Seeing the invisible killer

In 12 general hospitals in Ethiopia, children with severe pneumonia and low blood oxygen who received locally made bubble CPAP had a treatment failure rate of 0.8%, against 3.4% on standard low-flow oxygen, a cut of about three-quarters ⁷. No child on bubble CPAP died in hospital. The device was built from parts costing less than US$2, excluding the oxygen source ⁷.

Hypoxaemia, too little oxygen in the blood, cannot be seen by eye. A child can look tired and breathe fast, and a health worker cannot tell by looking whether that child is at 96% or at 84%. The expanded review examines detection, oxygen delivery and respiratory support, including earlier trials that found harm as well as benefit.

The burden

An estimated 4.9 million children under five died worldwide in 2024 ¹. Sub-Saharan Africa and Southern Asia together accounted for more than 80% of those deaths ¹. After decades of progress, the total has stagnated at roughly 4.8-4.9 million a year since 2022 ¹. WHO lists pneumonia among the leading causes, alongside preterm birth complications, birth asphyxia, diarrhoea and malaria, all largely preventable or treatable ¹.

WHO's pneumonia fact sheet puts the toll at 740,180 children under five in 2019: 14% of all under-five deaths, and 22% of deaths in children aged 1 to 5 ². Deaths are highest in southern Asia and sub-Saharan Africa ².

The Global Burden of Disease (GBD) 2021 study, led by the Institute for Health Metrics and Evaluation, gives a newer estimate for lower respiratory infections (pneumonia and bronchiolitis). It estimated 502,000 deaths in children younger than 5 in 2021 (95% uncertainty interval 406,000-611,000) ³. About half of these, 254,000, occurred in countries with a low Socio-demographic Index ³. Streptococcus pneumoniae was the single largest cause of lower respiratory infection deaths across all ages ³.

The trend is down. From 1990 to 2019, the global all-age death rate from these infections fell by 41.7% ³. The two sets of numbers differ because they use different methods and years, but both say the same thing: hundreds of thousands of young children still die of pneumonia each year, most of them in Africa and South Asia.

Hypoxaemia is how many of those children die. A Murdoch Children's Research Institute (MCRI) meta-analysis of 213 studies and 601,757 patients, mostly from sub-Saharan Africa and South Asia, found hypoxaemia in 24.5% of admitted newborns and 12.1% of admitted children ⁴. Patients with hypoxaemia had 4.84 times the odds of death (95% CI 4.11-5.69) ⁴.

The Lancet Global Health Commission on medical oxygen security, led from MCRI, puts the gap in stark terms. In low- and middle-income countries, only 30% of the 299 million people who need oxygen for acute medical or surgical conditions receive adequate oxygen therapy, with the lowest access in sub-Saharan Africa ¹². The Commission calls this 70% coverage gap larger than the gaps for HIV/AIDS (23%) and tuberculosis (25%) treatment ¹².

How it works

Every cell in the body needs oxygen. The lungs move it from the air into the blood. In pneumonia, the small air sacs of the lung fill with fluid and pus, and less oxygen gets through. A young child compensates by breathing faster and harder. When that is not enough, blood oxygen falls. The brain, heart and other organs start to fail.

Measuring it. A pulse oximeter clips onto a finger or toe and shines red and infrared light through the skin. Blood carrying oxygen absorbs light differently from blood without it. The device turns that difference into a number, the oxygen saturation, or SpO2. WHO recommends hospital admission and oxygen for children with pneumonia whose SpO2 is below 90% ⁵.

This is why oximetry matters. Signs such as blue lips appear late and are easy to miss, especially in poor light or on darker skin. The oximeter turns a hidden danger into a number that any trained health worker can act on.

Treating it. The standard treatment is low-flow oxygen: oxygen from a cylinder, a concentrator or a piped supply, given through small prongs in the nose. For many children that is enough.

Some children with severe pneumonia need more help. Bubble CPAP (continuous positive airway pressure) keeps a steady gentle pressure in the airways, which holds the tiny air sacs open so they can take up oxygen. The "bubble" is simple: the child breathes out through a tube whose end sits under water in a bottle. The depth of the tube sets the pressure. The Ethiopian version was assembled locally from basic parts ⁷.

