Estimated Cost-effectiveness of Solar-Powered Oxygen Delivery for Pneumonia in Young Children in Low-Resource Settings

Key Points Question Is solar-powered oxygen delivery (solar-powered O2) a cost-effective intervention for use in children younger than 5 years with hypoxemia in low-resource settings? Findings This economic evaluation compared the costs and health outcomes of solar-powered O2 with (1) null case with no oxygen, (2) grid-powered oxygen concentrators, and (3) fuel generator-powered concentrators. Use of solar-powered O2 was cost-effective relative to the null case and grid-powered concentrators and was cost-saving relative to fuel generator-powered concentrators. Meaning The results of this economic evaluation suggest that solar-powered O2 is a cost-effective intervention for pediatric patients with hypoxemia in low-resource settings.


Introduction
Hypoxemia is present in 10% to 15% of children admitted to hospitals globally. 1 Pneumonia, the leading cause of childhood mortality outside the neonatal period, is a common cause of hypoxemia. 2,3 Based on a meta-analysis of 13 studies involving 13 928 children with pneumonia, hypoxemia is a strong predictor of mortality, increasing the risk of dying 5-fold. 4 Although bacterial pneumonia is the leading cause of hypoxemia, other pathogenic and congenital pathologies may also lead to hypoxemia as a final common pathway to respiratory failure. 5 Regardless of etiology, hypoxemia requires treatment with supplemental oxygen. Improved oxygen systems reduce pneumonia mortality by an estimated 35%, but access remains unreliable in low-and middle-income countries (LMICs). 6 Given that pneumonia is responsible for approximately 900 000 childhood deaths annually, access to oxygen is an important public health issue. 7,8 Although oxygen is included on the World Health Organization (WHO) list of essential medicines, 9 it may not be available in hospitals and health centers in LMICs because of cost and/or logistical challenges. 10,11 During the current COVID-19 pandemic, oxygen needs globally and in low-resource settings are expected to increase, exacerbating the gap in availability. Methods currently used in low-resource settings include compressed oxygen cylinders and grid-powered oxygen concentrators. 12,13 Cylinders require supply chains linking oxygen production plants to hospitals, which may be compromised by poor road conditions, costs of transportation, and weak supply chain management. 12,13 Oxygen losses due to leakage can also affect the cost-effectiveness and reliability of oxygen cylinders. 14,15 Oxygen concentrators, though shown to be more costeffective and user-friendly than cylinders, depend on a reliable and uninterrupted supply of electricity, which is often unavailable in resource-constrained settings. 16 A previous systematic review showed that 26% of health facilities in sub-Saharan Africa reported no access to electricity, and only 28% of centers reported reliable access. 17 Power outages lasted a median of 7% of the time monitored in a study from western Kenya (range, 1%-58%). 16 In that study, facilities experienced a median of 7 power outages per week (interquartile range, 7-16 outages) lasting a median of 17 minutes each (interquartile range, 11-27 minutes). 16 Solar-powered oxygen delivery (solar-powered O 2 ) has been shown to be an effective solution for supplemental oxygen delivery in low-resource settings. 18,19 Solar-powered oxygen delivery has been described in detail previously and implemented at 2 hospitals in Uganda to successfully treat children with hypoxemia. 18,19 In brief, photovoltaic cells installed on the roofs of hospitals collect solar energy, which is stored as electricity in a battery bank, then used to power an oxygen concentrator for production of medical-grade oxygen. 18 The efficacy of solar-powered O 2 was demonstrated in a proof-of-concept pilot study and a randomized clinical trial that showed clinical noninferiority compared with cylinder oxygen. 18,19 Solar-powered oxygen delivery has several advantages, including low operating costs, consistency and reliability through grid-power outages, being userfriendly for hospital staff, reduced oxygen waste, and reduced carbon footprint owing to exclusive use of freely available inputs of solar energy and air. 18,19 Having demonstrated that solar-powered O 2 is a feasible, safe, and effective solution to the oxygen gap in LMICs, 18,19 we now seek to answer whether solar-powered O 2 is a cost-effective intervention for treating pediatric patients with hypoxemia in low-resource settings. We followed the WHO Choosing Interventions That Are Cost-Effective (WHO-CHOICE) methodology and the associated guidelines for performing a generalized cost-effectiveness analysis. 20 One of the main benefits of this approach is the use of a "null" case, wherein the effects of all currently available interventions are removed, allowing for more effective comparison between different interventions. 20 We hypothesized that solar-powered O 2 would be cost-effective relative to the null case (no oxygen), using the gross domestic product (GDP) per capita of target LMICs as a costeffectiveness threshold. Secondary analyses compared solar-powered O 2 with oxygen concentrators powered by grid electricity and fuel generators. These analyses may more closely approximate the decision facing administrators and policy makers on the use of solar-powered O 2 .

