Drain, Paul K. MPH; Smith, Jennifer S. PhD; Hughes, James P. PhD; Halperin, Daniel T. PhD; Holmes, King K. MD, PhD
In 2000, HIV infection/AIDS caused >5% of all deaths world-wide, as the world’s fourth leading cause of mortality, and became the leading cause of lost disability-adjusted life years for females. 1 Although HIV infection/AIDS was the leading cause of mortality in sub-Saharan Africa, in some other world regions, HIV infection /AIDS failed to appear within the top 15 causes of mortality. 1 In many developed countries, rates of HIV infection have decreased in association with programs that target reduction in risky sexual behaviors and sexually transmitted infection morbidity and provide widespread access to voluntary counseling and testing, prophylaxis for and treatment of HIV-related opportunistic infections, and, recently, multidrug antiretroviral therapy. 2,3 However, in most developing countries, access to HIV prevention and treatment programs has been extremely limited, contributing to disparities between developed and developing countries in seroprevalence and mortality rates of HIV infection. 4–7 Analyses of socioeconomic and developmental correlates of country-specific HIV seroprevalence might increase our understanding of the influence of societal characteristics on HIV transmission, reinforce individual-level studies, and suggest potential population-level interventions.
We used country-specific age-standardized HIV seroprevalences per 100 adults 15–49 years old and number of children (younger than 15 years of age) living with HIV infection for the beginning of the year 2000 from the Joint United Nations Programme on HIV/AIDS. 8 Countries with <0.1% of adults infected with HIV were assumed to have an adult seroprevalence of 0.05%. Child seroprevalences were calculated by dividing a country’s estimated number of children living with HIV infection in the year 2000 8 by its child population in the year 2000. 9 Country-specific data were collected for the 167 countries listed in the Joint United Nations Programme on HIV/AIDS report.
The United Nations Development Programme 10 provided the Human Development Index data, which measure a country’s development status on the basis of life expectancy, educational attainment, and adjusted real income. After summarizing the global distribution of population and HIV infection/AIDS cases among adults and children for 167 countries (Fig. 1), the 45 countries classified as high human development (“developed countries”) were excluded, restricting subsequent analyses to the 122 countries classified as low and medium human development (“developing countries”) (Table 1).
Country-level statistics were compiled from various sources for 81 variables, including age of the HIV epidemic (estimated as the number of years before the year 2000 that the first case of HIV or AIDS was reported by each country), development (7 indicators), sexual behavior (12), reproductive health (12), infectious diseases (5), health services (15), economic (11), education (4), population (11), religion (2), and percent of males circumcised (Appendix 1). Major data sources included the United Nations Development Programme, 10,11 the World Health Organization, 12 the United Nations Children’s Fund, 13 United Nations Statistics Division data for 2000 and 2001, 14 the Joint United Nations Programme on HIV/AIDS, 8 the World Bank, 15,16 Save the Children, 17 Macro International, 18 and others. 19–24 All population statistics were estimated for the year 2000 and obtained from the United Nations Population Division. 9 Countries were categorized in geographic regions based on an International Agency for Research on Cancer classification system. 24 The US Central Intelligence Agency provided each country’s predominant religion and percentage of the population predominantly Christian or Muslim. 25 Data on herpes simplex virus 2 seroprevalence, generally representing the overall prevalence among reproductive-age, non–high-risk women, were obtained from a recent review of published literature. 26 Cleland and Ferry 19 provided observations for female age at first sexual intercourse and first marriage and polygamy to supplement Macro DHS data; however, data were not attainable for all countries for most variables.
Male circumcision prevalence was the only variable not obtained from a published source. Halperin and Bailey 27 supplied male circumcision data by using the same method and expanding on their previous study. To minimize potential misclassification, countries were classified as having low (<20%), medium (20–80%), or high (>80%) male circumcision rates.
