This chapter contains reproduced material from Waldron et al.1,2 These are Open Access articles distributed in accordance with the terms of the Creative Commons Attribution (CC BY 4.0) licence, which permits others to distribute, remix, adapt and build upon this work, for commercial use, provided the original work is properly cited. See: https://creativecommons.org/licenses/by/4.0/. The text below includes minor additions and formatting changes to the original text.
The PRECISE study was an embedded mechanistic sub-study within the Biomarker-guided duration of Antibiotic Treatment in Children Hospitalised with confirmed or suspected bacterial infection (BATCH) trial (ISRCTN No:11369832). In the PRECISE study, we aimed to identify if a PCT-guided antibiotic algorithm is beneficial, harmful or ineffective in certain subgroups of paediatric patients with suspected or confirmed bacterial infection. We aimed to identify subgroups that do not benefit from the intervention, so potential harm is avoided, resources are not wasted, and treatment for subgroups that do benefit is optimised.
The host-response to bacterial infection is highly heterogeneous; there is therefore a need for stratified treatment strategies instead of the traditional ‘one-size-fits-all’ approach. A precision medicine approach can improve patient outcomes by identifying endotypes in whom the intervention is likely to be beneficial, harmful or ineffective. From a clinical point of view, it is important to identify subgroups in whom a PCT-guided antibiotic algorithm might potentially be harmful, so that alternative treatments and diagnostic strategies could be tried.
Embedding mechanism evaluation within an RCT can provide important information about the underlying mechanisms of an intervention. A growing body of evidence suggests that clinical trial designs in sepsis should include biomarkers which reflect the status of the host immune response.3,4 This could identify molecular endotypes for whom future targeted adjunctive therapies might be beneficial.
The endotypes to be examined in the PRECISE study were stratified by endothelial dysfunction (MR-proADM) and host immune response (host response of three blood biomarkers: TRAIL, IP-10 and CRP). These biomarkers have immunomodulatory (MR-proADM, IP-10), immunosuppressive and immunoregulatory effects (TRAIL), as well as playing a part in both the adaptive and innate host immune response to infection (TRAIL, IP-10, CRP).5–11 This provides potential for targeted patient management and enhanced clinical trial design.
The health problem being addressed in the BATCH trial
Sepsis is defined as life-threatening organ dysfunction caused by a dysregulated host response to infection.12 Sepsis causes many non-specific symptoms and signs that can also be caused by a large number of conditions that may or may not be due to infection, and that may or may not require immediate or urgent treatment. The problem for clinicians is the difficulty in distinguishing bacterial sepsis from other conditions presenting with non-specific signs and symptoms. Prompt administration of antibiotics reduces mortality, but indiscriminate antibiotic use increases antimicrobial resistance, resulting in increased costs in hospitalised patients.13,14 Severe sepsis accounts for approximately 45% of intensive care unit (ICU) bed-days and 33% of hospital bed-days, representing a significant resource burden in the NHS. Not all children admitted with bacterial infection will meet the criteria for sepsis, but they could still have serious infection, requiring IV antibiotics for several days.
Limitations of currently used biomarkers
Blood tests currently used in the NHS, such as CRP do not reliably differentiate between serious bacterial infection (SBI) and inflammation, and show a delayed response (12–24 hours) to bacterial infection. Procalcitonin (PCT) is a biomarker released in response to inflammatory stimuli including bacterial infections, with very high levels produced in SBI.15 In contrast to CRP, PCT rises early (within 6–12 hours) and peaks early, falling rapidly in response to effective antimicrobial therapy. This makes blood PCT potentially a better biomarker for monitoring progression of SBI and response to antimicrobial therapy, and for informing initiation, change or discontinuation of antimicrobial therapy. There are no evidence-based biomarker-guided algorithms to support antimicrobial stewardship (AMS) in adults or children, and the BATCH trial sought to determine if PCT can help deliver shorter, yet equally safe duration of antibiotics to treat SBI in children.
