Every year 900 000 children under the age of 5 die of pneumonia . Preventative and curative measures in the community and within health facilities are effective in reducing the burden of pneumonia, and their coverage needs to be increased . However, mortality for children with respiratory failure, not responding to standard treatment, continues to be high, and there may be a role of continuous positive airway pressure (CPAP) as an additional therapy.
CPAP has an established role for the management of respiratory distress syndrome in preterm babies , and is commonly used in the treatment of bronchiolitis in older infants . Multiple studies of bubble CPAP in low resource settings have documented mortality reductions in special care baby units [5,6], but highlighted the resource implications, and importance of broader contextual factors in implementation .
Conversely, the literature has been less clear on the role of CPAP for pneumonia, particularly in district and secondary hospitals in low- and middle-income countries (LMICs), where collectively, the greatest number of cases are managed. Two systematic reviews and meta-analyses have reported on the effectiveness of CPAP for children with severe pneumonia in LMICs [8,9] finding mixed results on mortality and trend towards higher adverse events rates in CPAP compared to standard low-flow oxygen. Both reviews noted the heterogeneity in populations and context and the likely implications these factors have on safety and efficacy. And both reviews recommended further randomised controlled trials, with Sessions suggesting that CPAP implementation in resource-limited settings be restricted to well-staffed intensive care or high dependency settings until further data are available .
This review aimed to increase understanding of the patient selection and contextual factors that influence effectiveness and safety of CPAP for children with severe pneumonia. We also aimed to review other factors around implementation including feasibility, sustainability, and acceptability in resource-limited settings, in order to inform global recommendations and highlight gaps in the literature for future implementation research.
We conducted a systematic search of medical databases Medline, Embase, and PubMed for all relevant articles between January 1, 2000 to August 19, 2020. We mapped search terms to medical subject headings where possible, using Boolean operators to combine searches into our final systematic search query. We used synonyms of “pneumonia”, “CPAP”, and “child” to target our search strategy, with oversight from an experienced Health Service Librarian to ensure all relevant papers were identified. We also searched reference lists of all included references for eligible studies. The specific search terms used for our Medline search and further details of the search strategy, information sources and data collection processes are included in Appendix S1-S2 in the Online Supplementary Document.
Assessment of study eligibility
Two reviewers (CW and RS) independently screened the titles and abstracts of all returned studies. We obtained full text for studies that were screened in by either reviewer, and the same two reviewers independently assessed them for inclusion. We included studies involving CPAP for children (aged 28 days to 4 years) with World Health Organization (WHO) pneumonia/severe pneumonia conducted in a LMIC. We focussed on more recent studies (published after the year 2000) to maximise contextual relevance and excluded studies that focussed on a neonatal population (Table S1 in the Online Supplementary Document). We did not exclude studies of CPAP in known viral aetiology of lower respiratory infection if these children fulfilled the WHO definition of pneumonia (cough or difficulty breathing with tachypnoea or chest wall indrawing). We resolved disagreements by discussion and, where appropriate, review by a third reviewer (HG).
Data management, extraction, and synthesis
We used a standardised data extraction form to extract data relevant to our review. Two reviewers (CW and RS) independently extracted data from each eligible study and entered data into an Excel spreadsheet (Microsoft, Redmond, US). We resolved disagreements by discussion and contacted study authors where appropriate to resolve any uncertainties. We extracted data on the study population (age, sex, comorbidities), CPAP details (type, humidification, oxygen source, air/oxygen mix, etc.), clinical guidelines, monitoring and supportive care (starting pressure and FiO2, decision-making algorithm, fluids/feeding, etc.), health service characteristics (level of care, staffing, etc.), and outcomes (mortality, adverse events, treatment failure, etc.) (detail in Appendix S1 in the Online Supplementary Document).
We used tables and narrative synthesis to describe context, intervention, and population and triangulate this with the reported outcome and safety data (using the effect measures reported in each original study) We did not conduct a meta-analysis of CPAP outcomes, as this has been reported previously  and the focus of this study was to understand contextual and population factors. Heterogeneity in study outcomes, populations and contextual factors reported in included studies prevented quantitative synthesis of potential associations. Rather, we triangulated context, intervention, and population data with outcome data to enable qualitative synthesis. We graded levels of evidence for key findings using Oxford Levels of Evidence framework, where 1 is the highest level of evidence (typically systematic review of trials) and 5 is the lowest level of evidence (typically mechanism-based reasoning) .