What has to be in place. A reading is only the start. For a child in a health centre, low SpO2 means urgent referral, transport, a hospital with oxygen, and staff who know how to give it. For a child in hospital, it means a reliable oxygen source, working equipment and nurses who check the child often. The Commission describes the needed system across production, storage, distribution, supply, clinical use, coordination, regulation and financing ¹². Each link can break.

What the evidence shows

2015–2019: respiratory support helped in one setting and harmed in another

In a 2015 trial at icddr,b in Dhaka, Bangladesh, 225 children with severe pneumonia and hypoxaemia received bubble CPAP, low-flow oxygen or high-flow oxygen. Deaths were 4% with bubble CPAP and 15% with low-flow oxygen. The trial stopped early, which makes the size of the benefit less certain ¹⁵.

A 2019 trial in a Malawian district hospital gave a different result. Among 644 children with severe pneumonia and high-risk conditions, mortality was 17% with bubble CPAP and 11% with low-flow oxygen (risk ratio 1.52, 95% CI 1.02–2.27). Care was delivered without daily physician supervision ¹⁶. The populations and care environments differed, so the contrast cannot isolate which difference caused harm. It does rule out assuming that a low-cost CPAP device is beneficial in every ward.

A separate 2019 stepped-wedge trial in 12 Nigerian hospitals also complicates the story. A full oxygen-system package did not improve child mortality compared with the preceding pulse-oximetry phase (adjusted odds ratio 1.06, 95% CI 0.77–1.46); neonatal mortality was higher (1.45, 1.04–2.00). The authors noted low statistical power and major economic changes affecting hospitals ¹⁷. Installing an oxygen system and demonstrating an additional survival benefit are different outcomes.

2021: COAST tested oxygen strategies

COAST randomised children in Uganda and Kenya to respiratory-support strategies. Among those with oxygen saturation of 80–91%, liberal oxygen did not significantly change 48-hour mortality versus permissive hypoxaemia with rescue oxygen below 80% (adjusted odds ratio 1.16, 95% CI 0.49–2.74) ¹⁸. Recruitment stopped early, well below the planned sample. The broad interval does not establish equivalence or justify withholding oxygen in routine care; it does show that oxygen thresholds have been tested, with important uncertainty remaining.

Ethiopia: locally made bubble CPAP in general hospitals

The trial was led by the Armauer Hansen Research Institute, with funding from SIDA Sweden and Grand Challenges Ethiopia ⁷. It was a cluster-randomised trial in 12 general (secondary) hospitals. Six hospitals used bubble CPAP as first-line support; six used standard low-flow oxygen ⁷. The trial is registered as NCT03870243.

Between June 2021 and July 2022, it enrolled 1,240 children aged 1-59 months with severe pneumonia and hypoxaemia, 620 in each group ⁷. Treatment took place on ordinary paediatric wards, with about one nurse per six patients and one general practitioner per 18 patients ⁷.

The primary outcome was treatment failure: very low oxygen despite treatment, a need for mechanical ventilation, death, or leaving hospital against medical advice. It occurred in 5 of 620 children (0.8%) on bubble CPAP and 21 of 620 (3.4%) on low-flow oxygen (adjusted RR 0.24, 95% CI 0.07-0.87) ⁷. Six children died in hospital, all in the low-flow group (p=0.031) ⁷. No serious adverse events were attributed to bubble CPAP ⁷.

Caveats: the trial was open-label, and overall mortality was low, eight deaths among 1,240 children (0.6%) ⁷. The authors note low HIV, malaria and malnutrition rates in the study area, and call for implementation research in higher-mortality settings ⁷. When deaths within 72 hours of leaving against medical advice were added, the difference in deaths was no longer significant (RR 0.14, 95% CI 0.02-1.16) ⁷.