The context of the PRECISE study in terms of current practice
Current practice on duration of IV antibiotics for SBI depends largely on local practice and clinician preference, with little evidence to guide duration. The BATCH trial had the potential to impact the clinical care of hospitalised children with confirmed or suspected SBI, which currently accounts for a large proportion of antibiotic use in hospitalised children.
In children hospitalised with confirmed or suspected bacterial infection, the BATCH trial demonstrated that the addition of a PCT-guided algorithm to usual care (UC) is non-inferior in terms of safety, but does not reduce duration of IV antibiotics, and is not cost-effective. In the presence of robust AMS programmes to reduce antibiotic use, a PCT-guided algorithm added little value.1,16 A total of 1949 children were recruited; 977 were randomised to the PCT group [427 female (43.7%), 550 male (56.3%)], and 972 to the UC group [478 female (49.2%), 494 male (50.8%)]. The PCT-guided algorithm was non-inferior (with respect to a pre-defined absolute 5% non-inferiority margin) to usual standard care in terms of a composite safety outcome {risk difference −0.81% [upper 95% confidence interval (CI) bound: 1.11%]}, but duration of IV antibiotics was not significantly different between the PCT group (median 96.0 hours) and the UC group (median 99.7 hours) [hazard ratio (HR) 0.96 (95% CI 0.87 to 1.05)]. The intervention was not cost-effective as it was more expensive with no significant improvement in IV antibiotic duration, even though it resulted in a non-significant improvement in health-related quality of life.1,2,16
The PRECISE study takes this initiative one step further by seeking to make a step-change in the application of precision medicine by targeting PCT-guided antibiotic algorithms to those patients who, on the basis of host-response endotypes, are likely to benefit from the approach. Current practice in the management of SBI is to treat syndromic infections with a ‘one-size-fits-all’ approach; for example 10 days IV antibiotics for culture-negative meningitis, 3–4 days IV antibiotics for osteo-articular infections, 5–7 days IV antibiotics for severe pneumonia.17 Reliance on CRP leads to over-treatment as it is slow to fall in response to antibiotics.15 Given that the host-immune response, rather than the infecting bacterial load, is what determines clinical response, an approach which stratifies IV antibiotic duration on the basis of immune status, is likely to achieve the best outcomes for patients.
The knowledge gap: precision medicine for optimisation of clinical trials
Towards precision medicine in sepsis: a position paper from the European Society of Clinical Microbiology and Infectious Diseases was published in December 2018 and states that
key to a precision medicine approach is the accuracy of biomarkers to stratify patients into subgroups based on specific pathophysiological features. A biomarker can aid in a theranostic approach to select a specific therapy for an individual patient and simultaneously be used to measure response to treatment.18
It goes on to argue that AMS is an important opportunity for the successful implementation of precision medicine in sepsis. Drug-response phenotypes can be identified by examining interactions between sepsis endotypes and response to sepsis therapies in order to optimise clinical trials and ultimately diagnostic and care pathways and patient management. An endotype is a biological subtype defined by distinct pathophysiologic mechanisms within a phenotype (e.g. bacterial/viral host response, adaptive/innate host immune response, organ dysfunction/no organ dysfunction). It recommends validation studies to determine the robustness of such endotypes for sepsis prognosis.18 This study aimed to address that need by seeking to identify sepsis endotypes that might have better or worse outcomes with PCT-guided antibiotic algorithms, or whose outcomes could be improved by further risk stratification in PCT-guided antibiotic algorithms. Since the BATCH trial was funded, there has been new published data suggesting that two newer biomarkers/biomarker panels could further improve stratification of patients treated with IV antibiotics that may benefit from PCT-guided antibiotic decisions. The biomarkers selected for stratification in the PRECISE study have immunomodulatory (MR-proADM, IP-10), immunosuppressive and immunoregulatory effects (TRAIL), as well as playing a part in both the adaptive and innate host immune response to infection (TRAIL, IP-10, CRP). This provides potential for targeted patient management and enhanced clinical trial design.