Assessment of study quality and risk of bias
We assessed the quality and risk of bias of all included studies by using the Effective Public Health Practice Project (EPHPP) Quality Assessment Tool . Using this tool, two reviewers (CW and RS) independently rated study quality as strong, moderate or weak with respect to selection bias, study design, confounders, blinding, data collection method, withdrawals and dropouts, and a global rating. Where disagreements occurred, a third reviewer (HG) carried out a final assessment. Evidence for recommendations and conclusions was graded using the Oxford Centre for Evidence Based Medicine levels of evidence .
Our search identified 1105 articles, with 2 additional articles identified on review of references. After duplicates were removed 902 references were screened, including full text review of 41 publications, of which 18 were excluded (Figure 1): 9 because it was not possible to extract data specific to patients who received CPAP [12–24], 4 because it was not possible to extract data specific to post-neonatal patients [25–28], 3 because it was not possible to extract data specific to low and middle income countries [4,29,30], 1 because it was a pooled analysis which included studies in which most patients did not have pneumonia , and 1 because it was published prior to the year 2000 .
Figure 1. Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) flow diagram. CPAP – continuous positive airway pressure, LMIC – low- or middle-income country.
We included 23 studies in qualitative synthesis, including six individually randomised controlled trials (RCTs) [33–38], one cluster open-label crossover trial , twelve observational studies [40–51], one cost-effectiveness analysis , and three case reports (Table 1) [53–55]. Four studies reported on CPAP sustainability , health workforce implications , parental attitudes , or cost-effectiveness , but did not provide clinical data.
Table 1. Characteristics of included studies
ARDS – acute respiratory distress syndrome, CPAP – continuous positive airway pressure, CXR – chest radiography, GCS – Glasgow Coma Score, HDU – high dependency unit, HFLC – high-flow nasal cannula, LFNC – low-flow nasal cannula, NGT – nasogastric tube, PICU – paediatric intensive care unit, PJP – pneumocytis jiroveci pneumonia, PSNZ – Paediatric Society of New Zealand, RCT – randomised controlled trial, RDS – respiratory distress syndrome, TB – tuberculosis, RR – relative risk, OR – odds ratio, CI – confidence interval
Risk of bias
Of the 19 studies providing original clinical data, quality was assessed as weak in 7 studies [33,40,41,44,53–55], moderate in 10 [34,35,37,38,42,43,45–47,50] and strong in 2 studies (Table S2 in the Online Supplementary Document) [36,39].
Where were the studies conducted?
Studies took place in 2 low-income countries (Uganda and Malawi), 5 lower-middle income countries (Bangladesh, India, Ghana, Papua New Guinea, and Tanzania) and 1 upper-middle income country (Brazil) (Table 1).
Fourteen (61%) studies were predominately in urban tertiary hospitals [33–35,37,40,42–47,50,53,55], and 8 (35%) were in district level hospitals (mostly rural) [36,38,39,41,48,49,51,54]. Of the 14 tertiary hospital based studies, 6 took place in intensive care wards [33,34,37,40,42,44].
A total of 5050 children were enrolled in the 19 included studies that provided original clinical data (including case reports): 43.5% (2200) in a single large crossover study in two district hospitals in Ghana . This study aside, 72% (2052/2850) of the remaining children were studied in tertiary centres.
Which children have been studied?
While all included studies included a subset of children with pneumonia requiring CPAP, the populations included children of varying ages with bronchiolitis, all cause respiratory distress, respiratory failure, as well as WHO-defined severe or hypoxaemic pneumonia (Table 1).
Three randomised trials focussed on infants aged <12 months with bronchiolitis (total 165 participants), all conducted in tertiary hospitals settings in Brazil and India, including two in paediatric intensive care units (PICU) [33,35,37]. These studies all reported zero inpatient mortality in both CPAP and comparison groups (high-flow or standard oxygen therapy), small or no improvement in clinical signs, and similar rates of treatment failure and length of stay between CPAP and High Flow Nasal Cannula (HFNC) or standard care groups. Young infants with likely bronchiolitis were also reported in observational studies of WHO pneumonia in India (27%-31% of total population) [40,43], Malawi (53% of total population) , Papua New Guinea (1.6% of total population) , and in 3 case reports of successful CPAP treatment in Malawi and Tanzania [53–55].
