Solar-powered oxygen delivery for the treatment of children with hypoxemia: protocol for a cluster-randomized stepped-wedge controlled trial in Uganda

This study aims to demonstrate the efficacy of SPO₂ in children hospitalized with acute hypoxemic respiratory illness in Uganda. The study is a multicenter prospective evaluation of SPO₂, using a stepped-wedge cluster-randomized design.

Our primary aim is to compare the 48-h mortality among children aged < 5 years admitted with hypoxemia before and after the implementation of SPO₂ delivery at 20 hospitals in Uganda, adjusting for confounding effects of calendar time and site. As secondary aims, we will compare safety and efficacy outcomes among participants, monitor SPO₂ system functionality and operational costs, and build capacity among nurses to deliver O₂. The working hypothesis is that SPO₂ can decrease the mortality of children admitted with hypoxemia Additional files 1 and 2.

The study will be a stepped-wedge cluster RCT. This trial will not be blinded. Clusters (health facilities) will be randomly allocated to the timing of implementation of the SPO₂ system. The installation of SPO₂ will not be simultaneous but will proceed in accordance with the stepped-wedge design (Fig. 1). All health facilities will have a run-in period during which nurses will be trained in pulse oximetry, O₂ management, and the study protocol. Data collection (prospective enrolment and accurate electronic record-keeping) will be in place by the end of the run-in period. SPO₂ will be introduced one site at a time, on a monthly basis, until all 20 sites have SPO₂ installed. The timing of installation of SPO₂ will be random, with allocation concealment. Analysis will compare mortality before and after implementation of SPO₂, using linear mixed effects (LME) logistic regression, adjusting for the confounding effects of time (fixed effect) and site (random effect) [17]. Data collection will continue until the trial’s termination, considered to be one month after the installation of SPO₂ in the 20th site.

Uganda counts approximately 145,000 child deaths annually, of which 16% are attributed to pneumonia [4]. Uganda’s public healthcare system consists of health centers (levels II–IV), general/provincial hospitals, regional referral hospitals, and the National Referral Hospital. The majority of deaths occur in a limited number of general hospitals and level IV health centers where at least half have poor access to a stable electrical supply or O2 cylinders [18]. These represent ideal sites for early implementation of SPO₂ because of a context-specific need for an O₂ delivery system that does not rely on electrical power or cylinder distribution. We will enroll 20 sites in our cluster-randomized trial.

Health facilities were evaluated and selected based on the following criteria. Sites were included if they: (1) had pediatric inpatient services; (2) lacked consistent O₂ supply on pediatric wards; and (3) had adequate space and willingness to install solar panels, a battery bank, and O₂ concentrator on the hospital premises (i.e. on pediatric ward). Sites were excluded from the study if: (1) had no inpatient services for children; (2) had pre-existing, functional, and consistent O₂ delivery systems (cylinder or concentrator); (3) did not have space or were unwilling to install solar panels, a battery bank, and O₂ concentrator.

We conducted site visits to 44 health facilities across the country and interviewed key informants at each site. Based on site selection criteria, five sites were chosen from each of Uganda’s four geographic regions, for a total of 20 sites (Fig. 2). Additional “back-up” sites have also been identified, in case of difficulties with site re-engagement, site/personnel recruitment, or installation at one or more of the primary sites. Characteristics of the 20 chosen sites were as follows. While cylinder oxygen was available at 9/20 (45%) of sites, dedicated cylinders were not available on the pediatric wards of any health facility. Oxygen concentrators were present at 18/20 (90%) of sites, with only 4/18 (22%) of sites with dedicated concentrators available on pediatric wards. Concentrators were shared across wards in 11/18 (61%) sites. At the time of the interview, 11/18 (61%) concentrators were operational, and only 4/18 (22%) produced oxygen with ≥ 80% purity. Power outages were reported as frequent in 17/20 (85%) of sites. While 19/20 (95%) of sites had generators available, none turned them on when children needed oxygen. Because electricity is necessary to run oxygen concentrators, this suggested that access to a reliable power source is a major obstacle in delivering oxygen therapy. Overall, only 3/20 (15%) of sites had oxygen available for pediatric use.

Qualified site champions (local nursing leads) were recruited for each site to support SPO₂ implementation.