Precision medicine does not only pertain to genetic predisposition and epigenetic changes influencing choice of drug therapy, but with regards to sepsis, serves to improve understanding of the host-pathogen interaction and how a particular individual might benefit from current or novel targeted therapies and diagnostics.19 Precision medicine using baseline biomarkers has the potential to risk-stratify patients allowing accurate predictions of response and prognosis, which could guide clinicians’ decisions regarding targeted management, and employ particular adjunctive therapies to augment the host immune response. Diagnostics which are measured serially as part of an algorithm, such as that used in the BATCH trial can support timely clinical decisions about the management of antibiotics in patients with sepsis. In the future, baseline biomarker panels at the bedside could risk stratify patients presenting with sepsis to antibiotics and precision-directed therapy representing a shift from treating sepsis using a ‘one-size-fits-all’ approach to using a precision-directed approach, with could lead to an improvement in sepsis outcomes.20
There are three recent studies from different parts of the world which suggest that phenotyping sepsis allows for a precision approach to resuscitation and management. The study by Zhang et al. on adults in an ICU in China demonstrated that distinct sub-phenotypes of sepsis exist which showed different mortality outcomes and responses to fluid resuscitation.21 The study by Seymour et al. on adults with sepsis in the US identified four clinical phenotypes that correlated with host-response patterns and clinical outcomes, and simulations suggested these phenotypes may help in understanding heterogeneity of treatment effects.22 The study by Levin et al. in children, mostly from sub-Saharan Africa used physiological scores, haematological and biochemical data to describe three sub-groups with differing responses to bolus fluid administration.23
The PRECISE study aims to identify endotypes or sub-phenotypes of infection to facilitate optimisation of antibiotic dosing and duration (when, by how much and for how long). The embedded mechanistic sub-study within an existing RCT allowed the theranostic exploration of host–response biomarkers to guide judicious antibiotic use.
Mid-regional pro-adrenomedullin
Adrenomedullin (ADM) is a potent vasodilator with an important role in maintaining the microcirculation and endothelial function. ADM levels are unmeasurable biochemically because of a very short half-life, but mid-regional pro-adrenomedullin (MR-proADM) is a fragment of 48 amino acids which splits from the final proADM molecule in a ratio of 1 : 1 with ADM and represents the levels and activity of ADM, as a stable and reliable surrogate marker for ADM release.24
Mid-regional proadrenomedullin levels rise rapidly during the initial stages of sepsis and other infections and have demonstrated significant clinical utility in the early risk stratification of patients with sepsis.25 The persistence of high values or the reduction in clearance of MR-proADM at day 2 and day 5 following admission may indicate that it is persistently being released due to ongoing organ dysfunction, despite clearance of infection. Decreased clearance of MR-proADM is associated with death in sepsis.24 In a study of adult patients with sepsis, MR-ProADM, PCT, lactate and CRP were measured in plasma on days 1, 3 and 7 following admission to the ICU. MR-proADM was the only biomarker showing significant differences in concentration between survivors and non-survivors at all three time points and was the biomarker showing the best diagnostic accuracy for predicting mortality in ICU.26 A proof-of-concept RCT in adults with lower respiratory tract infection (LRTI) evaluated the adding of MR-proADM to an established guideline for clinical decisions, and demonstrated a shorter length of stay for patients in the intervention group.27
A multicentre trial of adults with LRTI demonstrated additional value of serial MR-proADM over baseline measurements. At day 3, using a cut-off of 1.5 nmol/l, the highest mortality was observed in patients with high/high or low/high levels at baseline and day 3 respectively. This suggests that MR-proADM could be used to stratify groups of patients in whom PCT-guided antibiotic duration can be aggressively applied (low risk), and those in whom a more cautious approach, or a combination of MR-proADM and PCT might be useful (high risk).28
In a secondary analysis of a large RCT of adult patients with sepsis and septic shock, mortality in patients with decreasing PCT concentrations (≥ 50% from baseline to day 4), was reduced significantly when patients had continuously low MR-proADM levels (mortality 17% to 1.8%), but increased significantly when patients had continuously high MR-proADM levels (mortality 17% to 53.8%). The cut-offs used in this study were: low (≤ 2.75 nmol/l) and high (> 10.9 nmol/l). Patients with decreasing PCT values of ≥ 50% from baseline to day 4, but continuously high MR-proADM levels had a significantly higher risk of developing nosocomial infection, and those with increasing MR-proADM levels were more likely to need source control procedures.29
A study from our group of febrile children attending the emergency department (ED) reported that high MR-proADM and high PCT were associated with increased likelihood of subsequent disease progression. Incorporating MR-proADM into clinical risk stratification may be useful in clinician decision-making regarding initiation of IV antibiotics, fluid resuscitation, and escalation to paediatric intensive care unit (PICU)/high dependency unit (HDU) admission.7
These studies suggests that serial measurements of MR-proADM might allow stratification of host response, allowing on one hand, earlier escalation of therapy in patients at risk of treatment failure, and on the other, de-escalation of therapy and early discontinuation of antibiotics in low-risk patients. Therefore, additional stratification of the BATCH cohort using MR-proADM may provide additional benefit to the PCT-guided algorithm.
