The present study was conducted using a nested matched-case design. From the Molecular Diagnosis and Risk Stratification of Sepsis (MARS) cohort, we selected ICU-acquired bloodstream infections caused by either one of the following three pathogen groups of interest: (1) GNR (including E. coli; K. pneumoniae; P. aeruginosa; Aeromonas; M. morganii; S. marcescens; S. maltophilia; H. influenzae; P. mirabilis; P. vulgaris), (2) enterococcal species (E. Faecalis; E. faecium), and (3) Candida species (C. albicans; C. glabrata; C. lusitaniae; C. krusei; C. parapilosis or C. species not further determined).

Events were considered to be ICU acquired if the following criteria were met: (1) a first occurrence of a positive blood culture at least 96 h after ICU admission and (2) no prior blood culture yielding the same pathogen for at least 30 days. Blood cultures yielding more than one pathogen were excluded. To avoid the selection of possible contaminants, positive cultures with an incubation period of > 72 h were ineligible for selection. Furthermore, enterococci were included only if enterococcal species had been isolated from at least two consecutive blood cultures taken on different days (whereas all GNR and Candida species isolated from blood cultures were considered true positives). Final exclusion criteria included abdominal surgery in the week prior to infection onset and the presence of chronic or acute pancreatitis, as these conditions may directly affect the reliability of the PSP biomarker [8].

Subsequently, we conducted a matched case analysis. For the matching procedure, available BSI cases due to Candida species were matched to individual BSI cases from the (larger) GNR group in a 1:1 ratio. Next, the resulting pairs were also matched to cases from the (largest) enterococci group. Matching criteria were (1) the presence of immunodeficiency at ICU admission; (2) length of ICU stay on the day of BSI onset (± 4 days), (3) use of renal replacement therapy 72 h before BSI onset, and (4) Sequential Organ Failure Score (SOFA) (± 4 points) 72 h before BSI onset. To this end, immunodeficiency was defined as the active use of immunosuppressive drugs upon ICU admission (including chronic or high-dose corticosteroids, calcineurin inhibitors, mycophenolate mofetil, and others), hematologic malignancy, active cancer (including chemotherapy or radiotherapy in the year prior to admission), neutropenia, or any other documented humoral or cellular deficiency. Renal replacement therapy was used as a matching criterion, as kidney dysfunction may directly affect PSP levels [18, 19].

We used consecutive daily samples that had been collected as part of the MARS cohort during three days prior to infection onset (i.e., the day on which a first positive blood culture was drawn). EDTA blood was spun at 1500 rpm for 15 min and supernatant plasma was stored at − 80 °C within a maximum of 4 h from sampling. PSP was subsequently measured using the CE-marked in-vitro diagnostic PSP capsule on the abioSCOPE® platform (Abionic SA, Epalinges, Switzerland). This ‘point-of care’ device measures PSP concentrations in plasma between 20 and 600 ng/mL within 8 min using nanofluidic immunoassay technology [20]. Samples yielding out-of-range values were diluted 1:4 and subsequently retested to achieve an accurate result.

We exploited the MARS cohort, in which consecutive critically ill patients with an expected ICU length of stay > 48 h had been prospectively enrolled in two tertiary mixed ICUs in The Netherlands. Participating centres were the University Medical Centre Utrecht (recruiting 2011–2019) and the Amsterdam University Medical Centre (recruiting 2011–2013). Selective digestive decontamination was part of standard care in both centres [21]. The MARS project was approved by the Medical Ethics Committees of both participating hospitals (protocol number 10-056). The current study was additionally reviewed by the UMCU Biobank Research Ethics Committee (protocol number 21-405).

Differences in patient and event characteristics were analysed using one-way analysis of variance (ANOVA), Kruskal–Wallis tests, or chi-square tests, as appropriate. To facilitate the analysis of temporal trends, a delta PSP value was calculated for each BSI case reflecting the absolute change from 72 h before infection (T-72) to infection onset (T0). Absolute and trend differences in PSP levels between groups were subsequently analysed using one-way ANOVA. For the latter analysis, all PSP values were log-transformed.

To gain detailed insight into how PSP responds to BSI onset, we used several explorative approaches, some of which were informed by post hoc findings. First, we investigated the concomitant rise and fall of other inflammatory markers, including CRP, white blood cell count, and fever. To this end, we defined a systemic inflammatory response as the occurrence of at least one of the following criteria in the 72 h preceding onset of infection: (1) a CRP ≥ 100 mg/L with at least a 20 mg/L increase compared to the previous day, (2) a white blood cell count ≥ 12 × 109/L with at least a 1 × 109/L increase compared to the previous day, (3) a white blood cell count ≤ 4 × 109/L with at least a 1 × 109/L decrease compared to the previous day, or (4) fever, defined as a temperature ≥ 38.3 °C with at least a 1 °C rise compared to the previous day. We then conducted a sensitivity analysis excluding BSI events that did not meet any of these criteria to investigate our hypothesis that patients lacking a systemic inflammatory response might be distorting the temporal trend. Second, since blood cultures were not systematically obtained on a daily basis, it is possible that pathogens were in fact already present in the bloodstream for some time before detection. To address this issue, we realigned the timelines of 72 cases of patients with a systemic inflammatory response by assuming that T0 (i.e., the true onset of BSI) coincided with the observed peak in CRP level if this peak occurred before a first positive culture. This is a conservative approach since CRP levels may take up to 50 h to reach their maximum in simple infection [22]. However, we assumed that a rise in CRP (in the absence of new surgical trauma) could at least approximate potential diagnostic delays. Finally, we constructed subgroups that were stratified by severity of disease (SOFA-score of 7 or higher) as measured on the day of BSI onset. A Wilcoxon rank sum test was then used to assess differences between groups.

We used R Studio 2023.12.1 Build 402 (R foundation for Statistical Computing, Vienna, Austria) for all statistical analyses. Continuous variables are presented as medians with interquartile ranges (IQRs). Discrete values are shown as counts with percentages. All reported p values are two-sided, and statistical significance was set at an α-value of 0.05.