Prediction of Disease Severity in Young Children Presenting With Acute Febrile Illness in Resource-limited Settings

Overview

Note that this is a study that is co-sponsored by Medecins Sans Frontieres, Spain, and the University of Oxford. The primary objective is to develop a risk prediction algorithm, combining measurements of host biomarkers and clinical features at the point-of-triage, for children with an acute febrile illness in resource-limited settings. The secondary objectives are to determine which host biomarkers, feasible for measurement at the point-of-care, are predictive of disease severity. Additionally to determine the optimal combination of clinical features (including demographics, anthropometric data, historical variables, vital signs, clinical signs and clinical symptoms), feasible for assessment by limited-skill health workers, that is predictive of disease severity. The tertiary objectives are to explore the impact of different methods of outcome classification on development of the risk prediction algorithm, and to explore the performance of the algorithm to predict disease severity in key presenting clinical syndromes and aetiologies.

Background Febrile illnesses are amongst the most common reasons that parents seek non-routine healthcare for their children and a proportion progress to severe disease with substantial risk of mortality. Sepsis, defined as acute life-threatening organ dysfunction caused by a dysregulated host response to infection, carries significant morbidity. Incidence is highest in the paediatric age-range, with an estimated over four million children developing sepsis each year. This estimate is necessarily conservative, as the burden of sepsis is greatest in low- and middle-income countries (LMICs) where population-level data are not readily available. Distinguishing febrile children that require referral or admission to hospital, from those who can safely be cared for in the community, is challenging. In many tropical settings rational triage is especially difficult: healthcare providers receive limited training, and many acute febrile syndromes are clinically indistinguishable yet have different disease trajectories and require different interventions. Particularly in conflict settings, referral decisions involve complex mechanisms, costs and risks to both patient and provider. Consequently, patient outcomes are compromised: children with severe disease go unrecognised, whilst those with milder illnesses are unnecessarily hospitalised. Healthcare providers are trained to perform systematic clinical assessments, which focus on eliciting symptoms and signs that predict poor outcomes. Whilst tools such as the World Health Organization's Integrated Management of Childhood Illness (IMCI) aim to support this, results are inconsistent and adherence is poor. Integration of several syndrome-specific algorithms is impractical for many limited-skill health workers. A recent systematic review concluded that validity of existing paediatric triage tools is uncertain and that none are likely to be reliable in resource-constrained environments. Numerous clinical severity scoring systems predict deterioration in hospitalised patients and recent attempts have been made to adapt these to the paediatric population. However, many require physiological variables that are only routinely measured in the intensive care unit (ICU) and simplified versions perform inconsistently. In addition, the generalisability of these tools to LMICs is uncertain and further work is required to explore their utility at the point-of-triage. Rationale Spot Sepsis will collect the necessary data to permit a multi-country evaluation of existing severity scoring systems at the point-of-triage, and determine whether a combinatorial approach, utilising simple-to-elicit clinical features and measurement of host biomarkers, holds potential for widespread public health impact. Context-appropriate tools that enable reliable disease severity assessment in febrile children have great potential to improve patient outcomes and ensure optimal allocation of scarce resources. Pathogen-specific diagnostics alone are not sufficient to improve clinical management of febrile illness: the majority of children in LMICs present to facilities with limited diagnostic support and even well-resourced research studies often only ascertain a microbiological cause in a minority of febrile patients. Furthermore, in many circumstances, even when a pathogen is identified a child's illness trajectory can remain unpredictable. Characterising the host response represents an alternative strategy, which has potential to be robust to heterogeneity in fever aetiology and applicable to all febrile children irrespective of whether a microbiological cause of their illness can be identified. Given wide-ranging pathogen (aetiology and/or inoculum) and host (genetic and/or immunological) diversity, tools that combine multiple parameters are likely to be most effective: a uniform pathophysiological disturbance is unlikely. The Sequential Organ Failure Assessment (SOFA) score is proposed as a parsimonious tool to risk stratify adults admitted to the intensive care unit (ICU) with suspected infection. Recent work has adapted SOFA for paediatric use (pSOFA), with encouraging results. However both SOFA and pSOFA are validated for patients admitted to the ICU. Whilst quick-SOFA (qSOFA), a simplified version of SOFA consisting of three simple clinical variables (mental status, respiratory rate and systolic blood pressure), performs well in hospitalised adults outside of the ICU, it appears less useful in LMICs, when adapted for use in children and outside the hospital setting. Of note, abnormal systolic blood pressure is known to be a late sign in unwell children and sphygmomanometers are often not present outside of hospital settings in LMICs. Recognising that different clinical parameters may be relevant for febrile children at the point-of-triage in LMICs, a recent modified-Delphi process identified 45 possible predictors of paediatric sepsis feasible for collection in resource-limited settings. Which of these might be most important is yet to be determined. Simple