AN INVESTIGATION ASSESSING THE FRACTION OF HUMAN-IMMUNODEFICIENCY-VIRUS INFECTION ATTRIBUTABLE TO ETHNIC-GROUP VARIATIONS IN RISK

Considerable speculation and a recent publication has suggested that m uch of the human immunodeficiency virus (HIV-1) infection in the subse t of the UK population represented by women having babies in Metropoli tan London is substantially imported from Africa. Other data are suppo rtive of this hypothesis. Ades et al. (1992) concluded that the fracti on of HIV seroprevalence, identified in a large London-based study usi ng neonatal blood from the dried blood spots on Guthrie cards, attribu table to mothers born in Africa was 92.4% with 95% Cl : 82-100%. This is an important observation which requires confirmation. This paper re ports on an attempt to do this with closely similar methodology using the Unlinked Anonymous Survey (UAS) of HIV-1 seroprevalence in pregnan t women attending antenatal centres, together with data from the Offic e of Population Censuses and Surveys (OPCS) on the country of origins of mothers delivering in 1991 at the 15 London centres included in the UAS. As in Ades' analysis there appeared to be a strong association b etween centre prevalences and the proportions of women of African orig in delivering at those centres. This was consistent with a high fracti on of seroprevalence attributable to such women, but it was also clear that the assumptions required for the estimation of the attributable fractions were not fully met. Analyses modified to justify the assumpt ions produced attributable fraction estimates ranging from 44% to 96% with tower 95% confidence limits 20% or less. There is clearly an asso ciation and it may result from a proportion of HIV infections being im ported to the UK from elsewhere, but the nature of the variation makes estimation of the magnitude of such effects somewhat speculative. Fur thermore, associations observed in such ecological analyses are not ge nerally adequate for inferring a causal relationship. The correct expl anation of these associations can only be identified by studies when i nformation on all the necessary variables is available at the individu al level.