Marseille 2007
| Abstract #510 - What Drives HIV Epidemics in Developing Countries? | ||||
|
Session: 16.2: Methodology Matters (Parallel) on Monday @ 14.00-16.00 in 5 Chaired by Graham Hart, Dominique Costagliola Authors: Presenting Author: Prof Eileen Stillwaggon - Gettysburg College, United States |
||||
| Additional Authors: Prof Larry Sawers, | ||||
| Aim: To demonstrate the possibilities and limitations of regression analysis for explaining cross-national differences in HIV prevalence and to suggest new models that explain the divergence in incidence in various regions. Rationale: To devise effective prevention policies and treatment protocols, we need a theory of disease causation and epidemic spread that includes relevant social/behavioral, economic, and biological factors influencing individual transmission and population dynamics. | ||||
| Method / Issue: The paper presents cross-national multivariate OLS regressions in which the dependent variable is HIV prevalence. The regressors include factors significantly correlated with HIV prevalence in past cross-national regression studies or indicated in the biomedical literature as cofactors of HIV transmission. These include social/behavioral factors (female literacy rates, female-to-male literacy ratios, percent Muslim, measures of or proxies for risky sexual behaviors, net migration, urbanization, etc.), economic factors (including measures of poverty and inequality, GDP per capita), and biological or health status factors that affect transmission rates (wasting, change in protein consumption or per capita food production, and prevalence of malaria, tuberculosis, schistosomiasis hematobium, and helminth infection). | ||||
| Results / Comments: Analysis of about 60 developing countries (for which complete data are available) indicates that many variables expected to be significantly correlated with HIV prevalence have significant t values. Estimated equations have high R-squares. Multivariate regression analysis provides strong support for the hypothesis that numerous social/behavioral, economic, and biological variables influence the course of the epidemic. Although the regression analysis presented generates good results, the authors argue that the model provides only partial explanation of HIV epidemics in developing countries. The first problem is multicollinearity, although instrumental variables and principal component factors reduced estimation bias. More important are specification errors. Inability to obtain measures of schistosomiasis prevalence, STI prevalence, or consistent cross-national measures of risky sexual behavior produces missing variable bias. Other missing variables include measures of iatrogenic transmission of HIV and prevalence of male circumcision. Another specification error is the inability of OLS to model, except in the most primitive fashion, the myriad biological interactions that produce nonlinearities. Multivariate regression analysis can suggest relevant variables but cannot fully explain the complexity of the epidemics or their divergent trajectories. | ||||
| Discussion: Epidemics have characteristics of complex, contingent systems in which successive iterations are determined by interactions of variables in previous rounds. Disease and treatment interactions have non-linear effects or positive feedbacks that are not well represented in regression analysis. Moreover, regression analysis is based on stochastic processes. Disease transmission at the individual level may be unpredictable, but it is not random. Multiple variables determine the risk of transmission and the trajectory of an epidemic. At the population level stochastic processes play some role, but that process probably cannot be represented with regression analysis. To represent the multiple and interacting forces determining the risk of HIV in a population and the trajectory of an epidemic through that population, it is necessary to employ a model that reflects nonlinearities and interactions. Models from evolutionary biology and other applications of complexity theory are proposed as better representations of HIV dynamics. | ||||
| Go Back | ||||
