Abstract
In their article ‘Horizontally-acquired HIV infection in Kenyan and Swazi children’ (Vol. 20, 2009), Okinyi et al. present an estimated prevalence and possible causes of horizontal HIV transmission in Swazi and Kenyan children. While we appreciate the attention to this neglected aspect of HIV transmission research, we have grave concerns about generalizations made based on two very small, poorly controlled data-sets. Swaziland has the highest prevalence of HIV worldwide at 26%. 1 Suggesting that 20% or more of HIV-positive children in Swaziland are infected through medical procedures has serious implications on the use of health-care services in Swaziland and unfairly calls into question the integrity of those fighting on the front-line of this sobering epidemic.
Using the 2006–2007 Swaziland Demographic and Health Survey (DHS) to evaluate the prevalence of horizontally acquired HIV in Swaziland has serious limitations. While the survey did follow rigorous standards to ensure a representative sample of the Swazi population, methods used to evaluate relationships between household members had no verification procedures. The mother–child relationships reported in the DHS survey are based on respondent reporting and not on records (e.g. birth certificates or child health cards), nor confirmed genetically. In a context where 38% of children do not live with their biological mothers, and where 18% of women live in polygamous households, 2 this is a critical limitation. Given the small sample of 11 discordant mother–child pairs among 50 HIV-infected children reported to be living with their mothers, even the mis-classification of one or two relationships makes a major difference to the calculated prevalence of presumed horizontal transmission.
In addition, the authors' methods for extrapolating estimates to the general Swazi population are erroneous. While they use DHS prevalence estimates for children aged 2–12 years living with their mothers, they generalize this to a Swazi population aged 0–14 years. The denominator applied to the DHS estimate under-represents the number of HIV-positive children in the general Swazi population by excluding children age 0–2 years, i.e. those who are most likely to have acquired HIV vertically and who have a 50% risk of mortality by their second birthday if they are not diagnosed and treated. 3 Further, we would argue that a population of orphans not living with their mothers is significantly different with regard to their HIV status and mode of transmission than children living with their mothers. These children are more likely to be HIV-positive and likely to have acquired the HIV vertically because maternal death implies high maternal viral load and HIV is more likely to be transmitted from mother to child with higher viral loads. 4
The Kenyan case-control study, whose findings on transmission correlates were crudely applied to Swaziland, also has serious methodological flaws. We find it concerning that no P values were reported for exposures since the possibility for statistical error is high in a sample of eight sibling pairs. Displaying results as percentages with this sample size is also misleading. Regarding the main outcome variable of ‘diversity of exposures’, we consider a paired t-test with no log transformation inappropriate for such a small sample size and question why the actual P value was not reported. Non-parametric methods would have been more appropriate. Finally, it is unclear why no multivariate analysis was conducted; attempts to control for confounding and interaction are limited to a statement that ‘cases and controls had similar reasons for most health-care-related exposures’.
To make conclusions on correlates of horizontal transmission in Swaziland based on a small Kenyan study with limited analysis is premature and potentially inaccurate. While we agree that iatrogenesis is a potentially significant contributor to horizontal HIV transmission, the implication that Swazi health-care facilities have unhygienic practices with no exploration of differences between modern and traditional practitioners is problematic. Using evidence on nosocomial HBV and HCV transmission to support their conclusions may exaggerate the HIV transmission risk since the risk of HBV and HCV transmission through percutaneous exposure is approximately six times greater than the risk of HIV transmission. The exploration of other potential determinants of horizontal infection was also weak. For example, applying a national sexual violence prevalence estimate to a sample of 11 children may be inaccurate; and the assumption that Swazi children under 12 have never engaged in sexual intercourse is incorrect, since studies suggest this may not be the case. 5,6
Unfortunately the publication of this study has detrimentally affected both health-care provision and research in Swaziland. The popular media has branded nurses as ‘killers’, accusing them of causing one in five paediatric HIV infections, 7 thereby compromising intensive efforts by the Ministry of Health to increase uptake of HIV-related services. Some facilities, questioning how study findings will be interpreted, have confiscated research instruments. While journal editors and authors cannot control how published studies are used, they are responsible for accurately describing the scope of their findings. Reporting that an estimated 20% of HIV-infected Swazi children aged 2–12 are horizontally infected and suggesting ‘iatrogenic HIV transmission is not uncommon in these children’ overreaches the limitations of the data-sets used. The findings should have been reported as hypothesis-generating observations rather than as definitive conclusions. The emphasis of the discussion could then have focused around the need for more rigorous research to precisely quantify the prevalence and elucidate the modes of horizontal transmission. The likelihood of sensationalized reaction to the findings could thus have been minimized. To correct the misinterpretation of these data we ask the authors to write a follow-up paper and press release clarifying their statistical methods and rectifying the generalized conclusions made.