Novel Tools to Improve Management of Paediatric Community-Acquired Pneumonia - ToolCAP

In 2019, lower respiratory tract infections (LRTI) caused 2.6 million deaths worldwide, making them the fourth leading cause of death overall. While the statistic is global the effect is concentrated in low-resource settings and among children, with pneumonia being the single largest infectious cause of death in children worldwide.

LRTI (including pneumonia) are the most common reason for sick children to present for acute outpatient care and current best practice pneumonia management guidelines advocate for all cases to receive a course of antibiotics. This relies on the World Health Organization's (WHO) Integrated Management of Childhood Infections (IMCI) guidelines, first developed in the 1990s - a time when providing access to antibiotics was a major goal for public health programs. Today, antibiotics are readily available in most countries in sub-Saharan Africa (SSA) and antibiotic over-prescription has become a public health crisis. The IMCI guidelines were re-iterated in 2014 but still rely on presumptive treatment based on clinical signs alone (cough, respiratory rate, and lower costal indrawing), as evidence to include additional diagnostic tests into the IMCI approach has been lacking. However clinical features inadequately distinguish bacterial infections and complications from common, self-limiting viral infections. Furthermore, recent evidence estimates the incidence of bacterial respiratory infections in SSA, using a combination of microbiology, chest x-ray (CXR), and clinical outcomes as reference standards, is as low as 2-4% in primary, and 23.3-31.6% in secondary, care. Therefore, approximately 9 out of 10 courses of antibiotics recommended by current guidelines for children with LRTI are estimated to be unnecessary.

Bacterial antimicrobial resistance (AMR), responsible for 1.27 million deaths in 2019 and with the highest burden in SSA, is thus of increasing concern; with nearly as many deaths as malaria and HIV combined. Inappropriately and excessively prescribing antibiotics represents one primary contributor of bacterial AMR. In SSA, more than 50% of children who are sick receive antibiotics when visiting health facilities with 80-90% of such antibiotics prescribed at the outpatient level and most deemed inappropriate. Most inappropriate antibiotic use occurs with respiratory infections due to systematic overprescription as outlined above. Antibiotic use and AMR are projected to increase over the next years, indicating urgent action. Accordingly, WHO declared AMR as "one of the biggest threats to global health, food security and development today.". Effective solutions to improve antibiotic stewardship for childhood infections in SSA primary care settings remain lacking, though its well recognized improved diagnostic and management processes are essential.

Lacking more effective tools to safely identify the minority of children requiring antibiotics, frontline clinicians in low resource settings therefore prescribe antibiotics to nearly every child, driving the above-described overuse and fuelling local AMR, whilst significantly overtreating cases of viral pneumonia. Besides the risk of AMR at the population level, reducing antibiotic prescriptions is also important for each individual patient. Antibiotic exposure early in life has been associated with an increased risk of health conditions, including asthma, allergic rhinitis, atopic dermatitis, autoimmune disease, obesity, and neurodevelopmental disorders. Improving IMCI diagnostic criteria to better identify children with pneumonia that would benefit from antibiotic therapy is therefore a WHO research priority.

Several tools have since been proposed (currently at different stages of diagnostic development) to both improve diagnostic pathways and improve appropriateness of antimicrobial prescriptions in low and middle income countries. These include the use of point of care C-reactive Protein (CRP), Procalcitonin (PCT), comprehensive electronic decision support algorithms, and more novel applications of established technologies. Two such technologies are Lung Ultrasound (LUS) and Lung Auscultation (LAusc). LUS is a well-established, near consumable-free, and non-invasive point-of-care respiratory exam. While LUS is less ubiquitous than the stethoscope, it's new portable and affordable ultrasound-on-a-chip design, pluggable into a mobile device, has the potential to be integrated into the standard clinical exam without incurring extra costs, time, radiation, or specialist consultation. These portable ultrasound devices have reached regulatory approval and are used in medical care across SSA. This together with increasing evidence showing its ability to effectively detect lung consolidation in pneumonia have made it an increasingly attractive tool for frontline clinicians; already becoming an established practice in the outpatient case management of children with respiratory infections in many high-resource settings as well as growing steadily in popularity in SSA.