Five studies included children with general respiratory distress (total 2763 participants), including 3 observational studies [40,41,45] and 2 trials comparing CPAP to delayed initiation of CPAP or standard care [38,39]. These studies included a range of ages and illnesses, with mortality ranging from 3% (typically younger infants with bronchiolitis) to 29% (high HIV-related pneumocystis carinii pneumonia). Three comparative studies from semi-rural hospitals in Ghana and Uganda suggested that CPAP was associated with respiratory improvement within a few hours of initiation [38,41], and possible mortality benefit for infants <1 year of age but not older children .
Two trials [34,36] and two observational studies [42,43] focussed on children with hypoxaemic or high-risk pneumonia (total 1967 participants), with mortality ranging from 3% to 14%. The trial from a tertiary PICU in Bangladesh reported reduced mortality with CPAP compared to high-flow or standard low-flow oxygen , but the trial in a rural district hospital in Malawi found increased mortality ; both trials were stopped early at interim analysis.
Four observational studies enrolled children with respiratory failure (failing standard oxygen therapy), mostly related to severe pneumonia. Mortality ranged from 33% to 55% (urban Papua New Guinea and Malawi) [46,47,50], except for one Indian study involving children with influenza-like illness during the swine flu epidemic (zero mortality) . These studies reported improved respiratory signs and did not include a comparison group.
Across all these studies, older children (especially >12 months), children with HIV, severe acute malnutrition, multi-organ failure, or other comorbidities generally fared worse on CPAP that those without [39,46,50]. Early response to CPAP (specifically correction of hypoxaemia) was a strong predictor of survival .
All but 3 of the studies [33,37,40] used a form of bubble CPAP (Table 2). Seven studies used commercially available bubble CPAP systems [36,38,39,45–47,53]. In 5 studies a bubble CPAP circuit was improvised by using nasal prong tubing, cut with one limb submerged under a column of water, the depth determining set pressure [41,56–59]. Depths between 5 and 10 cm were used in included studies. Other modalities included a flow-inflating device , and CPAP via a mechanical ventilator [33,37].
Table 2. CPAP details of included studies
CPAP – continuous positive airway pressure
*Cost of CPAP equipment only excluding oxygen provision and ongoing maintenance costs.
†Exchange rate Indian rupee: US$, 74:1.
Oxygen was supplied by oxygen concentrators or cylinders in most studies, with wall or central pressurised O2 available in 3 studies [38,39,43]. Fifteen systems allowed blending of air and oxygen, 10 included humidification, and only one heating . Two studies used mask interfaces [33,40]. The remaining used exclusively nasal prongs, a combination of prongs and nasopharyngeal catheters [35,50] or prongs and masks [36,37].
Staffing and clinical practice context
Of 19 studies providing original clinical data, only 3 observational studies from urban hospitals in Africa reported nurse to patient ratios (range = 1:2 to 1:40) (Table 1) [46,47,50]. Similarly, few studies reported doctor supervision, generally involving a mid-level doctor or medical officer available onsite during the day and on-call overnight with at least daily senior medical ward round [39,46,47,50]. The notable exception was one trial in rural Malawi that was nurse-led (doctor available by phone for advice only) and was stopped early due to excess mortality in the CPAP arm .
Studies typically required nursing review within 1-2 hours of commencing CPAP or changing settings [34–41,43,45,47], then just 2-4 times per day [36,38,39,41,44–47,50], with only three studies from tertiary hospitals in India and Bangladesh continuing 1-2 hourly observations [34,40,43]. Many studies used pulse oximetry [33–44,46,47,50], and sometimes blood gas analysis and/or formal severity scores [34,35,37,41,43,44,47], to monitor clinical response, but frequency of monitoring was rarely documented.
Nasogastric or orogastric tubes (NGT/OGT) for gastric decompression and/or feeding were commonly included in study protocols but adherence to this was poorly reported [34,36,37,40,41,46,47,50,54,55]. In one study although 96% of patients met criteria for NG tube insertion, only 16% received one . Chest radiography (CXR) was available in most facilities [34–37,41,43,44,47], but not all [39,46].
Equipment and training
Training was typically directed at nurses [34,36–39,41,47], often described as interactive/hand-on/practical [36,38,41], included some test of competence [36,38,41], and follow-up supervision or re-training [36,38,39].
Few studies reported on equipment function, maintenance, or sustainability, and those that did found practical challenges. An RCT from a PICU in Bangladesh reported equipment failure in 20% of cases during the trial . During a crossover trial in 2 small hospitals in Ghana, CPAP was stopped early due to loss of power in 1 patient despite backup generators provided by the study team . Six patients ran out of oxygen while on CPAP during an observational study in a regional hospital in Uganda .
