Mid-Regional Pro-Adrenomedullin in combination with Paediatric Early Warning Scores for risk stratification of febrile children presenting to the Emergency Department:Secondary Analysis of a Non-Prespecified United Kingdom Cohort Study

In this issue, Lenihan and colleagues (1) evaluate the utility of adding the blood biomarkers mid-regional pro-adrenomedullin (MR-proADM) and procalcitonin (PCT) to the Pediatric Early Warning Score (PEWS) to risk stratify children who present to the emergency department (ED) with fever. They performed an unplanned secondary analysis of a study of 1183 febrile children < 16 years of age who underwent blood sampling as part of their ED evaluation. The outcomes included intensive care unit (ICU) admission, need for fluid resuscitation, and presence of definite or probable bacterial infection as defined by an algorithmic approach. Primary study enrollment occurred at a single center between 2010–2012, and blood samples were banked at that time and later used for this analysis (2).

The authors’ attempt to identify prognostic biomarkers in a broad cohort of febrile children is laudable, as febrile illness continues to be one of the most common reasons for children to seek emergency care, and recognition of sepsis/organ dysfunction and determination of infectious etiology are ongoing challenges for emergency providers. In addition, the concept of using biomarkers to augment a clinical warning score is a strategy that has precedent in the literature (3,4) and is likely to prove even more successful in coming years, particularly as increasingly granular clinical data and “omic” level biological data becomes feasible to use in the clinical space.

Similar to prior work in adults and children with sepsis and lower respiratory tract infections (5,6,7) these investigators found that elevations of both PCT and MR-proADM were associated with PICU admission, fluid resuscitation, and presumed bacterial infection, while elevation of MR-proADM in the setting of low PCT was associated only with fluid resuscitation. These results may be explained by the fact that the association between MR-proADM and disease severity is likely less etiology-dependent than PCT. In adults with CAP, MR-proADM levels were similar among patients infected with viruses, typical bacteria, and mixed infections, with no association between MR-proADM and etiology (8). This contrasts with PCT which tends to be elevated more frequently in bacterial infections, as virus-stimulated macrophage synthesis of interferon-alpha inhibits TNF synthesis and PCT expression (9, 10). As most febrile illnesses in children are viral and children can have severe illness due to viral infections, MR-proADM will theoretically be elevated in severe disease (e.g., need for fluid resuscitation) regardless of etiology.

There are several challenges to implementing MR-proADM into routine practice. First, it is not widely available. While this study and others suggest possible utility of MR-proADM, it cannot be used unless it is demonstrated to be clinically acceptable and widely available. Second, unified cutpoints have not been defined, particularly for children. In one study of adults, a cutpoint of 1.54 nmol/L was found to best discriminate disease progression, while a cutpoint of 0.9 nmol/L was used for disposition decisions (11). In one pediatric study evaluating the use of MR-proADM to predict severe disease, a cutpoint of 0.66 nmol/L was found to be the best discriminator of disease severity (6), while in the present study by Lenihan, a cutpoint of 0.7 nmol/L was used. Until cutpoints are established and the assay becomes more generalizable, implementation of MR-proADM will remain difficult.

This study has several challenges inherent to retrospective designs evaluating risk of severe illness in children with fever. First, the study population was limited to children having bloodwork performed in the ED. Thus, there is already an element of clinical judgment involved in determining need for blood-based diagnostics (i.e., confounding by indication) that may make it difficult to generalize their findings to all children with fever seeking emergency care. Second is the question of what to do when data elements are missing or present in different frequencies, as elements such as vital signs and laboratory results are often missing in a non-random fashion, which could lead to biased results. These investigators presumed that missing values were normal, which is a reasonable approach based on routine clinical practice, and is likely the best course of action given the limitations of retrospective data. However, tools such as sensitivity analyses can be helpful to model additional patterns of missingness to assess the degree of bias introduced by missing data. Third is the question of which outcome to use, with the goal of choosing actionable items that are also patient-centered. Because mortality is typically quite low in ED-based studies, investigators often choose one outcome that seeks to identify hemodynamic and/or organ dysfunction risk and a second that seeks to identify pathogen source or etiology. Although there are likely distinct biologic elements in each of these pathways, deciding which organ dysfunction outcomes to choose may be challenging, unless there are specific biologic pathways being interrogated. These authors chose ICU admission and need for fluid resuscitation as proxies for organ dysfunction, given that organ dysfunction is a requirement for ICU admission at their institution. However, both of these outcomes are somewhat subjective, often representing clinician concern for organ dysfunction as opposed to being a direct measure. Fluid resuscitation is particularly challenging as an outcome measure in this space because although fluid certainly is a treatment for shock, it is also a treatment for dehydration that may not indicate organ dysfunction.

Another important limitation of this work is around the timing of when prediction (i.e., measurement of PEWS and biomarkers) occurred and when the outcomes of fluid administration and ICU disposition decisions were made. It is important when evaluating studies attempting to stratify risk of critical illness to determine whether they are identifying risk in children who already have organ dysfunction upon ED arrival (and thus there is nothing to predict since the outcome is already present) vs. predicting organ dysfunction in children who do not yet show signs. The latter can be challenging in single center work because organ dysfunction is relatively rare and larger numbers are often needed to make meaningful conclusions. In this study, the authors concluded that MR-proADM was more likely to be associated with organ dysfunction (12) and PCT was more likely but not exclusively to be associated with infection type (13), but this was not specifically stated or tested. The analysis seemed to use these markers interchangeably, as opposed to testing associations of each marker with distinct outcomes.

Similar challenges around missing data arise when considering infection type, as typically in clinical practice not all patients have comprehensive viral and bacterial testing performed. This, while clinically appropriate, can make it difficult to determine a true reference standard of bacterial vs viral etiology. This is particularly true in this study, where the authors used an elevated inflammatory marker (C-reactive protein >60 mg/L) to define probable bacterial infection in those with negative cultures, and a low inflammatory marker (C-reactive protein <60mg/L) to define likely viral infection. This can lead to circular reasoning when interpreting PCT results, and brings serious questions to the results around infectious source.

Two important elements to consider when contemplating bringing clinical risk stratification tools to the bedside are the clinical goals of the tool and the population in which the tool is likely to be used. For example, the authors highlight the fact that low biomarker scores in this context may be helpful in identifying low risk children who perhaps do not need ICU care, as opposed to identifying high risk children who may benefit from additional precision-based care. Because of the relative infrequency of severe organ dysfunction in the ED setting, these broad-based populations are key to developing prognostic enrichment tools, but may be less valuable for predictive enrichment strategies (14). This highlights a key area for future collaboration between ED and ICU investigators--where prognostic strategies in the ED setting are important for identifying the spectrum of illness and the frequency of high risk patients, and where predictive enrichment occurs once a patient requires ICU care and can be used to guide novel precision-based therapies.