Study design and setting
The study is an unblinded, randomised, superiority trial with parallel assignment and a 1:1 allocation ratio. We used an adaptive design strategy (see ‘Sample size and power using adaptive design methods’ in the Methods and analysis section).
We are conducting the trial in a district hospital. District hospitals are where the majority of children in low-resource settings are treated for severe pneumonia.12 The study is designed to be representative of everyday care in order to optimise the generalisability of our results, while maintaining the scientific rigour of a trial.
Salima District Hospital (SDH) in Malawi is the project site. SDH is a 250-bed government facility; the paediatric department admits approximately 7000 children annually. The site was chosen, with input from the Malawi Ministry of Health, since SDH is similar to other Malawi district hospitals in terms of clinical staff and laboratory and radiology capabilities.
Recruitment and enrolment
All children admitted to the paediatric ward between the ages of 1 and 59 months are first triaged by study staff. Study staff triage patients 24 hours/day. Children identified with WHO severe pneumonia are then consented and screened for eligibility by study staff.
Inclusion/exclusion criteria
Children aged 1–59 months are enrolled if their caregiver provides written informed consent, they meet WHO-defined severe pneumonia criteria (figure 1) and have one of three high-risk conditions: severe hypoxaemia, HIV infection or exposure, and/or severe malnutrition.4
Figure 1Inclusion criteria.
We defined severe figure 6yhypoxaemia as a peripheral oxyhaemoglobin saturation (SpO2) <90% as measured non-invasively with a pulse oximeter (Masimo Rad5 with LNOP® Y-I Multisite wrap sensor). HIV infection is a positive HIV DNA PCR for children <12 months or a positive HIV antibody test for children ≥12 months.4 In accordance with Malawi guidelines, children <24 months old who are uninfected but have an HIV-infected mother, irrespective of breast feeding status, are considered HUE.4 Severe malnutrition is a weight for height of ≤3 SD from the median and/or bilateral pedal oedema.4 In this study, children with a mid-upper arm circumference (MUAC) <115 mm are also classified as having severe malnutrition.4 We included children with a MUAC <115 mm, including those between 1 and 6 months, because they have been found to have an increased mortality risk when hospitalised for pneumonia.9 24
Exclusion criteria are limited to previous study enrolment or psychosocial conditions that would interfere with the study.
Randomisation and blinding
Simple 1:1 randomisation is conducted at enrolment using sequentially numbered sealed opaque envelopes. Due to the nature of the bCPAP intervention this study is unblinded for study staff and caregivers.
Intervention
Children randomised to the trial’s study arm are initiated onto bCPAP with oxygen, while those randomised to the trial’s control arm are initiated onto low-flow oxygen only. All other treatment is the same for both arms and follows WHO and Malawi guidelines for severe pneumonia and other conditions.4 Children randomised to the control arm and non-study patients do not receive bCPAP as this is not standard of care in Malawi. In addition, invasive ventilation is not available as part of routine care in Malawi, and is therefore not available for study patients who are in either arm.
Equipment and support
bCPAP is delivered using the Fisher & Paykel Bubble CPAP system. This system is a validated non-invasive ventilator device that provides warmed humidification and pressure control. bCPAP flow is driven using Newlife Intensity Airsep oxygen concentrators capable of delivering 10 liters per minute (LPM) and 90%–97% fractional inspired oxygen concentration to the bCPAP system via oxygen tubing. In our experience, we need oxygen flows of between 6 and 8 LPM to achieve 5–8 cmH20 bCPAP pressure.
bCPAP is a closed circuit capable of delivering pressure to a patient (see figure 2). The bCPAP circuit consists of four parts: (1) inspiratory limb, (2) nasal interface, (3) expiratory limb and (4) sterile water reservoir. The inspiratory limb connects to the nasal interface, which is placed on the child.
Figure 2bCPAP design. bCPAP, bubble continuous positive airway pressure.
We use two types of nasal interfaces based on the patient’s size: (1) nasal prongs and (2) unvented nasal masks. We elected to not use full-face masks for delivering bCPAP given the increased safety risks of aspiration and secretions with full-face masks.