Cost-effectiveness Analysis
This economic evaluation was completed from January 12, 2020, to February 27, 2021. The decision analytic framework used was a cost-effectiveness comparison between 2 scenarios: the intervention (solar-powered O 2 ) and a comparator condition. For the primary analysis, the comparator condition was the null case (no oxygen); for secondary analyses, the comparator conditions were grid-powered concentrators or fuel generator-powered concentrators.
The setting for implementation of solar-powered O 2 was a single rural or remote health facility with inpatient pediatric services in an LMIC without prior available medical oxygen. 10 Costeffectiveness of solar-powered O 2 was assessed from health care sector and societal perspectives. 21,22 A time horizon of 10 years was used. We followed the Consolidated Health Economic Evaluation Reporting Standards (CHEERS) guideline in reporting our findings (eAppendix and eMethods in the Supplement). Ethics approval was granted by the Health Research Ethics Board at the University of Alberta. The cost-effectiveness analysis used parameters that were derived from the literature and past experience installing the systems. There were no patient-specific data here; therefore, patient consent was not required or relevant.

Health Outcomes and Costs
The published literature was used when possible to estimate input parameters for health outcomes and costs ( Table 1; eTable 1 in the Supplement). 23,24,26,31,32,34 When published data were not available, we used data from our own experience implementing and evaluating solar-powered O 2 in Uganda (Table 1). 18,19 The GDP deflator method derived from method 2 by Turner and colleagues 36 was used to adjust for inflation and convert costs to a single base year (2019). We used 2019 as the base year because the most recent GDP deflator statistics were available up to 2019. 36 The GDP deflator for a given period reflects the average annual rate of inflation in the economy as a whole during that period.
Gross domestic product deflators are available from the World Bank. 37 Local costs were adjusted using local inflation rates before converting to US dollars. 36 For conversion of local currency to US dollars, we used historical conversion rates. 38 The real costs of solar-powered O 2 components, consumables, and equipment for alternative oxygen delivery methods are shown in Table 1 and eTable 1 in the Supplement.
With respect to nonmedical costs, we included opportunity costs and direct costs. Opportunity costs were the wages for 1 caregiver for the duration of the hospitalization and were based on the household income in Uganda. 39 Direct costs included travel, accommodation, and food for 1 caregiver for the duration of the hospitalization. Food cost was calculated as the difference between the daily cost of purchasing food and the cost of food in the home environment if the child was not hospitalized. 33,35 The outcome (health effect) of interest was the number of disability-adjusted life-years (DALYs) saved with solar-powered O 2 . The DALYs represent a widely used public health metric of disease burden. The WHO advocates the use of DALYs for generalized cost-effectiveness analyses and recommends this methodology for comparability. 20 The DALYs lost due to a disease refers to the combination of years of life lost (YLL) due to premature mortality and years of life lost due to disability (YLD), which accounts for the loss of health by applying a disability weighting. In the context of this study, we focused on YLL, under the assumption that otherwise healthy children who recover from pneumonia will not have long-standing disability. In the case of fatal childhood pneumonia, YLL were calculated as the difference between the life expectancy for patients (based on vital statistics) and the age at death.
All DALYs were calculated using the following formulas: For the DALY calculation, we neglected the YLD, such that YLL accounted for all the DALYs lost. This was based on the assumption that children who recover from pneumonia do not have residual morbidity. 40,41 For our base case scenario, both health outcomes and costs were discounted at 3% following the WHO-CHOICE recommendations. 20 Discounting was performed using a discounting factor (DF) given by the following formula 20 :

Calculation of Cost-Effectiveness
The comparison between the 2 scenarios used the incremental cost-effectiveness ratio (ICER) to assess the trade-off between improved health outcomes and increased costs. The ICER was defined as the difference in cost between interventions, divided by the difference in their effect (DALYs saved): The threshold for cost-effectiveness was assumed to be the GDP per capita in representative LMICs. 42 We used the GDP per capita of Uganda, where solar-powered O 2 was pioneered, and the lowest GDP per capita in the world (South Sudan, GDP of $220) for maximum stringency.