Uganda: an oxygen package across 31 facilities

MCRI worked with Ugandan partners, including Makerere University and the Ministry of Health, on a health-system package based on Uganda's National Scale Up of Medical Oxygen Implementation Plan 2018-2022 ⁸. It covered 24 level IV health centres and seven general or regional referral hospitals in the Busoga and North Buganda regions ⁸.

The package included clinical training, mentorship and supervision; pulse oximeters and oxygen cylinders; support for biomedical engineers; and oxygen supply strategies and guidelines ⁸. It was funded by the Bill & Melinda Gates Foundation and ELMA Philanthropies ⁸.

The study included 71,997 admitted newborns and children ⁸. The primary outcome was pulse oximetry on the first day of admission. It rose from 23.7% before the intervention to 87.7% after (adjusted OR 40.10, 95% CI 37.38-42.93) ⁸. The authors estimate that an additional 1,623 newborns and 1,266 children received oxygen during the 16-month post-intervention period than would have been expected without the package ⁸.

Caveats: this was a before-after study with no control group. In-hospital deaths among children were 1.9% before and 1.6% after; among newborns, 7.8% before and 8.2% after ⁸. The study showed that screening can scale, not that the package cut deaths.

How much low oxygen is dangerous: the North Carolina meta-analysis

A University of North Carolina team updated a 2015 review of hypoxaemia and death in children under five with lower respiratory infection in low- and middle-income countries ⁵. It combined 44 studies, mostly from Africa and Asia, with 33 studies and 155,633 children in the main analysis ⁵.

Any hypoxaemia carried 4.36 times the odds of death (95% CI 3.52-5.39) ⁵. An SpO2 below 90% carried 4.75 times the odds (95% CI 3.42-6.58) ⁵. Children at 90-94%, often treated as acceptable, still had 2.27 times the odds of death compared with those at 95-100% (95% CI 1.22-4.25) ⁵.

Caveats: heterogeneity was substantial, and 8 of the 33 studies had a high risk of bias, though a sensitivity analysis without them gave similar results ⁵. These are associations. They do not show that giving oxygen at 90-94% would save lives.

West Africa: oximetry in primary care (AIRE)

The AIRE project, run by ALIMA, Inserm and Terre des hommes with Unitaid funding, added routine pulse oximetry to Integrated Management of Childhood Illness (IMCI) consultations in 16 primary health centres in Burkina Faso, Guinea, Mali and Niger ⁶. From July 2021 to July 2022 it followed 1,998 severely ill children for 14 days ⁶.

Of these, 7.1% had severe hypoxaemia (SpO2 below 90%) and 10.5% had moderate hypoxaemia (90-93%) ⁶. Day-14 mortality was 26.1% with severe hypoxaemia, 7.5% with moderate hypoxaemia and 2.3% without ⁶. After adjustment, the odds of death were 9.34 times higher with severe hypoxaemia (95% CI 5.08-17.16) and 2.32 times higher with moderate hypoxaemia (95% CI 1.16-4.64) ⁶.

Oximetry did prompt action: 83.8% of severely hypoxaemic children had a referral decision and 82.3% were hospitalised ⁶. But only 34.5% received oxygen ⁶. The study was observational and registered retrospectively (PACTR202206525204526) ⁶.

TIMCI India and Tanzania: the honest null

The largest trial of the period tested oximeters in primary care head-on. Swiss TPH and PATH ran it with Unitaid funding, in 172 primary care facilities: 106 in India and 66 in Tanzania ¹⁰. It is registered as NCT04910750.

Facilities were randomised to pulse oximetry or control; in Tanzania a third arm added a digital clinical decision support algorithm (CDSA) ¹⁰. Between March 2022 and April 2023 the trial enrolled 157,677 sick children ¹⁰.

There were two primary outcomes. Severe complications (death, delayed hospitalisation, or hospitalisation without referral) by day 7 were rare: 0.1% in control facilities among children aged 2-59 months, against 0.3% in the oximetry arm (adjusted risk difference 0.1%, 95% CI 0.0-0.2) ¹⁰. Referred hospitalisations within 24 hours did not rise ¹⁰. The oximetry-plus-CDSA arm showed no significant differences either ¹⁰.