For the BATCH cohort, in the low MR-proADM group, the PCT-guided antibiotic algorithm might suggest stopping IV antibiotics after 3 or 4 days, whereas in the control group, the low MR-proADM might suggest low risk of organ dysfunction and a prompt switch to oral antibiotics after 2 days. Conversely, as demonstrated in the RCT of adults with sepsis,29 the high MR-proADM group, the PCT-guided antibiotic algorithm might recommend stopping IV antibiotics if PCT decreases by ≥ 80%, yet there could still be significant risk of organ dysfunction and nosocomial infection, which could lead to patient harm if antibiotics were stopped too soon. The additional information which stratification could provide is important for patients (as it could protect them from harm), and important for clinicians as it gives them increased confidence to stop IV antibiotics early without adverse effects, or conversely to treat for longer to improve patient outcomes.
Host proteins comprising TRAIL, IP-10 and CRP
A host-protein signature including both viral- and bacterial-induced proteins, TNF-related apoptosis-inducing ligand (TRAIL), interferon gamma-induced protein-10 (IP-10) and CRP, was shown to be superior to other lab parameters and protein biomarkers in differentiating between bacterial and viral infection at diagnosis.30 Patients with an index test score below 0.35 were classified as likely viral or other host immune response, between 0.35 and 0.65 as equivocal, and above 0.65 as likely bacterial host immune response.30 Unfortunately, we were unable to use the proprietary signature from MeMed Ltd. (MeMed BV, Tirat Carmel, Israel) as it required serum samples and we only collected plasma samples in the BATCH trial. We therefore investigated the interaction between randomisation arm each of the proteins separately.
TRAIL is a member of the tumour necrosis factor (TNF) superfamily of cytokines and plays an important role in the induction of cell death by apoptosis, thereby inhibiting virus spread from infected cells. TRAIL is induced in response to a wide range of viral infections but is reduced in bacterial infections.8 TRAIL has been demonstrated to exert immunosuppressive and immunoregulatory effects which are important for the innate-to-adaptive immunity transition.8
Interferon gamma-induced protein-10 secretion is induced by IFN-γ and other endogenous cytokines, including TNF-α. Exogenous lipopolysaccharide stimulation can also play a secondary role in inducing IP-10 secretion.31 Levels of IP-10 have been shown to be elevated in many viral infections, including severe influenza infection, suggesting the role of this chemokine in immune response to viral infection.32 IP-10 or CXCL10 exerts its biological effects by binding to CXCR3, a seven trans-membrane spanning G protein-coupled receptor. It is chemoattractant for CXCR3-positive cells, including macrophages, dendritic cells, natural killer cells and activated T lymphocytes. The powerful chemotactic action of CXCL10 on activated lymphocytes allows it to modulate both innate and adaptive immunity. IP-10 has been demonstrated to play a role in both bacterial and viral infections including Mycoplasma infection, Legionnaire’s disease and tuberculosis as well as in rhinovirus, respiratory syncytial virus, coronavirus, coxsackie virus, hepatitis virus B and C and dengue.33
The TRAIL, IP-10 and CRP host-protein signature exhibited the highest diagnostic performance for distinguishing between bacterial and viral aetiologies in adult and paediatric patients presenting with respiratory infections and fever without source.34,35 The signature was significantly more accurate than CRP alone, PCT, and routine laboratory parameters in febrile children presenting to the ED.36 Of particular relevance to the BATCH trial, the study by Srugo et al.,36 which was conducted in Switzerland and Israel, recruited children that had received antibiotics for up to 48 hours. These cases are routinely observed in real-life clinical settings and their inclusion enhances the generalisability of the signature.36 The BATCH cohort of children fall into this category. The host-protein signature has never previously been trialled in patients to assess clinical utility in guiding IV antibiotic duration.