Statistical Analysis
To evaluate the association of uncertainty with cost-effectiveness, we conducted univariate sensitivity analyses in which a single key input parameter was varied throughout the plausible range while maintaining other parameters at their base case values. The resulting variation in the ICER was displayed as a tornado plot (eMethods in the Supplement). Additionally, a probabilistic sensitivity analysis was performed. Input parameters were randomly sampled from their assumed probability distributions (Table 1) to assess stability of the calculated ICER when multiple input parameters were varied simultaneously. The resulting incremental costs, incremental health outcomes (DALYs saved), and ICERs were plotted on a cost-effectiveness plane and used to generate a cost-effectiveness acceptability curve. Further details are provided in the eMethods in the Supplement.
We used bootstrap analysis to sample the costs and health outcomes concurrently, using the probability distributions of the input variables. We generated multiple estimates of the ICER and its component variables, and we used these to calculate the 95% CI for each variable (2.5th percentile and 97.5th percentile). Analyses were performed using R statistical software, version 3.6.2 (R Core Team).

Direct Medical Costs of Solar-Powered O 2
Under the base case assumptions, installation of a solar-powered O 2 system at a single hospital required a capital cost of $12 411. This cost comprised photovoltaic cells ($3930, at $2.92/W) 27 ,

Nonmedical Costs of Solar-Powered O 2
The societal perspective adds the expected costs incurred by the families of patients (

Health Outcomes and ICER
For a hospital with 431 pneumonia admissions per year, the system could treat 869 hypoxemic patients over 10 years (see Table 1

Sensitivity Analysis
We examined the sensitivity of our ICER estimate to variations in the key input variables. The ICER estimate was most sensitive to the number of children presenting with pneumonia and the mortality rate of pneumonia (Figure 1). The effects of component costs on the ICER (unit cost of photovoltaic panels and batteries) were small.
In a detailed 1-way sensitivity analysis for 4 selected input variables, the ICER was inversely proportional to parameters used to compute DALY saved (Figure 2A and B), including the number of children treated over the life of the system and the case fatality rate of children presenting with pneumonia. The ICER was favorable (<$604 per DALY saved) when the number of patients with pneumonia exceeded 15 per year and when the case fatality rate exceeded 0.3%. In contrast, the ICER varied linearly with component costs (Figure 2C and D) and was insensitive to changes in the component costs over a plausible range of parameter inputs.
In a probabilistic multiway sensitivity analysis, the ICER was favorable (<$604 per DALY saved) in 99.7% of simulations ( Figure 3A). 25 At an alternative threshold of $220, corresponding to the lowest GDP per capita of any country globally (South Sudan), solar-powered O 2 remained costeffective in 97.8% of simulations. 25 The cost-effectiveness acceptability curve ( Figure 3B) showed that, at a willingness to pay of $136 per DALY saved, the likelihood of the intervention being costeffective was 95%.

Comparison to Other Methods of Oxygen Delivery
The direct medical cost of grid-powered oxygen concentrators over 10  Supplement). The probabilistic multiway sensitivity analysis and cost-effectiveness acceptability curve are shown in eFigure 2 in the Supplement.
Compared with fuel generator-powered concentrators, solar-powered O 2 did not save lives or DALYs but was associated with a cost saving of $7120 during the life of the equipment. Accounting for uncertainties in the parameters, this estimate had a wide 95% CI, ranging from a cost saving of $59 876 to an excess cost of $11 673 (eTable 3 in the Supplement).

