Objectives: Clinical algorithms can be helpful in decisions about treatment and feeding options in infancy, but have had limited exposure to real data. This analysis uses data from a large clinical trial to test such algorithms, and thereby develop a successor which performs usefully in poor countries with high HIV-prevalence.
Methods: The ZVITAMBO trial followed 14 110 mother-baby pairs through infancy. 32% of mothers were HIV-positive. Infants were HIV tested regularly using DNA PCR. Clinical signs were evaluated in terms of identifying HIV-infection at 6 weeks, 6 and 12 months, using Zimbabwean, South African, and WHO generic adaptations of the WHO integrated management of childhood illness HIV algorithm. A modification, in which HIV-exposed infants are first divided into being at least or less than median weight-for-age, was derived and evaluated.
Results: At 6 weeks 65% of all infected babies are less than median weight-for-age. Adding at least two clinical signs reduces sensitivity to 20% but those identified are 1.5 (95% CI 1.1-2.1) times more likely to die by 6 months than other infected infants. At 6 months, 86% of uninfected babies of HIV-infected mothers can be identified by selecting those whose weight is greater than median or, if less than median, who exhibit <2 clinical signs.
Conclusions: In poor settings a simple clinical algorithm can identify children with probable HIV infection, especially those at risk of early death, who can then be referred for further testing and care, including highly active antiretroviral therapy. Most infants who are HIV-uninfected can be identified at 6 months and provided with support to maintain infection-free survival, including focussed infant-feeding counselling.