Statistical analyses concentrated on the 122 developing countries and were weighted by each country’s adult or child populations to adjust for the precision of sampling methods. All regression analyses used a robust variance to account for unmeasured ecologic and population differences. 28 Subanalyses were also conducted without India, China, or Bangladesh, because of their large populations and often outlying indicator values, to determine their impact on the results. Stata Version 6 was used for conducting all statistical analyses. 29
Differences in mean HIV seroprevalences by geographic region and predominant religion were analyzed using a 1-way analysis of variance, and subsequent pairwise comparisons used Bonferroni multiple-comparison tests. Both adult and child HIV seroprevalences were natural log (ln)–transformed to create more normal distributions for regression analyses. Univariate linear regression statistics examined each independent variable with ln HIV seroprevalences as the dependent variable. Variables are presented in the tables by order of the R2 value, signifying the strength of correlations between independent variables and HIV seroprevalence obtained from univariate regression analyses. Reproductive health measures were regressed with child HIV seroprevalences as dependent variables. Variables measured on a continuous scale were explored for linear fit, abnormal patterns, and residual distribution. All variables with <30 observations, such as herpes simplex virus seroprevalence, were examined for asymmetric distribution of observations.
Countries were categorized into approximate tertiles as having low (≤100), medium (101–1500), and high (>1500) HIV seroprevalences per 105 adults. 8 The mean and SD for each independent variable were separately summarized among the low and high HIV seroprevalence groups. Low versus high mean values for each variable were analyzed for statistical significance using 2-sample t tests for independent samples with unequal variances.
A multivariate analysis of covariance model for ln adult HIV seroprevalences included the major geographic and religious categories and continuous variables with >90 observations and significant univariate associations (P < 0.10) with HIV seroprevalence either initially or upon exploration. China was excluded from the multivariate analysis because of its large population weight, previously reported low rates of sexually transmitted infections, 30,31 and uncertainty about reported rates of HIV infection. 32 Indicators influenced by the HIV epidemic, such as life expectancies, population growth rates, and tuberculosis cases, were excluded from the model. From each group of highly correlated variables, only 1, which appeared to best capture the common factor, was included. For example, the 4 measures of immunization rates were highly correlated; therefore, only diphtheria and tetanus toxoids and pertussis immunization was included. In addition, because the percent of males circumcised was colinearly correlated with the proportion of Muslim populations, these 2 variables were independently examined in the model. Nonsignificant variables (P > 0.05) were removed 1 at a time from the full model in a backward fashion and reexamined in the final model. Several other variables, with a significant univariate association and between 50 and 90 observations, such as age at first sexual intercourse for females, teenage birth rate, and Gini index (a measure of increasing income inequality), were individually examined in the full model. Countries included in the final model were explored for any selection biases or differences from countries excluded by the model.
Country-specific adult HIV seroprevalence (cases per 105 adults) ranged from 50 to 35,800 for the 167 countries (data not shown), averaging 1229 for the 122 developing countries and 314 for the 45 developed countries. The mean adult HIV seroprevalence for the 37 least-developed countries (4525) was nearly 15 times higher than that for the 45 most-developed countries. The adult HIV seroprevalence for the 85 medium-developed countries was 759 (1082 without China).
The global distribution of adult population and adult HIV infection/AIDS cases by human development category (least, medium [with separate results for India and China], and most developed) is shown in Figure 1. The 37 least-developed countries, with 10% of the world’s adult population, had 44% of the total adult cases. The 83 medium-developed countries, excluding India and China, had 32% of the adult population and 39% of the world’s adult HIV infection/AIDS cases. India, with 16% of the population, had 11% of the cases. China, with 23% of the global adult population, reported only 2% of adult cases. The 45 most-developed countries had 18% of the adult population and only 5% of adult HIV infection/AIDS cases.
Country-specific child HIV seroprevalence (cases per 105 children) ranged from 0.19 to 1540 for 159 countries with available data (data not shown). The mean child HIV seroprevalence was 80 for 116 developing countries and 11 for 43 developed countries. The mean prevalence for the 35 least-developed countries (237) was >22 times higher than that for the 43 most-developed countries. The child HIV seroprevalence for the 81 medium-developed countries was 43 and 61 without India and China.