Discussion
In resource-limited settings, solar-powered O 2 has been previously shown to be safe and effective and to run reliably off the grid for the treatment of young children with hypoxemia. 18,19 The results of our analysis suggest that solar-powered O 2 is also cost-effective relative to the null case (no oxygen), cost-effective relative to grid-powered concentrators, and cost-saving relative to fuel generatorpowered concentrators.
We calculated an ICER of solar-powered O 2 of $20 per DALY saved, relative to the null case (no oxygen). If Uganda's GDP ($604) is used as a threshold, solar-powered O 2 is a cost-effective investment for health facilities with no prior oxygen. In other LMICs, we expect solar-powered O 2 to A, Nonlinear relationship between number of pneumonia cases and ICER. Solar-powered O 2 was most cost-effective at high-volume facilities. B, Nonlinear relationship between pneumonia mortality and ICER. Due to differences in referral patterns, resources, and capacity for management, mortality in childhood pneumonia may vary between sites. Solar-powered O 2 was most likely to be cost-effective at high mortality facilities. ICER estimate varies linearly and was relatively insensitive to uncertainties in unit costs of C, photovoltaic (PV) panels and D, batteries. Of these, a change in the unit cost for batteries had the largest effect on ICER. Ah indicates ampere hour; DALY, disability-adjusted lifeyear. be cost-effective because the ICER was less than $220, the lowest GDP per capita globally (South Sudan), in 97.8% of simulations ( Figure 3A). 25,43 A previous study found that the ICER of cylinder oxygen (an alternative method of oxygen delivery) was $54 per DALY saved relative to the null case. 14 Solar-powered oxygen delivery appears to be more cost-effective; however, the ICER for cylinder oxygen was well within the limits of uncertainty of our estimate for ICER of solar-powered O 2 (95% CI, $2.83-$206), and differences in methods and assumptions between this previous study and ours could confound this comparison. This ICER can also be situated within a suite of other nonalternative childhood pneumonia interventions, such as pneumonia case management ($73 per DALY saved), pneumococcal conjugate vaccine ($100 per DALY saved), and Haemophilus influenzae type b vaccine ($202 per DALY saved). 30,[44][45][46] Our analysis also suggested that solar-powered O 2 is cost-effective relative to grid-powered concentrators ($140 per DALY saved) and cost-saving relative to fuel generator-powered concentrators (estimated $7120 lower cost).
The ICER estimates (solar-powered O 2 vs null case) were most sensitive to parameters related to the DALYs saved (eg, patient volume and mortality, Figure 2). The ICER is inversely proportional (y   18 The dashed line shows an alternative threshold of $220/DALY saved, the GDP per capita of South Sudan, lowest in the world. A total of 99.7% and 97.8% of simulations were cost-effective using these 2 thresholds, respectively. B, Cost-effectiveness acceptability curve suggests 95% CIs that solar-powered O 2 will be cost-effective beyond a willingness-to-pay threshold of $136/ DALY saved.

Limitations
Our study has several limitations. Our findings depend on the accuracy of the input parameters.
Some parameters were based on few data (eg, the relative risk reduction in mortality with improved oxygen availability), 6 and some were taken from our own experience implementing solar-powered O 2 in Uganda. 18,19 The ICER was sensitive to parameters that vary between health facilities, such as patient volume, case fatality rate, and consistency of grid electricity; therefore, our findings should be applied with caution to facilities that differ substantially from our base case assumptions. To mitigate this limitation, we used 1-way and multiway sensitivity analyses to describe the variation in the ICER with uncertainties in the inputs. Our model did not include contingencies such as surge demand (eg, respiratory virus outbreaks) and system failures (eg, solar-powered O 2 battery depletion). These circumstances would be expected to increase the ICER through increased mortality (eg, insufficient oxygen supply) or costs (eg, backup cylinder oxygen). The choice of the comparator group would affect the ICER estimate. To provide several perspectives on the ICER, we used several comparators: null case with no oxygen (primary analysis), grid-powered oxygen concentrators, and fuel generator-powered concentrators (secondary analyses). Our DALY calculation did not include years lived with disability since children who survive an acute episode of hypoxemic severe pneumonia are expected to be discharged without permanent disability. 40,41 The time horizon of our analysis was 10 years 20 ; however, a longer time horizon could be more sensitive to variability in costs (eg, maintenance and equipment replacement costs) and stochastic events such as system failures and demand surges. Discounting of health outcomes is controversial. 20 We used a 3% discount rate without age-weighting for our base case but provided a sensitivity analysis that included no discounting for health outcomes. 20 The threshold used for cost-effectiveness in our study was based on GDP per capita; however, there has been some criticism of this methodology. 48 Finally, whereas oxygen has utility for many clinical situations, our analysis focused specifically on oxygen therapy for inpatients younger than 5 years with hypoxemia. We therefore caution against extrapolating our findings to other clinical conditions. Our analysis is relevant to rural or remote hospitals in LMICs with a pediatric inpatient ward that can be served with a single oxygen concentrator and should not be applied to other settings. Additional details of the assumptions and limitations of the analysis can be found in the eMethods in the Supplement.

Conclusions
The results of this economic evaluation suggest that solar-powered O 2 is a cost-effective intervention relative to the null case (no oxygen) for treating children younger than 5 years with hypoxemia when compared with the GDP per capita of target LMICs. Solar-powered oxygen delivery also appears to be cost-effective relative to grid-powered concentrators and cost-saving relative to fuel generatorpowered concentrators. Given the magnitude of pediatric pneumonia deaths, estimated at 900 000 per year, 2 a life-saving and cost-effective intervention such as solar-powered O 2 could represent an important tool toward improvements in global child survival.