Cofactor infections and HIV epidemics in developing countries: implications for treatment

Abstract

This article shows that the burden of certain tropical disease infections, after controlling for other factors, is positively correlated with HIV prevalence. Using cross-national data and multivariate linear regression analysis, we investigate the determinants of HIV prevalence in low- and middle-income countries. We begin with social and economic variables used in other cross-national studies and then incorporate data on parasitic and infectious diseases endemic in poor populations, which are found to be strongly and significantly correlated with – and are potent predictors of – HIV prevalence. The paper concludes by arguing that treating tropical diseases may be a cost-effective add-on to HIV-prevention and -treatment programs, thus slowing the spread of HIV in disease-burdened populations.

Keywords:

• HIV
• cofactors
• malaria
• parasites
• STIs

Notes

1. The data were drawn from UNAIDS (2006, 2007) (adult HIV prevalence in 2006); World Bank (World Development Indicators, Gini coefficient, per capita GDP-PPP, urbanization); UNDP (2005) (adult literacy rate, contraceptive use, malaria prevalence); US Department of State (2004) (percent Muslim); USAID (n.d.) (median age at first sex, female); World Bank (1997, pp. 318–324) (age of the epidemic); and World Health Organisation (2004) (DALYs for each condition). Countries with per capita income in 2003 of less than US$12,000 on a purchasing power parity basis were included in the analysis. Rather than imputing missing values, countries with missing data were dropped from the analysis, leaving a data set of 80 countries. The present study follows established practice by using the log form of HIV prevalence and per capita income, which reduces the influence of outliers and generally leads to more robust and efficient estimates when analyzing highly skewed variables such as these.

2. It would have been preferable to use HIV incidence to describe how the epidemic is unfolding. Prevalence includes historical information and is confounded by greater survival in countries with effective antiretroviral delivery programs. But incidence data are generally not available.

3. All measures of literacy and school enrollment in our sample are highly correlated and perform similarly as regressors. In our sample, the adult literacy rate and the female-to-male literacy ratio are virtually identical in a statistical sense (with a simple correlation of 0.93). Using the latter variable, however, gives the misleading impression that the variable measures gender discrimination; hence the present study uses the more straightforward variable, the adult literacy rate.

4. In a forthcoming article, the authors demonstrate that much of the elevated prevalence of HIV in southern Africa can be explained by the higher prevalence of endemic diseases in that region.

5. Because the vast majority of deaths from malaria are in infants, the DALYs associated with malaria are extremely high. That measure, however, does not reflect the burden of malaria among sexually active adults, unlike the DALYs from other diseases. We have therefore chosen to use malaria prevalence.

6. To check against the possibility of spurious correlation (that diseases in general or tropical diseases specifically are correlated with HIV prevalence rather than the ones for which there is evidence of an association), we added about 30 other diseases one at a time to the basic model. The only ones significantly (at the 95% or better level) and positively correlated with HIV prevalence were opportunistic infections (such as tuberculosis), which could not be included in the regressions due to endogeneity, and trachoma (other than immune dysregulation, we are not aware of any reason why trachoma might lead to HIV infection, but its statistical association of HIV suggests that further research may be fruitful).

7. Two important reasons why these coefficients might overstate the importance of cofactor infections are endogeneity (reverse causality) and specification errors (especially missing variables positively correlated with the cofactor disease variables).

8. The sexual behavior surveys reflect a bias in data collection common in AIDS policy documents. The assumption that the AIDS epidemics in sub-Saharan Africa result from something distinctive about African sexuality influences the decision to expend greater efforts to collect data on sexual behavior in African countries than in other parts of the world.