The global distribution of children and children living with HIV infection/AIDS was also related to human development and had even more disparity than the adult distribution (Fig. 1). The 35 least-developed countries, with 19% of the world’s child population, had 66% of the total child cases. The 79 medium-developed countries, excluding India and China, had 45% of the child population and 22% of children living with HIV infection/AIDS. India, with 7% of the child population, had 6.6% of cases. China, with nearly 7% of the child population, reported only 0.2% of cases. The 43 most-developed countries had 22% of the child population and only 5.5% of child cases.
Geography and Religion
For developing countries, both geographic region and predominant religion were independently associated with adult HIV seroprevalence (P < 0.001). Although HIV seroprevalences differ significantly across the region, Africa had the highest adult HIV seroprevalence with an average of 6700 (Table 2). Latin America and the Caribbean and Asia had the next highest adult seroprevalences with 593 and 346, respectively. The mean adult HIV seroprevalence for developing countries in Europe (292) was slightly less than the mean seroprevalence for all developed countries. The mean adult HIV seroprevalence for the Middle Eastern region (10) was nearly 700 times lower than the average for the African region.
The distribution of the mean adult HIV seroprevalence differed significantly between predominant religions (Table 3). Countries whose predominant religion was classified as indigenous African, all of which are located in sub-Saharan Africa, had a mean adult seroprevalence of 5893. Predominantly Christian countries, with an average of 3074, had the second highest average seroprevalence. Predominantly Muslim and Hindu countries had mean adult seroprevalences of 1086 and 691, respectively. Predominantly Buddhist countries had the lowest mean seroprevalence (247), although this increased to 1197 without inclusion of China.
Associations between HIV seroprevalence and other variables are presented in Table 4 together with comparisons between developing countries with the lowest (≤100) versus the highest (>1500) HIV seroprevalence. Countries with more years since the beginning of the HIV/AIDS epidemic had higher HIV seroprevalences.
Less-developed countries, as measured by the Human Development Index rank, had significantly higher HIV seroprevalences. Of the 3 variables used to calculate “human development,” longer life expectancy was significantly associated with lower HIV seroprevalences; high educational attainment and gross national product per capita were both higher in the low HIV seroprevalence countries and showed trends toward significance. However, associations between HIV and life expectancy are partially explained by recalculations of life expectancies to reflect increasing HIV seroprevalence in many developing countries. Of other development indicators, more human development, such as access to safe water, was generally associated with lower HIV seroprevalences.
Of sexual behavior variables, more polygamy and higher divorce rates appeared strongly related to higher HIV seroprevalence, with limited observations. Countries with earlier ages at female sexual debut, more births to women younger than 20 years of age, higher total fertility, and more females having nonregular sexual partners had significantly higher HIV seroprevalences. Initial analyses suggested that HIV seroprevalence was not significantly higher in countries with younger female age at first marriage or first delivery. However, the associations between rates of HIV infection and younger female ages at first marriage and at first delivery became significant in subanalyses that excluded Bangladesh, which has a low HIV prevalence and extremely young women entering marriage and motherhood. Countries with the highest HIV seroprevalence also had a 3-fold higher percentage of females with nonregular sex partners than did countries with lowest HIV seroprevalences and had a smaller percentage of males using condoms with nonregular sex partners.
For reproductive health measures (Table 4), analyses were conducted with child HIV seroprevalence as the dependent variable. Contraception was associated with lower HIV seroprevalences both for children (Table 4) and for adults (women using any contraception: coefficient = −0.041, R2 = 0.26, P ≤ 0.001; women using modern contraception: coefficient = −0.034, R2 = 0.21, P ≤ 0.001). Countries with low HIV seroprevalence generally had lower infant mortality rates and fewer infants with low birth weights. Countries with more frequent reports of exclusive breast-feeding during a child’s first 4 months had lower HIV seroprevalences among children, while countries with more breast-feeding in the first hour after birth and longer duration of breast-feeding had higher child HIV seroprevalences.