A number of studies have compared LUS to CXR and/or Computerised tomography (CT) for pneumonia diagnosis. A multi-centre trial in Europe including 362 adult patients with pneumonia compared LUS performed by a specialist to a diagnosis of pneumonia made by CXR and in some cases CT. The authors reported a sensitivity of 93.4% (95% Confidence Interval (CI) 89.2-96.3%) and a specificity of 97.7% (95%CI 89.2-99.6%). Most studies in children (performed primarily in high-income countries and tertiary care settings) have found high diagnostic accuracy when comparing LUS to CXR, using either CXR, CT, or Magnetic Resonance Imaging (MRI) as reference standards, with a recent meta-analysis suggesting an overall sensitivity of 95% (91-97%) and specificity of 96% (90-99%). A smaller number of studies have investigated using LUS to assist in determining the etiological causes of the consolidations. Although these have shown a possibility of differentiating viral from bacterial infections, the reference standards used were nasopharyngeal swab carriage, clinical syndromes, and other clinically available data. A notable limitation of this study was therefore the lack of a microbiologically confirmed diagnosis, making it hard to make definitive statements about the ability of LUS to differentiate. This lack of reliable reference standards in paediatric pneumonia is a well-documented challenge, most notably discussed in the Pneumonia Etiology Research for Child Health (PERCH) study, and thus limits the value of etiology focused diagnostic accuracy studies. Although the first studies on the use of LUS to detect lung consolidation in children in LMICs are promising, as seen in Nepal, Peru, and Egypt, they remain limited in scope and context, with no evidence on the clinical impact of LUS assisted diagnoses on antimicrobial use or health outcomes, particularly in a primary care setting. This highlights the importance of the assessment of LUS integrated into care to assess for real changes in treatment outcomes through an interventional study, rather than observational diagnostic accuracy studies.

Studies assessing the potential of LUS in the diagnosis of pulmonary tuberculosis (TB) in children are equally limited. With paediatric TB diagnosis remaining a huge challenge in low resource settings, many cases still remain undiagnosed and untreated - an issue being addressed though the development of pragmatic TB Treatment Decision Algorithms for settings with and without CXR. These Treatment Decision Algorithms are now included in the operational handbook accompanying the WHO guidelines on the management of TB in children and adolescents. More evidence is however needed on the potential role of LUS as an alternative diagnostic modality in the context of these TB Treatment Decision Algorithms.

Regarding operability of LUS in SSA by novice health workers, LUS performance evidence shows it is possible to capture images/video with minimal training and achieve high diagnostic quality, as evidenced in both South Sudan and Bangladesh, though concerns remain regarding diagnostic quality outside research settings. Image acquisition aside, interpretation and inter-user bias remain the largest concern, this however makes them excellent candidates for automated interpretation by objective computer-aided pattern detection. Using deep learning our collaborating team at École polytechnique fédérale de Lausanne (EPFL) have found excellent preliminary results in Swiss cohorts, matching and out-performing expert evaluation for risk stratification (Area Under the Receiver Operating characteristic Curve (AUROC) >80%) and diagnosis (AUROC >90%) of COVID-19 pneumonia (unpublished data).

As with LUS, LAsuc is another established tool being applied in a novel way. While evidence for the potential of LAusc for paediatric pneumonia is limited, some preliminary evidence on the predictive capacity of other respiratory sounds is emerging; for instance, the application of deep learning was able to predict diagnosis from breath and cough sounds collected on a mobile application with an AUC of around 80%. Another group achieved above 95% sensitivity and specificity on discriminating COVID-19 coughs from other pathologies as well as healthy patients. A study in South Africa included 33 culture-confirmed TB patients with healthy controls and found a diagnostic accuracy of 75%. This doesn't however speak much to the more pressing question of differentiating TB from non-TB pneumonia. While these are extremely promising results, no such trials have been performed in TB endemic areas, particularly challenging the specificity of LAusc for the detection of either pneumonia or TB. There is also as yet no regulatory approval for the use of AI powered LAusc devices in clinical practice.

This study therefore seeks to enable adoption of LUS into the routine management of IMCI clinical pneumonia through addressing key evidence gaps for translation into policy and practice, which includes generating evidence on the possible impact on health outcomes, namely: Does the integration of LUS into IMCI-based management guidelines better identify those children that would most benefit from antibiotic treatment, thus reducing antibiotic overuse without compromising health outcomes?

All paediatric patients (>60 days - 12 years) presenting to the ER or Outpatients Department (OPD) in participating study sites will be triaged as per routine care. Study staff will screen all presenting patients and consecutive patient recruitment will be done for those meeting all inclusion criteria. Those showing signs of severe disease or WHO IMCI Danger signs on triage (and hence meeting exclusion criteria) will be referred immediately to the treating clinicians as per local policy.

For those meeting inclusion criteria, sufficient time to read the informed consent form and make an informed decision will be given to the patients/parents/caregivers before inclusion in the study. Information covered will include the nature of the study, its purpose, the procedures involved, the expected duration, the potential risks and benefits and any discomfort it may entail.