In a large study of childhood life-threatening infectious diseases across Europe (including the hospital where the CI is based in England), with over 2800 children, laboratory tests did not identify a causative pathogen in over half of children despite the introduction of increasingly sensitive and precise diagnostic techniques. This means that in more than 50% of paediatric patients admitted to hospital with SBI, decisions on the need, type and duration of antimicrobial therapy thus have to be made without clear guidance from microbiological findings.37 About 10% of children in that cohort had infection of either definite viral or probable viral aetiology. This suggests that of the BATCH cohort with suspected bacterial infection, being treated with IV antibiotics, there may be a significant proportion with viral aetiology, in whom a PCT-guided antibiotic algorithm may be of limited value. If a bacterial/viral test could be performed at admission or at randomisation, then it could identify a sub-group of patients in whom antibiotics could and should be withheld altogether or stopped after 24 hours (high likelihood of viral infection). The current PCT-guided algorithm would probably recommend stopping soon after randomisation, that is, after 2 or 3 days of IV antibiotics. In the high likelihood of bacterial infection sub-group, the current PCT-guided algorithm might recommend stopping after 5 days of IV antibiotics, whereas if there is very high likelihood of invasive bacterial infection, it might be safer to give 7–10 days IV antibiotics.
Why this research would lead to improvements in patient care
Precision medicine has been applied successfully to other fields including small molecule inhibitors in certain cancers and monoclonal antibody therapy in allergic asthma, whereby it targets specific subgroups of individuals with a disease. A tailored approach (1) minimises the trial and error approach that often results in delays in administering effective therapy; (2) minimises adverse drug reactions (ADRs); and (3) allows more accurate monitoring of response to therapy. Improvement in the stratification of patients with sepsis will provide new opportunities for current therapies within specific subgroups, allowing the targeting of specific pathways in the correct group of patients with sensitive end points. Implementation of improved risk stratification will allow better monitoring of response to antimicrobial therapy, and identify sub groups that might benefit from adjunctive therapies.38 Better targeting of diagnostic algorithms, such as PCT-guided antibiotic duration, could improve outcomes.
For the BATCH cohort, further characterisation of endotypes with a predominantly bacterial or viral host response, or with significant organ dysfunction, will help identify patients who are more or less likely to benefit from a PCT-guided antibiotic algorithm. Precision medicine moves away from a ‘one-size-fits-all’ approach, where the clinical utility and effectiveness of a PCT-guided antibiotic algorithm are tested in all children on IV antibiotics for suspected or confirmed bacterial infection, to one where clinical utility and effectiveness are tested in defined endotypes, in order to determine the patient sub groups in whom such a management strategy will yield the greatest benefit to patients or could cause harm. Conversely, in those sub groups were there is no benefit, costs could be saved.
From a parental perspective, if treatment can be more patient specific and IV antibiotics identified sooner as being effective, then patients can potentially stop IV antibiotics (or switch to oral antibiotics) quicker and be discharged from hospital earlier, than the current treatment pathways allow. This will lessen the emotional and financial strains experienced by parents and carers while their child is hospitalised. While the main BATCH trial used PCT samples serially as part of an algorithm, in the PRECISE substudy we were unable to measure serial biomarkers in the laboratory due to technical issues with the assays, and therefore only the baseline values were used in the model.