Abstract
Our objective was to evaluate HIV prevalence and identify risk factors for HIV infection among women attending the antenatal clinic (ANC) at a large public hospital in Kisumu town, western Kenya. Between June 1996 and November 1997, in the context of a study to determine the effect of placental malaria on mother-to-child transmission of HIV in western Kenya, HIV-1 antibody testing was offered to women with a singleton uncomplicated pregnancy of 32 weeks' gestation attending the ANC. Women were interviewed using a structured questionnaire and had a fingerstick blood sample collected for haemoglobin (Hb), malaria smears, and HIV antibody testing. Overall HIV seroprevalence was 26.1% (743/2844) (95% confidence interval (CI): 24.5-27.7) and in bivariate evaluation was significantly associated with anaemia (Hb< 11 g/dl) (risk ratio (RR) 1.8), malarial parasitaemia (RR 1.6), fever (axillary temperature 37.5 C at screening) (RR 1.6), a history of being treated for either vaginal discharge (RR 1.5) or tuberculosis (RR 1.6), reported alcohol consumption (RR 1.6), being an unmarried multigravida (RR 2.2) or a history of the most recent child having died (RR 2.0). Poisson regression analysis for all women identified 5 significant factors independently associated with HIV seropositivity: anaemia (adjusted RR 1.7; 95% CI 1.3-2.0), malarial parasitaemia (adjusted RR 1.7; 95% CI 1.4-2.0), a history of being treated for vaginal discharge (adjusted RR 1.5; 95% CI 1.1-2.0), fever (adjusted RR 2.0; 95% CI 1.3-3.2) and reported alcohol consumption (adjusted RR 1.6; 95% CI 1.1-2.5). Multigravidae women whose most recent child had died were also more likely to be HIV seropositive (adjusted RR 1.9; 95% CI 1.7- 2.8). Only 5.5% (156/2844) of the women had none of these risk factors, of whom 12% (18/156) were HIV(+). Even though the model containing the 5 identified factors fitted the data well (goodness-of-fit xi2=18.41, P =0.10), its collective capacity to predict HIV infection was poor; while 74% of the truly positive women were correctly predicted positive by the model, 52% of the truly negative women were misclassified. Among pregnant women attending the ANC in western Kenya, we were unable to identify a subgroup at risk of HIV infection using non-serological information, indicating that wherever possible universal access to voluntary HIV counselling and testing would be preferable to targeted screening.