PRB

(September 2006) Not so long ago, many wondered if the population of Africa would be “wiped out” by the HIV/AIDS epidemic. As serious as the situation was and is, those doomsday projections were unwarranted. But part of this understandable confusion resulted from a lack of solid information on the spread of the disease. That, in turn, resulted from a virtually complete lack of vital statistics such as births and deaths.

While vital statistics are still largely absent in many countries, many more fragmentary pieces of evidence now allow for more reliable measurements. And with this new data comes a new controversy: The HIV/AIDS situation in Africa may not be as serious as once thought. But, just as earlier fears were an overreaction, this revelation may lead to an underreaction. Both sides can benefit from a short history of how estimates of the epidemic have been derived.

Using Sentinel Site Data

The first testing sites were often called “sentinel” sites, and that is indeed what they were—a type of early warning system. These sites and clients included pregnant women coming to prenatal clinics and patients coming to clinics for treatment for sexually transmitted diseases. One advantage of sentinel sites is the inexpensive testing that can be conducted regularly as part of routine medical care.

But sentinel site data presented a significant statistical problem. In most countries, groups that had been tested were not representative of the total population. Such groups might have an urban “bias” in that testing might have been done in urban centers. The government of the Indian state of Tamil Nadu, for example, has pointed out that, since about 40 percent of pregnant women use private hospitals, testing only in government hospitals may overstate HIV prevalence in the state.

Beginning with estimates for 1997, UNAIDS used sentinel site data and other measurements to estimate national prevalence levels, because those sources were the only figures available. Estimates and methods were often revised as the number of sites increased or appeared in different areas of a country. In their efforts, UNAIDS cooperated closely with the Health Studies Branch of the U.S. Census Bureau’s International Programs Center, which had been compiling a comprehensive database of HIV prevalence data from sentinel sites and other sources since the mid-1980s.

Since 2001, however, a new source of information and data on HIV prevalence levels has become available for a growing number of countries: demographic and health surveys with nationally representative samples.

DHS Surveys Offer More Reliable Prevalence Estimates

The Demographic and Health Survey (DHS) program, conducted by ORC Macro in cooperation with national agencies, tests survey respondents about HIV. To date, 14 such DHS surveys, along with HIV/AIDS Indicator Surveys (AIS)—all in Africa—have been completed, and all but one indicated that HIV prevalence is likely lower than previously estimated. (The Washington Post reported on this in an April 2006 article written by Craig Timberg.) On average, the DHS figures were about 20 percent below UNAIDS sentinel site-based estimates.

The table illustrates the existing UNAIDS prevalence estimate for the 14 countries, prevalence as measured by the DHS, and the percentage of respondents who were actually tested. This last figure is obviously of interest since testing for infection could have triggered a high refusal rate, yet the percentage of respondents tested was generally quite good. ORC Macro found no major demographic differences between the tested and nontested groups. In all cases, the proportion of males tested was lower than females, sometimes quite a bit lower.

Percent of HIV-Positive Adults, Ages 15-49

Source: UNAIDS, 2002, 2004, and 2006; and DHS/AIS surveys.

UNAIDS now uses available DHS results. The DHS also includes much previously unavailable information, such as information on women who are not pregnant, on men, and from remote rural areas. Including men in the DHS was especially important since prevalence rates for men in Africa were, on average, one-half that for women.

Of the higher-prevalence countries, DHS prevalence for Kenya was quite different from previous estimates. In Kenya, 6.7 percent of adults ages 15 to 49 were found to be HIV-positive in the 2003 DHS. Previously, UNAIDS had estimated it was 15.0 (with a range from 12.0 to 18.0 percent). Zambia, previously estimated at about 22 percent, was about 16 percent in the DHS, noticeably closer in this instance. Surprisingly, Uganda ‘s HIV-prevalence rate was 6.4 percent in the AIS, but only 4.1 percent in the earlier UNAIDS estimate. Uganda has been widely praised for its far-ranging HIV/AIDS programs, and its prevalence level was thought to be declining. The higher AIS rate does not necessarily mean that prevalence is not declining, only that it is doing so from a higher level. UNAIDS estimated Kenya’s prevalence rate at a time when the trend in HIV prevalence was still upward, but subsequent testing of pregnant women indicated that the trend was now downward.

Even a Low Rate Is Still Too High

Perhaps Pali Lehohla, the statistician-general of South Africa, put it best when, in discussing varying levels of HIV estimates in that country, said that “sensationalizing differences in estimation of HIV prevalence rates is not useful.”

Measuring HIV infection on a nationally representative basis is a vitally important part of any HIV-prevention program and is especially valuable in tracking program effectiveness. But politicizing various estimates may only serve to damage prevention efforts.

Any country with what seems to be a “low” rate of one or two percentage points has a serious HIV/AIDS problem. Is HIV/AIDS still a terrible public health threat to Africa? It certainly is.

Carl Haub is a senior demographer at PRB and holds the Conrad Taeuber Chair of Population Information.