ANALYSIS

Low blood oxygen sharply raises death risk in children with pneumonia in poorer countries

A meta-analysis of more than 155,000 children finds mortality is elevated even at oxygen saturations of 90–94%, above the threshold at which WHO advises hospital oxygen.

Odds of death vs a less-hypoxaemic reference group, by thresholdAny hypoxaemia vs none: 4.36; SpO2 <90% vs 90–100%: 4.75; SpO2 90–94% vs 95–100%: 2.2703.57Any hypoxaemia vs none4.36SpO2 <90% vs 90–100%4.75SpO2 90–94% vs 95–100%2.27
Odds of death vs a less-hypoxaemic reference group, by threshold
GroupValue (value)
Any hypoxaemia vs none4.36 (3.52 to 5.39)
SpO2 <90% vs 90–100%4.75 (3.42 to 6.58)
SpO2 90–94% vs 95–100%2.27 (1.22 to 4.25)
Odds of death vs a less-hypoxaemic reference group, by threshold Odds ratios from the meta-analysis. Each bar uses a different reference group, shown after the label; whiskers are 95% confidence intervals. Source: The Lancet Global Health

Low blood-oxygen levels are one of the clearest warning signs that a child with a chest infection may die, and a new systematic review sharpens the picture of how much risk attaches to which oxygen level. Published in The Lancet Global Health, the analysis pools studies of children under five with lower respiratory infections (LRIs) in low-income and middle-income countries (LMICs) [s1].

The headline is not that low oxygen is dangerous — that has been known for years — but where the danger begins. The authors report elevated mortality even in a band of oxygen saturations that current WHO guidance treats as not yet requiring hospital oxygen [s1].

What was pooled

The review updated a 2015 analysis, searching six databases for studies published between 1 January 2015 and 18 June 2025 and combining the newly eligible studies with those from the original review [s1]. It identified 26 new studies that met inclusion criteria and combined them with 18 from the original review, for 44 studies published between 1993 and 2024 [s1]. Most were from Africa (25 of 44, or 57%) or Asia (19 of 44, or 43%), with some spanning multiple locations [s1]. The primary meta-analysis drew on 33 of those studies and 155,633 participants [s1]. The protocol was registered on PROSPERO (CRD42023433946) [s1].

The numbers

Hypoxaemia of any threshold was associated with higher odds of death from LRI than no hypoxaemia, with an odds ratio of 4.36 (95% confidence interval 3.52 to 5.39) [s1].

Broken down by oxygen level, the pattern is graded. For an oxygen saturation (SpO2) below 90% compared with 90–100%, the odds ratio for death was 4.75 (95% CI 3.42 to 6.58) [s1]. WHO recommends hospitalisation and oxygen administration for patients with SpO2 below 90%, so a raised risk in this band is expected [s1].

The finding that carries a policy implication is the next one up. For an SpO2 of 90–94% compared with 95–100% — a band above the WHO oxygen threshold — mortality risk was more than twice as high, with an odds ratio of 2.27 (95% CI 1.22 to 4.25) [s1].

What the authors take from it

Their interpretation is measured: SpO2 below 90% strongly predicts mortality in children with LRIs in LMICs, and children with an SpO2 of 90–94% also have elevated risk, which suggests that paediatric LRI and pneumonia treatment algorithms should consider management at this hypoxaemia threshold [s1]. That is a call to reconsider a cut-off, not a claim that raising oxygen at 90–94% has been shown to save lives — a distinction the study design cannot cross, because it measures association, not the effect of treatment.

The limits, stated plainly

The authors are direct about the weaknesses. Heterogeneity across the analyses was substantial, with an I-squared statistic of 64–85% — meaning the individual studies' results varied more than chance alone would explain, so the pooled odds ratios are summaries of a scattered literature rather than a single consistent effect [s1]. Eight of the 33 studies in the primary meta-analysis (24%) had a high overall risk of bias; a sensitivity analysis restricted to studies at low or moderate risk of bias yielded similar results, which is reassuring but does not remove the underlying variability [s1].

There is also the structural limit of an observational pooling: children with lower oxygen are sicker in other ways too, and an odds ratio cannot fully separate the oxygen level from everything that travels with it. The value of the 90–94% finding is that it flags a group worth studying and possibly managing differently — the trial-grade question of whether acting at that threshold changes outcomes is the one that follows.

Why it matters

Oxygen is a treatment that many hospitals in high-income settings take for granted and many in LMICs cannot reliably supply. A pulse oximeter reading is one of the cheapest severity signals available, and where the cut-off for concern is set determines who gets escalated. This review argues the current line may be drawn a few percentage points too low for the children who most need the margin [s1]. It stops short, appropriately, of claiming that treating at the higher threshold has been shown to save lives — that is the trial the association points toward, not a conclusion the association can deliver on its own.

This article summarises a single meta-analysis and does not offer medical advice.

Sources

Sources

  1. Hypoxaemia and mortality in children with lower respiratory infection in low-income and middle-income countries: systematic review and meta-analysisThe Lancet Global Health , September 1, 2026

More on

Related coverage