The authors conclude that referral barriers, unequal oxygen access and hospital care quality must be addressed before these tools can improve child outcomes ¹⁰.

TIMCI Kenya and Senegal: fewer antibiotics

A companion pre-post study led by Swiss TPH introduced oximeters and CDSAs in primary care in Kenya and Senegal ⁹. It enrolled 50,580 sick children (NCT05065320) ⁹.

Urgent referrals did not change ⁹. Only 11.4% of children aged 2-59 months with a reading that called for urgent referral were actually referred ⁹. But antibiotic prescribing fell by 22.6% (95% CI 18.3-26.9) in children aged 2-59 months and by 14.6% (8.7-20.6) in infants aged 1-59 days, while caregiver-reported recovery at day 7 stayed stable ⁹.

Caveats: there was no control group. Uptake was 75% in Kenya and 40% in Senegal, where a long health-worker strike disrupted the work ⁹.

Malawi: testing oxygen at 90-93%

At Salima District Hospital, a Johns Hopkins-led pilot (NoGoLO2, NCT06176664) randomised children with pneumonia and SpO2 of 90-93% to no oxygen, low-flow oxygen or high-flow nasal oxygen ¹¹. It enrolled 21 of 29 eligible children, had an average of 0.8 protocol deviations per participant, and followed all children to day 14 ¹¹. It showed only that a full trial is feasible.

Results at a glance

LimitsWhat didn't work (or the limits)

Oximeters alone did not save lives. TIMCI in India and Tanzania found no survival benefit ¹⁰. Oximeters, with or without decision support, did not increase referred hospitalisations or reduce deaths ¹⁰. This does not mean oximeters are useless. It means that a number on a screen does not help a child unless someone acts on it and there is somewhere to send that child.

Referral rarely follows detection. In Kenya and Senegal, only about one in nine children aged 2-59 months with a referral-level reading was referred ⁹. The authors point to low hypoxaemia prevalence in primary care and barriers to referral ⁹. Families may face transport costs, distance and fees. Health workers may doubt the reading or know the hospital has no oxygen.

Detection without oxygen. In AIRE, two in three severely hypoxaemic children never received oxygen ⁶. Among moderately hypoxaemic children, only 7.1% did ⁶. The Commission found pulse oximeters in only 54% of general hospitals in low- and middle-income countries, and oxygen therapy in only 58% ¹². In primary health-care facilities, both are nearly entirely unavailable ¹².

Oximetry is not yet routine. The Commission found pulse oximetry was done in only 19% of people presenting to general hospitals in these countries, and almost never at primary level ¹². It also flags a need to improve oximeter accuracy in people with darker skin and in infants and young children ¹².

The bCPAP safety boundary matters. The Ethiopian result must be read alongside the higher mortality in the earlier Malawi trial ⁷¹⁶. Patient selection, clinical oversight, monitoring and escalation capacity need evaluation; device price alone cannot establish safe delivery.

The Uganda study measured screening, not survival. Its before-after design shows oximetry coverage rising, not deaths falling ⁸.

Optimal thresholds remain uncertain. COAST tested strategies in children at 80–91%, with rescue treatment and imprecise mortality estimates ¹⁸. The specific observational risk at 90–94% ⁵⁶ and the small NoGoLO2 pilot ¹¹ do not establish the best treatment threshold for that whole range.

Cost and scale

The devices are cheap. The Ethiopian team built bubble CPAP from parts costing less than US$2, excluding the oxygen source, against US$5,000 to US$18,000 for commercial bCPAP devices ⁷.

The system is not. The Commission estimates that closing the acute oxygen coverage gap in low- and middle-income countries would need an additional US$6.8 billion a year, or US$34.0 billion over five years ¹². That figure covers all ages and excludes pandemic surges and long-term oxygen therapy ¹².

Operating costs, such as maintenance, spare parts, power and staff, account for 50-80% of total oxygen system costs, yet have received little investment ¹². Previous efforts, including COVID-19 spending, largely delivered equipment without the systems and people to keep it running ¹².