Even less is known about medium- and long-term sustainability of equipment or health care worker skills. A 16-month follow-up study of the Ghana trial found that 13% (1/8) of CPAP machines and 63% (5/8) of oxygen concentrators were non-functional and detected a significant drop-off in skills and knowledge in the staff who had been trained since the end of the study (despite some retraining having taken place) . A case report from rural Tanzania highlighted challenges to sustained CPAP practice due to unavailability of components and increased demands on clinical staff .
Comfort and complications
CPAP adverse effects were inconsistently reported, with minor complications (eg, mucosal dryness, nasal skin trauma, mild gastric distension) ranging from <10% [35,36,38–41,43,44,46] to >50% [37,50] of participants. Studies that compared CPAP to HFNC or low-flow oxygen generally found higher rates of minor complications with CPAP [36,37]. Aspiration (0%-1.2%) [34,36,41,46] and pneumothorax (0%-1.8%) [34–39,41,43,44] events were uncommon.
Costs were reported in 7/19 studies providing clinical data: three studies reported the costs of modifying an existing oxygen (+/− air) source to provide bubble CPAP (range US$3.40 to 6.80 without humidifier) [35,41,44], and four studies reported the cost of a low cost CPAP device and consumables (US$350 to 400) – but none included personnel, training, or maintenance costs (Table 2) [38,39,45,53]. In a cost-effectiveness analysis, Kortz inputted one-off and ongoing CPAP equipment and costs, training costs, but not costs for extra personnel or staffing time . In this analysis CPAP for severe pneumonia cases costed US$64 per patient more than standard care, with a favourably cost per Disability Adjusted Life Years (DALY) averted compared to other pneumonia interventions.
Reporting of the personnel and time-cost of CPAP is a gap in the literature. Recent work from Sessions et al found that health care workers performed more tasks and took more time (118 vs 83 minutes to initiate and 4.6 vs 1.5 minutes to adjust) with CPAP treatment relative to low flow oxygen . They estimated this could lead to an additional 164 hours of work per month hospital if CPAP use was expanded to all patients with severe pneumonia (with or without comorbidity).
This review aimed to understand patient selection and contextual factors that influence the effectiveness and safety of CPAP for children with severe respiratory infection that would meet WHO’s definition of severe pneumonia. We found heterogeneity in patient population, contexts, and intervention that limits the ability to pool results. However, while there are gaps, there are currently sufficient data and expert opinion to outline practical recommendations.
Which children benefit from CPAP?
Our understanding of the mortality benefits of CPAP for severe respiratory distress in children in low and middle income countries has been shaped by 3 RCTs demonstrating significant benefit in reducing clinical failure and mortality , no benefit in all-cause mortality , and increase in mortality . Hypotheses behind the reasons for these conflicting results highlight the importance of context and study design in assessing the potential role of CPAP [60–62].
In this review we document the heterogeneity of patient populations studied across both RCTs and observational studies, reflected in the large differences in the proportion of children who died in the studies. The reasons are multifactorial, but importantly reflect differences in case definition, inclusion and exclusion criteria. Predictably, studies of bronchiolitis (ie, single organ pathology) based in intensive care or highly monitored units, with access to mechanical ventilation, had the lowest (zero) mortality. Conversely, studies that enrolled children with respiratory failure – defined as hypoxaemia or respiratory distress worsening or not improving with low flow oxygen – documented mortality in 50% of children. So, while all studies enrolled patients that would fulfil WHO’s definition of severe pneumonia this encompasses a wide range of populations from infants with (mostly) self-limiting bronchiolitis to older children with complicated pneumonia and multi-organ failure .
This review highlights that our evidence base for post-neonatal CPAP is heavily skewed by data in young children, with most studies enrolling children with mean/median ages <18 months. This is reflected in both studies exclusively of bronchiolitis [33,35,37], and in studies of pneumonia enrolling a significant proportion (27%-54%) of infants with bronchiolitis [40,43,45]. In others, it reflected the higher risk of severe disease and deterioration in the first year of life [47,50].
Table 3 summarises the current evidence using the Oxford Levels of Evidence framework . In keeping with studies from high resource settings , there is evidence that CPAP reduces respiratory distress and improves oxygenation in bronchiolitis, and possibly reduces mortality. The current literature most supports the use of CPAP in infants <1 year with likely bronchiolitis, who have hypoxaemia or respiratory distress not responding to low flow oxygen and preserved respiratory drive and upper airway reflexes, in settings with low HIV prevalence and malnutrition and adequate staffing. For this population, it would make sense to align (at least the CPAP set-up) with what is being recommended for neonatal CPAP [64,65]. Box 1 provides an example checklist to assess appropriateness of implementing CPAP in any given setting.
