Extracorporeal Membrane Oxygenation (ECMO) has gained widespread acceptance as rescue treatment for refractory cardiovascular and respiratory failure in adults and children (1–4). The use of ECMO has extended to increasingly complex patient populations, such as patients with malignancies (5, 6) and patients with chronic medical comorbidities or those requiring extracorporeal cardiopulmonary resuscitation (ECPR) (7, 8). This complex patient group is subject to high risk of developing complications and more likely to be colonized with multi-drug resistant organisms (9).

Infections represent one of the most common complications in patients managed on ECMO (10). A recent review of the literature revealed an incidence of ECMO-associated infections between 10.1 and 116.2 per 1,000 ECMO days (11). Nosocomial infections acquired on ECMO have been described in up to 64% of adult and 30% of neonatal ECMO runs and are associated with increased ECMO duration, prolonged mechanical ventilation, prolonged hospital stay, and increased mortality; particularly in presence of invasive fungal infections (12).

Clinical signs of infection in patients on ECMO are non-specific and often masked, making a diagnosis of infection challenging (10). During ECMO, respiratory rate and temperature are usually artificially controlled, and ECMO perfusion, sedation, and cardiovascular drugs alter heart rate and blood pressure. In addition, ECMO-triggered systemic inflammation can mimic features of sepsis. A definite diagnosis of infection therefore relies on microbiological cultures which commonly take up to 48–72 h to analyse (13). Blood culture performance depends on bacterial load, and results can be falsely negative in patients exposed to prophylactic or therapeutic antibiotics (14). This diagnostic uncertainty may lead to unnecessary treatment with antibiotics and may promote antibiotic resistance (15). At the same time, missing a diagnosis of a nosocomial infection will delay initiation of effective antimicrobial treatment and potentially contribute to excess ECMO mortality and morbidity (16, 17).

Laboratory markers of infection have gained widespread acceptance to guide decisions to start and stop antibiotics in critical care (18). Extracorporeal circuits and shock-related acute phase responses lead to non-specific increases in infection markers, further confounding the diagnosis of infection in this population (19). In addition to the traditional infection markers White Cell Count (WCC) and C-reactive protein (CRP), a large randomized-controlled trial in adult general ICU patients demonstrated superiority of a Procalcitonin (PCT) guided approach (20). Similarly, PCT guided therapy in neonates with suspected early-onset sepsis allowed to reduce the duration of antimicrobial therapy (21, 22). While observational studies in critically ill children suggest superiority of PCR over CRP, adequately powered RCTs are lacking (23). In contrast, the available evidence supporting the use of infection markers on ECMO is very limited: Only four studies across 65 adult and 47 neonatal/pediatric patients reported on the accuracy of markers to diagnose infection in patients already on ECMO (19, 24–26).

We hypothesized that PCT would be superior to CRP and WCC to diagnose infection in neonates and children treated with ECMO. For this purpose, we assessed the diagnostic accuracy of daily measured PCT, CRP, and WCC to diagnose infection in 65 consecutive ECMO runs at our center.

We performed a retrospective, single center cohort study of all consecutive neonatal and pediatric ECMO runs at the Queensland Children's Hospital Pediatric Intensive Care Unit (PICU). Ethics approval was obtained by the institutional human research and ethics committee including waiver of consent (HREC/17/QRCH/8). This study was reported following the current Standards for Reporting Diagnostic accuracy studies (STARD) guideline (27).

We included ECMO cases in neonates, children and adolescents aged <18 years with initiation of ECMO between 01/01/2015 and 31/12/2016. All types of runs were included [venovenous (VV), and venoarterial (VA), extracorporeal cardiopulmonary resuscitation (ECPR), cardiac, respiratory]. Two authors independently reviewed the institutional patient data management system (Metavision, iMDsoft, USA) of all patients treated with ECMO during the study period to ascertain which patients had a microbiologically proven infection, suspected infection or no infection on each ECMO day.

