The role of health care in the spread of HIV/AIDS in Africa: evidence from Kenya

Abstract

In 2006, Deuchert and Brody1 presented findings from an analysis of the 2003 Kenya Demographic and Health Survey (KDHS) suggesting that prophylactic tetanus toxoid (TT) injection during pregnancy is a vector for HIV. The letter from Schmid et al.2 raised important programmatic issues but both the original analysis and response have overlooked a critical aspect of the design.3 The KDHS sample design requires the use of sample weights and survey design variables to generate accurate variance estimates and test statistics. We were able to replicate Deuchert and Brody's analysis only by using unweighted data. However, when properly weighted data and the design variables are used, the association between HIV prevalence and prophylactic TT injections was not found to be statistically significant.

The KDHS is a household survey designed to produce national estimates as well as separate estimates for each of the eight provinces, and urban and rural areas.4 The two-stage sample design consists of 400 clusters, stratified by urban and rural areas in the first stage; and 10,000 households selected in the second stage. The survey was completed by 8195 women and 3578 men; 3151 women and 2851 men were tested for HIV. This analysis is limited to 1616 women who gave birth in the previous five years; who knew whether or not they had received a TT injection during their most recent pregnancy; and for whom HIV test results were available.

To further assess the association between TT injection and HIV prevalence, we calculated HIV prevalence and ran regression models for additional variables of interest. No statistically significant association was found between HIV prevalence and TT injections, condoms used as contraceptive method, condoms used at last intercourse, number of partners in the last 12 months, ever receiving money or gifts in exchange for sex or current working status. Unadjusted analysis found several significant behavioural or demographic variables associated with HIV, including province, religion and years of sexual activity, however, after adjusting for age, rural or urban residence, and other variables in the model, only current marital status, wealth index and having had a sexually transmitted disease (STD) or symptoms of an STD remained significant.

The Deuchert and Brody unweighted odds ratios and 95% confidence intervals are presented in bold in Table 1. Their regression analysis, adjusted for age and provincial HIV prevalence, shows a statistically significant difference in the prevalence of HIV by TT status. However, this difference was not statistically significant when we replicated the analysis using the weighted data and design variables using the SUDAAN software package.5 Analysis was completed with and without North Eastern Province, which had no cases of HIV in the DHS 2003 sample and was not part of the previous DHS samples.