We will include patients presenting to the selected sites meeting the following inclusion criteria: (1) age < 5 years; (2) hypoxemia (SPO₂ < 92%) based on non-invasive pulse oximetry taken within 24 h of admission, on room air; and (3) warrant hospital admission based on clinical judgement. Patients will be excluded from the study if: (1) measured SPO₂ ≥ 92%; (2) they can be managed as an outpatient; (3) parents or guardians unwilling to provide written informed consent to participate in study; (4) known cyanotic cardiac condition; and (5) diagnosis of hypoxic ischemic encephalopathy in a neonate.

Screening procedures will be as follows: children aged < 5 years presenting to the participating facilities will be screened for hypoxemia using a portable pulse oximeter if they have cough and/or difficulty breathing. Past medical history will be reviewed with the parent or guardian for known cyanotic cardiac condition and/or hypoxic ischemic encephalopathy. If the patient meets eligibility criteria, the parent or guardian will be approached for informed consent.

The random order of site installation will be generated before the trial. Site names will be written on paper and placed in sequentially numbered sealed opaque envelopes. Once a month, at the time of site selection for the next SPO₂ installation, the next envelope will be opened to reveal the site. The envelope will be signed and dated, and all envelopes and records will be kept for quality monitoring purposes.

Randomization will be in blocks of four. Each block will include one site from each of four geographic regions of the country (Northern, Eastern, Western, and Central Regions). The order of the four regions within a block will be random; the selection of a site within the region will be randomly sampled without replacement. Thus, after each sequential block of four, the sites will be balanced by region.

For eligible participants, the parent or legal guardian will be approached for consent to participate in the study (Additional file 3). If granted, the patient will be admitted to the pediatric ward, where there will be a SPO₂ system installed or not, according to random timing of installation at each site. All patients will receive standard care for their underlying disease, including antibiotics for pneumonia, intravenous fluids as necessary, blood transfusion as necessary, antipyretics, and any other medical therapy required. The study will ensure there are no stock-outs of essential medications or equipment. Basic demographic and clinical data will be collected from the case admission record, and patients will be followed during their hospital admission. The primary outcome, death at 48 h after admission, will be recorded. Other outcomes could include death after 48 h, discharge, and transfer to another facility. Death at 48 h (versus other) will later be analyzed as a binary variable. Patients discharged or transferred to another facility will be followed up by telephone at 48 h after admission to assess status.

During hospitalization, secondary outcomes will be recorded, including vital signs at admission and daily thereafter until discharge. Oxygen saturations will be monitored on a 4-hourly basis, during oxygen administration and after. Oxygen utilization (flow rate, duration) will be recorded throughout hospitalization. The need for oxygen therapy will be assessed daily, using standardized criteria for weaning oxygen. The length of stay among survivors will be recorded, as well as the final patient disposition (discharged without disability, discharged with disability, transferred to another facility, or death). The participant flow is illustrated in Fig. 3.

Our study’s primary endpoint is mortality at 48 h after admission. We will also evaluate several secondary outcome measures, including in-hospital mortality (time to death), length of hospital stay, duration of supplemental O₂ therapy (time to wean O₂), O₂ delivery system failure(s), cost (installation, servicing, and maintenance), baseline and retained nursing skills in managing oxygen therapy, etiology (based on polymerase chain reaction of dried blood sample and nasopharyngeal swabs), as well as diagnostic and prognostic host biomarkers (Additional file 4).

Data will be routinely monitored on each site by a dedicated data collection officer or study nurse, who will be responsible for recording patient information, follow-up, and uploading data to trial databases through KoBoCollect on tablet PCs. During entry, any missing data will be noted on an Errors/Omission log which will be checked regularly by the study team to fill in the missing data. Once uploaded, study data will be securely stored in a locked room and on a password-protected device, accessible only to study personnel. The study monitor or other authorized representatives may inspect all documents and records required to be maintained by the principal investigator; the study site will permit access to such records.

Adverse events (AEs) and unintended effects of SPO₂ will be monitored, reported, and collected by clinical staff, and will be managed according to national and local standard of care. AEs will be monitored continuously during the study by a dedicated nursing staff. A running log of AEs will be kept and reviewed periodically to allow the study team to assess if any patterns of AEs emerge in real time. All study deaths will require completion of a serious AE form, which will be made available to ethics boards and regulatory authorities within seven days of the event.