Of infectious disease variables, herpes simplex virus 2 seroprevalence, cervical cancer incidence, and tuberculosis prevalence were associated with HIV seroprevalence; this association was significantly higher for countries with the highest HIV seroprevalences. Few countries had type-specific herpes simplex virus 2 serologic data, but they were evenly distributed and not geographically limited. Countries with the highest HIV seroprevalence had higher prevalences of malaria, which was itself highly correlated with geographic region.
Almost all indicators of better health services in Table 4 were significantly associated with lower HIV seroprevalence. Countries with more access to essential medications, lower child mortality rates, and better delivery of health services had lower HIV seroprevalences.
For economic variables, countries with more income inequality, as measured by the Gini index, had significantly higher HIV seroprevalences. Countries with the highest HIV seroprevalences had lower gross national products and tended to spend less on health per capita.
Although none of the education indicators were significantly associated with HIV seroprevalence, countries with the highest HIV seroprevalence had more illiterate adult males and females than countries with high seroprevalence and tended to have less total schooling.
Of population indicators, higher HIV seroprevalences were associated with higher birth rates, population growth rate, younger population median age, and larger percentages of population younger than 25 years of age.
All religious variables had strong associations with HIV prevalence in subanalyses, when excluding India and China, which are neither predominantly Muslim nor predominantly Christian. Countries with large Muslim populations had substantially lower HIV seroprevalences, whereas those with large Christian populations had generally higher HIV seroprevalences.
Male circumcision was not significantly associated with HIV seroprevalence when including China, which had low male circumcision practice and low HIV seroprevalence. However, subanalyses excluding China showed that countries with >80% of males circumcised had significantly lower HIV seroprevalence than countries with <20% of males circumcised. This association was not independent of religion, sexual practices, or other confounders and was not established as causal.
In the multivariate model, higher HIV seroprevalences were associated with a lower percent of males circumcised, older HIV epidemics, geographic region, and younger populations, (percent of population younger than 25 years of age) (Table 5). The relationship between HIV and female adult illiteracy changed from a positive to a negative correlation coefficient in the multivariate model. HIV seroprevalence was negatively associated with the number of doctors and the percentage of children fully immunized for diphtheria and tetanus toxoids and pertussis, suggesting better health services in general are associated with lower seroprevalences and lower HIV seroprevalences. These 7 variables accounted for 84% of variations in adult HIV seroprevalence for the 89 countries included in the multivariate model. These 89 countries did not appear systematically different from the 33 countries excluded because of missing information. Variables with fewer observations showed no trend toward significance when included in the full model.
Both geography and religion were independently associated with adult HIV seroprevalence for 122 developing countries. In Africa, HIV seroprevalence was 10–20 times higher than those in Latin America, Asia, and Europe and 67 times higher than that in the Middle East. Middle Eastern and predominantly Buddhist countries tended to have the lowest seroprevalences, while African and predominantly indigenous African and Christian countries generally had the highest seroprevalences. After excluding China, male circumcision prevalence and higher percentages of Muslim populations had colinearity; both were associated with lower HIV seroprevalences. Countries with the highest HIV seroprevalence tended to have older HIV epidemics, younger female ages at sexual debut and first delivery, higher total fertility, and more women with nonregular sex partners. The associations of lower child HIV seroprevalence with more exclusive breast-feeding in the first 4 months and with shorter total duration of breast-feeding are consistent with observational studies, 33,34 but these relationships may have residual confounding and should not be interpreted as causal. Contraception use by married women was associated with lower HIV seroprevalences among children. Countries with high HIV seroprevalences had 2-fold higher herpes simplex virus 2 seroprevalence and 3-fold higher cervical cancer incidence among women. Almost all health service measures were positively associated with lower HIV seroprevalences, and countries with low HIV seroprevalence generally had longer life expectancies, less child mortality, more immunization coverage, and better access to essential medications and health care personnel. Of economic spending measures, greater income inequality but not less health spending was associated with higher HIV seroprevalences. Countries with high HIV seroprevalence tended to have younger more illiterate populations with less access to safe water. The multivariate model indicated that countries with more male circumcision, younger HIV/AIDS epidemics, and a lower percentage of the population younger than 25 years of age had lower HIV seroprevalences.