The ECMO service provides centralized ECMO support for the state of Queensland and northern New South Wales including ECPR, cardiac, and respiratory ECMO for all neonatal and pediatric age groups. The default ECMO setup included centrifugal pumps, a shunt containing a filter for slow continuous ultrafiltration, in- and outlet pressure monitors, and continuous measurement of venous and arterial saturations, hemoglobin, and ECMO flow. Steroids are not routinely used. Antimicrobial prophyaxis using cephazolin is used for the first 24 h after cannulation, and in children with open chest for the duration of the chest remaining open.

Infections are one of the most common complications for patients on ECMO and contribute to higher morbidity and mortality (12). While the benefit of rapid identification and treatment of sepsis has been broadly recognized in critically ill patients (16, 33), accurate and timely diagnosis of infections on ECMO poses a major challenge, and the evidence to guide best practice in this area remains scarce. Due to profound activation of inflammatory pathways on patients requiring ECMO and aggravated by the extracorporeal circuit itself, it is unclear to what extent widely used infection (or inflammation) markers provide clinically meaningful diagnostic accuracy in this patient group (10). We assessed the diagnostic accuracy of PCT, CRP and WCC to diagnose bacterial infections in 65 neonatal and pediatric ECMO runs. While CRP performed best, its accuracy was only moderate with an AUC below 0.8. PCT did not discriminate reliably in our cohort. Neither changes in CRP or PCT, nor combinations of biomarkers improved the performance of the diagnostic approach.

Our findings contrast with self-reported practice in a recent survey of 147 intensivists from 91 ECMO centers across 25 countries, which identified that rising CRP or PCT were the most commonly used criteria to diagnose infection (34). Our study demonstrates that patients are exposed to antibiotics on almost 90 percent of ECMO days, indicating an urgent need for improved markers or infection prediction models. Previous studies in patients on ECMO showed that WCC, absolute neutrophil count, immature to total white cell ratio, platelet count, lactate and fibrinogen were not predictive of bloodstream infections (35–38). In adult general ICU patients with sepsis and septic shock, PCT has been shown to improve the diagnosis of bacterial infections compared to traditional infection markers such as WCC or CRP, allowing to enhance antimicrobial stewardship by reducing duration of antibiotic therapy with potential mortality benefit (20, 22). A recent literature review identified 59 full text articles reporting on infections on ECMO, of which only four investigated the role of infection markers (11). These previous studies on infection markers on ECMO were very small, including <30 runs, and yielded inconclusive results (19, 24, 25, 39): Rungatscher et al. reported that neither CRP or PCT were predictive of infection in 20 children on ECMO. In contrast, a recent report on 27 neonatal and pediatric ECMO runs where the proportion of positive cultures was very high (19% BSI) observed a sensitivity and specificity of a PCT threshold of 0.5 ng/ml to diagnose infection of 92%, and 43%, respectively. In two studies on 27 and 41 adult patients, a PCT cut-off of 2 ng/ml was reported as optimal to discriminate presence of infection vs. inflammation on ECMO, however was only predictive of infection on VA ECMO but not on VV ECMO. All studies were retrospective single center studies applying different definitions of infection, they reported on one or two markers rather than comparing CRP, PCT, and WCC, and none adjusted for repeated measurements.

Our findings are concordant with a number of studies showing that WBC have poor to moderate performance to discriminate infection in children (30, 37, 40, 41). In addition, it is possible that our approach to dichotomizing WBC (in order to capture both abnormal high, as well as abnormal low WBC) may have affected results. Of note, the diagnostic accuracy to diagnose infections on ECMO decreased when CRP and WBC were combined—which may be explained by overlaying areas of poor specificity with poor sensitivity. In addition, a multitude of non-infectious factors on ECMO may affect both high and low WBC counts, such as stress, steroids, massive transfusion, and chylus loss.