It is still good value. The Commission judges medical oxygen as cost-effective as routine childhood immunisation ¹². It urges governments to include pulse oximetry and oxygen in universal health coverage schemes, so that costs do not fall on families ¹².

Who pays and delivers. In the studies reviewed here, funding came from Unitaid (TIMCI and AIRE) [6,9,10], the Bill & Melinda Gates Foundation and ELMA Philanthropies (Uganda) ⁸, and SIDA Sweden and Grand Challenges Ethiopia (bCPAP) ⁷. Delivery ran through ministries of health and public facilities: Uganda's national oxygen scale-up plan ⁸, Ethiopia's general hospitals ⁷, and government primary care in India, Tanzania, Kenya and Senegal [9,10].

Rollout status. Fewer than 30 countries have developed national oxygen plans, as called for in WHO's 2023 resolution on increasing access to medical oxygen ¹². The Commission asks all governments to do so by 2030 ¹². It also calls on donors to support the new Global Oxygen Alliance and to replenish the Global Fund with a strong oxygen-access mandate ¹².

What to watch

Oxygen at 90-93%. NoGoLO2 showed that a full trial comparing no oxygen, low-flow and high-flow oxygen at SpO2 90-93% is feasible in Malawi ¹¹. A definitive trial would settle whether current thresholds should change.

Better oximeters for small children. A Johns Hopkins-led trial with Stellenbosch University and Karolinska Institutet (NCT05914324) is testing two new paediatric pulse oximeters in outpatient care in Cape Town, South Africa ¹³. Its primary outcome is the proportion of children whose oxygen saturation is correctly managed ¹³. It enrolled 936 children and its primary completion date was April 2026 ¹³. Results are pending.

Bubble CPAP beyond Ethiopia. A Yale University feasibility trial in Pakistan (NCT07620054), with the US National Heart, Lung, and Blood Institute, will test a team-based care strategy for bubble CPAP in children with severe pneumonia ¹⁴. It is not yet recruiting; its estimated start is April 2028 ¹⁴.

Open questions. Can referral from primary care be made to work, through transport, fee waivers or better links to hospitals [9,10]? Can oxygen be brought closer to children, into health centres where the Commission finds it almost absent ¹²? And can oximeters be made accurate for infants and children with darker skin ¹²?

National plans. Watch which ministries publish national oxygen plans ahead of the Commission's 2030 target ¹².

What it means

Hypoxaemia is common among sick children in hospital, and it multiplies the risk of death, including at levels often considered moderate [4-6]. A pulse oximeter makes it visible.

But a reading is only useful if something happens next. TIMCI shows that handing out oximeters, on their own, does not reduce deaths ¹⁰, and AIRE shows most hypoxaemic children in West African primary care did not get oxygen ⁶. Uganda improved screening ⁸ and Ethiopia reduced treatment failure ⁷, but Nigeria found no additional child-survival benefit from its full oxygen package ¹⁷, and Malawi found harm with bCPAP ¹⁶. Those endpoints and settings should not be combined into a single claim that every package saves lives.

For ministries of health: plan oximetry, referral and oxygen supply together, in a national oxygen plan ¹². Uganda's national scale-up plan shows that near-universal screening in hospitals is achievable within two years ⁸. Budget for running costs, not only equipment ¹².

For funders: pay for systems, not boxes. The Commission's finding that operating costs are 50-80% of the total, and have been neglected, is the guide ¹². The Ethiopian trial shows that low-cost, locally made respiratory support can work outside tertiary centres ⁷. The Kenya and Senegal data show oximeters bring an added benefit in reducing unneeded antibiotics ⁹.

For clinicians and nurses: measure SpO2 in every sick child, treat below 90%, and watch closely those at 90-94%, who face higher risk [5,6]. Bubble CPAP needs an appropriate clinical protocol, patient selection and close monitoring; the Malawi harm signal prevents a general recommendation based on the Ethiopian trial alone ⁷¹⁶.

For researchers: the next questions are whether children at 90-94% benefit from oxygen ¹¹, and how to make referral from primary care actually happen [9,10].