The expiratory limb connects the patient’s nasal interface to a reservoir of sterile water, and is inserted into the reservoir of water at a depth equivalent to the desired pressure (eg, 5 cm depth is equivalent to 5 cm H20 pressure).
bCPAP is a closed system. However, bCPAP pressure can be lost at the nasal interface if the seal is broken or via the child’s mouth. We confirm the desired bCPAP pressure with a digital manometer at least twice per day and after all pressure or flow adjustments.
We use the Newlife Intensity Airsep oxygen concentrator with nasal prongs to deliver low-flow oxygen for children enrolled in the control arm. The Airsep concentrator is recommended by the WHO for use in low-resource settings and delivers up to 10 LPM. We use the Caire Sure Flow to split oxygen flow from a single concentrator for multiple patients. With the Airsep concentrator, the Caire Sure Flow can provide 2 LPM of oxygen for up to five patients.
A back-up generator guarantees continuous electrical supply.
Respiratory support protocol
Children triaged with severe pneumonia are immediately placed on low-flow oxygen as is standard care for SDH. Children remain on low-flow oxygen during consent, screening for eligibility and randomisation. Because the study enrols 24 hours/day, this process generally takes less than 60 min.
Children randomised to the control arm are continued on low-flow oxygen. Those children randomised to the bCPAP arm are initiated on bCPAP. Both bCPAP and low-flow oxygen initiation parameters are age dependent.
Enrolled children are reviewed every 4–6 hours until discharge from the hospital. Additional evaluations are conducted as needed based on clinical course.
For children on bCPAP, those between the ages of 30 and 59 days are started on a pressure level of 7 cmH20; children >2 months old are started on 8 cmH20. For those randomised to low-flow oxygen, children 30–59 days old on low-flow oxygen are started on 0.5 LPM; children >2 months old are started on 2 LPM, per WHO guidelines.4
Children are weaned in a stepwise process based on vital signs and danger signs during the daytime respiratory status rounds. Children need to be without multiple respiratory danger signs or severe hypoxaemia for at least 24 hours, prior to any incremental respiratory support weaning (bCPAP or oxygen). Figures 3 and 4 outline our weaning protocols for bCPAP and low-flow oxygen.
Figure 3Weaning protocol for bCPAP. bCPAP, bubble continuous positive airway pressure.
Figure 4Weaning protocol for low-flow oxygen. LPM, liters per minute.
Antibiotics
All enrolled children are started on parenteral benzylpenicillin and gentamicin at standard WHO doses and schedule.4 HIV-infected and HUE children <12 months old are also started on enteral cotrimoxazole and prednisone for presumptive Pneumocystis jirovecii infection.4 Patients who remain with fever in combination with persistent respiratory or general danger signs or who develop a new respiratory or general danger sign after 3 days of benzylpenicillin and gentamicin are switched to ceftriaxone.
Patients with altered mental status at any time are administered meningitic doses of parenteral ceftriaxone. Any child unable to be weaned off of respiratory support after 5 days of treatment is transitioned to ceftriaxone.
Children >12 months old with HIV infection or HUE are started on cotrimoxazole and prednisone after 3 days of benzylpenicillin and gentamicin if they persist with fever, in combination with persistent respiratory or general danger signs, or have a new respiratory or general danger sign.
Patients receive at least 5 days of parenteral benzylpenicillin and gentamicin. Children switched to ceftriaxone receive a total of at least 5 days of ceftriaxone. All children initiated on cotrimoxazole complete a 21-day course and a 5-day course of prednisone.
Feeding
Children with any general danger sign, 3 or more respiratory danger signs, or apnoea or grunting in isolation are not allowed to eat by mouth. Instead, these children receive a nasal gastric tube for feeding of expressed breast milk or formula every 2 hours at a standardised amount calculated to provide appropriate calories and fluid. Children with severe malnutrition receive nasal gastric feeds following the WHO protocol for severe malnutrition.4
Children able to eat by mouth and without severe malnutrition receive standard hospital food or food brought by their family. Children with severe malnutrition and able to eat by mouth receive malnutrition feeds per WHO recommendations.4
Additional treatments
All children are evaluated for malaria using a rapid blood test. Patients who are malaria positive are treated with a 3-day course of artesunate monotherapy per Malawi guidelines.
Children are also evaluated for anaemia via bedside haemoglobin measurement. Children without severe malnutrition are transfused with packed red blood cells for a haemoglobin level <6 mg/dL; children with severe malnutrition are transfused for a haemoglobin level <4 mg/dL. Blood product availability is typical of Malawi district hospitals and is often unreliable.
