than 7 days. Biomarkers average trajectories were visualized accordingly. All statistical analyses were performed using R version 4.0.2.
Biomarkers dynamics and progression toward sepsis
To explore the dynamics of serial biomarker measurement throughout the study period, we modeled biomarker courses in the 3 days prior to EAC sepsis diagnoses by fitting a linear mixed-effect model with PSP, CRP, and PCT as response and the following explanatory variables: patient-specific random effect, group, day-to-event and group by day-to-event interaction as fixed-effects (estimates are provided in Additional file 1: Table 3). According to Van Breukelen  “testing absence of group by time interaction is equivalent to testing the hypothesis of no group effect on the change” in a response variable over days.
Assessment of biomarker accuracy in diagnosing sepsis
The joint performance of the biomarkers in the diagnosis of sepsis on the day it occurred as per EAC decision was assessed using receiving operating characteristic (ROC) analysis. To account for differences among study sites and incidences of nosocomial sepsis, a linear mixed-effects logistic regression model was used with the center serving as a random intercept and biomarkers serving as fixed events to determine diagnostic accuracy, sensitivity, specificity, and the positive and negative likelihood ratios (estimates are provided in Additional file 1: Table 4) . Cut-off values of PSP, CRP and PCT have been estimated without adjusting for center-effect.”