No interim data analysis or trial stopping rules are planned for this trial. Oxygen is known to be an essential and life-saving therapy and the efficacy of the SPO₂ system has been shown to be non-inferior compared to conventional oxygen therapy in the Ugandan context [15]. The stepped-wedge design and commitment to participating sites demands that SPO2 be installed at all sites, such that trial discontinuation at midpoint is not feasible. Because we do not plan to halt the trial before enrolling the planned number of clusters, we have not planned for a Data Monitoring Committee (DMC).

In our trial, we have a cluster-randomized design. The primary dependent variable is the 48-h mortality. The independent (predictor) variable of interest is the exposed/unexposed status of an individual patient (fixed effect), which will depend on whether the patient presents to a participating site before or after installation of SPO₂. Furthermore, we have covariates which will be modeled as both fixed and random effects. We will use a LME logistic regression model to examine the effect of SPO₂ on mortality while adjusting for changes in mortality over time (fixed effect) and variability in mortality between sites (random effect). We will use R [19] and lme4 [20] to model the binary outcome (survival) as a function of SPO₂ treatment, calendar time, and site. As fixed effects, we will enter SPO₂ treatment (before or after SPO₂ installation) and calendar time, without interaction term, into the model. We will model site as a random effect. We will determine the statistical significance of the SPO₂ treatment effect comparing models with and without the SPO₂ treatment term. If the model fit is significantly improved with inclusion of the treatment term, at α = 0.05 level of significance, we will conclude that the SPO₂ treatment effect is statistically significant. We will estimate the 95% confidence interval of the treatment effect (odds ratio) from the coefficient of the LME logistic regression model.

Secondary outcomes will be assessed using descriptive and comparative statistics, as appropriate. Where possible, LME models accounting for the clustered data structure and effects of calendar time will be used to determine differences between SPO₂-treated and untreated patients, with necessary adjustments being made to allow for multiple comparisons within secondary outcomes.

We will include 20 health facilities and a total of 2400 hospitalized patients. This sample size was calculated using a computer simulation, modeling trial conditions and applying the planned statistical analysis. Although analytic methods for sample size calculation for cluster-randomized stepped-wedge trials have been published [21], these require estimates of nuisance parameters (e.g. intra-cluster correlation coefficient) which were not available in the Ugandan context, such that the validity of this method was unknown.

For simulation parameters, we used data from the Demographic and Health Information System (DHIS) 2015 for Uganda. This database contains information on the number of patients with clinical pneumonia admitted to each hospital and the number of fatal cases, allowing us to estimate site-specific pneumonia mortality. In a computer simulation, we randomly selected 20 representative sites from the DHIS and used the number of patients admitted with pneumonia at each site to estimate the incidence of hypoxemic patients (13% [22]) that would be enrolled in a hypothetical clinical trial. The monthly number of hypoxemic pneumonia admissions was modeled as a Poisson distribution [23]. Monthly random allocation of SPO₂ to the 20 sites was simulated, according to the stepped-wedge design. The outcome of each patient at 48 h was modeled as a Bernoulli trial with probability of fatal outcome equal to the site-specific mortality. We used DHIS 2015 data to estimate the baseline mortality assuming a mortality reduction of 35% after installation of SPO₂ [13]. Having generated simulated trial data, we ran the planned statistical analysis, fitting a LME logistic regression model, as described above. Mortality was modeled as a function of SPO₂ treatment, with adjustment for covariates of calendar time (fixed effect) and site (random effect), and the p value for the treatment effect was calculated. We repeated the simulation > 5000 times and determined the proportion of trials that correctly detected a difference between patients receiving SPO₂ and those not receiving SPO₂ (statistical power). Using this simulation strategy, 20 sites enrolling a total of 2400 patients would provide > 80% power to detect a 35% mortality benefit of SPO₂ at the α = 0.05 level of significance [17, 24]. Data collection could feasibly be completed within 24 months (Fig. 4).

Figure 4a shows the statistical power as a function of the number of sites and the duration of enrolment. Each power estimate was based on at least 100 hypothetical trials in a computer simulation. Approximately 20 sites were required, enrolling patients for two years, to achieve sufficient statistical power. Figure 4b shows individual trial simulations, illustrating increasing power with increasing sample size. Figure 4c shows the sensitivity of the study power to input parameters: baseline mortality and mortality reduction with SPO₂.