Burden of STIs and Benefits of Control
On the basis of an independent analysis of cross-country data on STI prevalence, we believe that WHO may have underestimated the burden of STIs relative to that of HIV and other diseases (www.fic.nih.gov/dcpp/gbd.html). Adjusting the WHO estimates on the basis of our calculations increases the estimate of years of life lost burden by about 18.1 percent and the overall estimate of disability-adjusted life years (DALYs) lost by about 8.2 percent.
DALYs Gained from Effectively Preventing or Treating STIs
In the first edition of this volume, chapter 20 presented estimates of the so-called static and dynamic burdens of preventing or curing a single case of an STI and of HIV (Over and Piot 1993). We reproduce those estimates in table 17.2 for four of the STIs.2 The static benefit column estimates the average number of DALYs saved for a single person by curing or preventing his or her own case of each disease. These estimates are based on specific assumptions regarding the distribution of incidence, case-fatality rates, and severity across age ranges. Although updating these estimates to 2004 and varying them by region would ideally be possible, we have not found any more recent data. If the person who is cured of an STI ceases to be sexually active, the static benefit would be the only benefit of curing or preventing his or her case; however, most individuals who have contracted an STI remain sexually active and are therefore likely to communicate that STI to others. An STI prevented or cured in a sexually active person will prevent additional cases in that person's sex partners, in the sex partners' partners, and so on. Thus, the dynamic benefit columns indicate the magnitude of those additional benefits.
[Table .]
The key finding is that preventing or curing a case of any of the STIs in a core group member generates approximately 10 times the dynamic benefits of the same intervention in a person in a noncore group. This result is driven by the assumption that a member of a noncore group has a new sexual contact every 50 days, or about seven new contacts per year, whereas a member of a core group has 10 times as many contacts. Within this model, the results are proportional to the frequency of partner change, so that the dynamic benefits of curing or preventing a case in a sex worker who has two partners a day would be approximately 10 times as great as for a member of the core group in table 17.2. The implication is that preventing or curing a case of syphilis in a sex worker can result in up to 1,600 DALYs of benefit, a health effect that is likely to be competitive with any discussed in the other chapters in this volume.
Impact of STIs on HIV
The preceding discussion does not address the possibility that STI infections increase HIV transmission. On this point the evidence is mixed, with a study in Mwanza, Tanzania (Grosskurth and others 1995), demonstrating a statistically significant 40 percent reduction in HIV incidence attributable to an STI intervention, while two studies (Kamali and others 2003; Wawer and others 1999) in Uganda failed to show any such effect. Recent reanalyses (Orroth 2003) of the data from these studies suggest that the effect of an STI intervention on an HIV epidemic will vary depending on the sexual activity and resulting prevalence of STIs among those being treated.
None of the randomized controlled trials of the effect of STI treatment on HIV prevention exclusively targeted the most sexually active people in the community. In Mwanza, Tanzania, and Masaka, Uganda, treatment was provided to those who sought it at health care clinics. In Rakai, Uganda, treatment was given to all adults in all households, regardless of whether the individual complained of STI symptoms. Data on the prevalence of HIV infection in the three communities suggest that the HIV and STI epidemics were both at an earlier stage in Mwanza and were therefore more concentrated among those more sexually active. Thus, the people who became symptomatic and sought treatment were among the most sexually active people in Mwanza, and treating them would therefore have had a greater effect on HIV incidence than would treating an average person in the two Ugandan sites. Conversely, in the more generalized epidemics in Uganda, a larger proportion of new infections occurred within stable HIV-1 serodiscordant couples.
An alternative, less rigorous way to test for the effect of STI prevalence on HIV infection is to study the cross-sectional correlation in ecological data. In a replication of an earlier study (Over 1998), we have attempted to explain urban HIV prevalence in a cross-country sample by the prevalence of syphilis and gonorrhea seven years earlier after controlling for six other potentially confounding variables. The results of these cross-country regressions are presented in table 17.3.3 Columns (1) and (2) of table 17.3 present the results of regressions estimated on the subsets of countries for which data are available on all eight explanatory variables and on the dependent variable (2002 urban HIV prevalence). Their specifications differ only by the replacement of the prevalence of syphilis as an explanatory variable in column (1) with the prevalence of gonorrhea in column (2). Columns (3) and (4) repeat the same two regressions by replacing missing values of the two prevalence rates with estimates that are based on a regression of these rates on the other variables in the regression. This procedure expands the samples dramatically from 56 and 38 to 181 and 180, respectively.4
[Table .]
In interpreting these regressions, note first that all four specifications explain more than half of the variance in 2002 urban HIV prevalence, a remarkably good fit for cross-sectional regressions. In all these specifications, the lagged value of an STI prevalence is a statistically significant predictor of HIV prevalence approximately seven years later. The coefficient for gonorrhea is larger than the coefficient for syphilis and is more statistically significant in the augmented sample, though less so in the basic sample.
After the age of the epidemic is controlled for, several other variables contribute to explaining the variation in HIV prevalence. These variables include national income per capita (richer countries have lower infection rates), the percentage of the population that is Muslim (a higher percentage is associated with lower infection rates), and the ratio of males to females in the sexually active age range (higher ratios are associated with higher infection rates).
The major difference between the regressions using the augmented sample and those using the basic sample is in the statistical significance of the estimated coefficient of income inequality as measured by the Gini coefficient. When the sample is expanded to take advantage of the available data, the coefficient stabilizes at about 5.3 and is statistically significant at the 0.01 probability value, suggesting that an increased degree of income inequality is associated with increased HIV infection even after controlling for STI prevalence. This result lends support to the idea that income inequality is just as important as poverty in setting the stage for HIV transmission.
As with the results of any ecological or cross-sectional analysis, questions of attribution and interpretation arise. Is the statistically significant coefficient of syphilis or gonorrhea capturing a biological effect of an STI on increasing the transmission probability during sexual intercourse? Or is the coefficient instead simply reflecting the fact that greater sexual activity spreads all STIs, including gonorrhea, syphilis, and HIV? Is the coefficient of the percentage of the population that is Muslim capturing differential sexual activity or the prevalence of male circumcision, which is increasingly recognized as biologically protective? A biological interpretation of both the STI and the Muslim coefficients is suggested by the fact that the variable urban male-to-female ratio is probably already capturing much of the variation in the most risky sexual behavior: the practice of prostitution.
Increasing the availability of treatment for STIs and for HIV infection reduces the prevalence of the former and increases the prevalence of the latter. Thus, this statistical relationship between STI prevalence and HIV prevalence, even if once valid, will no longer obtain. Under current conditions, estimating the effect of a change in the prevalence rate of an STI on the incidence rate of HIV would be more relevant.
If we assume that in 2002 the HIV epidemic was approaching equilibrium in many urban settings and that prior to antiretroviral treatment the median duration of the illness was about 10 years, the prevalence of HIV infection is approximately equal to 10 times the incidence rate. Thus, a 10 percentage point increase in the prevalence of syphilis or gonorrhea is estimated to increase the incidence of HIV by 0.27 percentage points for syphilis and 0.57 for gonorrhea. For comparisons, the Mwanza trial found that a reduction in the prevalence of male urethritis of 0.6 percent was associated with a decrease of 0.7 percent in the incidence of HIV (Grosskurth and others 1995). Thus, the present study suggests an effect about one-fourth as strong as that of the Mwanza study.