A previous ecologic analysis by Over 35 of HIV seroprevalence among urban high-risk and low-risk groups using data from the US Census Bureau examined 13 variables for 72 developing countries. This study found that age of the epidemic, gross national product per capita, and Gini index were most strongly correlated with HIV seroprevalence after adjusting for other variables, while male circumcision, condom availability, and public health expenditures as a percent of the gross national product had no significant relation to HIV seroprevalence. Because this study found a significant relationship of HIV seroprevalence with educational disparity between sexes, we also examined the sex gap in illiteracy and years of schooling but found no association with adult HIV seroprevalence (data not presented). Our study differed from the study by Over by using more recent Joint United Nations Programme on HIV/AIDS data (1997 vs. 2000), including 70% more countries, utilizing country-level HIV seroprevalence estimates from the Joint United Nations Programme on HIV/AIDS instead of estimated rates for urban high- and low-risk populations that were published by the US Census Bureau, and analyzing additional variables.
Ecologic studies have several limitations. First, although ecologic analyses can describe population-level rates of disease occurrence in relation to population-level socioeconomic and developmental patterns, they cannot estimate individual risk or causal associations, because variables are not measured at the individual level and the temporal sequence of events is unknown. 36 Second, HIV seroprevalences reflect exposures that may have occurred several years earlier, and those analyses do not reflect how behavioral, economic, and demographic measures may have changed during this time. Third, validity of country-level data varies, and some data, such as those describing sexual behaviors, were not recorded for many countries, particularly countries in the Middle East. Where reliable data on male circumcision were lacking, prevalence of circumcision for some countries was estimated by examining the percentage of a country’s population that was Muslim; the percentage of Muslims and circumcision rates were not independent. Finally, our analysis could not include measures of several key potential influences on the HIV/AIDS epidemic, such as characteristics of sexual networks, injection drug use, unsafe injections, and male homosexual behaviors. Nonetheless, findings from this ecologic analysis can help to inform decisions concerning needs for additional data, allocation of health care resources, and design of prevention programs.
This study highlights potential contributions of several behavioral and biologic factors that help determine differences in HIV seroprevalence between developing countries. Our ecologic analyses support observations from a systematic comparison of population-level data from 4 African cities, 37–39 which suggested that differences in the rate of genital herpes and male circumcision explain much of the differences in HIV seroprevalence between the cities. A recent reanalysis of the 4-city data found that the reported number of lifetime partners and male circumcision were the main independent predictors of regional variations in the prevalence of HIV infection. 40 In addition, our analysis suggests the potential independent importance of several other behavioral, economic, and development factors. Our findings that countries with high HIV seroprevalence have higher proportions of women with nonregular sex partners and fewer proportions of men using condoms with nonregular partners are consistent with the examples of countries such as Thailand, Uganda, and Zambia, which have had reductions in the incidence of HIV infection concurrently with apparent reductions in such sexual behavior risks. 41–44
In conclusion, ecologic analyses of existing data portray a complex set of sociobehavioral, economic, developmental, and biologic factors as potential determinants of HIV seroprevalence in developing countries. Although these findings may not provide many new insights into risk factors for HIV infection, global ecologic analyses reinforce previous observations based on individual-level analyses of HIV risk determinants. These analyses also illustrate the wealth of ecologic data available that could be used in conjunction with more detailed studies of individuals of high-risk, low-risk, and bridge populations using multilevel analyses of risk determinants for HIV infection. In addition, the finding of fewer births attended by skilled personnel, but more midwives, in countries with high HIV seroprevalence suggests both challenges and feasible channels for reducing mother-to-child transmission of HIV. Finally, the fact that countries with the highest HIV seroprevalences tended to have lower total health expenditure, had fewer nurses and doctors, and provided more limited access to essential medications highlights the challenges in providing antiretroviral therapy to HIV-infected patients in developing countries.
