HIV-related pneumonia is the main cause of paediatric hospital admissions in southern Africa. We aimed to measure predictors of treatment failure and the cause of non-responsive pneumonia in children admitted to hospital with severe pneumonia in Durban, South Africa.

We investigated 358 children aged 1–59 months who presented with WHO-defined severe or very severe pneumonia. Children were recruited irrespective of HIV status and started on a standard antimicrobial regimen of benzylpenicillin and gentamicin. All infants also received high-dose trimethoprim-sulfamethoxazole. The primary outcome measure was treatment failure at 48 h.

242 (68%) children were HIV infected, 41 (12%) HIV exposed, uninfected, and 75 (21%) HIV uninfected. Failure to respond by 48 h was predicted by age under 1 year (adjusted odds ratio 6·38, 95% CI 2·72–14·91, p<0·0001), very severe disease (2·47, 1·17–5·24, p=0·0181), HIV status (HIV infected 10·3, 3·26–32·51; HIV exposed, uninfected 6·02, 1·55–23·38; p=0·0003), and polymicrobial disease (one organism 2·06, 1·05–4·05; two organisms 10·75, 4·38–26·36; p<0·0001) on logistic regression analysis. All children with three organisms failed treatment. 72/110 treatment failures had at least two organisms isolated. Three of nine HIV-exposed, uninfected infants, 29/74 HIV-infected, but no HIV-uninfected infants who failed study therapy had Pneumocystis jirovecii pneumonia.

For children younger than 1 year, the WHO guidelines are inadequate and need to be revised since both HIV-infected and HIV-exposed, uninfected infants had more treatment failures than did HIV-uninfected infants. Polymicrobial disease is an important reason for treatment failure, and we need to identify rapid low-cost diagnostic methods to assist clinicians.