Background
Severe infections are leading causes of morbidity and mortality among hospitalized patients [
1,
2]. Early detection of life-threatening infection is crucial to improving outcomes [
1]. To date, none of the circulating blood biomarkers or signatures of the immune response that have been investigated [
3,
4] detect life-threatening infection quickly enough and with an acceptable certainty [
5]. Two markers [C-reactive protein (CRP) and procalcitonin (PCT)] are widely used [
6‐
10] despite their sub-optimal performance [
11].
Pancreatic stone protein (PSP) is a pro-inflammatory mediator that binds to polymorphonuclear cells and triggers their activation in vitro (reviewed in [
12]); it is a recently described biomarker of infection whose performance has been thoroughly evaluated in several patient populations and in different clinical settings, including emergency rooms (ERs) [
13] and intensive care units (ICUs) [
14‐
18]. PSP was able not only to diagnose infection [
14,
15,
18] but also to characterize its severity [
16,
17] as well as to predict its outcome [
16,
17,
19]. Nevertheless, a clinically useful PSP threshold level has not yet been defined.
We aimed to perform an individual patient level meta-analysis of published data to determine the performance of PSP in detecting infection, to propose a threshold value for that purpose and to validate it across heterogeneous populations.
Discussion
Since its first description as a potential infection biomarker in patients after trauma [
14], pancreatic stone protein has repeatedly been shown to perform better than CRP and at least as well as PCT in identifying patients with infection [
13‐
15,
17,
18]. However, in contrast to PCT [
6,
26], no cut-off threshold value has yet been defined for its eventual clinical use.
We were able to obtain raw data from five small observational studies and did a meta-analysis at the individual patient level to explore the performance of PSP in detecting infection. The eligible studies were performed in different countries across Europe and included acutely ill patients from ERs or ICUs. The resulting cohort of 631 hospitalized adult patients encompassed an important proportion (42%) of patients without infection, which makes it the largest analysis of this nature on PSP.
Because we observed low heterogeneity of PSP effect despite strong heterogeneity in the baseline risk of infection among studies, we decided not to adjust for heterogeneity in the models presented in the main manuscript, because it may not be of clinical relevance (see Additional file
5: Supplemental section for models adjusting for heterogeneity). In the trade-off—adjusting for heterogeneity and improving performance of the biomarker versus addressing a clinical need and computing a unique biomarker threshold—we chose to omit adjustment for heterogeneity, especially as we wanted to determine a PSP cut-off value and compare it to those of PCT and CRP currently largely used at the bedside [
6‐
10].
We applied a stepwise approach to evaluate the performance of PSP and compared it to those of CRP and PCT for detecting infection in adult hospitalized patients presenting to the ER or the ICU. In all explored scenarios, PSP achieved a statistically significant better performance compared to CRP and PCT. Using the Youden’s index approach, we found a PSP cut-off value of 44.12 ng/ml, which is approximately four-fold higher than the upper value previously determined in 61 healthy adult volunteers (min. 4.0, max 18.3, median 10.8 ng/ml) [
27]. Using the same methodology, we found a CRP cut-off value of 99.5 mg/l and a PCT cut-off value of 0.2 ng/ml. Moreover, because there is much less overlap in PSP levels between infected and non-infected patients, PSP could guide decisions to start antibiotic treatment, as opposed to PCT that is currently rather used in ICU patients for antibiotic de-escalation [
8,
11,
28].
The inclusion of post-trauma and post-operative populations with a very low baseline risk of infection might have limited the ability of CRP and PCT to diagnose infection. Previous studies and meta-analysis exploring the diagnostic value of PCT for the detection of patients with infection have indeed reported that PCT accuracy and cut-off values are highly dependent on the clinical settings and the baseline risks for infection [
6,
26]. On the other hand, this might simply have highlighted the possible advantage of PSP over PCT and CRP in particular situations characterized by severe non-infectious systemic inflammatory states. Two studies including patients post-burn trauma recently confirmed this observation [
29,
30]. Interestingly, PSP levels rose up to 72 h before the clinical diagnosis of infection, confirming its potential role as an early biomarker of infection.
We further explored the value of biomarkers combinations. Combining PSP with CRP increased the accuracy of detecting infection from an AUC 0.81 to 0.90, a value that is usually considered as very good [
31] and among the highest AUCs reported for this setting [
4,
8,
9]. Interestingly, adding CRP to PSP markedly increases the sensitivity while not decreasing the specificity achieved by PSP alone (Additional file
1: Table S11).
This study has several strengths, which makes its results potentially generalizable. First, it was possible to include individual patient data of all published studies that evaluated the value of PSP to detect infection in hospitalized patients before March 2019. Second, it encompasses important clinical settings—ICUs and ERs—where early detection of infection is of the utmost importance in order to guide rapid management [
1]. Third, the raw data come from studies performed in several centres in the UK, Spain and Switzerland, reducing the risk of centre-effect bias. On the other hand, the measurement of PSP level was consistent throughout all studies, minimizing the risk of methodological bias. Finally, yet importantly, a balanced proportion of infected and non-infected patients was included in this analysis.
The main limitation of our study is the heterogeneity of the included populations, mixing community- and healthcare-acquired infections: some populations were included from medical emergency admissions with the goal to confirm infection at admission, while other patients were trauma or surgical patients who were likely to develop infection during hospitalization. This translates into an important heterogeneity in baseline risk of infection among the studies. Moreover, the five studies included patients for whom suspected or documented infection was the main reason of referral, which may limit the generalization of the findings in other circumstances. In addition, because only one study considered patients in the ER, we could not investigate subgroup of patients from ERs versus ICUs.
A further important limitation of primary research in infection biomarker evaluation is the lack of an objective gold standard definition of infection. Nevertheless, each of the five eligible studies set out clear criteria and procedures, all consistent with one another, using a synthesis of available clinical, radiological and microbiological evidences to assign each case as infection or not. Furthermore, the fact that we find low heterogeneity of PSP effect across the range of studies performed is reassuring that our fundamental observation is not undermined by variability in case definition. Moreover, the ability of PSP to diagnose infection is confirmed across all studies.
Our study opens new possibilities for the future use of PSP, including being used as a tool to optimize antibiotic stewardship program, as a biomarker to guide the decision to start antibiotics, or as inclusion or stratification criteria for future studies of severe infections. PSP is now available at the point of care (POC) with a turnaround time of less than 5 min [
12] making it highly suitable for situations where time to antibiotic is essential.
The next step would be to validate these results in independent cohorts of patients. As baseline risk of infection impacts the performance of a diagnostic biomarker and its corresponding threshold, the validation process would require to perform independent studies in selected settings (primary care, ER, ICU), each including at best patients with a homogenous baseline risk of infection. Such a thorough validation process has been performed for PCT [
6,
26] in large independent cohorts of patients, each with different baseline risks of infection and is still required before PSP to be routinely used in the clinic.
The benefit of serial measurement of PSP for the early detection of sepsis in intensive care unit patients has been evaluated in a prospective trial using the new POC technology [
32]. This patient population would be the first ideal cohort to validate the defined threshold in a homogenous patient population as well as the value of combining PSP to CRP. Both biomarkers can be now measured using point-of-care technologies; however, correction factors will be required while using new measurement assays and technologies. Finally, our methodology can be used to further evaluate the value of PSP in detecting the severity of the infection as well as its prognosis and to define for thresholds for that purpose as well.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.