Table 3. Conclusions from included studies
Box 1. Assessing suitability of CPAP
Below are some questions to assess whether CPAP would be a suitable intervention in each setting. If the answer to any of these questions is no, it should be addressed prior to reconsidering.
1. Are there adequate numbers of staff day and night so that patients on CPAP:
– are continuously monitored with adjustments made to the CPAP as needed to maintain efficacy (eg, maintaining an adequate nasal seal to maintain pressures);
– will receive any additional necessary management required (eg, nasogastric tube for decompression of stomach distension and nasogastric feeding or intravenous fluids);
– clinical deterioration will be quickly identified, and appropriate steps taken
2. Are all nursing and medical staff who will be caring for patients on CPAP:
– adequately trained in CPAP use, including knowledge of indications for use, how to set up CPAP, how to maintain it
– aware of the potential complications (eg, pneumothorax, nasal pressure areas, risk of unrecognised respiratory failure) and have skills in how to recognise and manage such complications safely and effectively
– is there a plan for ongoing monitoring and training to account for loss of knowledge and skills and / or staff turnover
3. Is adequate equipment available for using CPAP?
– Ensure CPAP will not negatively impact on care of other patients (eg, are there sufficient oxygen sources available that giving one patient CPAP will not limit access to other patients to standard oxygen therapy)
– Ensure infection control and equipment sterilisation procedures are in place
– Ensure technician/health care worker trained in day-to-day trouble shooting and maintenance
– Ensure adequacy of power supply or back-up source of power
4. Are adequate monitoring tools available?
– Monitoring charts of vital signs with thresholds for responses and escalation
– Clinical guidelines on the use of CPAP and care of patients with acute respiratory disease
– Pulse oximetry
– Audit processes
Future research should aim to further understand the subset of patients, fulfilling WHO’s definition of severe pneumonia and presenting outside of tertiary and intensive care units, who would most benefit from CPAP. And to determine clinical predictors that can be used to identify these children to further clarify guidelines on eligibility for safe and effective CPAP. Conversely, it would be also important to predict subsets of children who would do badly on CPAP, and understand the aetiology of CPAP failure or death. Building on the evidence base of minimal standards to sustain safe and effective CPAP will assist wider adoption. Research on CPAP use should ideally report all above reviewed contextual factors to provide readers with a better understanding of the intervention, the environment in which it was introduced, and the population in whom it was used. Ideally, costs should include total systems cost (including staffing, training, consumables, power, maintenance and repairs), and not just equipment costs, so that facilities can evaluate feasibility and cost-effectiveness in considering CPAP adoption. When CPAP is introduced into a setting significantly different from those studied in the literature, it is best to do so in the context of a quality improvement evaluation, such that efficacy can be properly reviewed in that setting prior to wider uptake. An evaluation and reporting framework for CPAP studies has been proposed (Table 4) .
Table 4. Reporting framework of CPAP studies adapted from Duke 2019 
Our review was intentionally broad, to capture all CPAP-related studies from low- and middle-income countries, and enabled us to capture 7 studies not included in previous systematic reviews [8,9]. We restricted our search to CPAP, rather than other non-invasive ventilation (NIV) strategies, to make our findings most applicable to low-resource settings where low-cost CPAP is likely to be more feasible than strategies requiring a mechanical ventilator. We also limited our search to English-language articles and we are aware that this excluded some studies from Latin America [67,68]. Despite this, our review included a broad range of settings across Africa and Asia-Pacific regions and the findings will be highly relevant to health managers and policy-makers who are seeking to enhance hospital care for children with severe pneumonia.
CPAP is a safe intervention in settings resourced with intensive monitoring and supportive care, and the strongest evidence for a benefit of CPAP is in infants (less than 1 year) with bronchiolitis. The available published evidence and clinical experience can be used to help facilities assess appropriateness of implementing CPAP, guide health workers in refining selection of patients most likely to benefit from it, and provide a framework for components of safe and effective CPAP therapy. There is scope for future research to better detail and standardise the reporting of context, studied populations and CPAP set-up to assist policy makers and health workers in adapting research findings to their local setting.
Thanks to Poh Chua, research librarian, for substantial technical support in setting up and running the database searches, and Helen Thomson and Haset Samuel for administrative support.
Data availability: All relevant data is included in the manuscript or supplemental material.