The authors thank Robert Bailey for assistance in classifying countries by male circumcision prevalence; David Wilson, Antonio de Moya, and Jeff Marck for reviewing classification of countries by male circumcision rates; Jamie Robinson for assisting in the review of published data on type-specific herpes simplex virus 2 serology; and Ron Nelson and Mary Fielder for assisting with the manuscript preparation.
1. Murray CJL, Lopez AD, Mathers CD, et al. Global Burden of Disease 2000 Project: aims, methods, and data sources [Global Programme on Evidence for Health Policy Discussion paper no. 36]. Geneva: World Health Organization; 2001.
2. Joint United Nations Programme on HIV/AIDS. Report on the Global HIV/AIDS Epidemic 2002. Geneva: UNAIDS; 2002.
3. Centers for Disease Control and Prevention. Diagnosis and reporting of HIV and AIDS in states with HIV/AIDS surveillance—United States, 1994–2000. MMWR Morb Mortal Wkly Rep. 2002;51:595–598.
4. Schwartlander B, Stover J, Walker N, et al. AIDS: resource needs for HIV/AIDS. Science. 2001;292:2434–2436.
5. Attaran A, Sachs J. Defining and refining international donor support for combating the AIDS pandemic. Lancet. 2001;357:57–61.
6. Joint United Nations Programme on HIV/AIDS. Report on the Global HIV/AIDS Epidemic 2002. Geneva: UNAIDS; 2002.
7. Joint United Nations Programme on HIV/AIDS. New UNAIDS report warns AIDS epidemic still in early phase and not leveling off in worst affected countries. Press release July 2, 2002. Geneva: UNAIDS; 2002.
8. Joint United Nations Programme on HIV/AIDS. Report on the Global HIV/AIDS Epidemic June 2000. Geneva: UNAIDS; 2000.
9. United Nations Population Division. World Population Prospects: The 2000 Revision. New York: United Nations; 2001.
10. United Nations Development Programme. Human Development Report 2000. New York: Oxford University Press; 2000.
11. United Nations Development Programme. Human Development Report 2001. New York: Oxford University Press; 2001.
12. World Health Organization. The World Health Report 2000, Health Systems: Improving Performance. Geneva: World Health Organization; 2000.
13. United Nations Children’s Fund. The State of the World’s Children 2000. New York: UNICEF; 2000.
14. United Nations Statistics Division. Demographic and social indicators. Available at: www.un.org/Depts/unsd/social/
. Accessed on December 5, 2001.
15. World Bank. World Development Indicators 2000. Herndon, VA: World Bank; 2000.
16. World Bank. World Development Report 2000/2001. Herndon, VA: World Bank; 2001.
17. Save the Children. State of the World’s Newborns, A Report from Saving Newborn Lives. Washington, DC: Save the Children; 2001.
18. Macro International. Demographic and health surveys. Washington, DC: Measure DHS+; 2000. Available at: www.measuredhs.com/data/indicators
. Accessed on December 13 and 14, 2001.
19. Cleland J, Ferry B, eds. Sexual Behaviour and AIDS in the Developing World. London: Taylor and Francis; 1995.
20. Murray C. WHO issues new healthy life expectancy rankings. Press release June 4, 2000. Geneva: World Health Organization; 2000.
21. The Johns Hopkins School of Public Health. Population Reports, Oral Contraceptive—An Update. Baltimore: Johns Hopkins University Press; 2000.
22. World Health Organization. Hepatitis C—global prevalence (update). Wkly Epidemiol Rec. 1999;49:425–427.
23. World Health Organization. WHO estimates of health personnel. World Health Organization Statistical Information Systems; 2001. Available at: www3.who.int/whosis
. Accessed on December 13, 2001.