Children with dehydration and shock are treated following the recent WHO guidelines for fluid resuscitation.13 Children with altered mental status have their glucose measured and are treated with dextrose if hypoglycaemic per WHO recommended thresholds.4
Sample size and power using adaptive design methods
This study will accrue 900 children (450 children per study arm) with pneumonia and high-risk comorbidities. Mortality assumptions for this study were derived from published literature and programmatic data from Malawi.
We estimate true mortality rates of 14.7% with standard care and 6.1% for bCPAP cases. With a 95% follow-up rate and α=0.05, the power (1 – β) is estimated to be 99%. Assuming mortality of 14.7% in the standard care arm, the sample is sufficient to reject the null hypothesis with 80% probability if the mortality rate in the bCPAP arm is 8.7% or lower.
The study cohort is further divided into two subgroups of high-risk conditions (figure 5). Distributions for subgroups were reassessed using adaptive design methods from data collected during the first 10 months of the study. All effect estimates were determined a priori.
Figure 5Projected enrolment.
Severe hypoxaemia without comorbidity subgroup: 300 (150 per arm) children with severe hypoxaemia but without HIV infection, HUE or severe malnutrition.
Power analysis subgroup 1: we estimate 13% mortality for standard care and 4% for bCPAP cases.8
With 95% follow-up, equal allocation to the standard care and bCPAP conditions, and the hypothesised true mortality rates described, these analyses are powered at 80%.
Comorbidity subgroup: 600 children (300 per arm) with HIV infection, HUE or severe malnutrition.
Power analysis subgroup 2: we estimate 15.6% mortality for standard care and 7.1% for bCPAP cases based on relative contributions by each subgroup extrapolated from observational data. The observational data reveal approximately:
10% of children are infected with HIV, with assumed standard care mortality of 30% and bCPAP mortality of 14%.7 9
30% of children have HUE, with assumed standard care mortality of 12% and bCPAP mortality of 5%.7
60% of children are severely malnourished (HIV uninfected and HIV unexposed), with assumed standard care mortality of 15% and bCPAP mortality of 7%.5
With 95% follow-up, equal allocation to the standard care and bCPAP conditions, and the hypothesised true morality rates described, these analyses are powered at 89%. Assuming mortality of 15.6% in the standard care arm, the sample is sufficient to reject the null hypothesis with 80% probability if the mortality rate in the bCPAP arm is 8.3% or lower.
Data collection, quality assurance and data management
All study staff are certified in Good Clinical Practice and were trained on study protocols and data collection. Staff complete refresher-training courses every 4–6 months and receive regular supervision.
Trial data are collected directly into a secure REDCap (Research Electronic Data Capture) database using encrypted tablets. REDCap is a secure, web-based application designed to support data capture for research studies. Paper case report forms are used for back-up when necessary.
Statistical methodology and analysis
Logistic regression will be used in the primary and interim outcome analyses. Hospital pneumonia outcome (pneumonia cure vs died) will be the primary endpoint; study arm (bCPAP vs standard care) will be the treatment variable in primary and interim outcome analyses.
We will employ the O’Brien-Fleming stopping rule to address any inflated type II error stemming from multiple comparisons.25 p Values (and Z-score cut-offs) for two-sided tests used in the first interim, second interim and final analyses will be .0006 (3.438), .0151 (2.431) and .0471 (1.985), respectively.26
If any significant imbalance between groups on baseline covariates is present then results of logistic regression analyses controlling for these variables will also be done. If missing data do not appear to be missing completely at random, multiple imputation methods will be used to model the missinginess mechanism based on observed data and to evaluate whether the study findings vary when accounting for missinginess.
Secondary analyses will examine whether treatment effect varies by baseline patient characteristics. As with the primary analyses, hospital pneumonia cure versus died before hospital pneumonia cure will be the outcome in logistic regression with independent variables including group assignment, patient characteristic and an interaction between group assignment and patient characteristic.
Separate models will be run for each patient characteristic of interest, including primary risk category—comorbidity (HIV infection or HUE, and/or severe acute malnutrition) and severe hypoxaemia. We will also examine whether severe hypoxaemia modifies the effect of treatment for children with a comorbidity through the use of interaction terms.