In our study, PCT was elevated in more than 75% of measurements to levels above the threshold commonly applied to diagnose infection in non-ECMO patients, and was unable to discriminate between confirmed or suspected infection and no infection. However, PCR values were available only for 43% of ECMO days, potentially biasing the findings. Nonspecific PCT increase has been observed in a range of settings such as ischemia, reperfusion injury or following cardiopulmonary bypass (24, 25, 42). This likely reflects the inflammatory response to the ECMO circuit immediately following cannulation and may be further aggravated by acute organ dysfunction, shock, and reperfusion injury. Exposure to non-epithelialized extracorporeal surfaces rapidly leads to systemic inflammation as a result of profound activation of cellular and humoral components of the innate immune system (43). We therefore hypothesize that the observed trend for PCT levels to decrease over time during ECMO runs may reflect the progression of the initial inflammatory response to ECMO. While infections are more commonly acquired after the first week on ECMO, our observations suggest that different infection marker thresholds may need to be considered depending on time spent on ECMO. However, our study was not powered to define infection marker thresholds specific for each week on ECMO, and the observed inter-patient variability was very high.

This study has several limitations. Due to the retrospective design, the definition of infection relied on a combination of microbiological findings, medical notes, and the initiation of antibiotic therapy for at least 5 days or until death. As clinical decision making was not blinded to infection markers, the reference standard was not totally independent of the index tests (incorporation bias), as requested by STARD (27). Moreover, the rate of confirmed BSI was low and the majority of infection episodes were classified as VAP. This may have impaired the quality of the reference standard, as differentiation of colonization and infection is challenging in pulmonary cultures. Aligned with CDC recommendations, positive endotracheal cultures in the absence of clinical or radiological signs were considered as colonization, however we cannot rule out that some of these findings resulted ultimatively in a true infection. We did not have access to colony forming unit counts. The application of VAP definitions to ECMO patients harbors further challenges as non-specific alterations on chest X-ray are common in patients with ultra long-protective ventilation on ECMO (10). Given the limitations of relying on microbiologically confirmed infections, additional analyses were performed on confirmed and suspected infections. In terms of index test measurements, while the institutional protocol during the study included CRP, PCT, and WCC in addition to blood cultures to be performed regularly, there was no protocol in place to define timing on the commencement or cessation of antibiotics at the institution, nor to define suspected infection prospectively. While WCC and CRP were performed almost daily, we identified fewer PCT values, which reduced the power of the study to analyze associations with PCT. Finally, we did not investigate immature to total ratio.

Our findings indicate that current infection markers are neither sensitive nor specific enough to reliably diagnose infection in neonates and children on ECMO. Difficulty in diagnosis combined with altered pharmacokinetics of medications on ECMO makes early effective treatment challenging. New biomarkers, such as host gene expression-based signatures, have been suggested to improve the guidance on antibiotic treatment decisions (44, 45). These markers have great potential to improve ECMO outcomes by facilitating earlier diagnosis of infection. Future research should develop and validate prediction models that estimate the risk of having a bacterial or fungal infection, based on all the available information including clinical signs and new infection markers. Implementation studies of these prediction models should then test whether the use of the models helps reduce unnecessary antibiotic exposure, resulting in improved antimicrobial stewardship at non-inferior safety.

In conclusion, while CRP performed best to diagnose infection for children on ECMO, its diagnostic accuracy was only moderate. Increases in WCC, and combinations of CRP and WCC or PCT, although commonly used in practice, did not result in improved diagnostic performance. Larger prospective studies are needed to test novel markers of infection to improve recognition of infection in patients on ECMO.

The datasets presented in this article are not readily available because requests for additional analyses on the data need clearance from the institutional ethics board. Requests to access the datasets should be directed to the corresponding author.

The studies involving human participants were reviewed and approved by HREC/17/QRCH/8. Written informed consent to participate in this study was provided by the participants' legal guardian/next of kin.

LJS designed the study, supervised analyses, interpreted the data, and wrote the first draft and the final manuscript. ID contributed to study design, collected all the data, performed initial analyses, and wrote with LJS the first draft. LPS and AS performed statistical analyses and contributed to manuscript writing. AM, EH, and NA were involved in study design, data interpretation, and manuscript writing. All authors read and approved the final manuscript.

LJS was supported by a Practitioner Fellowship of the National Health and Medical Research Council of Australia and New Zealand, and by the Children‘s Hospital Foundation, Brisbane, Australia. This study was performed at the Queensland Children's Hospital, Brisbane, Australia.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.