24. Ferlay J, Bray F, Pisani P, et al. GLOBOCAN 2000: Cancer Incidence, Mortality, and Prevalence Worldwide
. International Agency for Research on Cancer CancerBase no. 5, version 1.0 ed. Lyon: IARC Press; 2001. Available at: www-dep.iarc.fr
. Accessed on February 8, 2001.
25. US Central Intelligence Agency. The World Factbook 2001. Washington, DC: US Government Printing Office; 2000.
26. Smith JS, Robinson J. Herpes simplex virus infections: a review of seroprevalence worldwide. J Infect Dis. 2002;186(suppl 1):S3–28.
27. Halperin DT, Bailey RC. Male circumcision and HIV infection: 10 years and counting. Lancet. 1999;354:1813–1815.
28. StataCorp. Stata Statistical Software: Release 6.0. Chapter 23.11. College Station, TX: Stata Corporation; 1999.
29. StataCorp. Stata Statistical Software: Release 6.0. College Station, TX: Stata Corporation; 1999.
30. Cohen M, Henderson G, Aiello P, et al. Successful eradication of sexually transmitted diseases in the People’s Republic of China: implication for the 21st century. J Infect Dis. 1996;174(suppl 2):S223–S229.
31. Li H, Jin S, Xu H, et al. The decline of mortality rates of cervical cancer and a plausible explanation in Shandong, China. Int J Epidemiol. 2000;29:398–404.
32. Zhang KL, Ma SJ. Epidemiology of HIV in China. BMJ. 2002;324:803–804.
33. Coutsoudis A, Pillay K, Kuhn L, et al. Methods of feeding and transmission of HIV-1 from mothers to children by 15 months of age. AIDS. 2001;15:379–387.
34. Rousseau CM, Nduati RW, Richardson BA, et al. Longitudinal analysis of human immunodeficiency virus type 1 RNA in breast milk and of its relationship to infant infection and maternal disease. J Infect Dis. 2003;187:741–747.
35. Over M. The effects of societal variables on urban rates of HIV infection in developing countries: an exploratory analysis. In: Aimsworth M, Fransen L, Over M, eds. Confronting AIDS: Evidence from the Developing World. Washington, DC: The World Bank; 1999.
36. Morgenstern H. Uses of ecological analysis in epidemiologic research. Am J Public Health. 1982;72:1336–1344.
37. Buvé A, Caraël M, Hayes RJ, et al. The multicentre study on factors determining the differential spread of HIV in four African cities: summary and conclusions. AIDS. 2001;15(suppl 4):S127–S131.
38. Auvert B, Buvé A, Ferry B, et al. Ecological and individual level analysis of risk factors for HIV infection in four urban populations in sub-Saharan Africa with different levels of HIV infection. AIDS. 2001;15(suppl 4):S15–S30.
39. Ferry B, Caraël M, Buvé A, et al. Comparison of key parameters of sexual behaviour in four African urban populations with different levels of HIV infection. AIDS. 2001;15(suppl 4):S41–S50.
40. Auvert B, Ferry B. Modeling the spread of HIV infection in four cities of sub-Saharan Africa. 14th International AIDS Conference, Barcelona, Spain, July 7–12, 2002.
41. Nelson KE, Celentano DD, Eiumtrakol S, et al. Changes in sexual behavior and a decline in HIV infection among young men in Thailand. N Engl J Med. 1996;335:297–303.
42. Mbulaiteye SM, Mahe C, Whitworth JAG, et al. Declining HIV-1 incidence and associated prevalence over 10 years in a rural population in south-west Uganda: a cohort study. Lancet. 2002;360:41–46.
43. Agha S. Declines in casual sex in Lusaka, Zambia: 1996 – 1999. AIDS. 2002;16:291–293.
44. Bessinger R, Akwara P, Halperin D. Sexual Behavior, HIV and Fertility Trends: a Comparative Analysis of Six Countries; Phase I of the ABC Study. Washington, DC: Measure Evaluation/United States Agency for International Development, August 2003.
© 2004 Lippincott Williams & Wilkins, Inc.