clinical assessments (for example, the Lambaréné Organ Dysfunction Score \[LODS\]) predict in-hospital mortality in children with febrile illnesses in sub-Saharan Africa (SSA). In children with severe malaria performance of these assessments improves when combined with measurement of certain host biomarkers, such as those reflecting endothelial and immune activation, including angiopoietin-2 (Ang-2) and soluble triggering receptor expressed on myeloid cells-1 (sTREM-1). Importantly, this contrasts with laboratory indicators such as venous lactate, which although predictive of mortality in children with febrile illness, does not improve discrimination of clinical assessment.30,31 This suggests that whilst conventional laboratory indicators may be crude biochemical surrogates for clinical assessment, certain biomarkers may reflect sub-clinical endothelial and immune activation, and have potential to improve identification of children at risk of severe febrile illness. Most studies to date have focused on malaria in African children. Limited data exists from Asia, and non-malarial febrile illnesses are comparatively understudied. However, recent work suggests endothelial and immune activation play a central role in the pathophysiology of severe infection across a spectrum of microbial aetiologies, including dengue, bacterial sepsis and influenza. Furthermore, whilst most studies have enrolled hospitalised children, a recent study recruiting adults attending Tanzanian outpatient clinics, suggests that this approach could aid triage when patients first present to care from the community. Assumptions, limitations and generalisability The fundamental assumption underlying this approach is that there are final common pathways to severe febrile illness and that markers of these pathways will be elevated and measurable at a sufficiently early point in the course of an illness, to meaningfully contribute to patient assessment and triage. Whilst this is known for clinical parameters such as vital signs, a growing body of evidence supports this thesis: endothelial and immune activation appear to be part of a final common pathway to severe illness. Furthermore, alterations in microvascular physiology occur early in the disease course of common childhood febrile illnesses. This study will recruit children aged \> 28 days and \< 5 years presenting with an acute febrile illness. Hence, the major limitations are the exclusion of neonates and older children, and the decision to base the eligibility criteria around fever. This will limit the ability to develop a parsimonious tool for all children presenting with suspected infection in the community. The decision to limit recruitment to children aged \> 28 days and \< 5 years reflects the fact that, (a) all febrile neonates require further assessment, (b) outside the neonatal range, the biggest burden of disease affects children under the age of five, and (c) including older children would require substantially greater resources to ensure power to examine interaction of predictive performance with age. The decision to limit the study to children with an abnormal temperature or short history of fever reflects the fact that, (a) these inclusion criteria are expected to capture the majority of children with suspected sepsis (TuNDRA study, Angkor Hospital for Children \[OxTREC 512-17; 287 NECHR\], unpublished data), and (b) limiting this initial study to the febrile population will constrain heterogeneity, increasing likelihood of success whilst still addressing an important public health problem. This study aims to develop a risk prediction algorithm to improve disease severity assessment of febrile children in rural and/or remote community settings and decentralised models of care. However, derivation of a prediction algorithm requires a certain number of 'outcome events' (episodes of severe illness), and hence recruiting febrile children presenting at peripheral levels of the health system would be logistically and financially challenging. To overcome this challenge, Spot Sepsis will recruit patients presenting to mid-level health facilities, for example, district and/or provincial hospitals, where severe illness occurs more frequently. However, the investigators recognise that this compromise risks a potential loss of generalisability to community settings (the ultimate intended-use setting for the algorithm). To mitigate this risk, the sites selected for Spot Sepsis serve as a primary point of healthcare access for a predominantly rural and/or underserved population, the demographics of which are representative of patients presenting to lower levels of care. Hence the primary difference between the Spot Sepsis sites and eventual intended-use sites, is the frequency with which children at risk of severe illness attend, rather than systematic differences in their demographic characteristics (acknowledging that there can be differences in patients presenting to hospital outpatient departments compared to community care settings). This will maximise the chance of successful out-of-sample validation and generalisability of the tool to community settings. NOTE In addition, a greater understanding of biomarker kinetics is crucial. To investigate this further the investigators will perform an exploratory nested case-control sub-study at the Cambodia site to investigate whether the kinetics of the biomarker response differs between hospitalised children who develop severe illness and hospitalised controls. Furthermore, findings that the angiopoietin axis can remain activated beyond apparent clinical resolution, and conveys excess post-discharge mortality risk, is concordant with high post-discharge mortality observed in adults with bacterial sepsis and in hospitalised children in LMICs. If confirmed, measuring biomarkers of endothelial and immune activation at discharge could help prioritise at-risk children for follow-up. At the Cambodia site biomarker levels will be measured at discharge and patient follow-up extended to six months to determine if children at risk of poor outcomes can be readily identified prior to discharge and prioritised for outpatient or community-based follow-up.